Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

A Personalized Navigation Route Recommendation Strategy Based on Differential Perceptron Tracking User’s Driving Preference

A Personalized Navigation Route Recommendation Strategy Based on Differential Perceptron Tracking... Hindawi Computational Intelligence and Neuroscience Volume 2023, Article ID 8978398, 14 pages https://doi.org/10.1155/2023/8978398 Research Article A Personalized Navigation Route Recommendation Strategy Based on Differential Perceptron Tracking User’s Driving Preference Pengzhan Chen , Jihua Wu, and Ning Li Taizhou University, School of Intelligent Manufacture, Taizhou 318000, Zhejiang, China Correspondence should be addressed to Ning Li; wwwningning@126.com Received 9 August 2022; Revised 3 November 2022; Accepted 7 December 2022; Published 4 January 2023 Academic Editor: Maciej Lawrynczuk Copyright © 2023 Pengzhan Chen et al. Tis is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. With the increasing frequency of autonomous driving, more and more attention is paid to personalized path planning. However, the path selection preferences of users will change with internal or external factors. Terefore, this paper proposes a personalized path recommendation strategy that can track and study user’s path preference. First, we collect the data of the system, establish the relationship with the user preference factor, and get the user’s initial preference weight vector by dichotomizing the K-means algorithm. Te system then determines whether user preferences change based on a set threshold, and when the user’s preference changes, the current preference weight vector can be obtained by redefning the preference factor or calling diference perception. Finally, the road network is quantized separately according to the user preference weight vector, and the optimal path is obtained by using Tabu search algorithm. Te simulation results of two scenarios show that the proposed strategy can meet the re- quirements of autopilot even when user preferences change. Path recommendation systems usually recommend 1. Introduction paths based on the optimization of distance or travel time In recent years, with the rapid development of information cost functions. However, the shortest or fastest route is and intelligent transportation systems, artifcial intelligence usually not chosen by the driver, so the driver may have (AI) has been widely used in transportation tools to provide diferent recommendations when traveling. a comfortable travel experience for drivers and passengers Te personalized route recommendation strategy for [1–3]. Te vehicle Internet tends to share data, enabling each driver’s route selection preference has attracted much vehicles to exchange learning experiences and improve attention. Te personalized route recommendation strategy decision-making capabilities. Te individual models are not only meets people’s personalized needs but also solves the Blythe’s paradox, which is widely existed in the current trained on the basis of the data collected. By collecting these learning models from all vehicles, a comprehensive model transportation network. can be further developed. Tis is of the great signifcance to the application of intelligent transportation system (ITS) in autonomous driving and trafc control. 1.1. Literature Review. At present, the driver can get the current driving status of the vehicle through the OBD data. At present, the existing path navigation and recommen- dation software in China, such as the Baidu map application In reference [9], the authors proposed a monitoring system consisting of OBD and GPS, designated area, pricing and the Josiah Goddard map application, are all popular recommendation software, mainly based on the shortest time scheme, and the relationship between other related policies. Te OBD device is used to monitor engine operation and [4, 5] and the shortest distance [6–8], has been unable to meet measure fuel consumption and emissions reliably. In the growing demand for personalized tourism. 2 Computational Intelligence and Neuroscience and a genetic algorithm is used in this study to solve the low reference [10], OBD is used to analyze the driving behavior data through the vehicle preloading equipment and the efcacy and accuracy in robot path planning. Experimental results show that the proposed method can efciently factors that afect the safe driving and establish the logistic expression model. In reference [11], OBD is used to monitor improve robot path planning. In [22], Chen et al. proposed and prompt hydrocarbon (HC) emissions caused by a failure a deep reinforcement learning algorithm for path planning, of the vehicle emission control system. In this paper, the which has the comprehensive reward function of dynamic OBD data are used to correlate with the driving user’s path obstacle avoidance and goal approaching. Te results show selection preferences. Ten, the user’s initial preference that the method can avoid moving obstacles in the envi- weight vector is obtained by the bisecting K-means algo- ronment, complete the planning task, and has a high success rate. rithm clustering. In reference [12], Xiong et al. proposed a model predictive control optimal path optimization and Te personalized needs of users are not mentioned in these articles. In [23], Long et al. developed a novel route tracking framework. First, the relationship between the vehicle and the reference road is established in the road recommendation system to provide real-time personalized route recommendation for self-driving tourists, according coordinate system, and then, the safe lane under the high constraint environment is established by using the multi- to the specifc preferences of users personalized access layer search method; path boundary and vehicle dynamics routes, not only to save the total tour time but also to meet constraints are introduced to provide optimal control their specifc travel preferences. In [24], a dynamic route instructions. guidance method for driver’s personalized needs is pro- In reference [13], the unsupervised classifcation posed. Te user preference weight is given artifcially, method based on the bisecting k-means algorithm is ap- which has strong subjectivity and lacks objectivity and accuracy, but it provides a reference for the combination of plied to the data obtained in low-energy consumption proofread measurement of the online gas analyzer and personalized needs and path planning. In reference [25], a personalized path decision algorithm based on user pref- microsensor. Te good agreement with data from sensors validated the efectiveness of the proposed method. In [14], erence is proposed, which lays a good theoretical foun- dation and framework for this study. But there are also Zhao and othersproposed a reviewed block-based bag of words model using the bisection K-means clustering limitations. Te one-time sample trajectory data used in method which could signifcantly accelerate the process of this paper needs a lot of complex mathematical calculations codebook generation. In reference [15], a new algorithm to obtain the data needed for clustering. Te corresponding based on graph mining is proposed and the bisecting k- weight of user driving style obtained by fuzzy c-means means algorithm is used to fnd frequent terminology clustering is fxed and one-sided, while in real life, the user collection for the document. Te Tabu search algorithm is a driving preferences are not always the same and most of them are staged, so it cannot refect the user’s personalized typical shortest path algorithm, which is used to calculate the shortest path from one node to other nodes and is driving preferences. widely used in path planning. In reference [16], the Tabu search algorithm is proposed to solve the problem of Web service composition. It adds a periodic diversifcation step, 1.2. Contribution. As mentioned above, most traditional which is kept on the diversifcation step at the usual path planning schemes improve the speed and precision of completion point. Tis approach combines path relinking path planning algorithmically, while ignoring the efect of with dynamic diversifcation strategies, providing more user preferences [26–30] on path selection. In scenarios that opportunities for future research. In [17], the Tabu search combine user preferences and path planning, user prefer- algorithm is used to improve the route planning strategy in ences are unchanged by default. In real life, however, user urban areas under trafc congestion, which not only saves preferences are not always the same. At diferent times, user preferences may change periodically due to various internal driving time but also reduces fuel consumption. In [18], the Tabu search algorithm is used to identify the current to- or external factors, such as age, physical condition, climate, and work. Te type of preference may also change over time. pology of the network and help identify the shortest path from the point of failure to the nearest operation source. At the same time, when the preference type is unchanged, Te model improves efciency consumption by 23% and the corresponding preference weights may also change. bandwidth lifetime by 16%. How to make navigation more Terefore, this paper proposes a personalized path efective has been a research hotspot. In [19], Xie et al. planning strategy that can track and study user preferences. proposes a new method of combining global path planning First, the preference type is associated with the OBD data with local path planning, to provide an efcient solution for and a threshold is used to select the user’s current preference the unmanned surface vehicle (USV) path planning despite type. Ten, the initial preference weight vector is obtained by the changeable environment. Te method solves the path the split-k-average algorithm. Finally, the diferential per- ceptron is used to track and adjust the preference weights in planning problem with variable environments and is verifed by simulations and experiments. In [20], Mei real time, so that the optimal path can be recommended to meet the user’s needs when the user’s preferences change. In proposes an optimal tour guide path planning model based on an ant colony algorithm. Te experimental results show this paper, Nanchang is taken as the reference city of road that the proposed model and the optimal path planning network quantifcation, and the Tabu search algorithm is algorithm are more optimized. In [21], rough set theory used to verify the efectiveness of the strategy. Computational Intelligence and Neuroscience 3 In the light of the above, the contributions of the present comfort, and safety. Among them, the highest attention is document are summarized as follows: paid to safety. In this paper, we choose these four preference factors as user preference sets. (1) To associate OBD with user preferences and get the To determine the type of user preference factors, OBD initial preference weight vector by clustering data and user preference are used to establish a connection (2) Considering the periodic change of preference, the in this paper [31]. Te main function of OBD is to supervise change of two kinds of preference is summarized the status of components related to emission control during (3) Tracking the user preferences and fne-tuning the the actual use of vehicles. In this paper, OBD status in- formation, geographic location information, and trip record weights through the diferential perceptron information are used to correlate user preference factors. (4) Personalized quantization of road network, using the Te correlation process is as follows: Tabu search algorithm to plan the optimal path (1) In the driving process, the greater the average speed (5) Taking Nanchang as a reference city, the simulation experiment is carried out to verify the efectiveness of of the user, the more the user attaches importance to time. In this paper, v is selected to be the relevance the proposed strategy quantity (RQ) of a time-based user and v in any period time can be read directly from the server 2. System Implementation Review through the OBD terminal device. A personalized path recommendation system that can track (2) Te average fuel consumption (AFC) in the driving and study the user’s path selection preferences is proposed in process is the most important indicator for economic this paper, as shown in Figure 1. Te system mainly includes users. Te lower the value, the higher importance data preprocessing module, initial preference module, op- users attach to the economy. Terefore, AFC is timization and adjustment module, and path generation chosen as the associated quantity of economic users in this paper. As there is no specifc calculation module. When the change of user preferences has been detected, the system will make use of the latest OBD data in method for AFC in the standard OBD II and EOBD the storage space to study independently ofine until a path protocol, it is necessary to use OBD related data for that can meet the user’s current path selection preferences estimation. has been found. (3) Te change rate of the relative position of the A brief process of tracking and studying the path accelerator pedal (CR ) is closely related to the ap recommendation system is shown in Figure 2. In which, comfortable user. Te lower the (CR ) value, the ap three preference factors of time, economy, and comfort higher the user’s attention to the comfort. Users are taken as the user’s initial path selection preferences, who pay attention to comfort have good driving and the path points marked in black are path ofset points. stability. Tey step on and loose slowly, and give Te frst case is that the preference factors change, the oil smoothly. Terefore, CR is chosen as the ap latest recorded OBD data are needed to redetermine them, correlation quantity of comfortable users in this and the initial driving preference weight is obtained by paper. Moreover, it can be obtained directly from clustering. In the second case, the diference perceptron the server by modifying the interval time pa- needs to be used for learning and correction because the rameters of the standard OBD II and EOBD preference factors remain unchanged and the corre- protocol. sponding weight values between the preference factors (4) In the process of driving, excessive driving speed will change. Te third case is that in some road sections, the bring a threat to the safety of users. Terefore, the planned path is not consistent with the actual driving maximum traveling speed of this period is taken as path. Since, the trafc network is real-time and dynamic, the correlation quantity of safe users in this paper. emergencies, such as trafc lights, pedestrians, and ac- However, it can be read directly from the OBD cidents, will afect the coincidence rate. Tus, in the actual terminal device. situation, the planning path and the driving path are not exactly consistent. Terefore, within the allowable error Te calculation process of each associated quantity is range, in this case, it is considered that the user preference shown in Table 1. remains unchanged and it is considered as a sudden Considering the laws and regulations related to safe situation. driving and the opinions of professionals in the automobile industry, the threshold value of the average value of daily correlation volume is set, as shown in Table 2. 3. Path Recommendation System Te maximum three items of δ are selected as the user’s 3.1. Data Preprocessing Module. Diferent self-driving users preference factors by the relevant degree δ of correlation have diferent path selection preferences, so we conducted a and threshold calculation. As k � 3, the clustering efect is questionnaire on the surrounding self-driving users and obvious, and to explain the change of preference factors obtained the survey data shown in Figure 3. From Figure 3, conveniently, three preference factors are selected as the we can see that four preference factors are afecting the route user’s path selection preference each time. Te relevant selection of self-driving users, namely, time, economy, degree calculation formula is defned as follows: 4 Computational Intelligence and Neuroscience MODULE Data preprocessing module Initial habit module Path generation module Optimization and adjustment module Collecting User with certain OBD data habit factors Route recommendation system Figure 1: Te illustration of path recommendation. Initial Preference weight Methods:Bisecting K-means clustering Economy Time Comfort Data: OBD Phenomena: there are great differences between paths Inducement: preference factor AA H B J B A J B path1 path1 changed path1 Solution : Clustering with current C J KDD C T L C T L D path2 path2 path2 OBD data EE X YZZ F S N F E S N X F path3 path3 path3 Phenomena: there are differences AA H B J B A J B path1 path1 path1 between paths C J KDD C T L Inducement: preference weights C T L Z path2 path2 path2 changed EE X Y Z F S N V F S N M F path3 path3 path3 Solution : Calling differential perceptron AAHHBB A H B path1 path1 path1 CCJJ L D C J L D path2 path2 path2 Phenomena: path is basically EXX Y Z F E H Z F X H Z F path3 path3 unchanged path3 Inducement: habits are unchanged Solution : Remaining unchanged Path after Output Initial path habits changing Figure 2: Te illustration of recommendation under diferent circumstances. m − n examples and the bisecting k-means algorithm is used in | | i i ⎧ ⎪ δ � clustering in this paper. To facilitate the clustering and efect i m display of OBD data, the data should be normalized before < 2m , i ∈ T, E, C, S􏼁. (1) i i clustering. Te partial results of some user OBD sampling ⎪ N 1 data processed are shown in Table 3. ⎪ n � 􏽘 n i ij v, AFC, and CR are used to establish the three-di- N ap j�1 mensional coordinate system, and then, the normalized data are divided into the bisecting K-means clustering, and the initial preference weight vector of users is obtained 3.2. Initial Habit Module. After the user preference factor is according to the clustering results. In order to make the determined, the initial preference weight vector is obtained clustering efect signifcant, here, select Δt � 1 min to be by clustering. Time, economy, and comfort are taken as clustered, and the results are shown in Figure 4. Computational Intelligence and Neuroscience 5 4% 24% 36% 21% 15% Time Safety Economy Other Comfort Figure 3: Results of habit factor sampling for surrounding car users. Table 1: Te OBD data processing algorithm. Among them, n , n , and n are the numbers of coor- 1 2 3 dinate points belonging to the three clustering centers Algorithm 1: OBD data processing individually. Input: OBD data, V � 0.91, M � 28.27, R � 8.314, P � 6.17, E M PG G � 4.536, R � 14.7 PP A Output: v, AFC, CR , V ap max 3.3. Path Generation Module. Tis part is mainly divided 1: Begin into two parts. Te frst part describes how to individualize 2: While car running and quantify the road network to establish a model for 3: Read and storage the values of v, CR and V every ap max solving the optimal path. Te second part shows how to 6 seconds from OBD data calculate the optimal path by simulation. 4: Read the values of, L, R , M , and I every PM AP AT 6 seconds from OBD data 5: Calculate AFC 6: M � (R × M /I /120) × (V /100) × E × M /R 3.3.1. Road Network Quantifcation. To simplify the road AF PM AP AT E D M 7: AFC � (R × P × G × v/3600 × M ) × (1 + L) model and calculation, referring to the regulations of A G PP AF 8: Storage AFC highway technical engineering standard, urban road design 9: End while code, urban road network planning index system, Nanchang 10: End urban trafc planning, and combining with the actual characteristics of the road network, Baidu map and the results of feld investigation and verifcation, the roads in As can be seen from the fgure, the value of red Nanchang are divided into the following fve categories. To marked points v-axis is generally higher than the other sum up, the road conditions in Nanchang are actual. Te two categories. Due to the highest attention to time, the actual situation of each road is shown in Table 4. value v of time users is generally large. Terefore, the red In order to unify the quantitative standard and simplify marker class is a time-based feature point and Te values the calculation of the optimal path with a genetic algorithm, of the AFC-axis of the points marked in green are gen- the data in Table 3 are normalized and correlated. Te erally smaller than those of the other two categories. process is as follows: Economic users pay the highest attention to fuel con- 1 300 1 4 ⎧ ⎪ sumption, and the value of AFC is generally very small. Cos t � 􏼒0.6∗ + 0.1∗ + 0.3∗ 􏼓, ⎪ 8 MS RND RS Terefore, the green marker is the economic character- istic point. Te value of CR -axis of blue marked points ap ⎪ is generally smaller than the other two categories. ⎪ Cos t � EC, Comfortable users pay the most attention to driving (3) stability and the value of CR is generally very small. ⎪ ap ⎪ 1 1 Cos t � ∗ , Terefore, the blue marker is the comfort feature point. ⎪ 8 RE After the bisecting K-means clustering converges, the ⎪ number of coordinate points belonging to three clus- ⎪ MS tering centers is normalized to get the initial user pref- ⎩ Cos t � 0.5 ∗ + 0.5∗ RS. erence weight vector w � (w , w , w ). Te normalization 1 2 3 formula is as follows: Te data processed by formula (3) are shown in Table 5. Based on the comprehensive consideration of diferent w � (j � 1, 2, 3). (2) driving preferences and simplifed calculation, the unit cost n + n + n 1 2 3 function is designed as follows: 6 Computational Intelligence and Neuroscience Table 2: Correlation threshold. Relevant quantity v (km/h) AFC (L/100 km) CR (%) V (km/h) ap max Treshold 40 8 10 60 Table 3: Partial OBD data after normalization. Time v AFC CR V ap max 08:08:34:25 0.1250 0.0090 0.0588 0.3600 08:08:40:25 0.0500 0.0068 0.5294 0.2010 08:08:46:25 0.2500 0.1109 0.4706 0.5212 08:08:52:25 0.3500 0.1267 0.8824 0.5842 08:08:58:25 0.3562 0.1154 0.5882 0.6124 08:08:64:25 0.6250 0.7467 0.6471 0.7653 08:09:00:25 0.8875 0.9887 0.4706 0.9102 1 1 0.8 0.8 0.6 0.6 – – v v 0.4 0.4 0.2 0.2 0 0 1 1 0.8 0.8 0.5 0.5 0.6 0.6 0.4 0.4 AFC 0.2 AFC 0.2 CR CR 0 0 ap 0 0 ap (a) (b) Figure 4: Bisecting K-means clustering results (above is before clustering and below is after clustering). optimal path becomes a single source shortest path problem U � f c , c , · · · , c � 􏽘 􏼁 ω c . (4) k k 1 2 m i i shown in Figure 5. i�1 Te process of using the Tabu search algorithm to get the optimal path of the model shown in Figure 5 is shown in In this paper, each user has three main driving prefer- Table 6. As can be seen from Table 6, the fnal optimal path is ences, that is, the other three preference weights are set to 1 ⟶ 4 ⟶ 7 ⟶ 10 ⟶ 13 ⟶ 15 ⟶ 16 ⟶ 20, and zero, making m � 3, so the cost of personalized quantif- the total cost consumption is 88. cation of the K segment road is as follows: U � L × f c , c , c 􏼁 � L × 􏽘 ω c . (5) Tk k k 1 2 3 k i i 3.4. Optimization and Adjustment Module. In order not to i�1 afect the user’s self-driving experience, it is stipulated that the system will conduct tracking and studying at night every day, and the preference tracking process is shown in 3.3.2. Tabu Search Algorithm. After the recommendation Table 7. Due to the existence of emergencies, the consis- system personalizes and quantifes the road network tency between the planned path and the actual driving path according to the user preference weight vector [32], the will not always be 100%. In the system, a threshold pa- model becomes a classic problem of fnding the optimal rameter is set for the coincidence degree. Before training solution. In this paper, the Tabu search algorithm is used to every day, the system will flter the OBD data of the day and calculate the optimal path according to the user’s preference. eliminate the data whose coincidence degree is higher than Here, the model diagram shown in Figure 5 is selected to the threshold value. Ten, the system recalculates the illustrate how the algorithm calculates the optimal path. In correlation degree δ of four correlation quantities Figure 5, the cost consumption values of each point between according to the remaining OBD data of the day in the 1 and 20 are calculated, respectively, by formula (5). After storage space and sorts them from small to large. If the personalized quantifcation of the road network, fnding the correlation quantity corresponding to the minimum value Computational Intelligence and Neuroscience 7 Table 4: Performance of road networks at all levels. Time RND (km/km ) RS (v/c) RE (a.u.) MS (km/h) EC (RMB/km) Expressway 0.42 0.42 0.40 100 0.74 Main road 1.31 0.68 0.35 60 0.51 Sub road 1.60 0.84 0.30 50 0.58 Landscape road 0.38 0.62 0.65 50 0.54 Business zone 1.75 1.15 0.22 40 0.62 Table 5: Road network costs at normalized levels. Cost Road classifcations Cos t Cos t Cos t Cos t T E C S E 0.31 0.74 0.31 0.32 M 0.40 0.51 0.36 0.49 S 0.47 0.58 0.42 0.58 L 0.36 0.54 0.19 0.58 B 0.62 0.62 0.57 0.74 1 8 15 11 8 13 6 8 16 9 15 2 14 7 24 20 14 Figure 5: Road network model after personalized quantifcation. Table 6: Tabu search algorithm solution process. Routes Total consumptions 1 ⟶ 1 0 1 ⟶ 2 8 1 ⟶ 3 12 1 ⟶ 4 10 1 ⟶ 3 ⟶ 5 17 1 ⟶ 4 ⟶ 6 14 1 ⟶ 4 ⟶ 7 17 1 ⟶ 3 ⟶ 5 ⟶ 8 24 1 ⟶ 4 ⟶ 6 ⟶ 9 23 1 ⟶ 4 ⟶ 7 ⟶ 10 35 1 ⟶ 4 ⟶ 6 ⟶ 9 ⟶ 11 38 1 ⟶ 3 ⟶ 5 ⟶ 8 ⟶ 12 49 1 ⟶ 4 ⟶ 7 ⟶ 10 ⟶ 13 49 1 ⟶ 4 ⟶ 7 ⟶ 10 ⟶ 14 55 1 ⟶ 4 ⟶ 7 ⟶ 10 ⟶ 13 ⟶ 15 55 1 ⟶ 4 ⟶ 7 ⟶ 10 ⟶ 13 ⟶ 15 ⟶ 16 63 1 ⟶ 4 ⟶ 7 ⟶ 10 ⟶ 13 ⟶ 17 60 1 ⟶ 4 ⟶ 7 ⟶ 10 ⟶ 13 ⟶ 17 ⟶ 18 68 1 ⟶ 4 ⟶ 7 ⟶ 10 ⟶ 14 ⟶ 19 85 1 ⟶ 4 ⟶ 7 ⟶ 10 ⟶ 13 ⟶ 15 ⟶ 16 ⟶ 20 88 8 Computational Intelligence and Neuroscience Table 7: Te habit tracking processing algorithm. Table 8: Te improved diferential perceptron algorithm. Algorithm 2: Habit tracking processing Algorithm 3: Improved diferential perceptron Input: Te latest OBD data, current habit weight vector w Input: OBD data, habit weight vector ′ ′ Output: new habit weight vector w � (ω , ω , ω ) Output: Improved new habit weight vector w � w w w ′ 1 ′ 2 ′ 3 1 2 3 1: Begin 11: While n ! � 2000 or ‖w − w‖> ε 2: Calculate δ with the latest OBD data, i ∈ T, E, C, S 12: For each road classifcation do 3: If RQ of δ min unchanged 4: w � Improved diferential perceptron (the last OBD 13: Calculate the total length of the section in AP and PP data, w) respectively 5: Else 6: Obtain w by reclustering the OBD data of the other AP AP three RQ ⎧ ⎪ L � 􏽘 L Ti Ti (7) End if j�1 14: i ∈ T, E, C, S PP AP (8) End ⎪ L � 􏽘 L Ti j j�1 15: End for of δ changes, the system determines that the preference 16: For each habit factor do factor changes, and clusters the latest data of the other three correlation quantities to get the initial weight vector again. 17: Calculate the total cost in AP and PP respectively On the contrary, the system judges that the preference AP factor has not changed, and calls the diferential perceptron U � L 􏽘 ω c ⎧ ⎪ APi i i ⎪ Ti to fne-tune the weight vector. Because of the relatively i�1 18: i ∈ T, E, C, S ⎪ PP small number of data samples on the same day, direct U � L 􏽘 ω c ⎩ PPi Ti i i i�1 clustering by ignoring the previous data will enlarge the 19: Calculate weight change volume respectively efect of change. Data clustering together will cover up 20: ∆ω � η(􏽐 U − 􏽐 U )i ∈ T, E, C, S APi PPi smaller weight changes. Terefore, in this paper, we choose 21: Calculate and normalize the new driving style weight to call the diferential perceptron to fne-tune to improve 22: ω � ω + Δω j � 1, 2, 3 j j the accuracy. 23: End for 24: n � n + 1 Te setting of the diferential perceptron function is 25: Re-quantify road network with w shown in Table 8. Taking OBD data and current weight 26: Replan the optimal path with the Tabu search algorithms vector as input, the maximum number of iterations is set as 27: End while 2,000. Te total length of fve types of roads is calculated for 28: Return w � w w w 1 2 3 the actual path and the planned path, respectively, and the total cost corresponding to the three preference factors is calculated for the two cases. Ten, the change of the weight 4.1. Experiment 1. Te preference factor of self-driving user of each preference factor is calculated. Among which, η is the A is economy, comfort, and safety, and the corresponding learning efciency, take 0.003 to obtain the weight vector initial weight vector is obtained by the clustering algorithm, closest to the current user path selection preference through which is w � (0.48, 0.14, 0.38). After the preference factor of repeated iterative learning. user A changes, the planned path and the actual driving path are shown in the red path and green path in Figure 6, re- 4. Experiment spectively. It can be seen from the fgure that due to the change of preference factors, the coincidence of the planning Two types of preference change are proposed in this paper: path and the actual path are very low, only 0.785%, indi- one is the change of user preference factor and the other is cating that the current planning strategy can no longer meet the change of the weight of the same preference factor. To the personalized traveling needs of users. verify the efectiveness of the strategy proposed in this paper, After the personalized quantifcation of the road network Nanchang city is taken as a reference City, and simulation w � (0.48, 0.14, 0.38), the specifc generation values of experiments are carried out in two scenarios, respectively. In economic consumption (TC), comfort consumption (EC), scenario 1, the user preference factor changes. Assuming and safety consumption (CC) in two cases are shown in that the initial preference factor of user A is economy, Table 9, respectively. comfort, and safety, then due to work reasons, time is more Te total cost C of the actual path and the planned total important than the economy, so the preference factor path can be obtained through classifcation and integration becomeseconomy, comfort, and safety. In scenario 2, the of the data of the road type in Table 9 and is shown in weight of the same user preference factor changes. Suppose Table 10. As can be seen from Table 10, due to the change of that the initial preference factor of user B is time, economy, preference type, the path types in the two cases are diferent, and comfort. Ten, due to the fnancial crisis of the family, and the total cost is also very diferent, which is 12.737. user B pays more attention to the economy and less attention When the system is tracking and studying at night, the to comfort. In these cases, 90% of the threshold value is used correlation δ of four preference factors is calculated by to determine whether it is an emergency or not. formula (1). Among them, the correlation quantity of Computational Intelligence and Neuroscience 9 River School Landscape road Train station Sub road Expressway Business zone Main road Park Figure 6: Te initial planning path and the actual driving path before the adjustment in Experiment 1 (the left one is the model diagram and the right one is the actual map). Table 9: Adjustment of the frst two paths in Experiment 1. Paths Road names Types ERL (km) EC CC SC Wugong Mountain Avenue M 3.4 0.832 0.171 0.633 Xiangyun Avenue M 7.6 1.861 0.383 1.415 Changnan Avenue M 13.4 3.280 0.675 2.495 Initially planned route Changdong Avenue M 9.4 2.301 0.474 1.750 Ziyang Avenue M 6.5 1.591 0.328 1.210 Ziyang East Avenue M 2.6 0.637 0.131 0.484 Aviation City Avenue S 0.51 0.142 0.030 0.112 Wugong Mountain Avenue M 3.3 0.808 0.166 0.615 Circumferential Expressway E 60.4 21.454 2.621 7.345 Actual path Ziyang East Avenue M 0.79 0.193 0.040 0.147 Aviation City Avenue S 0.51 0.142 0.030 0.112 Table 10: Te total cost of the frst two paths before adjusting in Experiment 1. Paths Types L (km) U C Tk total M 42.90 20.652 Initially planned route 20.936 S 0.51 0.284 M 4.09 1.969 Actual path E 60.4 31.420 33.673 S 0.51 0.284 minimum δ changes from the previous time to economy. As can be seen from Figure 7, after tracking and Terefore, the system determines that the user A’s prefer- studying, the consistency between the planned path and the ence factor changes, and the current preference factor actual path is greatly improved, from 0.785% to 98%. Among changes to time, comfort, and safety. Te current weight them, the specifc cost data in the two cases after adjustment vector is obtained by the bisecting K-means clustering. Fi- are shown in Table 11. nally, the planning path calculated by the Tabu search al- Te total cost C of the actual path and the planned total gorithm using personalized quantitative road network is path can be obtained through classifcation and integration shown in the red path in Figure 7. of the data of the road type in Table 11 and is shown in 10 Computational Intelligence and Neuroscience River School Landscape road Train station Sub road Expressway Business zone Main road Park Figure 7: Te adjusted initial planning path and actual driving path in Experiment 1 (the left diagram is the model diagram and the right one is the actual map). Table 11: Data of two paths after being adjusted in Experiment 1. Paths Road names Type ERL (km) TC CC SC Wugong Mountain Avenue M 3.3 0.937 0.166 0.243 Initially planned route Circumferential Expressway E 60.7 13.360 2.634 2.914 Liu Cheng Street S 0.82 0.274 0.048 0.071 Wugong Mountain Avenue M 3.3 0.937 0.166 0.243 Circumferential Expressway E 60.4 13.294 2.621 2.899 Actual path Ziyang East Avenue M 0.79 0.224 0.040 0.058 Aviation City Avenue S 0.51 0.170 0.030 0.044 Table 12 As can be seen from Table 12, the path types in the Te total cost C of the actual path and the planned total two cases are the same, and the total cost diference is only path can be obtained through classifcation and integration of 0.081, which indicates that the system can still meet the the data of the road type in Table 13 and is shown in Table 14. user’s path planning needs through tracking and adjusting It can be seen from the table that, due to the change of after the user’s preference factor changes. preference weight, the path types in the two cases are the same, but the length is diferent, and the total cost diference is 2.149. 4.2. Experiment 2. Te preference factor of self-driving user B Similarly, when the system is tracking and studying at is time, economy, and comfort, and the corresponding initial night, the correlation δ of four preference factors is weight vector is w � (0.14, 0.44, 0.42). After user B’s prefer- recalculated by formula (1). Among them, the correlation ence weight changes, the planned path and the actual driving quantity of the corresponding minimum δ value does not path are shown in red and green paths in Figure 8, respectively. change, which is safety. Terefore, the system will determine Due to the change of preference weight, the consistency that the user B preference type has not changed but the between the planned path and the actual path is not high, weight changes, and call the diferential perceptron to fne- which is 76.64%. It shows that the current planning strategy tune, and fnally, get the current weight vector does not fully meet the user’s personalized travel needs. To w � (0.15, 0.51, 0.34). Finally, the planning path calculated improve the degree of coincidence, we need to further by the Tabu search algorithm using personalized quantita- improve the accuracy of the weight vector. tive road network w is shown in the red path in Figure 9. After the personalized quantifcation of the road network As can be seen from Figure 9, after tracking and w � (0.14, 0.44, 0.42), the specifc generation values of time studying, the consistency between the planned path and the consumption (TC), economic consumption (TC), and actual path has improved, from 76.64% to 84.58%. Among comfort consumption (EC) in two cases are shown in Ta- them, the specifc cost data in the two cases after adjustment ble 13, respectively. are shown in Table 15. Computational Intelligence and Neuroscience 11 Table 12: Total cost of two paths after adjustment in Experiment 1. Paths Types L (km) U C Tk total M 3.3 1.346 Initially planned route E 60.7 18.908 20.647 S 0.82 0.393 M 4.09 1.668 Actual path E 60.4 18.815 20.728 S 0.51 0.245 River School Landscape road Train station Sub road Expressway Business zone Main road Park Figure 8: Te initial planning path and the actual driving path before adjustment (the left one is the model diagram and the right one is the actual map). Table 13: Te data of the frst two paths before the adjustment in Experiment 2. Paths Road names Types ERL (km) TC EC CC Wugong Mountain Avenue M 3.4 0.190 0.763 0.514 Xiangyun Avenue M 8.1 0.454 1.818 1.225 Riverside Avenue S 5.6 0.368 1.429 0.988 Hongcheng Road M 3.7 0.207 0.830 0.559 Jinggangshan Avenue M 0.9 0.050 0.202 0.136 Initially planned route Eight One Avenue M 1.5 0.084 0.337 0.227 Beijing West Road L 2.3 0.116 0.546 0.184 Beijing East Road L 5.8 0.292 1.378 0.463 Ziyang Avenue M 6.5 0.364 1.459 0.983 Ziyang East Avenue M 1.5 0.084 0.337 0.227 Aviation City Avenue S 0.51 0.034 0.130 0.023 Wugong Mountain Avenue M 3.4 0.190 0.763 0.514 Xiangyun Avenue M 8.1 0.454 1.818 1.225 Changnan Avenue M 8.7 0.487 1.952 1.315 Nanlian Road M 2.4 0.134 0.539 0.363 Jinggangshan Avenue M 4.8 0.269 1.077 0.726 Actual path Eight One Avenue M 1.5 0.084 0.337 0.227 Beijing West Road L 2.3 0.116 0.546 0.184 Beijing East Road L 5.8 0.292 1.378 0.463 Ziyang Avenue M 6.5 0.364 1.459 0.983 Ziyang East Avenue M 1.5 0.084 0.337 0.227 Aviation City Avenue S 0.51 0.034 0.130 0.023 12 Computational Intelligence and Neuroscience Table 14: Te total cost of the frst two paths before the adjustment in Experiment 2. Paths Types L (km) U C Tk total M 25.6 11.328 Initially planned route L 8.1 2.979 17.418 S 6.11 3.111 M 36.9 16.328 Actual path L 8.1 2.979 19.567 S 0.51 0.260 River School Landscape road Train station Sub road Expressway Business zone Main road Park Figure 9: Te initial planning path and the actual driving path before adjustment (the left one is the model diagram and the right one is the actual map). Table 15: Adjusted data of two paths in Experiment 2. Paths Road names Types ERL (km) TC EC CC Wugong Mountain Avenue M 3.4 0.204 0.884 0.416 Xiangyun Avenue M 8.1 0.486 2.107 0.991 Changnan Avenue M 6.4 0.384 1.665 0.783 Yingbin North Avenue M 3.9 0.234 1.014 0.477 Fuhen Road M 1.6 0.096 0.416 0.196 Hongcheng Road M 1.1 0.066 0.286 0.135 Initially planned route Jinggangshan Avenue M 0.9 0.054 0.234 0.110 Eight One Avenue M 1.5 0.090 0.390 0.184 Beijing West Road L 2.3 0.124 0.633 0.149 Beijing East Road L 5.8 0.313 1.597 0.375 Ziyang Avenue M 6.5 0.390 1.691 0.796 Ziyang East Avenue M 0.79 0.047 0.205 0.097 Aviation City Avenue S 0.51 0.036 0.151 0.073 Wugong Mountain Avenue M 3.4 0.020 0.884 0.416 Xiangyun Avenue M 8.1 0.049 2.107 0.991 Changnan Avenue M 8.7 0.052 2.263 1.065 Nanlian Road M 2.4 0.014 0.624 0.294 Jinggangshan Avenue M 4.8 0.029 1.248 0.588 Actual path Eight One Avenue M 1.5 0.009 0.390 0.184 Beijing West Road L 2.3 0.124 0.633 0.149 Beijing East Road L 5.8 0.313 1.597 0.375 Ziyang Avenue M 6.5 0.039 1.691 0.796 Ziyang East Avenue M 1.5 0.084 0.337 0.227 Aviation City Avenue S 0.51 0.036 0.151 0.073 Computational Intelligence and Neuroscience 13 Table 16: Adjusted total cost of two paths in Experiment 2. Paths Types L (km) U C Tk total M 34.19 15.129 Initially planned route L 8.1 3.191 18.580 S 0.51 0.260 M 36.9 16.328 Actual path L 8.1 3.191 19.779 S 0.51 0.260 Te total cost C of the actual path and the planned Data Availability total path can be obtained through classifcation and integration Te data used to support the fndings of this study are of the data of the road type in Table 15 and is shown in available from the corresponding author upon reasonable Table 16. As can be seen from Table 16, the path types in the request (pzchen@tzc.edu.cn). two cases are the same, and the total cost diference is 1.199. Compared with studying before adjustment, the diference Conflicts of Interest in total cost decreases and the coincidence increases. It shows that the system can also meet the user’s path planning Te authors declare that they have no conficts of interest. needs by tracking and adjusting when the user’s preference weight changes. Acknowledgments 5. Conclusion and Outlook Tis research work was supported by the National Nature Science Foundation under Grant no. 62163014. A personalized path recommendation strategy that can track and study the user’s path selection preferences is proposed in References this paper. In the case that the self-driving users do not know their preferences and their preferences will change peri- [1] A. Forestiero and G. Papuzzo, “Recommendation platform in odically, the system frst determines the main driving Internet of Tings leveraging on a self-organizing multiagent approach,” Neural Computing & Applications, vol. 34, no. 18, preference factors of users through the collected OBD data Article ID 16049, 2022. and obtains the initial preference weight vector through the [2] A. Forestiero, “Heuristic recommendation technique in In- bisecting K-means clustering. When the preference changes, ternet of Tings featuring swarm intelligence approach,” the system judges according to the relevant parameters and Expert Systems with Applications, vol. 187, Article ID 115904, adjusts the preference weight vector according to the situ- ation. Ten, the road network model is quantifed according [3] A. Forestiero and G. Papuzzo, “Agents-based algorithm for a to the weight vector. Finally, the Tabu search algorithm is distributed information system in internet of things,” IEEE used to calculate the optimal path. In the two scenarios, Internet of Tings Journal, vol. 8, no. 22, Article ID 16548, through the comparison between the planned path and the actual path and the path before and after tracking and [4] J. Ma and G. Sun, “Mutation ant colony algorithm of milk-run vehicle routing problem with fastest completion time based studying, the efectiveness of the strategy proposed in this on dynamic optimization,” Discrete Dynamics in Nature and paper is verifed. Te results show the consistency between Society, vol. 2013, no. 1, Article ID 418436, 6 pages, 2013. the planned path and the actual path has improved from [5] J. Guo, Y. Wu, X. Zhang et al., “Finding the ‘faster’ path in 76.64% to 84.58% after tracking and studying the user’s vehicle routing,” IET Intelligent Transport Systems, vol. 11, personalized driving behavior. Te adjusted total cost of no. 10, pp. 685–694, 2017. initially planned route is 18.580, while actual path 19.779. [6] A. Idri, M. Oukarf, A. Boulmakoul, K. Zeitouni, and A. Masri, Te path types of the two cases are the same, and the total “A distributed approach for shortest path algorithm in dy- cost diference is only 0.081. It is proved that the strategy can namic multimodal transportation networks,” Transportation also meet the personalized traveling needs of users when Research Procedia, vol. 27, pp. 294–300, 2017. their preferences change. [7] Y. Zhang, S. Song, Z. J. M. Shen, and C. Wu, “Robust shortest To make the Tabu search algorithm not easily fall into path problem with distributional uncertainty,” IEEE Trans- actions on Intelligent Transportation Systems, vol. 19, no. 4, local optimum when calculating the optimal path, the road pp. 1080–1090, 2018. network model in this paper is relatively simple. However, [8] P. Faizian, M. A. Mollah, X. Yuan, Z. Alzaid, S. Pakin, and when the actual road network is larger and the route is more M Lang, “Random regular graph and generalized de bruijn complex, the algorithm is easy to fall into local optimum. At graph with $k$ -shortest path routing,” IEEE Transactions on the same time, only static factors rather than the real-time Parallel and Distributed Systems, vol. 29, no. 1, pp. 144–155, trafc situations in the road network are considered in this paper. In the future work, we should improve the accuracy of [9] Y. Wu and D. Sperling, “Using on-board diagnostic and global planning path algorithm or change the way of road network positioning system to price emissions from on-road heavy- modeling, and add some dynamic factors to make the duty vehicles,” Journal of Transportation Engineering, Part A: strategy proposed in this paper more universal. Systems, vol. 144, no. 8, 2018. 14 Computational Intelligence and Neuroscience [10] Y. Pan, “Using OBD-II data to explore driving behavior [27] Y. J. Song and J. M. Kim, “Characterization of privacy based model,” in Proceedings of the IEEE-ICASI 2017, Sapporo, on context sensitivity and user preference for multimedia context-aware on IoT,” Multimedia Tools and Applications, Japan, July 2017. [11] P. Kalita, “Power train and emission control: allocation vol. 78, no. 5, pp. 5355–5366, 2019. [28] W. Wu, R. Zhang, and L. Liu, “A personalized network-based procedure by OBD-II system for automotive technology,” IOP recommendation approach via distinguishing user’s prefer- Conference Series: Materials Science and Engineering, vol. 211, ence,” International Journal of Modern Physics B, vol. 33, Article ID 012019, 2017. no. 6, Article ID 1950029, 2019. [12] L. Xiong, Z. Fu, Z. Qian, B. Leng, D. Zeng, and Y. Huang, [29] J. Han and H. J. Suk, “Exploring user’s preference on the color “Model predictive trajectory optimization and tracking in of cavity and lighting of an oven product,” Archives of Design highly constrained environments,” International Journal of Research, vol. 32, no. 2, pp. 19–29, 2019. Automotive Technology, vol. 23, no. 4, pp. 927–938, 2022. [30] H. Liu, H. Ma, Y. Chang, Z. Li, and W. Wu, “Target com- [13] A. Caron, N. Redon, P. Coddeville, and B. Hanoune, munity detection with user’s preference and attribute sub- “Identifcation of indoor air quality events using a K-means space,” IEEE Access, vol. 7, Article ID 46583, 2019. clustering analysis of gas sensors data,” Sensors and Actuators [31] G. Wei, “Research on parametric indicators of driving habits B: Chemical, vol. 297, Article ID 126709, 2019. based on OBD technology,” Ergonomics, vol. 23, no. 2, [14] C. Zhao, X. Li, and Y. Cang, “Bisecting k-means clustering pp. 38–44, 2017. based face recognition using block-based bag of words [32] L. Israel Pinto and P. Bertemes Filho, “New quantitative model,” Optik - International Journal for Light and Electron indicator for measuring the quality of road,” IEEE Latin Am. Optics, vol. 126, no. 19, pp. 1761–1766, 2015. Trans, vol. 14, no. 2, pp. 582–585, 2016. [15] S. Syed and A. Mohammad, “Preceding document clustering by graph mining based maximal frequent termsets preser- vation,” Te International Arab Journal of Information Technology, vol. 16, no. 3, pp. 364–370, 2019. [16] F. Glover and Z. Lu, ¨ “Focal distance tabu search,” Science China Information Sciences, vol. 64, no. 5, Article ID 150101, [17] M. Gmira, M. Gendreau, A. Lodi, and J. Y. Potvin, “Tabu search for the time-dependent vehicle routing problem with time windows on a road network,” European Journal of Operational Research, vol. 288, no. 1, pp. 129–140, 2021. [18] E. Kharati and M. Khalily-Dermany, “Determination of the multicast optimal route for mobile sinks in a specifed deadline using network coding and tabu search algorithm in wireless sensor networks,” Iranian Journal of Science and Technology, Transactions of Electrical Engineering, vol. 45, no. 2, pp. 447–459, 2021. [19] S. Xie, P. Wu, H. Liu et al., “A novel method of unmanned surface vehicle autonomous cruise,” Industrial Robot: Inter- national Journal, vol. 43, no. 1, pp. 121–130, 2016. [20] Y. Mei, “Study on the application and improvement of ant colony algorithm in terminal tour route planning under Android platform,” Journal of Intelligent and Fuzzy Systems, vol. 35, no. 3, pp. 2761–2768, 2018. [21] Y. Wang and Q. Liu, “Robot path planning method based on rough set theory and a genetic algorithm,” IEEE Trans. Agro Food Ind. Hi-Tech, vol. 28, no. 1, pp. 1972–1976, 2017. [22] P. Chen, J. Pei, W. Lu, and M. Li, “A deep reinforcement learning based method for real-time path planningand dy- namic obstacle avoidance,” Neurocomputing, vol. 497, pp. 64–75, 2022. [23] L. Long, J. Xu, S. S. Liao, and H. Chen, “A real-time per- sonalized route recommendation system for self-drive tourists based on vehicle to vehicle communication,” Expert Systems with Applications, vol. 41, no. 7, pp. 3409–3417, 2014. [24] Q. Long, “Dynamic route guidance method facing driver’s individual demand,” J. Cent. South Univ, vol. 44, no. 5, pp. 2125–2129, 2013. [25] P. Zhan, “Path planning strategy for vehicle navigation based on user habits,” Appl. Sci.-Basel, vol. 8, no. 3, pp. 1–24, 2018. [26] Y. Ito, T. Ogawa, and M. Haseyama, “Accurate estimation of personalized video preference using multiple users’ viewing behavior,” IEICE - Transactions on Info and Systems, vol. 101, no. 2, pp. 481–490, 2018. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Computational Intelligence and Neuroscience Hindawi Publishing Corporation

A Personalized Navigation Route Recommendation Strategy Based on Differential Perceptron Tracking User’s Driving Preference

Download PDF
14 pages

Loading next page...
 
/lp/hindawi-publishing-corporation/a-personalized-navigation-route-recommendation-strategy-based-on-Y4Ef0fTa3z

References (32)

Publisher
Hindawi Publishing Corporation
ISSN
1687-5265
eISSN
1687-5273
DOI
10.1155/2023/8978398
Publisher site
See Article on Publisher Site

Abstract

Hindawi Computational Intelligence and Neuroscience Volume 2023, Article ID 8978398, 14 pages https://doi.org/10.1155/2023/8978398 Research Article A Personalized Navigation Route Recommendation Strategy Based on Differential Perceptron Tracking User’s Driving Preference Pengzhan Chen , Jihua Wu, and Ning Li Taizhou University, School of Intelligent Manufacture, Taizhou 318000, Zhejiang, China Correspondence should be addressed to Ning Li; wwwningning@126.com Received 9 August 2022; Revised 3 November 2022; Accepted 7 December 2022; Published 4 January 2023 Academic Editor: Maciej Lawrynczuk Copyright © 2023 Pengzhan Chen et al. Tis is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. With the increasing frequency of autonomous driving, more and more attention is paid to personalized path planning. However, the path selection preferences of users will change with internal or external factors. Terefore, this paper proposes a personalized path recommendation strategy that can track and study user’s path preference. First, we collect the data of the system, establish the relationship with the user preference factor, and get the user’s initial preference weight vector by dichotomizing the K-means algorithm. Te system then determines whether user preferences change based on a set threshold, and when the user’s preference changes, the current preference weight vector can be obtained by redefning the preference factor or calling diference perception. Finally, the road network is quantized separately according to the user preference weight vector, and the optimal path is obtained by using Tabu search algorithm. Te simulation results of two scenarios show that the proposed strategy can meet the re- quirements of autopilot even when user preferences change. Path recommendation systems usually recommend 1. Introduction paths based on the optimization of distance or travel time In recent years, with the rapid development of information cost functions. However, the shortest or fastest route is and intelligent transportation systems, artifcial intelligence usually not chosen by the driver, so the driver may have (AI) has been widely used in transportation tools to provide diferent recommendations when traveling. a comfortable travel experience for drivers and passengers Te personalized route recommendation strategy for [1–3]. Te vehicle Internet tends to share data, enabling each driver’s route selection preference has attracted much vehicles to exchange learning experiences and improve attention. Te personalized route recommendation strategy decision-making capabilities. Te individual models are not only meets people’s personalized needs but also solves the Blythe’s paradox, which is widely existed in the current trained on the basis of the data collected. By collecting these learning models from all vehicles, a comprehensive model transportation network. can be further developed. Tis is of the great signifcance to the application of intelligent transportation system (ITS) in autonomous driving and trafc control. 1.1. Literature Review. At present, the driver can get the current driving status of the vehicle through the OBD data. At present, the existing path navigation and recommen- dation software in China, such as the Baidu map application In reference [9], the authors proposed a monitoring system consisting of OBD and GPS, designated area, pricing and the Josiah Goddard map application, are all popular recommendation software, mainly based on the shortest time scheme, and the relationship between other related policies. Te OBD device is used to monitor engine operation and [4, 5] and the shortest distance [6–8], has been unable to meet measure fuel consumption and emissions reliably. In the growing demand for personalized tourism. 2 Computational Intelligence and Neuroscience and a genetic algorithm is used in this study to solve the low reference [10], OBD is used to analyze the driving behavior data through the vehicle preloading equipment and the efcacy and accuracy in robot path planning. Experimental results show that the proposed method can efciently factors that afect the safe driving and establish the logistic expression model. In reference [11], OBD is used to monitor improve robot path planning. In [22], Chen et al. proposed and prompt hydrocarbon (HC) emissions caused by a failure a deep reinforcement learning algorithm for path planning, of the vehicle emission control system. In this paper, the which has the comprehensive reward function of dynamic OBD data are used to correlate with the driving user’s path obstacle avoidance and goal approaching. Te results show selection preferences. Ten, the user’s initial preference that the method can avoid moving obstacles in the envi- weight vector is obtained by the bisecting K-means algo- ronment, complete the planning task, and has a high success rate. rithm clustering. In reference [12], Xiong et al. proposed a model predictive control optimal path optimization and Te personalized needs of users are not mentioned in these articles. In [23], Long et al. developed a novel route tracking framework. First, the relationship between the vehicle and the reference road is established in the road recommendation system to provide real-time personalized route recommendation for self-driving tourists, according coordinate system, and then, the safe lane under the high constraint environment is established by using the multi- to the specifc preferences of users personalized access layer search method; path boundary and vehicle dynamics routes, not only to save the total tour time but also to meet constraints are introduced to provide optimal control their specifc travel preferences. In [24], a dynamic route instructions. guidance method for driver’s personalized needs is pro- In reference [13], the unsupervised classifcation posed. Te user preference weight is given artifcially, method based on the bisecting k-means algorithm is ap- which has strong subjectivity and lacks objectivity and accuracy, but it provides a reference for the combination of plied to the data obtained in low-energy consumption proofread measurement of the online gas analyzer and personalized needs and path planning. In reference [25], a personalized path decision algorithm based on user pref- microsensor. Te good agreement with data from sensors validated the efectiveness of the proposed method. In [14], erence is proposed, which lays a good theoretical foun- dation and framework for this study. But there are also Zhao and othersproposed a reviewed block-based bag of words model using the bisection K-means clustering limitations. Te one-time sample trajectory data used in method which could signifcantly accelerate the process of this paper needs a lot of complex mathematical calculations codebook generation. In reference [15], a new algorithm to obtain the data needed for clustering. Te corresponding based on graph mining is proposed and the bisecting k- weight of user driving style obtained by fuzzy c-means means algorithm is used to fnd frequent terminology clustering is fxed and one-sided, while in real life, the user collection for the document. Te Tabu search algorithm is a driving preferences are not always the same and most of them are staged, so it cannot refect the user’s personalized typical shortest path algorithm, which is used to calculate the shortest path from one node to other nodes and is driving preferences. widely used in path planning. In reference [16], the Tabu search algorithm is proposed to solve the problem of Web service composition. It adds a periodic diversifcation step, 1.2. Contribution. As mentioned above, most traditional which is kept on the diversifcation step at the usual path planning schemes improve the speed and precision of completion point. Tis approach combines path relinking path planning algorithmically, while ignoring the efect of with dynamic diversifcation strategies, providing more user preferences [26–30] on path selection. In scenarios that opportunities for future research. In [17], the Tabu search combine user preferences and path planning, user prefer- algorithm is used to improve the route planning strategy in ences are unchanged by default. In real life, however, user urban areas under trafc congestion, which not only saves preferences are not always the same. At diferent times, user preferences may change periodically due to various internal driving time but also reduces fuel consumption. In [18], the Tabu search algorithm is used to identify the current to- or external factors, such as age, physical condition, climate, and work. Te type of preference may also change over time. pology of the network and help identify the shortest path from the point of failure to the nearest operation source. At the same time, when the preference type is unchanged, Te model improves efciency consumption by 23% and the corresponding preference weights may also change. bandwidth lifetime by 16%. How to make navigation more Terefore, this paper proposes a personalized path efective has been a research hotspot. In [19], Xie et al. planning strategy that can track and study user preferences. proposes a new method of combining global path planning First, the preference type is associated with the OBD data with local path planning, to provide an efcient solution for and a threshold is used to select the user’s current preference the unmanned surface vehicle (USV) path planning despite type. Ten, the initial preference weight vector is obtained by the changeable environment. Te method solves the path the split-k-average algorithm. Finally, the diferential per- ceptron is used to track and adjust the preference weights in planning problem with variable environments and is verifed by simulations and experiments. In [20], Mei real time, so that the optimal path can be recommended to meet the user’s needs when the user’s preferences change. In proposes an optimal tour guide path planning model based on an ant colony algorithm. Te experimental results show this paper, Nanchang is taken as the reference city of road that the proposed model and the optimal path planning network quantifcation, and the Tabu search algorithm is algorithm are more optimized. In [21], rough set theory used to verify the efectiveness of the strategy. Computational Intelligence and Neuroscience 3 In the light of the above, the contributions of the present comfort, and safety. Among them, the highest attention is document are summarized as follows: paid to safety. In this paper, we choose these four preference factors as user preference sets. (1) To associate OBD with user preferences and get the To determine the type of user preference factors, OBD initial preference weight vector by clustering data and user preference are used to establish a connection (2) Considering the periodic change of preference, the in this paper [31]. Te main function of OBD is to supervise change of two kinds of preference is summarized the status of components related to emission control during (3) Tracking the user preferences and fne-tuning the the actual use of vehicles. In this paper, OBD status in- formation, geographic location information, and trip record weights through the diferential perceptron information are used to correlate user preference factors. (4) Personalized quantization of road network, using the Te correlation process is as follows: Tabu search algorithm to plan the optimal path (1) In the driving process, the greater the average speed (5) Taking Nanchang as a reference city, the simulation experiment is carried out to verify the efectiveness of of the user, the more the user attaches importance to time. In this paper, v is selected to be the relevance the proposed strategy quantity (RQ) of a time-based user and v in any period time can be read directly from the server 2. System Implementation Review through the OBD terminal device. A personalized path recommendation system that can track (2) Te average fuel consumption (AFC) in the driving and study the user’s path selection preferences is proposed in process is the most important indicator for economic this paper, as shown in Figure 1. Te system mainly includes users. Te lower the value, the higher importance data preprocessing module, initial preference module, op- users attach to the economy. Terefore, AFC is timization and adjustment module, and path generation chosen as the associated quantity of economic users in this paper. As there is no specifc calculation module. When the change of user preferences has been detected, the system will make use of the latest OBD data in method for AFC in the standard OBD II and EOBD the storage space to study independently ofine until a path protocol, it is necessary to use OBD related data for that can meet the user’s current path selection preferences estimation. has been found. (3) Te change rate of the relative position of the A brief process of tracking and studying the path accelerator pedal (CR ) is closely related to the ap recommendation system is shown in Figure 2. In which, comfortable user. Te lower the (CR ) value, the ap three preference factors of time, economy, and comfort higher the user’s attention to the comfort. Users are taken as the user’s initial path selection preferences, who pay attention to comfort have good driving and the path points marked in black are path ofset points. stability. Tey step on and loose slowly, and give Te frst case is that the preference factors change, the oil smoothly. Terefore, CR is chosen as the ap latest recorded OBD data are needed to redetermine them, correlation quantity of comfortable users in this and the initial driving preference weight is obtained by paper. Moreover, it can be obtained directly from clustering. In the second case, the diference perceptron the server by modifying the interval time pa- needs to be used for learning and correction because the rameters of the standard OBD II and EOBD preference factors remain unchanged and the corre- protocol. sponding weight values between the preference factors (4) In the process of driving, excessive driving speed will change. Te third case is that in some road sections, the bring a threat to the safety of users. Terefore, the planned path is not consistent with the actual driving maximum traveling speed of this period is taken as path. Since, the trafc network is real-time and dynamic, the correlation quantity of safe users in this paper. emergencies, such as trafc lights, pedestrians, and ac- However, it can be read directly from the OBD cidents, will afect the coincidence rate. Tus, in the actual terminal device. situation, the planning path and the driving path are not exactly consistent. Terefore, within the allowable error Te calculation process of each associated quantity is range, in this case, it is considered that the user preference shown in Table 1. remains unchanged and it is considered as a sudden Considering the laws and regulations related to safe situation. driving and the opinions of professionals in the automobile industry, the threshold value of the average value of daily correlation volume is set, as shown in Table 2. 3. Path Recommendation System Te maximum three items of δ are selected as the user’s 3.1. Data Preprocessing Module. Diferent self-driving users preference factors by the relevant degree δ of correlation have diferent path selection preferences, so we conducted a and threshold calculation. As k � 3, the clustering efect is questionnaire on the surrounding self-driving users and obvious, and to explain the change of preference factors obtained the survey data shown in Figure 3. From Figure 3, conveniently, three preference factors are selected as the we can see that four preference factors are afecting the route user’s path selection preference each time. Te relevant selection of self-driving users, namely, time, economy, degree calculation formula is defned as follows: 4 Computational Intelligence and Neuroscience MODULE Data preprocessing module Initial habit module Path generation module Optimization and adjustment module Collecting User with certain OBD data habit factors Route recommendation system Figure 1: Te illustration of path recommendation. Initial Preference weight Methods:Bisecting K-means clustering Economy Time Comfort Data: OBD Phenomena: there are great differences between paths Inducement: preference factor AA H B J B A J B path1 path1 changed path1 Solution : Clustering with current C J KDD C T L C T L D path2 path2 path2 OBD data EE X YZZ F S N F E S N X F path3 path3 path3 Phenomena: there are differences AA H B J B A J B path1 path1 path1 between paths C J KDD C T L Inducement: preference weights C T L Z path2 path2 path2 changed EE X Y Z F S N V F S N M F path3 path3 path3 Solution : Calling differential perceptron AAHHBB A H B path1 path1 path1 CCJJ L D C J L D path2 path2 path2 Phenomena: path is basically EXX Y Z F E H Z F X H Z F path3 path3 unchanged path3 Inducement: habits are unchanged Solution : Remaining unchanged Path after Output Initial path habits changing Figure 2: Te illustration of recommendation under diferent circumstances. m − n examples and the bisecting k-means algorithm is used in | | i i ⎧ ⎪ δ � clustering in this paper. To facilitate the clustering and efect i m display of OBD data, the data should be normalized before < 2m , i ∈ T, E, C, S􏼁. (1) i i clustering. Te partial results of some user OBD sampling ⎪ N 1 data processed are shown in Table 3. ⎪ n � 􏽘 n i ij v, AFC, and CR are used to establish the three-di- N ap j�1 mensional coordinate system, and then, the normalized data are divided into the bisecting K-means clustering, and the initial preference weight vector of users is obtained 3.2. Initial Habit Module. After the user preference factor is according to the clustering results. In order to make the determined, the initial preference weight vector is obtained clustering efect signifcant, here, select Δt � 1 min to be by clustering. Time, economy, and comfort are taken as clustered, and the results are shown in Figure 4. Computational Intelligence and Neuroscience 5 4% 24% 36% 21% 15% Time Safety Economy Other Comfort Figure 3: Results of habit factor sampling for surrounding car users. Table 1: Te OBD data processing algorithm. Among them, n , n , and n are the numbers of coor- 1 2 3 dinate points belonging to the three clustering centers Algorithm 1: OBD data processing individually. Input: OBD data, V � 0.91, M � 28.27, R � 8.314, P � 6.17, E M PG G � 4.536, R � 14.7 PP A Output: v, AFC, CR , V ap max 3.3. Path Generation Module. Tis part is mainly divided 1: Begin into two parts. Te frst part describes how to individualize 2: While car running and quantify the road network to establish a model for 3: Read and storage the values of v, CR and V every ap max solving the optimal path. Te second part shows how to 6 seconds from OBD data calculate the optimal path by simulation. 4: Read the values of, L, R , M , and I every PM AP AT 6 seconds from OBD data 5: Calculate AFC 6: M � (R × M /I /120) × (V /100) × E × M /R 3.3.1. Road Network Quantifcation. To simplify the road AF PM AP AT E D M 7: AFC � (R × P × G × v/3600 × M ) × (1 + L) model and calculation, referring to the regulations of A G PP AF 8: Storage AFC highway technical engineering standard, urban road design 9: End while code, urban road network planning index system, Nanchang 10: End urban trafc planning, and combining with the actual characteristics of the road network, Baidu map and the results of feld investigation and verifcation, the roads in As can be seen from the fgure, the value of red Nanchang are divided into the following fve categories. To marked points v-axis is generally higher than the other sum up, the road conditions in Nanchang are actual. Te two categories. Due to the highest attention to time, the actual situation of each road is shown in Table 4. value v of time users is generally large. Terefore, the red In order to unify the quantitative standard and simplify marker class is a time-based feature point and Te values the calculation of the optimal path with a genetic algorithm, of the AFC-axis of the points marked in green are gen- the data in Table 3 are normalized and correlated. Te erally smaller than those of the other two categories. process is as follows: Economic users pay the highest attention to fuel con- 1 300 1 4 ⎧ ⎪ sumption, and the value of AFC is generally very small. Cos t � 􏼒0.6∗ + 0.1∗ + 0.3∗ 􏼓, ⎪ 8 MS RND RS Terefore, the green marker is the economic character- istic point. Te value of CR -axis of blue marked points ap ⎪ is generally smaller than the other two categories. ⎪ Cos t � EC, Comfortable users pay the most attention to driving (3) stability and the value of CR is generally very small. ⎪ ap ⎪ 1 1 Cos t � ∗ , Terefore, the blue marker is the comfort feature point. ⎪ 8 RE After the bisecting K-means clustering converges, the ⎪ number of coordinate points belonging to three clus- ⎪ MS tering centers is normalized to get the initial user pref- ⎩ Cos t � 0.5 ∗ + 0.5∗ RS. erence weight vector w � (w , w , w ). Te normalization 1 2 3 formula is as follows: Te data processed by formula (3) are shown in Table 5. Based on the comprehensive consideration of diferent w � (j � 1, 2, 3). (2) driving preferences and simplifed calculation, the unit cost n + n + n 1 2 3 function is designed as follows: 6 Computational Intelligence and Neuroscience Table 2: Correlation threshold. Relevant quantity v (km/h) AFC (L/100 km) CR (%) V (km/h) ap max Treshold 40 8 10 60 Table 3: Partial OBD data after normalization. Time v AFC CR V ap max 08:08:34:25 0.1250 0.0090 0.0588 0.3600 08:08:40:25 0.0500 0.0068 0.5294 0.2010 08:08:46:25 0.2500 0.1109 0.4706 0.5212 08:08:52:25 0.3500 0.1267 0.8824 0.5842 08:08:58:25 0.3562 0.1154 0.5882 0.6124 08:08:64:25 0.6250 0.7467 0.6471 0.7653 08:09:00:25 0.8875 0.9887 0.4706 0.9102 1 1 0.8 0.8 0.6 0.6 – – v v 0.4 0.4 0.2 0.2 0 0 1 1 0.8 0.8 0.5 0.5 0.6 0.6 0.4 0.4 AFC 0.2 AFC 0.2 CR CR 0 0 ap 0 0 ap (a) (b) Figure 4: Bisecting K-means clustering results (above is before clustering and below is after clustering). optimal path becomes a single source shortest path problem U � f c , c , · · · , c � 􏽘 􏼁 ω c . (4) k k 1 2 m i i shown in Figure 5. i�1 Te process of using the Tabu search algorithm to get the optimal path of the model shown in Figure 5 is shown in In this paper, each user has three main driving prefer- Table 6. As can be seen from Table 6, the fnal optimal path is ences, that is, the other three preference weights are set to 1 ⟶ 4 ⟶ 7 ⟶ 10 ⟶ 13 ⟶ 15 ⟶ 16 ⟶ 20, and zero, making m � 3, so the cost of personalized quantif- the total cost consumption is 88. cation of the K segment road is as follows: U � L × f c , c , c 􏼁 � L × 􏽘 ω c . (5) Tk k k 1 2 3 k i i 3.4. Optimization and Adjustment Module. In order not to i�1 afect the user’s self-driving experience, it is stipulated that the system will conduct tracking and studying at night every day, and the preference tracking process is shown in 3.3.2. Tabu Search Algorithm. After the recommendation Table 7. Due to the existence of emergencies, the consis- system personalizes and quantifes the road network tency between the planned path and the actual driving path according to the user preference weight vector [32], the will not always be 100%. In the system, a threshold pa- model becomes a classic problem of fnding the optimal rameter is set for the coincidence degree. Before training solution. In this paper, the Tabu search algorithm is used to every day, the system will flter the OBD data of the day and calculate the optimal path according to the user’s preference. eliminate the data whose coincidence degree is higher than Here, the model diagram shown in Figure 5 is selected to the threshold value. Ten, the system recalculates the illustrate how the algorithm calculates the optimal path. In correlation degree δ of four correlation quantities Figure 5, the cost consumption values of each point between according to the remaining OBD data of the day in the 1 and 20 are calculated, respectively, by formula (5). After storage space and sorts them from small to large. If the personalized quantifcation of the road network, fnding the correlation quantity corresponding to the minimum value Computational Intelligence and Neuroscience 7 Table 4: Performance of road networks at all levels. Time RND (km/km ) RS (v/c) RE (a.u.) MS (km/h) EC (RMB/km) Expressway 0.42 0.42 0.40 100 0.74 Main road 1.31 0.68 0.35 60 0.51 Sub road 1.60 0.84 0.30 50 0.58 Landscape road 0.38 0.62 0.65 50 0.54 Business zone 1.75 1.15 0.22 40 0.62 Table 5: Road network costs at normalized levels. Cost Road classifcations Cos t Cos t Cos t Cos t T E C S E 0.31 0.74 0.31 0.32 M 0.40 0.51 0.36 0.49 S 0.47 0.58 0.42 0.58 L 0.36 0.54 0.19 0.58 B 0.62 0.62 0.57 0.74 1 8 15 11 8 13 6 8 16 9 15 2 14 7 24 20 14 Figure 5: Road network model after personalized quantifcation. Table 6: Tabu search algorithm solution process. Routes Total consumptions 1 ⟶ 1 0 1 ⟶ 2 8 1 ⟶ 3 12 1 ⟶ 4 10 1 ⟶ 3 ⟶ 5 17 1 ⟶ 4 ⟶ 6 14 1 ⟶ 4 ⟶ 7 17 1 ⟶ 3 ⟶ 5 ⟶ 8 24 1 ⟶ 4 ⟶ 6 ⟶ 9 23 1 ⟶ 4 ⟶ 7 ⟶ 10 35 1 ⟶ 4 ⟶ 6 ⟶ 9 ⟶ 11 38 1 ⟶ 3 ⟶ 5 ⟶ 8 ⟶ 12 49 1 ⟶ 4 ⟶ 7 ⟶ 10 ⟶ 13 49 1 ⟶ 4 ⟶ 7 ⟶ 10 ⟶ 14 55 1 ⟶ 4 ⟶ 7 ⟶ 10 ⟶ 13 ⟶ 15 55 1 ⟶ 4 ⟶ 7 ⟶ 10 ⟶ 13 ⟶ 15 ⟶ 16 63 1 ⟶ 4 ⟶ 7 ⟶ 10 ⟶ 13 ⟶ 17 60 1 ⟶ 4 ⟶ 7 ⟶ 10 ⟶ 13 ⟶ 17 ⟶ 18 68 1 ⟶ 4 ⟶ 7 ⟶ 10 ⟶ 14 ⟶ 19 85 1 ⟶ 4 ⟶ 7 ⟶ 10 ⟶ 13 ⟶ 15 ⟶ 16 ⟶ 20 88 8 Computational Intelligence and Neuroscience Table 7: Te habit tracking processing algorithm. Table 8: Te improved diferential perceptron algorithm. Algorithm 2: Habit tracking processing Algorithm 3: Improved diferential perceptron Input: Te latest OBD data, current habit weight vector w Input: OBD data, habit weight vector ′ ′ Output: new habit weight vector w � (ω , ω , ω ) Output: Improved new habit weight vector w � w w w ′ 1 ′ 2 ′ 3 1 2 3 1: Begin 11: While n ! � 2000 or ‖w − w‖> ε 2: Calculate δ with the latest OBD data, i ∈ T, E, C, S 12: For each road classifcation do 3: If RQ of δ min unchanged 4: w � Improved diferential perceptron (the last OBD 13: Calculate the total length of the section in AP and PP data, w) respectively 5: Else 6: Obtain w by reclustering the OBD data of the other AP AP three RQ ⎧ ⎪ L � 􏽘 L Ti Ti (7) End if j�1 14: i ∈ T, E, C, S PP AP (8) End ⎪ L � 􏽘 L Ti j j�1 15: End for of δ changes, the system determines that the preference 16: For each habit factor do factor changes, and clusters the latest data of the other three correlation quantities to get the initial weight vector again. 17: Calculate the total cost in AP and PP respectively On the contrary, the system judges that the preference AP factor has not changed, and calls the diferential perceptron U � L 􏽘 ω c ⎧ ⎪ APi i i ⎪ Ti to fne-tune the weight vector. Because of the relatively i�1 18: i ∈ T, E, C, S ⎪ PP small number of data samples on the same day, direct U � L 􏽘 ω c ⎩ PPi Ti i i i�1 clustering by ignoring the previous data will enlarge the 19: Calculate weight change volume respectively efect of change. Data clustering together will cover up 20: ∆ω � η(􏽐 U − 􏽐 U )i ∈ T, E, C, S APi PPi smaller weight changes. Terefore, in this paper, we choose 21: Calculate and normalize the new driving style weight to call the diferential perceptron to fne-tune to improve 22: ω � ω + Δω j � 1, 2, 3 j j the accuracy. 23: End for 24: n � n + 1 Te setting of the diferential perceptron function is 25: Re-quantify road network with w shown in Table 8. Taking OBD data and current weight 26: Replan the optimal path with the Tabu search algorithms vector as input, the maximum number of iterations is set as 27: End while 2,000. Te total length of fve types of roads is calculated for 28: Return w � w w w 1 2 3 the actual path and the planned path, respectively, and the total cost corresponding to the three preference factors is calculated for the two cases. Ten, the change of the weight 4.1. Experiment 1. Te preference factor of self-driving user of each preference factor is calculated. Among which, η is the A is economy, comfort, and safety, and the corresponding learning efciency, take 0.003 to obtain the weight vector initial weight vector is obtained by the clustering algorithm, closest to the current user path selection preference through which is w � (0.48, 0.14, 0.38). After the preference factor of repeated iterative learning. user A changes, the planned path and the actual driving path are shown in the red path and green path in Figure 6, re- 4. Experiment spectively. It can be seen from the fgure that due to the change of preference factors, the coincidence of the planning Two types of preference change are proposed in this paper: path and the actual path are very low, only 0.785%, indi- one is the change of user preference factor and the other is cating that the current planning strategy can no longer meet the change of the weight of the same preference factor. To the personalized traveling needs of users. verify the efectiveness of the strategy proposed in this paper, After the personalized quantifcation of the road network Nanchang city is taken as a reference City, and simulation w � (0.48, 0.14, 0.38), the specifc generation values of experiments are carried out in two scenarios, respectively. In economic consumption (TC), comfort consumption (EC), scenario 1, the user preference factor changes. Assuming and safety consumption (CC) in two cases are shown in that the initial preference factor of user A is economy, Table 9, respectively. comfort, and safety, then due to work reasons, time is more Te total cost C of the actual path and the planned total important than the economy, so the preference factor path can be obtained through classifcation and integration becomeseconomy, comfort, and safety. In scenario 2, the of the data of the road type in Table 9 and is shown in weight of the same user preference factor changes. Suppose Table 10. As can be seen from Table 10, due to the change of that the initial preference factor of user B is time, economy, preference type, the path types in the two cases are diferent, and comfort. Ten, due to the fnancial crisis of the family, and the total cost is also very diferent, which is 12.737. user B pays more attention to the economy and less attention When the system is tracking and studying at night, the to comfort. In these cases, 90% of the threshold value is used correlation δ of four preference factors is calculated by to determine whether it is an emergency or not. formula (1). Among them, the correlation quantity of Computational Intelligence and Neuroscience 9 River School Landscape road Train station Sub road Expressway Business zone Main road Park Figure 6: Te initial planning path and the actual driving path before the adjustment in Experiment 1 (the left one is the model diagram and the right one is the actual map). Table 9: Adjustment of the frst two paths in Experiment 1. Paths Road names Types ERL (km) EC CC SC Wugong Mountain Avenue M 3.4 0.832 0.171 0.633 Xiangyun Avenue M 7.6 1.861 0.383 1.415 Changnan Avenue M 13.4 3.280 0.675 2.495 Initially planned route Changdong Avenue M 9.4 2.301 0.474 1.750 Ziyang Avenue M 6.5 1.591 0.328 1.210 Ziyang East Avenue M 2.6 0.637 0.131 0.484 Aviation City Avenue S 0.51 0.142 0.030 0.112 Wugong Mountain Avenue M 3.3 0.808 0.166 0.615 Circumferential Expressway E 60.4 21.454 2.621 7.345 Actual path Ziyang East Avenue M 0.79 0.193 0.040 0.147 Aviation City Avenue S 0.51 0.142 0.030 0.112 Table 10: Te total cost of the frst two paths before adjusting in Experiment 1. Paths Types L (km) U C Tk total M 42.90 20.652 Initially planned route 20.936 S 0.51 0.284 M 4.09 1.969 Actual path E 60.4 31.420 33.673 S 0.51 0.284 minimum δ changes from the previous time to economy. As can be seen from Figure 7, after tracking and Terefore, the system determines that the user A’s prefer- studying, the consistency between the planned path and the ence factor changes, and the current preference factor actual path is greatly improved, from 0.785% to 98%. Among changes to time, comfort, and safety. Te current weight them, the specifc cost data in the two cases after adjustment vector is obtained by the bisecting K-means clustering. Fi- are shown in Table 11. nally, the planning path calculated by the Tabu search al- Te total cost C of the actual path and the planned total gorithm using personalized quantitative road network is path can be obtained through classifcation and integration shown in the red path in Figure 7. of the data of the road type in Table 11 and is shown in 10 Computational Intelligence and Neuroscience River School Landscape road Train station Sub road Expressway Business zone Main road Park Figure 7: Te adjusted initial planning path and actual driving path in Experiment 1 (the left diagram is the model diagram and the right one is the actual map). Table 11: Data of two paths after being adjusted in Experiment 1. Paths Road names Type ERL (km) TC CC SC Wugong Mountain Avenue M 3.3 0.937 0.166 0.243 Initially planned route Circumferential Expressway E 60.7 13.360 2.634 2.914 Liu Cheng Street S 0.82 0.274 0.048 0.071 Wugong Mountain Avenue M 3.3 0.937 0.166 0.243 Circumferential Expressway E 60.4 13.294 2.621 2.899 Actual path Ziyang East Avenue M 0.79 0.224 0.040 0.058 Aviation City Avenue S 0.51 0.170 0.030 0.044 Table 12 As can be seen from Table 12, the path types in the Te total cost C of the actual path and the planned total two cases are the same, and the total cost diference is only path can be obtained through classifcation and integration of 0.081, which indicates that the system can still meet the the data of the road type in Table 13 and is shown in Table 14. user’s path planning needs through tracking and adjusting It can be seen from the table that, due to the change of after the user’s preference factor changes. preference weight, the path types in the two cases are the same, but the length is diferent, and the total cost diference is 2.149. 4.2. Experiment 2. Te preference factor of self-driving user B Similarly, when the system is tracking and studying at is time, economy, and comfort, and the corresponding initial night, the correlation δ of four preference factors is weight vector is w � (0.14, 0.44, 0.42). After user B’s prefer- recalculated by formula (1). Among them, the correlation ence weight changes, the planned path and the actual driving quantity of the corresponding minimum δ value does not path are shown in red and green paths in Figure 8, respectively. change, which is safety. Terefore, the system will determine Due to the change of preference weight, the consistency that the user B preference type has not changed but the between the planned path and the actual path is not high, weight changes, and call the diferential perceptron to fne- which is 76.64%. It shows that the current planning strategy tune, and fnally, get the current weight vector does not fully meet the user’s personalized travel needs. To w � (0.15, 0.51, 0.34). Finally, the planning path calculated improve the degree of coincidence, we need to further by the Tabu search algorithm using personalized quantita- improve the accuracy of the weight vector. tive road network w is shown in the red path in Figure 9. After the personalized quantifcation of the road network As can be seen from Figure 9, after tracking and w � (0.14, 0.44, 0.42), the specifc generation values of time studying, the consistency between the planned path and the consumption (TC), economic consumption (TC), and actual path has improved, from 76.64% to 84.58%. Among comfort consumption (EC) in two cases are shown in Ta- them, the specifc cost data in the two cases after adjustment ble 13, respectively. are shown in Table 15. Computational Intelligence and Neuroscience 11 Table 12: Total cost of two paths after adjustment in Experiment 1. Paths Types L (km) U C Tk total M 3.3 1.346 Initially planned route E 60.7 18.908 20.647 S 0.82 0.393 M 4.09 1.668 Actual path E 60.4 18.815 20.728 S 0.51 0.245 River School Landscape road Train station Sub road Expressway Business zone Main road Park Figure 8: Te initial planning path and the actual driving path before adjustment (the left one is the model diagram and the right one is the actual map). Table 13: Te data of the frst two paths before the adjustment in Experiment 2. Paths Road names Types ERL (km) TC EC CC Wugong Mountain Avenue M 3.4 0.190 0.763 0.514 Xiangyun Avenue M 8.1 0.454 1.818 1.225 Riverside Avenue S 5.6 0.368 1.429 0.988 Hongcheng Road M 3.7 0.207 0.830 0.559 Jinggangshan Avenue M 0.9 0.050 0.202 0.136 Initially planned route Eight One Avenue M 1.5 0.084 0.337 0.227 Beijing West Road L 2.3 0.116 0.546 0.184 Beijing East Road L 5.8 0.292 1.378 0.463 Ziyang Avenue M 6.5 0.364 1.459 0.983 Ziyang East Avenue M 1.5 0.084 0.337 0.227 Aviation City Avenue S 0.51 0.034 0.130 0.023 Wugong Mountain Avenue M 3.4 0.190 0.763 0.514 Xiangyun Avenue M 8.1 0.454 1.818 1.225 Changnan Avenue M 8.7 0.487 1.952 1.315 Nanlian Road M 2.4 0.134 0.539 0.363 Jinggangshan Avenue M 4.8 0.269 1.077 0.726 Actual path Eight One Avenue M 1.5 0.084 0.337 0.227 Beijing West Road L 2.3 0.116 0.546 0.184 Beijing East Road L 5.8 0.292 1.378 0.463 Ziyang Avenue M 6.5 0.364 1.459 0.983 Ziyang East Avenue M 1.5 0.084 0.337 0.227 Aviation City Avenue S 0.51 0.034 0.130 0.023 12 Computational Intelligence and Neuroscience Table 14: Te total cost of the frst two paths before the adjustment in Experiment 2. Paths Types L (km) U C Tk total M 25.6 11.328 Initially planned route L 8.1 2.979 17.418 S 6.11 3.111 M 36.9 16.328 Actual path L 8.1 2.979 19.567 S 0.51 0.260 River School Landscape road Train station Sub road Expressway Business zone Main road Park Figure 9: Te initial planning path and the actual driving path before adjustment (the left one is the model diagram and the right one is the actual map). Table 15: Adjusted data of two paths in Experiment 2. Paths Road names Types ERL (km) TC EC CC Wugong Mountain Avenue M 3.4 0.204 0.884 0.416 Xiangyun Avenue M 8.1 0.486 2.107 0.991 Changnan Avenue M 6.4 0.384 1.665 0.783 Yingbin North Avenue M 3.9 0.234 1.014 0.477 Fuhen Road M 1.6 0.096 0.416 0.196 Hongcheng Road M 1.1 0.066 0.286 0.135 Initially planned route Jinggangshan Avenue M 0.9 0.054 0.234 0.110 Eight One Avenue M 1.5 0.090 0.390 0.184 Beijing West Road L 2.3 0.124 0.633 0.149 Beijing East Road L 5.8 0.313 1.597 0.375 Ziyang Avenue M 6.5 0.390 1.691 0.796 Ziyang East Avenue M 0.79 0.047 0.205 0.097 Aviation City Avenue S 0.51 0.036 0.151 0.073 Wugong Mountain Avenue M 3.4 0.020 0.884 0.416 Xiangyun Avenue M 8.1 0.049 2.107 0.991 Changnan Avenue M 8.7 0.052 2.263 1.065 Nanlian Road M 2.4 0.014 0.624 0.294 Jinggangshan Avenue M 4.8 0.029 1.248 0.588 Actual path Eight One Avenue M 1.5 0.009 0.390 0.184 Beijing West Road L 2.3 0.124 0.633 0.149 Beijing East Road L 5.8 0.313 1.597 0.375 Ziyang Avenue M 6.5 0.039 1.691 0.796 Ziyang East Avenue M 1.5 0.084 0.337 0.227 Aviation City Avenue S 0.51 0.036 0.151 0.073 Computational Intelligence and Neuroscience 13 Table 16: Adjusted total cost of two paths in Experiment 2. Paths Types L (km) U C Tk total M 34.19 15.129 Initially planned route L 8.1 3.191 18.580 S 0.51 0.260 M 36.9 16.328 Actual path L 8.1 3.191 19.779 S 0.51 0.260 Te total cost C of the actual path and the planned Data Availability total path can be obtained through classifcation and integration Te data used to support the fndings of this study are of the data of the road type in Table 15 and is shown in available from the corresponding author upon reasonable Table 16. As can be seen from Table 16, the path types in the request (pzchen@tzc.edu.cn). two cases are the same, and the total cost diference is 1.199. Compared with studying before adjustment, the diference Conflicts of Interest in total cost decreases and the coincidence increases. It shows that the system can also meet the user’s path planning Te authors declare that they have no conficts of interest. needs by tracking and adjusting when the user’s preference weight changes. Acknowledgments 5. Conclusion and Outlook Tis research work was supported by the National Nature Science Foundation under Grant no. 62163014. A personalized path recommendation strategy that can track and study the user’s path selection preferences is proposed in References this paper. In the case that the self-driving users do not know their preferences and their preferences will change peri- [1] A. Forestiero and G. Papuzzo, “Recommendation platform in odically, the system frst determines the main driving Internet of Tings leveraging on a self-organizing multiagent approach,” Neural Computing & Applications, vol. 34, no. 18, preference factors of users through the collected OBD data Article ID 16049, 2022. and obtains the initial preference weight vector through the [2] A. Forestiero, “Heuristic recommendation technique in In- bisecting K-means clustering. When the preference changes, ternet of Tings featuring swarm intelligence approach,” the system judges according to the relevant parameters and Expert Systems with Applications, vol. 187, Article ID 115904, adjusts the preference weight vector according to the situ- ation. Ten, the road network model is quantifed according [3] A. Forestiero and G. Papuzzo, “Agents-based algorithm for a to the weight vector. Finally, the Tabu search algorithm is distributed information system in internet of things,” IEEE used to calculate the optimal path. In the two scenarios, Internet of Tings Journal, vol. 8, no. 22, Article ID 16548, through the comparison between the planned path and the actual path and the path before and after tracking and [4] J. Ma and G. Sun, “Mutation ant colony algorithm of milk-run vehicle routing problem with fastest completion time based studying, the efectiveness of the strategy proposed in this on dynamic optimization,” Discrete Dynamics in Nature and paper is verifed. Te results show the consistency between Society, vol. 2013, no. 1, Article ID 418436, 6 pages, 2013. the planned path and the actual path has improved from [5] J. Guo, Y. Wu, X. Zhang et al., “Finding the ‘faster’ path in 76.64% to 84.58% after tracking and studying the user’s vehicle routing,” IET Intelligent Transport Systems, vol. 11, personalized driving behavior. Te adjusted total cost of no. 10, pp. 685–694, 2017. initially planned route is 18.580, while actual path 19.779. [6] A. Idri, M. Oukarf, A. Boulmakoul, K. Zeitouni, and A. Masri, Te path types of the two cases are the same, and the total “A distributed approach for shortest path algorithm in dy- cost diference is only 0.081. It is proved that the strategy can namic multimodal transportation networks,” Transportation also meet the personalized traveling needs of users when Research Procedia, vol. 27, pp. 294–300, 2017. their preferences change. [7] Y. Zhang, S. Song, Z. J. M. Shen, and C. Wu, “Robust shortest To make the Tabu search algorithm not easily fall into path problem with distributional uncertainty,” IEEE Trans- actions on Intelligent Transportation Systems, vol. 19, no. 4, local optimum when calculating the optimal path, the road pp. 1080–1090, 2018. network model in this paper is relatively simple. However, [8] P. Faizian, M. A. Mollah, X. Yuan, Z. Alzaid, S. Pakin, and when the actual road network is larger and the route is more M Lang, “Random regular graph and generalized de bruijn complex, the algorithm is easy to fall into local optimum. At graph with $k$ -shortest path routing,” IEEE Transactions on the same time, only static factors rather than the real-time Parallel and Distributed Systems, vol. 29, no. 1, pp. 144–155, trafc situations in the road network are considered in this paper. In the future work, we should improve the accuracy of [9] Y. Wu and D. Sperling, “Using on-board diagnostic and global planning path algorithm or change the way of road network positioning system to price emissions from on-road heavy- modeling, and add some dynamic factors to make the duty vehicles,” Journal of Transportation Engineering, Part A: strategy proposed in this paper more universal. Systems, vol. 144, no. 8, 2018. 14 Computational Intelligence and Neuroscience [10] Y. Pan, “Using OBD-II data to explore driving behavior [27] Y. J. Song and J. M. Kim, “Characterization of privacy based model,” in Proceedings of the IEEE-ICASI 2017, Sapporo, on context sensitivity and user preference for multimedia context-aware on IoT,” Multimedia Tools and Applications, Japan, July 2017. [11] P. Kalita, “Power train and emission control: allocation vol. 78, no. 5, pp. 5355–5366, 2019. [28] W. Wu, R. Zhang, and L. Liu, “A personalized network-based procedure by OBD-II system for automotive technology,” IOP recommendation approach via distinguishing user’s prefer- Conference Series: Materials Science and Engineering, vol. 211, ence,” International Journal of Modern Physics B, vol. 33, Article ID 012019, 2017. no. 6, Article ID 1950029, 2019. [12] L. Xiong, Z. Fu, Z. Qian, B. Leng, D. Zeng, and Y. Huang, [29] J. Han and H. J. Suk, “Exploring user’s preference on the color “Model predictive trajectory optimization and tracking in of cavity and lighting of an oven product,” Archives of Design highly constrained environments,” International Journal of Research, vol. 32, no. 2, pp. 19–29, 2019. Automotive Technology, vol. 23, no. 4, pp. 927–938, 2022. [30] H. Liu, H. Ma, Y. Chang, Z. Li, and W. Wu, “Target com- [13] A. Caron, N. Redon, P. Coddeville, and B. Hanoune, munity detection with user’s preference and attribute sub- “Identifcation of indoor air quality events using a K-means space,” IEEE Access, vol. 7, Article ID 46583, 2019. clustering analysis of gas sensors data,” Sensors and Actuators [31] G. Wei, “Research on parametric indicators of driving habits B: Chemical, vol. 297, Article ID 126709, 2019. based on OBD technology,” Ergonomics, vol. 23, no. 2, [14] C. Zhao, X. Li, and Y. Cang, “Bisecting k-means clustering pp. 38–44, 2017. based face recognition using block-based bag of words [32] L. Israel Pinto and P. Bertemes Filho, “New quantitative model,” Optik - International Journal for Light and Electron indicator for measuring the quality of road,” IEEE Latin Am. Optics, vol. 126, no. 19, pp. 1761–1766, 2015. Trans, vol. 14, no. 2, pp. 582–585, 2016. [15] S. Syed and A. Mohammad, “Preceding document clustering by graph mining based maximal frequent termsets preser- vation,” Te International Arab Journal of Information Technology, vol. 16, no. 3, pp. 364–370, 2019. [16] F. Glover and Z. Lu, ¨ “Focal distance tabu search,” Science China Information Sciences, vol. 64, no. 5, Article ID 150101, [17] M. Gmira, M. Gendreau, A. Lodi, and J. Y. Potvin, “Tabu search for the time-dependent vehicle routing problem with time windows on a road network,” European Journal of Operational Research, vol. 288, no. 1, pp. 129–140, 2021. [18] E. Kharati and M. Khalily-Dermany, “Determination of the multicast optimal route for mobile sinks in a specifed deadline using network coding and tabu search algorithm in wireless sensor networks,” Iranian Journal of Science and Technology, Transactions of Electrical Engineering, vol. 45, no. 2, pp. 447–459, 2021. [19] S. Xie, P. Wu, H. Liu et al., “A novel method of unmanned surface vehicle autonomous cruise,” Industrial Robot: Inter- national Journal, vol. 43, no. 1, pp. 121–130, 2016. [20] Y. Mei, “Study on the application and improvement of ant colony algorithm in terminal tour route planning under Android platform,” Journal of Intelligent and Fuzzy Systems, vol. 35, no. 3, pp. 2761–2768, 2018. [21] Y. Wang and Q. Liu, “Robot path planning method based on rough set theory and a genetic algorithm,” IEEE Trans. Agro Food Ind. Hi-Tech, vol. 28, no. 1, pp. 1972–1976, 2017. [22] P. Chen, J. Pei, W. Lu, and M. Li, “A deep reinforcement learning based method for real-time path planningand dy- namic obstacle avoidance,” Neurocomputing, vol. 497, pp. 64–75, 2022. [23] L. Long, J. Xu, S. S. Liao, and H. Chen, “A real-time per- sonalized route recommendation system for self-drive tourists based on vehicle to vehicle communication,” Expert Systems with Applications, vol. 41, no. 7, pp. 3409–3417, 2014. [24] Q. Long, “Dynamic route guidance method facing driver’s individual demand,” J. Cent. South Univ, vol. 44, no. 5, pp. 2125–2129, 2013. [25] P. Zhan, “Path planning strategy for vehicle navigation based on user habits,” Appl. Sci.-Basel, vol. 8, no. 3, pp. 1–24, 2018. [26] Y. Ito, T. Ogawa, and M. Haseyama, “Accurate estimation of personalized video preference using multiple users’ viewing behavior,” IEICE - Transactions on Info and Systems, vol. 101, no. 2, pp. 481–490, 2018.

Journal

Computational Intelligence and Neuroscience – Hindawi Publishing Corporation

Published: Jan 4, 2023

There are no references for this article.