Optimal Load Distribution in DG Sources Using Model Predictive Control and the State Feedback Controller for Switching Control
Optimal Load Distribution in DG Sources Using Model Predictive Control and the State Feedback...
Sayedi, Iman;Fatehi, Mohammad H.;Simab, Mohsen
2022-03-18 00:00:00
Hindawi International Transactions on Electrical Energy Systems Volume 2022, Article ID 5423532, 16 pages https://doi.org/10.1155/2022/5423532 Research Article Optimal Load Distribution in DG Sources Using Model Predictive Control and the State Feedback Controller for Switching Control 1 2 1 Iman Sayedi, Mohammad H. Fatehi , and Mohsen Simab Department of Electrical Engineering, Marvdasht Branch, Islamic Azad University, Marvdasht, Iran Department of Electrical Engineering, Kazerun Branch, Islamic Azad University, Kazerun, Iran Correspondence should be addressed to Mohammad H. Fatehi; mh_fatehi@kau.ac.ir Received 20 December 2021; Revised 9 February 2022; Accepted 22 February 2022; Published 18 March 2022 Academic Editor: Ci Wei Gao Copyright © 2022 Iman Sayedi et al. ,is 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. ,e distributed energy management of interconnected microgrids, which is based on model predictive control (MPC), relies on the cooperation of all the agents (i.e., microgrids). Model predictive control or MPC is widely used in industrial applications as an effective tool for dealing with multivariable limited control problems. MPC uses an explicit system model to predict the future horizon of the system and its outputs. ,is predictability allows calculating the optimal order of inputs to minimize output errors over a limited horizon, which is effected by the limitations of the system. ,is study presents a distributed economic model predictive control method using the new state feedback controller to control the switching of interface converters and compensate for the unbalanced and nonlinear loads. In this model, the islanding mode and the reconnection of the grid are considered to improve the transient behavior of the system to achieve steady-state power distribution. It has been proposed that it could obtain better results in predictive control, utilizing similarity transform in the state matrix and its modification. First, this model is simulated on distributed generation sources with power-sharing and local loads using the state feedback controller in MATLAB Simpower. ,en, the performance of the proposed method is evaluated, confirming that it is more reliable than the FS-MPC and DSVM-MPC methods. main reason behind the wide range of different dynamic 1. Introduction operations probably relates to various factors, including the A microgrid is a small-scale power grid designed for sup- existence of zeroes outside the stable region, unstable poles, plying electricity to end users. Under certain conditions, and long delays with indefinite and variable times. microgrids have the ability to inject their excess energy into Furthermore, it can overcome the uncertainties of the the main grid [1]. ,e most important challenge when using plant parameters because of load demand fluctuations and microgrids involves maintaining the system’s security and the errors of the implementation. ,e new method has been stability. Most renewable energy-based distributed genera- built based on new simple frequency domain conditions and tion (DG) plants need voltage source inverters (VSIs) to the whale optimization algorithm (WOA). ,is method is connect to the microgrid [2]. Nowadays, choosing the ap- utilized to design a robust proportional-integral-derivative propriate method for controlling industrial processes is (PID) controller based on the WOA in order to enhance the highly important. Moreover, control methods in the in- damping characteristics of the wind energy conversion dustry must have several characteristics, such as ease of use system [4]. New two methods of artificial intelligence (AI) by the operator, simple configuration, and cost-effectiveness techniques are used to design the model predictive con- [3]. Although using proportional-integral-derivative (PID) trollers (MPCs) with superconducting magnetic energy controllers has become common in the industry, it should be storage (SMES) and capacitive energy storage (CES) for load noted that industrial processes involve a wide range of frequency control (LFC) [5]. ,e model predictive control different operations, limiting the use of such controllers. ,e (MPC) algorithm is a method for dealing with such complex 2 International Transactions on Electrical Energy Systems presents a number of important points for determining the industrial processes. Implementing the predictive control method for electric power converters can be difficult due to optimal values of the weighting coefficients. In addition, [10] introduces a predictive control method without using the large computational load required to instantly solve optimization equations. To mitigate this problem, a number weighting coefficients in the induction motor drive. On the of practical solutions have been considered, such as the out- other hand, [11] shows that using discrete space vector of-line calculation of optimization equations and solving modulation in the FS-MPC control method enables these equations by evaluating finite state switching. ,e employing virtual switching states in the control algorithm latter is known as the finite state model predictive control in addition to the real switching states. ,is new technique is (FS-MPC) since it works using a finite set of possible states called the discrete space vector modulation-model predictive control (DSVM-MPC), using which the required sampling for the electric power converter switching. So far, the FS- MPC control method has been used in various applications, frequency is reduced, while the switching frequency of the converter is stabilized [12]. In addition to the advantages of e.g., as rectifier, inverter, motor control, and uninterruptible power supply (UPS). Based on the results of the relevant the FS-MPC method, this control method also provides other benefits, including fixed switching frequency and low studies, this control method has high performance ability for the optimal operation of the whole system, while there is no sampling frequency [13]. However, due to the use of the need to fine-tune the controller parameters. Despite these same algorithm and its discrete nature, it includes a limited capabilities, this method has two major drawbacks that limit and discrete number of converter vector space points. its use in industrial systems. ,e first drawback is that to ,erefore, to account for a wide and variable range of op- achieve high performance, and a high sampling frequency is erating points for the converter, more points are needed, required, leading to costly hardware and the need for high which may lead to slow dynamic performance, low accuracy, and distortion in the output of the converter. ,e economic computation power. Moreover, in this method, the switching frequency is variable, which increases the values of costs for the MPC and the heuristic approach in the con- sidered price scenarios are reported in the economic costs of the converter’s output filter and the volume, weight, and total cost of the converter. When FS-MPC control designs prosumers equipped with production units, energy storage systems, and electric vehicles. To this purpose, the predictive are implemented in laboratory experiments, a large volume of calculations needs to be performed during each sampling control manages the available energy resources by exploiting period, which causes a significant delay in the activation of future information about energy prices, absorption and the actuator signal. ,erefore, if the delay caused by the production power profiles, and electric vehicle (EV) usage, measurement, calculations, and the operation of the actuator such as times of departure and arrival and predicted energy is not considered in the design of the controller, it can cause consumption [14]. In the model predictive control and state poor controller performance. In this regard, [6] describes the feedback controller, the chosen states of the system are compared with their reference quantities to generate the reason for this delay and the method for compensating for it, which creates an interval between the moment the current is converter switching. With the method proposed, in addition to the fact that we have maintained system dynamics, this measured and the moment the new switching state is ap- plied. During this interval, the previous switching state is similarity transform should be used to evaluate all systems. present at the converter’s output, which causes a difference Hence, the following state transform matrix is considered for between the load current and the reference current, in- this system, and better results in terms of overshoot, rise creasing the current’s ripple. To solve this problem, [6] uses time, settling time, and steady state have been obtained with the prediction of the next two samples instead of the pre- this method, which are shown in the results of this study. diction of the next sample. Moreover, [7] proposes a new ,is study proposes a state feedback control method and predictive control method, called fast predictive control. In discusses the optimal design using the iteration method to this method, the volume of the calculations will be signif- define all possible switching states along the predictor’s horizon for the controller, thus mitigating the drawbacks of icantly reduced, and it can be used in multilevel converters that have a large number of control vectors. Another dis- the previous methods. ,erefore, finding an economically applicable model with the best state-space model is very advantage of the FS-MPC method is the variable switching frequency [8]. ,e variable switching frequency creates a important. A comprehensive review of DC microgrids can wide range of harmonics at the output of the converter, be found in [15, 16]. which can cause resonance and make filter design more difficult. In [9], the average switching frequency is kept 2. Study of a Microgrid Containing a DG Source constant by adding corrective terms to the cost function. In addition, [9] modifies the current predictive control scheme Figure 1 depicts the single-line diagram of a power system in such a way that the switching frequency can become containing two DG sources. ,e microgrid is connected to somewhat independent of the sampling frequency. Con- the main grid at the PCC point. Two sources, i.e., DG1 and figuring and selecting the weighting coefficients in the FS- DG2, are directly connected to the microgrid using the CB-3 MPC method will be a major challenge, which has a sig- and CB-4 circuit breakers, respectively. Both DGs have local nificant impact on the system’s performance. Configuring loads, which may be nonlinear and unbalanced. In addition, these coefficients is more time-consuming than adjusting the the microgrid may have a joint load as well, which is as- parameters of the PI controller in classical current control sumed to be balanced and located at a long distance from the and adjusting the hysteresis bandwidth. Furthermore, [10] DGs. One of the tasks of the DGs is to correct the imbalance International Transactions on Electrical Energy Systems 3 Rs CB–5 Ls Utility Load Vs g P Q G G CB-1 P Q R G2 G2 P Q R D1 G1 G1 v D2 PCC g1 g2 CB-2 CB-3 CB-4 P Q LC LC P Q LC P Q L1 L1 L2 L2 p1 v p2 PCC1 i Balanced L1 P Q 1 1 P Q Common 2 2 2 Load Nonlinear Nonlinear VSC VSC And And Compensator Unbalanced Compensator Unbalanced Local Load 2 Local Load 1 DG-1 DG-2 dc1 2dc Figure 1: ,e single-line diagram for the microgrid and the electricity grid, including two sources of distributed generation [31]. and the nonlinearity of the local load. Several control block of this controller is shown in Figure 2. Here, in order schemes have been introduced in order to ensure the proper to achieve the desired solutions, the value of h is set to operation of the microgrid in the grid connection mode or 0.001. It should be noted that i1, i , and v signals are cf cf the islanding mode [17]. Moreover, several control schemes measurable. have been introduced in order to ensure the proper oper- In [22], the authors present an adaptive optimization ation of the microgrid in the grid connection mode or the method for NMPC, where an extended Kalman filter (EKF) stand-alone mode [18]. A comprehensive review of dc is used for estimating the state variables. In [2], a system is microgrids can be found in [19–21]. investigated where the superheated steam temperature (SST) In the DG grid connection mode, the microgrid shares a is the main variable that must be controlled. To do so, several percentage of its local load with the main power grid, while categories of cascading PIDs have been used to control this the joint load is fully supplied by the main grid. During the variable. In [3], to adapt the model, neural networks are process of islanding, each DG source supplies its own local presented for controlling the model predictive control. In load, while the joint load is shared between the DG sources. [23], a fuzzy method based on Lyapunov fragment functions ,e mix powers drawn by the local loads include P + is used in the model predictive control. In [24], in order to L1 jQ and P + jQ . Moreover, the joint load draws the i make the system smart, generalized predictive control L1 L2 L2 LC current from the grid and it consumes the mix power (GPC) and a neural network model are used. In this method, P + jQ . a nonlinear neural network is used to extract the linear LC LC model of the system. In [25], horizon optimization is utilized Local loads at points PCC1 and PCC2 are connected to the DG sources with voltages of v and v , respectively. for predictive control using a genetic algorithm. In [26], the p1 p2 ,e real and reactive powers supplied by the DGs are prediction of the next two samples is used instead of pre- denoted by P1, Q1, P2, and Q2, respectively. It is assumed dicting the next sample. Moreover, [27] proposes a new that the microgrid is mainly resistive, and at the distribution predictive control method, called fast predictive control. In level, the impedances of the lines are denoted by R and this method, the volume of calculations will be significantly D1 R . reduced, and it can be used in multilevel converters that have D2 ,e main power supply is shown by V , while the feeder a large number of control vectors. In addition, [28] modifies inductance resistors are denoted by RS and LS, respectively. the current predictive control scheme in such a way that the ,e main power grid injects PG and QG powers into the switching frequency can become somewhat independent microgrid, and PS-PG and QS-QG powers supply the main from the sampling frequency. Increasing the sampling fre- quency improves the performance of FS-MPC, while re- load of the power grid.CB1 can disconnect the microgrid from the main power grid. ducing the ripple of the output current. However, increasing In the next sections, we will examine the performance the sampling frequency also increases the switching fre- of the system and the compensator in both grid con- quency, which ultimately leads to increased losses [29]. ,is nection and islanding modes. In all the modes mentioned study presents a predictive control method using the state in this section for the grid, a state feedback controller is feedback controller to control the switching interface con- used with a quadratic linear regulator to control the grid verters and compensate for the unbalanced and nonlinear and correct the unbalanced loads. ,e general diagram loads. 4 International Transactions on Electrical Energy Systems i (K) ref Minimization S (k) 3-phases of Cost I (K+1) Predictive Load Function Model Figure 2: ,e diagram of the finite state model predictive control for three-phase inverters [8]. When FS-MPC control schemes are implemented in 3. The Model Predictive Control (MPC) Method laboratory experiments, a large volume of calculations is ,e MPC method is an optimization problem where the cost performed in each sampling period, which causes a sig- function is minimized. Using the system model and the nificant delay in the activation of the actuator signal. values of the variables until time K, the state values are ,erefore, if the delay caused by the measurement, the predicted until time horizon K + N. Moreover, through the calculations, and the activation of the actuator is not con- optimization of the cost function, the first component of the sidered in the controller design, it can cause poor controller command sequence is applied at moment K+1. ,ese steps performance. In this regard, [26] describes the reason for are repeated for the next time steps. ,e cost function in- this delay and how to compensate for it. cludes the control objectives of the system, and its common terms include variables that need to follow a reference value [11]. 3.2. $e DSVM-MPC Controller. ,e main idea behind the According to equation (1), controlling these variables DSVM-MPC control method involves using other points in will be a function of the error between the predicted value the vector space. In the SVM modulation method, in ad- and its reference value. dition to the eight main switching states for the three-phase As can be seen in equation (1), this function can be the converter, other states can also be applied in the form of a size, the square, or the integral over a sampling time interval. linear combination of these base states. ,ese new states are called virtual states. However, because the cost function is k+1 calculated per each vector, it will only be possible to use a ∗ p (1) g � x (t) − x (t)dt. limited number of points in the vector space. In Figure 3, the real points (circles) and virtual points (squares) are shown for the switching of a three-phase converter [21] (Figure 3). 3.1. $e Finite State Model Predictive Control (FS-MPC). ,is control method also offers benefits such as a fixed ,is method uses the discrete nature of electronic power switching frequency and a low sampling frequency. How- converters in such a way that all converter voltage vectors are ever, due to using a similar algorithm and its discrete nature, tested in the cost function, and the vector that minimizes the it covers a limited and discrete number of points in the cost function is selected [20]. converter’s vector space [19]. Figure 2 shows the finite state model predictive control (FS-MPC). ,e algorithm for solving this method includes the following steps [26]: 3.3. $e Proposed DE-MPC Controller. ,e microgrid shown in Figure 1 is considered. It can be observed that when A : Load current measurement; entering the is landing mode and reconnecting the grid, the B : Prediction of the load current in the next samples for system’s response is not highly satisfactory since it takes a all possible switching states according to the following long time for the power distribution to reach a steady state. equation: To solve this problem, we first used a PSO algorithm and PID r × T T L S S control, and the issue was largely mitigated. In this section, (2) i (k + 1) � 1 − i(k) + v (k) − vs . p S i L L we try to design a distributed economic model predictive controller (DEMPC) to improve the transient state of the ,is equation is obtained by discretizing the voltage system. equation. ,e model predictive controller method is designed and C: ,e cost function is evaluated for each prediction. implemented based on the following three steps: In this method, the cost function is expressed as the (1) A model is used to predict the behavior of the control error between the reference current and the predicted variables for the next time step. current for each of the possible switching states ((3)). (2) A cost function is determined, including control objectives and the expected behavior of the g[n] � i − i S + i − i S . (3) αref αp i βref βp i system. D: ,e switching state that minimizes the cost function (3) ,e appropriate command is extracted by mini- is selected. mizing the value of the cost function. International Transactions on Electrical Energy Systems 5 V 1 Converter control region V (1,0,0) Converter real state V (1,0.5,0) Converter virtual state V (0.75,0,0) V (0.4,0,0) Figure 3: ,e vector space for switching of a power converter [21]. moment K + 2. ,ese types of calculations are called re- ,e model used for the prediction is a discrete-time model that can be represented in the form of state equations ceding horizon strategies. Figure 4 shows the performance of according to equations (4) and (5): the DE-MPC method. Using the system model and the values of the variables until time K, the state values until time x(k + 1) � Ax(k) + Bu(k), (4) horizon K + N are predicted. In addition, by optimizing the cost function, the first component of the command sequence y(k) � Cx(k) + Du(k). (5) is applied at moment K + 1. ,ese steps are repeated for the next time steps. In these equations, the vector x(k) denotes the cur- In this method, the number of switching states is defined rent values of the state variables, x(k + 1) is the future to control the microgrid, and based on this number and the prediction value for the state variables, u(k) denotes the proposed cost function, the switching operation is per- current values of the input variables, and y(k) is the formed. Furthermore, Figure 5 shows the state feedback vector of the current output values. In the next step, the controller diagram. Moreover, the proposed controller cost function must be determined. According to equation equations are explained in Section 3.4. (6), in this function, the reference values, values of the Among constraints that exist in this system, we could future states, and the future control commands are mention constraints in switching control S, which is con- considered. sidered in the equation below. ,e control signal, which is J � f(x(k), u(k), ....., u(k + N)). (6) defined by predictive control, is continuous. ,e variable S is denoted that to u signal. As it is mentioned, this signal takes It is noted that the constraints of the state equations are one of values +1 or −1. ,e principle of switching is pre- considered as follows. sented below: S.t: u ≤ u(k + i) ≤ u ; i � 0,. . ., N -1. min max u ∆u ≤ ∆u(k + i) ≤ ∆u ; i � 0,. . ., N -1. If u (k)> h then u � +1, min max u c (7) Y (k + i) ≤ y(k + i) ≤ y (k + i); i � 1,. . ., N min max p else If u (k)< − h then u � −1, ,e quantities of the prediction horizon are usually considered to be twofold the control horizon. ,at is in the in which h is a very small number. Choosing the h value simulation, N � 10 and N � 5 should be considered in this p u determines switching frequency, in a way that reference regard. values are tracked. After several sequences, when we ,e DE-MPC control method is an optimization achieved a more accurate model, we could choose this h problem in which the cost function is minimized. In this value even smaller. optimization, the system model and control objectives are ,e single-phase equivalent circuit of converted is shown considered for K+1 to K + N time steps. ,e result of this in Figure 6. Using this figure and in the presence of the LCL optimization is N consecutive commands. ,e first com- filter, the state vector in conventional methods is considered ponent of this command sequence is applied at moment as below: K + 1. Similarly, during this time, using the new measure- ment values, the optimization is performed for the next i i v x � . (8) f 1 cf moment, and the appropriate command is selected for the 6 International Transactions on Electrical Energy Systems Switching U (K-1) Sequence Z (t) є R U (t) є R S (K ) opt ref PI MPC Microgrid Algorithm - x (t) єR V V i i cf c cf l Figure 4: Principles of the operation of the predictive method. x = Λx + Γ uc + Γ v 1 2 PCC PCC i P ,Q 1 1 1 1ref v cf cf cf cfref K R i f f cf f C v + v cfref cf _ _ u,v dc1 Figure 5: ,e state feedback controller diagram block. P ,Q 1 1 L L f 1 cf C v PCC uV f cf dc Figure 6: Single-phase equivalent circuit VSC (LCL filter) [32]. Figure 6 shows the block diagram of the implemented state variables, it will be obtained the following descriptions control method. Using Figure 6 and considering selected for system state space: −R −1 ⎡ ⎢ ⎤ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ V ⎢ L L ⎥ ⎢ ⎥ ⎢ f f ⎥ dc . ⎢ ⎥ 0 ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎡ ⎢ ⎤ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ i ⎢ ⎥ i ⎢ ⎥ ⎡ ⎢ ⎤ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ f ⎢ ⎥ ⎢ L ⎥ ⎢ ⎥ ⎢ ⎥ f ⎢ ⎥ ⎢ ⎥ ⎥ ⎢ f ⎥ ⎡ ⎢ ⎤ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎡ ⎢ ⎤ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ 1 ⎥ ⎢ . ⎥ ⎢ 1 ⎥ ⎥ ⎥ ⎢ ⎥ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ • ⎢ ⎥ ⎢ ⎥⎢ ⎢ ⎥ ⎢ ⎥ ⎢ ⎢ ⎥⎢ ⎥ ⎢ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ − ⎥ ⎢ ⎥ ⎢ ⎥⎢ i ⎥ ⎢ ⎥ ⎢ ⎥ x � ⎢ ⎥ � ⎢ 0 0 ⎥⎢ ⎥ + ⎢ ⎥u + ⎢ ⎥v . (9) ⎢ ⎥ ⎢ ⎥⎢ 1 ⎥ ⎢ ⎥ ⎢ ⎥ 1 ⎢ i ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ PCC ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 1 ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ L ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ 0 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ L ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ 1 ⎥ ⎢ ⎥ 1 ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎣ ⎦ ⎢ ⎥ ⎢ ⎢ ⎥ ⎢ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎣ ⎦ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ . ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎣ ⎦ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ v ⎣ ⎦ ⎢ ⎥ ⎢ ⎥ v ⎢ ⎥ cf ⎢ ⎥ ⎢ ⎥ cf ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ 0 ⎢ ⎥ ⎢ 1 −1 ⎥ ⎢ ⎥ 0 ⎣ ⎦ C C f f International Transactions on Electrical Energy Systems 7 −1 Hence, the following state vectors are given that by x _ � C AC x + C Bu + C Cv � Λx + Γ u + Γ v . p p c p pcc 1 c 2 PCC equation (13). (13) dc ,e control rule is defined as follows: ⎢ ⎤ ⎥ ⎡ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ L ⎥ ⎢ ⎥ ⎢ f ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ u (k) � −Kx (k) − x (k). (14) ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ c 1 ref ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ B � ⎢ ⎥, ⎢ ⎥ 1 ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 0 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ In the above equation, K is a gain matrix, and x is the ⎢ ⎥ ⎢ ⎥ ref ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ reference vector. ,e gain matrix can be obtained using the ⎣ ⎦ DE-MPC control method and the proposed switching method. ,e control rule of equation (13) includes switching ⎢ ⎥ ⎡ ⎢ ⎤ ⎥ ⎢ ⎥ ⎢ ⎥ control, which is discussed in detail below. ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ Assuming complete control over u, a quadratic optimal ⎢ ⎥ ⎢ 1 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ − ⎥, B � ⎢ ⎥ ⎢ ⎥ linear steady state can be designed for this problem. As noted 2 ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ L ⎥ ⎢ ⎥ ⎢ ⎥ 1 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ (10) earlier, the control rule is as follows: ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎣ ⎦ 0 u � −Kx (k) − x (k). (15) 1 ref −R ,e control law discussed so far is for the system in −1 ⎢ 0 ⎥ ⎡ ⎢ ⎤ ⎥ ⎢ ⎥ ⎢ ⎥ ⎥ which the DGs have an output inductor. Alternatively, when ⎢ ⎥ ⎢ ⎥ ⎢ L L ⎢ ⎥ ⎢ f f ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ the DGs do not have an output inductance, the inductance ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ L1 is removed and the output filter is a simple LC filter. ,e ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 1 ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ system state is then modified follows: ⎢ ⎥ ⎢ ⎥ A � ⎢ 0 0 ⎥. ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ L ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 1 ⎥ T ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ x � i · v . (16) ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ 2 cf cf ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ∗ ⎢ 1 −1 ⎥ ⎢ ⎥ ⎢ ⎥ ⎣ ⎦ With respect to Figure 6, the reference for v v and given cf ∗ ∗ C C V and Ø , the current phasor through the capacitor C is f f f given by equation (14). In the following, it has been proposed that it could be ∗ ∗ I � wC V∗∠