Microgrid energy storage multi-agent distributed secondary control method based on MFAPC

Abstract

The invention relates to a micro-grid energy storage multi-agent distributed secondary control method based on MFAPC, and the method comprises the following steps: 1, obtaining a dynamic estimation value of a global state through a multi-agent consistency protocol based on a communication structure of a sparse communication network; step 2, adopting an adaptive power balance algorithm to realize SoC balance of the distributed ESS with different capacities; and step 3, accurately tracking expected voltage and power of the system by adopting non-linear single-input multi-output ESS model-free adaptive prediction control based on tight-format dynamic linearization, realizing stable operation of the microgrid and SoC balance of the ESS under the condition of not depending on microgrid topology, line resistance impedance and load requirements, and improving robustness of the system. 4, through analysis, the stability of system output tracking errors and global bounded input and bounded output of the controller is guaranteed; the method has the advantages that a model-free adaptive prediction control method is adopted, the prediction order is large, the accuracy is high, and the robustness is high.

Description

Technical Field

[0001] The present invention belongs to the technical field of microgrid control, and particularly relates to a multi-agent distributed secondary control method for microgrid energy storage based on MFAPC. Background Art

[0002] In recent years, with the increasing energy demand, the depletion of fossil fuels, and the increasingly serious environmental problems, renewable energy mainly composed of clean energy such as photovoltaic and wind power has integrated into the development of modern power grids with its green and efficient advantages and has become an important part of the power grid; a microgrid integrates distributed power sources, energy storage systems (ESSs), and loads, which is an effective way to locally consume renewable energy and reduce carbon emissions. At the same time, the characteristics of low construction cost and high operation efficiency of the microgrid have also attracted more extensive attention; however, due to the intermittency and instability of renewable energy, which will seriously affect the power quality of the system, an energy storage system with bidirectional energy flow characteristics is usually adopted in the microgrid to suppress the fluctuations of the output power of renewable energy. However, for the unbalanced state of charge (SoC), the ESS may be overcharged / overdischarged, affecting the service life of the battery. Therefore, it is necessary to design an appropriate coordinated control strategy to ensure proper power sharing among ESSs and achieve SoC balance;

[0003] For an islanded microgrid, traditional droop control can only distribute load power or current to each ESSs in a fixed ratio. However, in an actual microgrid, the situation of line impedance mismatch inevitably exists, making it difficult to achieve SoC balance of ESSs. To overcome the limitations of traditional droop control, microgrid hierarchical control is proposed. Among them, secondary control is used to achieve bus voltage restoration and reasonable power distribution, which can be divided into centralized control and distributed control. The centralized control method collects information of all distributed units and transmits the data to the central controller for optimization calculation in real time. However, the information involved in this control method is centralized and the calculation amount is large, seriously affecting the scalability and plug-and-play ability of the microgrid system; distributed secondary control obtains state information through data exchange between neighbors in a sparse communication network, reduces the dependence on the communication network, and has received extensive attention for improving the reliability and flexibility of the microgrid;

[0004] In recent years, multi-agent systems have been widely applied to distributed ESSs. In multi-agent control strategies, local agents only need to transmit information to their neighbors through a sparse communication network, and without a central controller collecting all information, the sharing of global information and the distributed coordinated control of each agent can be achieved. Existing secondary control of microgrids is usually based on PI controllers. However, the PI controllers of the existing technologies still require model information of the MG to design controller parameters, and there are problems such as large overshoot and difficulty in dealing with system disturbances. Therefore, it is very necessary to provide a multi-agent distributed secondary control method for microgrid energy storage based on MFAPC that adopts a model-free adaptive prediction control method, has a large prediction order, high accuracy, and strong robustness. Summary of the Invention

[0005] (1) Technical Problems

[0006] In view of the above-mentioned current situation of the existing technologies, the present application mainly addresses the following technical problems:

[0007] 1. For the energy storage system adopted in the existing technology in a microgrid, for an unbalanced state of charge, the ESS may experience overcharging / overdischarging, which affects the service life of the battery.

[0008] 2. The information involved in the centralized secondary control method in the existing technology is concentrated and the computational load is large, seriously affecting the scalability and plug-and-play ability of the microgrid system.

[0009] 3. For the secondary control of a microgrid based on a PI controller in the existing technology, model information of the MG is still required to design controller parameters, and there are problems such as large overshoot and difficulty in dealing with system disturbances.

[0010] (2) Technical Solutions

[0011] The purpose of the present invention is to overcome the deficiencies of the existing technologies, and to provide a multi-agent distributed secondary control method for microgrid energy storage based on MFAPC that adopts a model-free adaptive prediction control method, has a large prediction order, high accuracy, and strong robustness.

[0012] The purpose of the present invention is achieved as follows: A multi-agent distributed secondary control method for microgrid energy storage based on MFAPC, the method comprising the following steps:

[0013] Step 1: Based on the communication structure of a sparse communication network, obtain a dynamic estimate of the global state through a multi-agent consensus protocol (MADCP);

[0014] Step 2: Implement the SoC balance of distributed ESSs with different capacities by adopting an adaptive power balance algorithm;

[0015] Step 3: Adopt model-free adaptive predictive control (MFAPC) of a nonlinear single-input multi-output (SIMO) ESS based on compact format dynamic linearization (CFDL) to accurately track the desired voltage and power of the system, and achieve the stable operation of the microgrid and the SoC balance of the ESS without relying on the microgrid topology, line resistance impedance, and load demand, thereby improving the robustness of the system.

[0016] Step 4: Through analysis, ensure the system output tracking error and the globally bounded input-to-bounded output (BIBO) stability of the controller.

[0017] Furthermore, when obtaining the dynamic estimated value of the global state in Step 1, it is first necessary to establish a secondary control model for the islanded microgrid, specifically: The Coulomb counting method is used to estimate the state of charge of each ESS. where SoC i (0) is the state of charge value of the ESS at the initial moment; C maxi is the maximum capacity of the ESS; i Bi is the output current of the battery; η represents the ratio of the converter output voltage to the input voltage; ignoring the losses of the DC-DC converter, the above formula can be simplified to: where V Bi is the battery terminal voltage; P Bi is the output power of the ESS.

[0018] Furthermore, the multi-agent consensus protocol in Step 1 means that, without relying on a central controller and global communication, through the information exchange of adjacent agents, finally a certain variable of all agents tends to be consistent. Specifically: In the microgrid, the communication network among energy storages can be represented by a directed graph G=(V G ,E G ,A G ). Regarding the energy storage units in the microgrid as agents, they are marked as the point set V G ={v 1 ,v 2 ,...,v n}; The edge set represents the transmission link between agents; when (v i ,v j )∈E i , it indicates that agent i can receive the information of its neighbor agent j; The set of all adjacent nodes of node v i is represented by N i ={i∈V G (j,i)∈E G}; A G =|a ij |∈R M×M represents the adjacency matrix related to the directed graph G. When node vj When sending a message to node v i a = 1, conversely a ij = 0; D ij = diagd G = diagd ij | represents the degree matrix, and the Laplacian matrix is represented by L G = D G - A G and

[0019] Furthermore, in step 1, an adjacent communication network is established based on MADCP, and all ESSs achieve consistency through a protocol that depends only on local information and information from adjacent ESSs, realizing local estimation of global variables; the multi-agent dynamic consistency protocol can be expressed as: where x i (t) is the global variable output state of the i-th agent; and are the estimates of the global variable by agent i and agent j respectively; any measurement value x i change of agent i will directly affect the local estimated value while the change of the estimated value indirectly affects the estimates of other agents through the communication network, thus realizing the update of the estimated value for the entire network;

[0020] Then, the dynamic estimated value of the global state is obtained through MADCP for voltage regulation, power distribution, and SoC balancing of ESSs; the output data in MADCP is the local estimated value of the global state, including the estimated global output voltage v i0 , the global unit battery power reference and the global SoC reference SoC ref . Each ESS in the system receives its current estimate from adjacent ESSs and adaptively updates its local estimate using MADCP, thereby dynamically obtaining a consistent global state estimate.

[0021] Furthermore, the adaptive power balance algorithm in step 2 is specifically as follows: a model-free adaptive predictive controller is used to achieve voltage regulation and power distribution of ESSs, and the secondary control signal u i is added to the droop control expression V refi = V nom - k i P i as follows: V refi = V nom - k i P i + u i(9) Adjust the power distribution of each ESS according to the SoC information, based on the unit battery power reference value Design the adaptive compensation power for achieving the accurate distribution of the load power; the battery with a larger SoC is positive, and the battery with a smaller SoC is negative; the adaptive compensation power is defined as: In the formula, n is the acceleration factor; from the output power expression of the DC-DC converter it can be seen that the difference between SoC i and SoC ave determines the positive and negative of . When the SoC of the energy storage system is balanced, the compensation power is zero, and the adaptive power balance algorithm can be expressed as: In the formula, is the adjusted unit battery power reference value.

[0022] Furthermore, the non-linear single-input multi-output ESS in step 3 is specifically: in the microgrid, the working mode of the energy storage system satisfies the non-linear discrete-time system of SIMO, and can be written in the dynamic form of the SIMO non-linear system in the following general form: y(k + 1) = f(y(k), y(k - 1),..., y(k - n y ), u(k), u(k - 1),..., u(k - n u )) (12), where is the output of the system at time k; u(k) ∈ R is the input of the system at time k; n y and n u are both positive integers; is a known non-linear function.

[0023] Furthermore, the model-free adaptive predictive control scheme for the microgrid energy storage system in step 3 consists of a control law, a PJM estimation algorithm, a PJM prediction algorithm, and a reset mechanism; among them, the control law is specifically: for the SIMO non-linear discrete system, combine the characteristics of multi-step predictive control and model-free adaptive control, and construct a model-free adaptive predictive controller based on the CFDL data model, considering the following control input criterion function: In the formula, y * (k + i) = [V * (k + 1), i * (k + 1)] T ∈R 2 is the expected output of the system at time k + i, i = 1,..., N; λ > 0 is the weight coefficient;

[0024] Let Then the control input criterion function can be transformed into: Substitute equation into equation (19), and differentiate the resulting equation pair and set it equal to zero. After rearrangement, it can be obtained that Simplifying the matrix inversion operation in equation (20) gives Therefore, the control input at the current moment is: where

[0025] Furthermore, in the control algorithm in step 3, A 1 (k) and the time-varying pseudo-Jacobi parameter matrices Φ(k), Φ(k + 1),..., Φ(k + N u - 1) will be estimated by the PJM estimation algorithm and the PJM prediction algorithm; among them, the PJM estimation algorithm is specifically: constructing an estimation algorithm for the pseudo-Jacobi matrix, estimating Φ(k) using an improved projection algorithm, and setting the estimation criterion function for this time-varying parameter as follows: where is the estimated value of Φ(k), and μ > 0 is the weight coefficient;

[0026] Take the partial derivative of Φ(k) in equation (23) and set it to zero, that is, J′(Φ(k)) = 0, and using the matrix inversion lemma, the following PJM estimation algorithm can be obtained: where η ∈ (0, 2] is the step size coefficient; μ is the penalty factor.

[0027] Furthermore, the PJM prediction algorithm is specifically: the estimation sequence can be represented by an autoregressive model: where θ i (k) ∈ R m×m , i = 1,..., n p , n p is the appropriate order; therefore, the prediction algorithm is expressed as: Let respectively. For the parameter determination of θ(k), the following criterion function will be used to calculate θ(k): Taking the derivative of the above formula to obtain the optimal value, the following prediction algorithm can be obtained: where δ is the weighting constant, and δ ∈ [0, 1].

[0028] Furthermore, the reset mechanism specifically includes: the reset mechanism of the PJM estimation algorithm and the reset mechanism of the PJM prediction algorithm; among them, the reset mechanism of the PJM estimation algorithm is specifically: If or or If or Adopt a multi-layer recursive prediction algorithm, based on the existing parameters at time k Make predictions on the estimated values of Φ(k + 1),..., Φ(k + N u - 1);

[0029] The reset mechanism of the PJM prediction algorithm is specifically as follows: If or or If or

[0030] (III) Beneficial effects

[0031] 1. The present invention obtains the dynamic estimated value of the global state through the multi-agent consensus protocol, which is used for the voltage regulation, power distribution, and SoC balancing of the ESS;

[0032] 2. The secondary control strategy of the present invention can effectively eliminate the influence of line impedance on SoC balancing, and adopts an adaptive power balance algorithm to achieve SoC balancing of distributed ESSs with different capacities;

[0033] 3. The model-free adaptive predictive control scheme for the microgrid energy storage system of the present invention consists of a control law, a PJM estimation algorithm, a PJM prediction algorithm, and a reset mechanism, accurately tracks the desired voltage and power of the system, and realizes the stable operation of the microgrid and the SoC balancing of the ESS without relying on the microgrid topology, line impedance, and load demand, improving the robustness of the system

[0034] 4. Through strict theoretical analysis, it ensures the global bounded-input bounded-output stability of the system output tracking error and the controller. Description of the drawings

[0035] Figure 1 is a typical structure diagram of the islanded microgrid of the present invention.

[0036] Figure 2 is a simplified parallel model diagram of two groups of energy storage units of the present invention.

[0037] Figure 3 is a schematic diagram of the hierarchical control of the microgrid of the present invention.

[0038] Figure 4 is a model-free adaptive predictive control framework diagram of the present invention.

[0039] Figure 5 is a schematic diagram of the performance (SoC of each ESS) under the MFAPC control of the present invention.

[0040] Figure 6 Schematic diagram of the performance (output power of each ESS) under the MFAPC control of the present invention.

[0041] Figure 7 Schematic diagram of the performance (bus voltage) under the MFAPC control of the present invention.

[0042] Figure 8 Schematic diagram of the performance (SoC of each ESS) under the MFAPC control during communication failure of the present invention.

[0043] Figure 9 Schematic diagram of the performance (output power of each ESS) under the MFAPC control during communication failure of the present invention.

[0044] Figure 10 Schematic diagram of the performance (bus voltage) under the MFAPC control during communication failure of the present invention.

[0045] Figure 11 Schematic diagram of the performance (SoC of each ESS) under the PI control of the present invention.

[0046] Figure 12 Schematic diagram of the performance (output power of each ESS) under the PI control of the present invention.

[0047] Figure 13 Schematic diagram of the performance (bus voltage) under the PI control of the present invention Detailed implementation manners

[0048] The present invention will be further described below in conjunction with embodiments and / or drawings.

[0049] Embodiment 1

[0050] As Figures 1-13 shown, the multi-agent distributed secondary control method for microgrid energy storage based on MFAPC includes the following steps:

[0051] Step 1: Based on the communication structure of the sparse communication network, obtain the dynamic estimate of the global state through the multi-agent consensus protocol (MADCP);

[0052] In the present invention, ① Establishment of the secondary control model for the islanded microgrid: The typical structure of the islanded microgrid is as Figure 1As shown in the figure, it mainly includes a photovoltaic unit, an energy storage unit, and a load. The photovoltaic unit operates in the maximum power tracking mode by controlling the boost converter; each energy storage unit is connected to the microgrid system through a bidirectional DC / DC converter, which is usually used to ensure the dynamic energy balance and bus voltage stability of the system; however, due to problems such as line impedance mismatch and local load imbalance, traditional droop control methods are difficult to meet the requirements of precise system distribution. To enable the ESS to adaptively eliminate the output power difference according to the difference between its own and the global SoC, a more perfect secondary control strategy is needed to provide support.

[0053] During the charging and discharging process of the microgrid ESS, the SOC is an important indicator to measure the available capacity of the ESS. The Coulomb counting method is usually used to estimate the state of charge of each ESS. Among them, SoC i (0) is the state of charge value of the ESS at the initial moment; C maxi is the maximum capacity of the ESS; i Bi is the output current of the battery; η represents the ratio of the output voltage to the input voltage of the converter.

[0054] Ignoring the losses of the DC-DC converter, the above formula can be simplified to: Among them, V Bi is the battery terminal voltage; P Bi is the output power of the ESS; it can be seen from the above formula that the SoC adjustment of the ESS is achieved by changing its output power.

[0055] Since there is no problem of frequency and reactive power control in the microgrid system, only the bus voltage stability and power balance need to be maintained. In the hierarchical collaborative control system, the primary control layer is usually implemented based on P-V droop control with the bus voltage as the reference. The traditional P-V droop control expression is: V refi =V nom -k i P i (3), where V refi is the reference value of the ESS output voltage; V nom is the rated voltage of the microgrid; k i is the droop coefficient; P i is the output power of the DC / DC converter.

[0056] However, in an actual islanded microgrid, due to the existence of line impedance, it is difficult for traditional droop control methods to achieve precise power distribution, which in turn affects the balance of the energy storage SoC. For the convenience of explanation, two groups of ESSs operating in parallel are taken as an example for analysis, as Figure 2 shown, from Figure 2 it can be seen that the output power of the i-th DC-DC converter is: Taking the derivative of equation (2) gives: Taking ESS i and ESS j as an example, combining equations (3)-(5) can obtain the main factors affecting the SoC difference. As can be seen from equation (6), the SoC difference of different ESSs is mainly determined by the maximum energy storage capacity C max , the line impedance R line and the droop coefficient k i . When the energy storage capacity is relatively fixed, different line impedances will have a greater impact on the SoC balancing effect. Therefore, it is necessary to design a secondary control strategy to eliminate the influence of line impedance on SoC balancing.

[0057] ② Multi-agent dynamic consensus protocol: In a microgrid, the communication network among energy storages can be represented by a directed graph G = (V G , E G , A G ). Regarding the energy storage units in the microgrid as agents, they are marked as the point set V G = {v 1 , v 2 ,..., v n}; the edge set represents the transmission link between agents; when (v i , v j ) ∈ E i , it indicates that agent i can receive the information of its neighbor agent j; the set of all adjacent nodes of node v i is represented by N i = {i ∈ V G | (j, i) ∈ E G}; A G = |a ij | ∈ R M×M represents the adjacency matrix related to the directed graph G. When node v j sends a message to node v i , a ij = 1, otherwise a ij = 0; D G = diag|d ij | represents the degree matrix, and the Laplacian matrix is represented by L G = D G - A G , and

[0058] The Multi-Agent Dynamic Consensus Protocol (MADCP) means that without relying on a central controller and global communication, through the information exchange of adjacent agents, ultimately a certain variable of all agents tends to be consistent; based on the MADCP, the present invention establishes an adjacent communication network, and all ESSs achieve consensus through a protocol that only depends on local information and information from adjacent ESSs, realizing the local estimation of global variables; the Multi-Agent Dynamic Consensus Protocol can be expressed as: where, x i (t) is the global variable output state of the i-th agent; and are the estimations of the global variable by agent i and agent j respectively; in Equation (7), any measurement value x i change of agent i will directly affect the local estimation value while the change of the estimation value indirectly affects the estimations of other agents through the communication network, thereby realizing the update of the estimation value for the entire network.

[0059] When the communication network contains a spanning tree and the Laplacian matrix is balanced, there is: The above equation shows that the estimation values of the global variable by each ESS in the local distributed controller can accurately track the average value of the global variable, and the estimation values of all ESSs are consistent.

[0060] The dynamic estimation value of the global state is obtained through the MADCP for the voltage regulation, power distribution, and SoC balancing of ESSs; the output data in the MADCP is the local estimation value of the global state, including the estimated global output voltage v i0 , the global unit battery power reference and the global SoC reference SoC ref , where the unit battery power P i pu is equal to the battery output power divided by the charging capacity, that is, P i pu =P bi / C bi ; each ESS in the system receives its current estimation from adjacent ESSs and adaptively updates the local estimation using the MADCP, thereby dynamically obtaining a consistent global state estimation value.

[0061] Step 2: Adopt an adaptive power balance algorithm to achieve the SoC balance of distributed ESSs with different capacities;

[0062] In the present invention, data-driven secondary control: the proposed hierarchical control strategy is as Figure 3As shown in the figure, to improve the performance of the energy storage system and avoid the influence caused by the change of system model parameters, a model-free adaptive predictive controller is designed to achieve voltage regulation and power distribution of the ESS. The designed secondary control signal u i is added to Equation (3) as follows: V refi = V nom - k i P i + u i (9). Adjust the power distribution of each ESS according to the SoC information. Based on the unit battery power reference value design the adaptive compensation power to achieve the accurate distribution of load power; the storage battery with a larger SoC is positive, and the storage battery with a smaller SoC is negative; the adaptive compensation power is defined as: In the formula, n is the acceleration factor, taking 0.5.

[0063] It can be seen from Equation (4) that the difference between SoC i and SoC ave determines the positive and negative of . When the SoC of the energy storage system is balanced, the compensation power is zero. The adaptive power balance algorithm can be expressed as: In the formula, is the adjusted unit battery power reference value.

[0064] Step 3: Adopt the model-free adaptive predictive control (MFAPC) of the nonlinear single-input multi-output (SIMO) ESS based on compact form dynamic linearization (CFDL) to accurately track the desired voltage and power of the system, and achieve the stable operation of the microgrid and the SoC balance of the ESS without relying on the microgrid topology, line resistance impedance and load demand, and improve the robustness of the system.

[0065] In the present invention, in the microgrid, the working mode of the energy storage system satisfies the nonlinear discrete-time system of SIMO, and can be written in the general form of the SIMO nonlinear system dynamic form as follows: y(k + 1) = f(y(k), y(k - 1),..., y(k - n y ), u(k), u(k - 1),..., u(k - n u ))(12). In the formula, is the output of the system at time k; u(k) ∈ R is the input of the system at time k; n y and n u are both positive integers, which change with the scale of the ESS; is an unknown nonlinear function.

[0066] The following two assumptions are given for the system (12):

[0067] Assumption 1: The function f(·) is a smooth and continuous function, that is, the function has continuous partial derivatives with respect to each component of the control input u(k).

[0068] Assumption 2: The SIMO system (12) satisfies the generalized Lipschitz condition, that is, for k ∈ N + and Δu(k) ≠ 0, it satisfies ||Δy(k + 1)|| ≤ b||Δu(k)||, where b > 0 is a constant; ||·|| is the 2-norm.

[0069] In practical applications, the above two assumptions are reasonable for the microgrid energy storage system. Assumption 1 is a typical constraint condition in the control design of general nonlinear systems; Assumption 2 is an energy constraint condition determined by the inherent properties of the microgrid system. When the change in control input energy is limited, the change rate cannot be infinite.

[0070] Based on the above two assumptions of the microgrid energy storage system, the following theorem can be obtained:

[0071] Theorem 1: For the SIMO system (12) that satisfies Assumption 1 and Assumption 2, if Δu(k) ≠ 0 holds for any k, then there exists a time-varying matrix Φ(k) ∈ R called the pseudo-Jacobian matrix (PJM) m×m such that the system equation (12) can be equivalently transformed into the following CFDL data model: Δy(k + 1) = Φ(k)Δu(k) (13), where and for any time k, Φ(k) is bounded.

[0072] Based on the above incremental form of the CFDL data model (13) of the microgrid energy storage system, the following one-step forward output prediction equation can be derived: y(k + 1) = y(k) + Φ(k)Δu(k) (14). Based on the above equation, the N-step forward prediction equation can be further given as follows: The above equation consists of N equations. For the convenience of subsequent control algorithm calculations, the above equation is sorted out, where, Y Nm (k + 1) represents the N-step forward prediction vector of the system output; ΔU Nm (k) represents the control input increment vector; I m×m represents the m×m order identity matrix; 0 represents the m×m order zero matrix.

[0073] Therefore, equation (15) can be abbreviated as: Y Nm (k + 1) = E(k)y(k) + A(k)ΔU Nm(k)(16), if the control input satisfies Δu(k + j - 1) = 0, j > N u , then the prediction equation (16) becomes: where N u represents the control time domain constant;

[0074]

[0075] For the SIMO nonlinear discrete system, the characteristics of multi-step predictive control and model-free adaptive control are combined to design a model-free adaptive predictive controller based on the CFDL data model. Consider the following control input criterion function: In the formula, y * (k + i) = [V * (k + 1), i * (k + 1)] T ∈R 2 is the expected output of the system at time k + i, i = 1,..., N; λ > 0 is the weight coefficient, which is used to limit the change range of the control input.

[0076] Let then the control input criterion function can be transformed into: Substitute equation (17) into equation (19), and differentiate the obtained equation with respect to and set it equal to zero. After rearrangement, it can be obtained that Simplify the matrix inversion operation in equation (20) to obtain Therefore, the control input at the current time is: In the formula,

[0077] In the control algorithm equation (21), A 1 (k) and the time-varying pseudo-Jacobian parameter matrices Φ(k), Φ(k + 1),..., Φ(k + N u - 1) will be estimated by the PJM estimation algorithm and the PJM prediction algorithm for Φ(k) and Φ(k + i).

[0078] First, an estimation algorithm for the pseudo-Jacobian matrix is designed. The improved projection algorithm is used to estimate Φ(k). The estimation criterion function for this time-varying parameter is set as follows: In the formula, is the estimated value of Φ(k), μ > 0 is the weight coefficient, which limits the change of the PJM matrix estimated value within a reasonable range and improves the estimation accuracy.

[0079] Take the partial derivative of Φ(k) in equation (23) and set it equal to zero, that is, J′(Φ(k)) = 0, and use the matrix inversion lemma to obtain the following PJM estimation algorithm where η ∈ (0, 2] is the step coefficient. Introducing η makes the control algorithm more flexible and meets the requirements of system stability analysis; μ is the penalty factor to avoid the denominator being zero.

[0080] In practice, PJM is time-varying. To improve the tracking ability of the PJM estimation algorithm for time-varying parameters in a non-linear time-varying system, a reset mechanism is added to the PJM estimation algorithm. The specific reset methods are shown in Equations (25) and (26). If or or If or Only the estimated value of Φ(k) can be obtained through the PJM estimation algorithm (24), and the parameters Φ(k + 1),..., Φ(k + N u -1) at the (k + 1)-th and subsequent time instants cannot be calculated from the I / O data of the system. Therefore, a PJM prediction algorithm is designed, using a multi-layer recursive prediction algorithm, based on the existing parameters at the k-th time instant to predict the estimated values of Φ(k + 1),..., Φ(k + N u -1).

[0081] The estimated sequence can be represented using an autoregressive model (AR): where θ i (k) ∈ R m×m , i = 1,..., n p , n p is the appropriate order, and its value is usually selected as 2 - 7. Therefore, the prediction algorithm is expressed as: Let, respectively, For the determination of the parameter θ(k), the following criterion function will be used to calculate θ(k). Taking the derivative of the above equation and obtaining the optimal value, the following prediction algorithm can be obtained: where δ is the weighting constant and δ ∈ [0, 1].

[0082] The reset method of the PJM prediction algorithm is as shown in Equations (31) and (32). If or or If or The model-free adaptive prediction control scheme for the microgrid energy storage system consists of the control laws (21)(22), the PJM estimation algorithm (24), the PJM prediction algorithm (28)(30), and the reset mechanism (25)(26)(31)(32). The model-free adaptive prediction control framework diagram is as Figure 4As shown; in practical applications, the structure of the microgrid energy storage system is complex, the coupling between various parameters is strong, and it is greatly affected by the environment during operation. The model-free adaptive predictive control scheme proposed by the present invention does not require the physical model and structural parameters of the controlled system, and only requires the I / O data of the system to update the PJM estimation algorithm, PJM prediction algorithm and control rate in real time, and has the characteristics of large prediction order and high accuracy, and strong robustness at the same time.

[0083] Step 4: Through analysis, ensure the system output tracking error and the globally bounded input-to-bounded output (BIBO) stability of the controller.

[0084] In the present invention, stability analysis: Analyze the output tracking error and stability of the closed-loop microgrid system under the proposed secondary control strategy.

[0085] Assumption 3: The PJM matrix satisfies the diagonal dominance condition, that is b 1 ,b 2 represent two positive constants, and b 2 ≥b 1 (2α + 1)(n - 1), and all the element signs in Φ(k) are fixed.

[0086] Assumption 3 is a description of the input-output coupling relationship in the closed-loop data. For a SIMO nonlinear system with an unknown model, only the system I / O data is known, and the coupling between system variables is described by the diagonal dominant matrix Φ(k), which can reflect the relationship between the control input and the system output.

[0087] Lemma 2: Let A = (α ij ) ∈ C n×n , for any 1 ≤ i ≤ n, define the Gerschgorin disk as Then all the eigenvalues of matrix A satisfy

[0088] Theorem 2: For the discrete-time nonlinear system (16) satisfying Assumption 1, Assumption 2 and Assumption 3, Lemma 2 is based on Assumptions 1 - 3. If for the system (16), when y * (k + 1) = y * = const, and when the CFDL-MFAPC scheme is adopted, there always exists a positive number λ min > 0, such that when λ > λ min , the closed-loop control system can ensure: ① The tracking error of the system is convergent, that is ② The closed-loop system is BIBO stable, that is, the sequences {y(k)} and {u(k)} are bounded.

[0089] The present invention is a multi-agent distributed secondary control method for microgrid energy storage based on MFAPC. During use, the multi-agent dynamic consistency protocol (MADCP) proposed by the present invention can, without relying on a central controller and global communication, through adjacent agent information exchange, finally achieve the convergence of a certain variable of all agents; the present invention establishes an adjacent communication network based on MADCP, and all ESSs achieve consistency through a protocol that only depends on local information and information from adjacent ESSs, realizing the local estimation of global variables. Then, each ESS in the system of the present invention receives its current estimation from adjacent ESSs and adaptively updates the local estimation using MADCP, thereby dynamically obtaining a consistent global state estimation value; the model-free adaptive predictive control scheme of the present invention does not require the physical model and structural parameters of the controlled system, only requires the I / O data of the system to update the PJM estimation algorithm, PJM prediction algorithm, and control rate in real time, and has the characteristics of a large prediction order and high accuracy, and at the same time has strong robustness; the present invention has the advantages of adopting the model-free adaptive predictive control method, a large prediction order, high accuracy, and strong robustness.

[0090] Embodiment 2

[0091] As Figures 1-13 shown, the multi-agent distributed secondary control method for microgrid energy storage based on MFAPC, the method includes the following steps:

[0092] Step 1: Based on the communication structure of the sparse communication network, obtain the dynamic estimation value of the global state through the multi-agent consistency protocol (MADCP);

[0093] Step 2: Adopt the adaptive power balance algorithm to achieve the SoC balance of distributed ESSs with different capacities;

[0094] Step 3: Adopt the model-free adaptive predictive control (MFAPC) of the nonlinear single-input multi-output (SIMO) ESS based on the compact format dynamic linearization (CFDL) to accurately track the desired voltage and power of the system, and achieve the stable operation of the microgrid and the SoC balance of the ESS without relying on the microgrid topology, line resistance impedance, and load demand, and improve the robustness of the system.

[0095] Step 4: Through analysis, ensure the system output tracking error and the globally bounded input bounded output (BIBO) stability of the controller.

[0096] In the present invention, case analysis: ① Case description

[0097] To verify the effectiveness of the proposed distributed secondary control strategy of the present invention, a system simulation model was built based on the OPAL OP5700 platform and real-time tests of three cases were carried out; the islanded microgrid for testing consists of 3 ESSs, one photovoltaic and two loads; the parameters of the microgrid system are shown in Table 1, and the parameters of the MFAPC distributed controller are shown in Table 2.

[0098] Table 1 Microgrid system parameters

[0099]

[0100] Table 2 MFAPC distributed controller parameters

[0101]

[0102]

[0103] The system simulation process of this time includes 4 stages:

[0104] The first stage (0 - 4 s): Only the primary droop control is effective in the system, both Load 1 and Load 2 are connected to the system; the ESSs are in the discharging state

[0105] The second stage (4 - 15 s): When t = 4 s, the proposed secondary control strategy comes into effect;

[0106] The third stage (15 - 30 s): When t = 15 s, Load 1 is disconnected from the microgrid system, and the ESS changes from the discharging state to the charging state;

[0107] The fourth stage (30 - 40 s): When t = 30 s, the light intensity fluctuates. To maintain the stability of the system, the ESS changes from the charging state to the discharging state.

[0108] ② Result analysis

[0109] Case 1 verifies the effectiveness of the proposed MFAPC control strategy during the stable charge and discharge process of the ESS. The simulation results are as Figures 5-7 shown; in the first stage of the primary droop control process, due to the limitations of the traditional droop control and the inconsistency of the line impedance, the SoC starts to deviate from the same initial value, and the bus voltage deviation is large; after the proposed secondary control strategy of the present invention comes into effect in the second stage, the SoC gradually converges and finally reaches the equilibrium state, and the bus voltage quickly recovers; in the third stage and the fourth stage, the ESS has two charge and discharge state switches, Figures 6-7It shows the power distribution of the ESS and the recovery of the bus voltage during the state switching process. The power of the ESS is reasonably distributed according to its capacity, and the maximum voltage deviation of the bus voltage is within the allowable voltage deviation range of the system. Finally, it stabilizes near the rated voltage. The test results show that the secondary control strategy proposed in the present invention has good control effects on SoC balancing, bus voltage recovery, and reasonable power distribution.

[0110] Example 2 verifies the robustness of the proposed MFAPC control strategy against communication failures when the communication network topology changes. The simulation results are as Figures 8-10 shown. A ring communication topology is adopted among the ESSs in the microgrid. In the first and second stages, the communication network topology is normal. When t = 15 s, the communication line 1 - 3 is interrupted, and the communication topology is still in a connected state, and the system does not lose connectivity. The secondary control scheme proposed in the present invention is not significantly disturbed and can still reasonably achieve voltage recovery and power distribution. When t = 30 s, the communication lines 1 - 3 and 2 - 3 fail, and the communication topology is divided into two parts, and ESS3 is separated. However, the communication between ESS1 and ESS2 is maintained at this time, and the secondary control strategy is effective within it and can still achieve SoC balancing. The test results show that the secondary control strategy proposed in the present invention has strong robustness against communication failures.

[0111] Example 3 verifies the control performance of the proposed MFAPC strategy compared with the traditional PI control strategy. The simulation results are as Figures 11-13 shown. By comparing Figures 5-7 and Figures 11-13 it can be seen that the PI control strategy can achieve the balance of SoC, but its tracking performance of the output power of the ESS and the bus voltage is poor. In the third and fourth stages, compared with the MFAPC strategy, the output power and bus voltage of each ESS in the PI control strategy show a slower transient response. The reason is that the PI control algorithm needs to design a controller for each output separately, resulting in improper handling of the system coupling problem. The MFAPC algorithm can decouple the SIMO system by estimating the pseudo-Jacobi parameter PJM, making full use of future input and output information, and has better control performance.

[0112] The present invention is a microgrid energy storage multi-agent distributed secondary control method based on MFAPC. The multi-agent dynamic consistency protocol (MADCP) can, without relying on a central controller and global communication, through the information exchange of adjacent agents, finally achieve the convergence of a certain variable of all agents; an adjacent communication network is established based on MADCP, and all ESSs achieve consistency through a protocol, relying only on local information and information from adjacent ESSs to realize the local estimation of global variables. Each ESS in the system receives its current estimation from adjacent ESSs and adaptively updates the local estimation using MADCP, thereby dynamically obtaining a consistent global state estimation value; the model-free adaptive predictive control scheme does not require the physical model and structural parameters of the controlled system, only needs the I / O data of the system to update the PJM estimation algorithm, PJM prediction algorithm and control rate in real time, and has the characteristics of a large prediction order and high accuracy, and at the same time has strong robustness; the present invention has the advantages of adopting a model-free adaptive predictive control method, a large prediction order, high accuracy and strong robustness.

Claims

1. A multi-agent distributed secondary control method for microgrid energy storage based on MFAPC, characterized by: The method comprises the following steps: Step 1: Based on the communication structure of the sparse communication network, a dynamic estimate of the global state is obtained through a multi-agent consensus protocol; Step 2: Adopt an adaptive power balancing algorithm to achieve SoC balancing of distributed ESSs with different capacities; Step 3: Adopt model-free adaptive predictive control of nonlinear single-input multiple-output ESS based on tight-form dynamic linearization to accurately track the system expected voltage and power, achieve stable operation of the microgrid and SoC balance of the ESS without relying on the microgrid topology, line impedance and load demand, and improve the robustness of the system. Step 4: Through analysis, ensure the global bounded input-bounded output stability of the system output tracking error and the controller.

2. The microgrid energy storage multi-agent distributed secondary control method based on MFAPC as claimed in claim 1, characterized in that: When obtaining the dynamic estimated value of the global state in step 1, it is necessary to first establish a secondary control model of the isolated microgrid, specifically: using the Coulomb counting method to estimate the charge state of each ESS, Among them, SoC i (0) is the charge state value of ESS at the initial moment; C maxi is the maximum capacity of ESS; i Bi is the battery output current; η represents the ratio of the converter output voltage to the input voltage; ignoring the loss of the DC-DC converter, the above formula can be simplified to: Among them, V Bi is the battery terminal voltage; P Bi is the ESS output power.

3. The microgrid energy storage multi-agent distributed secondary control method based on MFAPC as claimed in claim 2, characterized in that: The multi-agent consensus protocol in step 1 refers to the process of achieving consistency of a certain variable of all agents through information exchange between adjacent agents without relying on a central controller and global communication. Specifically, in the microgrid, the communication network between energy storages can be represented by a directed graph G=(V G ,E G ,A G ) indicates that the energy storage unit in the microgrid is regarded as an intelligent agent, marked as a point set V G ={v1,v2,...,v n }; Edge set represents the transmission link between agents; when (v i ,v j )∈E i , indicating that agent i can receive information from its neighbor agent j; node v i The set of all adjacent nodes of is N i ={i∈V G (j,i)∈E G } indicates; A G =|a ij |∈R M×M Represents the adjacency matrix associated with the directed graph G. When the node v j To node v i When delivering a message ij =1, otherwise a ij =0;D G =diagd ij | represents the degree matrix, and the Laplacian matrix is ​​represented by L G =D G -A G indicates that, and 4. The microgrid energy storage multi-agent distributed secondary control method based on MFAPC as claimed in claim 3, characterized in that: The step 1 establishes a neighboring communication network based on MADCP. All ESSs achieve consistency through a protocol that only relies on local information and information from neighboring ESSs to achieve local estimation of global variables. The multi-agent dynamic consistency protocol can be expressed as: Among them, x i (t) is the global variable output state of the ith agent; and are the estimates of the global variable by agent i and agent j respectively; any measurement value x of agent i i Changes will directly affect the local estimate and Changes in the estimated value indirectly affect the estimates of other agents through the communication network, thereby enabling the entire network to update the estimated value; Then, MADCP is used to obtain dynamic estimates of the global state for ESS voltage regulation, power allocation, and SoC balancing. The output data in MADCP is a local estimate of the global state, including the estimated global output voltage v i0 , Global unit battery power reference and global SoC reference SoC ref ,Each ESS in the system receives its current estimate from neighboring ESSs and adaptively updates the local estimate using MADCP, thereby dynamically obtaining a consistent global state estimate.

5. The microgrid energy storage multi-agent distributed secondary control method based on MFAPC as claimed in claim 1, characterized in that: The adaptive power balancing algorithm in step 2 is specifically: using a model-free adaptive predictive controller to implement voltage regulation and power allocation of the ESS, and converting the secondary control signal u i Add the droop control expression V refi =V nom -k i P i The following is the case: V refi =V nom -k i P i +u i (9) Adjust the power allocation of each ESS according to the SoC information, based on the unit battery power reference value Design adaptive compensation power Used to achieve accurate distribution of load power; batteries with larger SoC Positive, SoC smaller battery Negative; adaptive compensation power Defined as: Where n is the acceleration factor; The output power expression of DC-DC converter is It can be seen that SoC i and SoC ave The difference determines When the energy storage system SoC is balanced, the compensation power is zero, and the adaptive power balancing algorithm can be expressed as: In the formula, It is the adjusted reference value of unit battery power.

6. The microgrid energy storage multi-agent distributed secondary control method based on MFAPC as claimed in claim 1, characterized in that: The nonlinear single-input multiple-output ESS in step 3 is specifically: in the microgrid, the working mode of the energy storage system satisfies the nonlinear discrete-time system of SIMO, which can be written as the following general form of the SIMO nonlinear system dynamic form: y(k+1)=f(y(k),y(k-1),...,y(kn y ),u(k),u(k-1),...,u(kn u ))(12), where is the output of the system at time k; u(k)∈R is the input of the system at time k; n y and n u All are positive integers; is a known nonlinear function.

7. The microgrid energy storage multi-agent distributed secondary control method based on MFAPC as claimed in claim 6, characterized in that: The model-free adaptive predictive control scheme for the microgrid energy storage system in step 3 is composed of a control rate, a PJM estimation algorithm, a PJM forecast algorithm, and a reset mechanism; wherein the control rate is specifically: for the SIMO nonlinear discrete system, the characteristics of multi-step predictive control and model-free adaptive control are combined to construct a model-free adaptive predictive controller based on the CFDL data model, considering the following control input criterion function: In the formula, y * (k+i)=[V * (k+1),i * (k+1)] T ∈R 2 is the expected output of the system at time k+i, i=1,...,N; λ>0 is the weight coefficient; make Then the control input criterion function can be transformed into: General Substituting into equation (19), the obtained equation for U Nu (k) is differentiated and set equal to zero, and then rearranged to obtain: Simplifying the matrix inverse operation in formula (20), we can get: Therefore, the control input at the current moment is: In the formula, 8. The multi-agent distributed secondary control method for microgrid energy storage based on MFAPC as claimed in claim 7, characterized in that: In the control algorithm in step 3, A1(k) and the unknown time-varying pseudo-Jacobi parameter matrix Φ(k), Φ(k+1), ..., Φ(k+N u -1), Φ(k) and Φ(k+i) are estimated by PJM estimation algorithm and PJM forecast algorithm; the PJM estimation algorithm is specifically: constructing an estimation algorithm of the pseudo-Jacobian matrix, using an improved projection algorithm to estimate Φ(k), and the estimation criterion function of the time-varying parameter is set as follows: In the formula, is the estimated value of Φ(k), μ>0 is the weight coefficient; Taking the partial derivative of Φ(k) in equation (23) and setting it to zero, that is, J′(Φ(k))=0, and using the matrix inversion lemma, we can obtain the PJM estimation algorithm as shown below: Where η∈(0,2] is the step size coefficient; μ is the penalty factor.

9. The microgrid energy storage multi-agent distributed secondary control method based on MFAPC as claimed in claim 8, characterized in that: The PJM forecast algorithm is specifically as follows: the estimated sequence can be represented by an autoregressive model: In the formula, θ i (k)∈R m×m ,i=1,...,n p , n p is of appropriate order; therefore, the prediction algorithm is expressed as: Separately, For the parameter determination of θ(k), the following criterion function will be used to calculate θ(k), Taking the optimal value of the above formula, we can get the following prediction algorithm: Where δ is a weighting constant and δ∈[0,1].

10. The microgrid energy storage multi-agent distributed secondary control method based on MFAPC as claimed in claim 7, characterized in that: The reset mechanism specifically includes: a reset mechanism of the PJM estimation algorithm and a reset mechanism of the PJM forecast algorithm; wherein the reset mechanism of the PJM estimation algorithm specifically includes: Using a multi-layer recursive forecasting algorithm, based on the existing parameters at time k For Φ(k+1),...,Φ(k+N u -1) is used to forecast the estimated value; The reset mechanism of the PJM forecast algorithm is as follows:

Images