Optimal control method for grid-connected retired power battery energy storage system

Abstract

This invention relates to the field of battery energy storage system control technology, specifically an optimized control method for a grid-connected retired power battery energy storage system. The method includes acquiring the voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment; based on these measurements, using an LSTM-MPC control strategy, obtaining the duty cycle of each single-phase submodule at the next moment; and importing this information into a carrier phase-shift modulation algorithm to generate single-phase submodule switching signals. This invention combines LSTM, MPC, and duty cycle modulation to achieve state of charge (SOC) tracking and power tracking for the battery pack under balanced grid voltage conditions. Under unbalanced grid voltage conditions, it considers the three-phase unbalanced operating characteristics of the system to suppress negative sequence current and power fluctuations, obtaining the optimal grid-connected current command and optimal output voltage command for the next moment, providing accurate data support for obtaining the single-phase submodule switching signals.

Description

Technical Field

[0001] This invention relates to the field of battery energy storage system control technology, and specifically to an optimized control method for a grid-connected retired power battery energy storage system. Background Technology

[0002] Currently, to address the challenges posed to power systems by new energy generation and power load fluctuations, constructing energy storage power stations to participate in power system optimization and dispatch has become an inevitable choice. Retired power battery energy storage systems, due to their ability to provide high energy density at relatively low investment costs, have made low-cost energy storage facilities based on the cascade utilization of retired power batteries a research hotspot. However, because the state of retired power batteries may undergo nonlinear changes, traditional energy storage system control strategies cannot be directly applied to retired power battery energy storage systems, making it difficult to ensure their safety during cascade utilization. Therefore, research on control strategies for retired power battery energy storage systems is of great significance.

[0003] The existing retired power battery energy storage system is a modular multilevel hybrid converter-battery energy storage system (MMHC-BESS). It can accurately control the switching state of a single sub-module containing a battery pack, and realize precise control of retired power battery packs in different states. It can be used as the topology of the retired power battery energy storage system, but it cannot achieve system control under multiple objectives. The existing retired power battery energy storage system control strategies include: (1) a multi-objective control strategy applicable to high voltage modular multilevel converter (MMC) technology, which does not consider the control of sub-modules containing battery packs and cannot be directly applied to the MMHC-BESS system; (2) MMC discretized mathematical model and model predictive control (MPC) strategy, which realizes multi-objective control that can be applied to the battery energy storage system, but does not consider the suppression of intra-phase circulating current; (3) an MPC optimized control strategy, which can realize power tracking and suppress power fluctuations and negative sequence current when the grid voltage is unbalanced, but does not consider the state of charge of the battery pack. The system does not track the state of charge (SOC) and does not take into account the difficulty in estimating the state of the battery pack due to the nonlinear change in the SOC value caused by battery aging. Summary of the Invention

[0004] This invention provides an optimized control method for a grid-connected retired power battery energy storage system, which overcomes the shortcomings of the prior art. It can effectively solve the problems of strong model dependence on the controlled object and inability to perform state of charge (SOC) tracking of the battery pack in existing optimized control methods for grid-connected retired power battery energy storage systems.

[0005] The technical solution of this invention is achieved through the following measures: an optimized control method for a grid-connected retired power battery energy storage system, comprising:

[0006] The voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment are collected. The voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment include: the grid-connected current measurement values ​​of the α-axis and β-axis at the current moment, the grid voltage measurement values ​​of the α-axis and β-axis at the current moment, the grid voltage measurement values ​​of the α-axis and β-axis at the current moment, the total output voltage measurement values ​​of the single-phase sub-modules of the α-axis and β-axis at the current moment, and the total output voltage measurement values ​​of the single-phase sub-modules at the current moment.

[0007] Based on the voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment, the duty cycle of each single-phase sub-module at the next moment is obtained by combining the LSTM-MPC control strategy. The LSTM-MPC control strategy includes a voltage and current prediction strategy based on a discretized mathematical model, a SOC prediction strategy based on an LSTM network model, and a current and voltage command optimization strategy based on a cost function.

[0008] The duty cycle of each single-phase submodule is imported into the carrier phase-shift modulation algorithm to generate the single-phase submodule switching signal;

[0009] The above steps are iterated within the set scheduling period to complete feedback correction and achieve optimal control.

[0010] The following are further optimizations and / or improvements to the above-mentioned technical solution:

[0011] The above-mentioned voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment, combined with the LSTM-MPC control strategy, yield the duty cycle of each single-phase sub-module at the next moment, including:

[0012] Based on the voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment, the predicted SOC, output voltage, grid current, and output power of each single-phase sub-module of the grid-connected converter under different switching states at the next moment are obtained using the LSTM network model and the discretized mathematical model.

[0013] In both cases of grid voltage balance and grid voltage imbalance, the minimum cost function is taken as the optimization control objective. The converter switching state with the minimum cost function value is selected as the input state at the next moment. Based on the optimal grid connection command acquisition model, the optimal grid connection current command and the optimal output voltage command of the α axis and β axis at the next moment are obtained.

[0014] The duty cycle of each single-phase submodule is calculated after the optimal grid-connected current command and the optimal output voltage command of the α-axis and β-axis at the next moment are transformed by coordinate transformation. The duty cycle is the ratio of the optimal output voltage command at the next moment to the predicted total output voltage of the single-phase submodule at the next moment.

[0015] Based on the voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment, the predicted SOC, output voltage, grid current, and output power of each single-phase submodule of the grid-connected converter under different switching states at the next moment are obtained using an LSTM network model and a discretized mathematical model, including:

[0016] The voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment are input into the discretized mathematical model set. The discretized mathematical models under all different switching states in the set are traversed to obtain the predicted output voltage and grid current values ​​of each single-phase submodule of the grid-connected converter under different switching states at the next moment. The predicted output power values ​​of each single-phase submodule of the grid-connected converter under different switching states at the next moment are also calculated. The discretized mathematical models include:

[0017] Grid-connected current prediction model:

[0018]

[0019] Among them, i α (k+1) and i β (k+1) represent the predicted grid-connected current values ​​along the α and β axes at time k+1, respectively; i g (k+1) is the predicted grid-connected current at time k+1; u α (k) and u β (k) represents the measured grid voltage values ​​along the α and β axes at time k; u j (k) represents the measured grid voltage at time k; u bat,α (k) and u bat,β (k) represents the total output voltage measurement values ​​of the single-phase submodules along the α and β axes at time k; u bat,j (k) represents the total output voltage measurement of the single-phase submodule at time k; i α (k) and i β(k) represents the grid-connected current measurements along the α and β axes at time k, respectively;

[0020] Output voltage prediction model for single-phase submodule:

[0021]

[0022] Among them, u bat,j (k+1) represents the predicted output voltage of the single-phase submodule at time k+1;

[0023] The predicted output voltage and grid current of each single-phase submodule are input into the LSTM network model to obtain the predicted SOC of each single-phase submodule under different switching states of the grid-connected converter at the next moment. The LSTM network model is obtained by training the LSTM neural network with several samples, which are the charging and discharging voltage and current data of retired power batteries.

[0024] As described above, when the grid voltage is balanced, the minimum cost function is used as the optimization control objective. The converter switching state with the minimum cost function value is selected as the input state for the next moment. Based on the optimal grid connection command acquisition model, the optimal grid connection current command and the optimal output voltage command for the α and β axes for the next moment are obtained. The cost function and the corresponding optimal grid connection command acquisition model are shown below:

[0025] Cost function:

[0026]

[0027] Among them, P ref P(k+1) and Q are the given reference active power and the predicted active power at time k+1, respectively; ref Q(k+1) represents the given reference reactive power and the predicted reactive power at time k+1, respectively. S SOC (k+1) represents the given reference SOC value and the predicted SOC value at time k+1, respectively;

[0028] The optimal grid connection command acquisition model includes:

[0029] Optimal grid-connected current command acquisition model:

[0030]

[0031]

[0032] in, and These are the optimal grid-connected current commands along the α and β axes at time k+1, respectively; P * (k+1) represents the optimal active power prediction value at time k+1; Q* (k+1) represents the optimal reactive power prediction value at time k+1; u α (k) and u β (k) represents the measured grid voltage values ​​along the α and β axes at time k;

[0033] Optimal output voltage command acquisition model:

[0034]

[0035] in, and These are the optimal output voltage commands for the α-axis and β-axis at time k+1, respectively. and These are the optimal grid-connected current commands along the α and β axes at time k+1, respectively; P * (k+1) represents the optimal active power prediction value at time k+1; Q * (k+1) represents the optimal reactive power prediction value at time k+1; u α (k) and u β (k) represents the measured grid voltage values ​​along the α and β axes at time k; T S The sampling period.

[0036] In the above-mentioned case of grid voltage imbalance, the minimum cost function is used as the optimization control objective. The converter switching state with the minimum cost function value is selected as the input state for the next moment. Based on the optimal grid connection command acquisition model, the optimal grid connection current command and the optimal output voltage command for the α and β axes for the next moment are obtained. The cost function and the corresponding optimal grid connection command acquisition model are shown below:

[0037] The cost function is shown below:

[0038]

[0039] Among them, i α- (k+1),i β- (k+1) represent the negative sequence components of the grid-connected current prediction value at time k+1 on the α and β axes, respectively; λ i , λ P , λ Q , λ SOC These are negative sequence current suppression, constant active power operation, constant reactive power operation, and constant SOC operation, respectively, and the sum of the weights of the four control objectives in the cost function is equal to 1.

[0040] The optimal grid connection command acquisition model includes:

[0041] Optimal grid-connected current command acquisition model:

[0042]

[0043] in, The predicted values ​​of the current commands on the α and β axes at time k+1; i α负 (k+1),i β负 (k+1) represents the predicted current command values ​​on the α and β axes at time k+1 under the current balance operation target; i αP (k+1),i βP (k+1) represents the predicted current command values ​​on the α and β axes at time k+1 under the target of constant active power operation; i αQ (k+1),i βQ (k+1) represents the predicted current command values ​​on the α and β axes at time k+1 under the target of constant reactive power operation.

[0044] Optimal output voltage command acquisition model:

[0045]

[0046] in, and These are the optimal output voltage commands for the α-axis and β-axis at time k+1, respectively. and These are the optimal grid-connected current commands along the α and β axes at time k+1, respectively; P * (k+1) represents the optimal active power prediction value at time k+1; Q * (k+1) represents the optimal reactive power prediction value at time k+1; u α (k) and u β (k) represents the measured grid voltage values ​​along the α and β axes at time k; T S The sampling period.

[0047] This invention, based on the topology of a retired power battery energy storage system, combines LSTM (Long Short-Term Memory) neural network, Model Predictive Control (MPC), and duty cycle modulation to disclose an optimized control method for a grid-connected retired power battery energy storage system capable of power point tracking (PPT) and State of Charge (SOC) tracking. The optimal output voltage is derived by combining a discrete mathematical model of the retired power battery energy storage system with a cost function, and then the switching on and off of the corresponding single-phase sub-modules is controlled using a carrier phase-shift modulation algorithm. Addressing the issue that traditional ampere-hour integration methods cannot achieve high-precision estimation of SOC under nonlinear changes due to battery aging in cost function construction, this paper uses real-time acquired voltage and current measurements, along with an LSTM network model and a discretized mathematical model, to predict the SOC, output voltage, grid current, and output power of each phase of the grid-connected converter under different switching states at the next moment. By combining LSTM, MPC, and duty cycle modulation, the paper achieves SOC and power tracking for the battery pack under balanced grid voltage conditions. For unbalanced grid voltage conditions, it considers the three-phase unbalanced operating characteristics of the system to suppress negative sequence current and power fluctuations, accurately obtaining the optimal grid current command and optimal output voltage command for the α and β axes at the next moment, providing accurate data support for obtaining the switching signals of single-phase submodules. Attached Figure Description

[0048] Appendix Figure 1 This is a flowchart of the method of the present invention.

[0049] Appendix Figure 2 This is a control block diagram of the LSTM-MPC control strategy in this invention.

[0050] Appendix Figure 3 This is a topology diagram of the retired power battery energy storage system in this invention.

[0051] Appendix Figure 4 This is the equivalent circuit diagram of a single-phase grid-connected retired power battery energy storage system in this invention.

[0052] Appendix Figure 5 This is a schematic diagram of the training process of the LSTM neural network in this invention.

[0053] Appendix Figure 6 This is an optimized control block diagram combining LSTM-MPC control and carrier phase-shift modulation for a decommissioned power battery energy storage system when the grid voltage is balanced, as described in this invention.

[0054] Appendix Figure 7 This is an optimized control block diagram combining LSTM-MPC control and carrier phase-shift modulation for a decommissioned power battery energy storage system when a grid imbalance fault occurs, as described in this invention.

[0055] Appendix Figure 8This is a graph showing the SOC prediction results of Embodiment 3 of the present invention.

[0056] Appendix Figure 9 This is a simulation waveform change curve of the system power tracking in Embodiment 3 of the present invention.

[0057] Appendix Figure 10 This is a simulation waveform change curve of the system charging and discharging switching when the power grid is balanced in Embodiment 3 of the present invention.

[0058] Appendix Figure 11 This is a waveform change curve of single-phase SOC simulation when the power grid is in balance in Embodiment 3 of the present invention.

[0059] Appendix Figure 12 This is a graph showing the power fluctuation change when the grid voltage is unbalanced in Embodiment 3 of the present invention.

[0060] Appendix Figure 13 This is a graph showing the voltage and current waveform changes when the power grid voltage is unbalanced in Embodiment 3 of the present invention.

[0061] Appendix Figure 14 This is a graph showing the change in negative sequence current suppression when the power grid voltage is unbalanced in Embodiment 3 of the present invention.

[0062] Appendix Figure 15 This is a graph showing the constant active power operation curve when the grid voltage is unbalanced in Embodiment 3 of the present invention.

[0063] Appendix Figure 16 This is a graph showing the constant reactive power operation curve when the grid voltage is unbalanced in Embodiment 3 of the present invention.

[0064] Appendix Figure 17 This is a graph showing the suppression curves of active and reactive power fluctuations when the power grid voltage is unbalanced in Embodiment 3 of the present invention.

[0065] Appendix Figure 18 This is a graph showing the SOC change curve when the grid voltage is unbalanced in Embodiment 3 of the present invention. Detailed Implementation

[0066] The present invention is not limited to the following embodiments, and the specific implementation can be determined according to the technical solution of the present invention and the actual situation.

[0067] The present invention will be further described below with reference to embodiments and accompanying drawings:

[0068] Example 1: As shown in the attached document Figure 1 As shown in the figure, an embodiment of the present invention discloses an optimized control method for a grid-connected retired power battery energy storage system, comprising:

[0069] Step S110: Collect the voltage and current measurement values ​​of the grid-connected retired power battery energy storage system at the current moment. The voltage and current measurement values ​​of the grid-connected retired power battery energy storage system at the current moment include: the grid-connected current measurement values ​​of the α-axis and β-axis at the current moment, the grid voltage measurement values ​​of the α-axis and β-axis at the current moment, the grid voltage measurement values ​​of the α-axis and β-axis at the current moment, the total output voltage measurement values ​​of the single-phase sub-modules of the α-axis and β-axis at the current moment, and the total output voltage measurement values ​​of the single-phase sub-modules at the current moment.

[0070] Step S120: Based on the voltage and current measurement values ​​of the grid-connected retired power battery energy storage system at the current moment, the duty cycle of each single-phase sub-module at the next moment is obtained by combining the LSTM-MPC control strategy. The LSTM-MPC control strategy includes a voltage and current prediction strategy based on a discretized mathematical model, a SOC prediction strategy based on an LSTM network model, and a current and voltage command optimization strategy based on a cost function.

[0071] Step S130: The duty cycle of each phase is imported into the carrier phase-shift modulation algorithm to generate the single-phase submodule switching signal;

[0072] Step S140: Iterate the above steps within the set scheduling period to complete feedback correction and achieve optimal control.

[0073] Example 2: This embodiment of the invention discloses an optimized control method for a grid-connected retired power battery energy storage system, including:

[0074] Step S210: Collect the voltage and current measurements of the grid-connected retired power battery energy storage system at the current time k. The voltage and current measurements of the grid-connected retired power battery energy storage system at the current time include: the grid-connected current measurements i along the α-axis and β-axis at the current time k. α (k) and i β (k) The grid-connected current measurement value i at time k. g (k) The measured grid voltage values ​​u along the α and β axes at time k. α (k) and u β (k) The measured total output voltage u of the single-phase submodules along the α and β axes at time k. bat,α (k) and u bat,β (k), the measured total output voltage u of the single-phase submodule at time k. bat,j (k).

[0075] Step S220: Based on the voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment, the duty cycle of each single-phase sub-module at the next moment is obtained by combining the LSTM-MPC control strategy. The LSTM-MPC control strategy includes a voltage and current prediction strategy based on a discretized mathematical model, a SOC prediction strategy based on an LSTM network model, and a current and voltage command optimization strategy based on a cost function.

[0076] The control block diagram of the above LSTM-MPC control strategy is attached. Figure 2 As shown, in conjunction with the appendix Figure 2 The above step S220 specifically includes:

[0077] Step S221: Based on the voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment, the SOC prediction, output voltage prediction, grid current prediction, and output power prediction of each single-phase sub-module of the grid-connected converter under different switching states at the next moment are obtained using the LSTM network model and the discretized mathematical model.

[0078] This step includes:

[0079] (i) Input the voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment into the discretized mathematical model set. Iterate through all the discretized mathematical models under different switching states in the discretized mathematical model set to obtain the predicted output voltage and grid current values ​​of each single-phase submodule of the grid-connected converter under different switching states at the next moment. Calculate the predicted output power of each single-phase submodule of the grid-connected converter under different switching states at the next moment. The discretized mathematical models include:

[0080] Grid-connected current prediction model:

[0081]

[0082] Among them, i α (k+1) and i β (k+1) represent the predicted grid-connected current values ​​along the α and β axes at time k+1, respectively; i g (k+1) is the predicted grid-connected current at time k+1; u α (k) and u β (k) represents the measured grid voltage values ​​along the α and β axes at time k; u j (k) represents the measured grid voltage at time k; u bat,α (k) and u bat,β (k) represents the total output voltage measurement values ​​of the single-phase submodules along the α and β axes at time k; u bat,j (k) represents the total output voltage measurement of the single-phase submodule at time k; i α (k) and iβ (k) represents the grid-connected current measurements along the α and β axes at time k, respectively;

[0083] Output voltage prediction model for single-phase submodule:

[0084]

[0085] Among them, u bat,j (k+1) represents the predicted output voltage of the single-phase submodule at time k+1;

[0086] The process of constructing a discretized mathematical model for a retired power battery energy storage system here includes:

[0087] (1) Analyze the topology of the retired power battery energy storage system. The specific topology of the retired power battery energy storage system is shown in the attached figure. Figure 3 As shown;

[0088] In a three-phase system, each phase consists of n cascaded sub-modules. Each phase is externally connected to an H-bridge, and the output is connected to an inductive filter and then fed into the power grid. R in the diagram... S,i L S,i i g,i , where i = a, b, and c represent the grid-connected equivalent resistance, inductance, and grid-connected current, respectively.

[0089] The operating state of the submodule depends on power switching devices S1 and S2 and switch K; during normal operation, K is open; when the submodule malfunctions, K immediately closes. When the submodule is activated, S1 is closed and S2 is open, and when i... g When i < 0, the submodule battery pack is in a discharging state; when i g When the voltage is >0, the submodule battery pack is in a charging state, and the submodule voltage is V. bat When the submodule battery pack S soc ≥80% or S soc When ≤20% of the submodules are disconnected, S1 is open and S2 is closed, at which time the submodule output voltage is 0.

[0090] (2) Establish a single-phase grid-connected model for retired power battery energy storage systems.

[0091] The equivalent circuit of a single-phase grid-connected retired power battery energy storage system is as follows: Figure 4 As shown. Where, u j (j=a,b,c), u bat,j (j=a,b,c) represent the grid-side voltage and the total external voltage of each phase submodule, respectively.

[0092] According to Kirchhoff's voltage law (KVL), its voltage equation is:

[0093]

[0094] Retired power battery energy storage systems can adopt an αβ coordinate system to design a grid-connected control strategy to achieve decoupled control of active and reactive power. The KVL and power equations in the αβ coordinate system are as follows:

[0095]

[0096]

[0097] Among them, i α i β u α u β These represent the components of the grid-connected current on the α and β axes, and the components of the grid voltage on the α and β axes, respectively; P represents active power; and Q represents reactive power.

[0098] (3) Establish a discretized mathematical model for retired power battery energy storage systems

[0099] The discretized mathematical model of the aforementioned retired power battery energy storage system includes a grid-connected current prediction model and a grid-connected output voltage prediction model, specifically with a sampling period of T. S By discretizing equations (1) and (2) using the forward Euler formula, the grid-connected current prediction model at time k+1 is obtained as follows:

[0100]

[0101] Among them, i α (k+1) and i β (k+1) represent the predicted grid-connected current values ​​along the α and β axes at time k+1, respectively; i g (k+1) is the predicted grid-connected current at time k+1; u α (k) and u β (k) represents the measured grid voltage values ​​along the α and β axes at time k; u j (k) represents the measured grid voltage at time k; u bat,α (k) and u bat,β (k) represents the total output voltage measurement values ​​of the single-phase submodules along the α and β axes at time k; u bat,j (k) represents the total output voltage measurement of the single-phase submodule at time k; i α (k) and i β (k) represents the grid-connected current measurements along the α and β axes at time k, respectively;

[0102] Output voltage prediction model for single-phase submodule:

[0103]

[0104] Among them, u bat,j (k+1) is the predicted output voltage of the single-phase submodule at time k+1.

[0105] (ii) Input the predicted output voltage and grid current of each single-phase submodule into the LSTM network model to obtain the predicted SOC of each single-phase submodule under different switching states of the grid-connected converter at the next moment. The LSTM network model is obtained by training the LSTM neural network with several samples, which are the charging and discharging voltage and current data of retired power batteries.

[0106] Traditional battery energy storage system control is based on the ampere-hour integral method to estimate the battery's state of charge (SOC). This method uses linear estimation of the battery model and assumes that the battery's maximum usable capacity and coulombic efficiency will not change with battery aging. However, in reality, due to the severe aging of retired power batteries, their maximum usable capacity and coulombic efficiency will change nonlinearly during secondary use. This leads to serious distortion of the SOC estimate of retired power batteries based on the ampere-hour integral method, which in turn causes the retired power battery energy storage control system to malfunction and even safety problems.

[0107] LSTM networks, with sufficient training data, suitable network structures, and appropriate training algorithms and parameter adjustments, can approximate any function. They are well-suited for state estimation during the nonlinear aging process of batteries, thus considering battery health status and timing characteristics in battery SOC prediction, dynamically correcting SOC prediction values ​​in real time, adapting to different battery aging conditions, and significantly improving the performance and reliability of retired power battery energy storage systems, extending battery life. Therefore, in this embodiment, an LSTM network model is obtained by training an LSTM neural network, and the SOC prediction values ​​of each phase of the grid-connected converter under different switching states at the next time step are obtained using the LSTM network model.

[0108] The training process of an LSTM network model includes:

[0109] (1) Data preprocessing. The battery input current and output voltage data and battery state of charge data are extracted into time series to form a training set, where X is the training data and Y is the corresponding label; the data is deduplicated, and for data that is completely repeated in the dataset, only one data is kept and the rest are deleted; the data is scaled to between 0 and 1 using the minimax normalization method, as shown in Equation (10);

[0110]

[0111] Where 1≤i≤2, x1 and x2 represent the battery input current and output voltage, respectively; x max x2 represents the maximum and minimum values ​​of the input.

[0112] (2) Network Training. The training process for the LSTM neural network is shown in the appendix. Figure 5 The training and testing data are divided into a 7:3 ratio, with 1,000,000 training sequences. The input consists of 100,000 normalized charge / discharge voltage and current data points from retired power batteries; the output is the predicted SOC value at the current moment. The attached figure shows the battery voltage u at time k. bat,j (k), Current i g (k) is taken as input, and the SOC value S of the battery at time k is S. SOC (k) is used as the output. After multiple tests and comparisons, considering the model's accuracy, generalization ability, and training speed, it was found that the optimal parameters for the LSTM neural network should be set as follows: 1 input layer, 3 hidden layers, 75 hidden layer nodes, 1 output layer, learning rate of 0.01, 1000 training iterations, training target error of 0.0001, and GELU activation function for the weight generation network.

[0113] In step S222, under both grid voltage balance and grid voltage imbalance conditions, the minimum cost function is taken as the optimization control objective. The converter switching state with the minimum cost function value is selected as the input state at the next moment. Based on the optimal grid connection command acquisition model, the optimal grid connection current command and the optimal output voltage command of the α axis and β axis at the next moment are obtained.

[0114] Step S222 above executes the optimized control of the decommissioned power battery energy storage system. The optimized control is divided into two cases: grid voltage balance and grid voltage imbalance, as detailed below:

[0115] (I) Optimized control during grid voltage balance

[0116] When the grid voltage is balanced, the optimized control of the decommissioned power battery energy storage system, combining LSTM-MPC control and carrier phase-shift modulation, is shown in the attached figure. Figure 6 As shown.

[0117] Cost function:

[0118]

[0119] Among them, P ref P(k+1) and Q are the given reference active power and the predicted active power at time k+1, respectively; ref Q(k+1) represents the given reference reactive power and the predicted reactive power at time k+1, respectively. S SOC (k+1) represents the given reference SOC value and the predicted SOC value at time k+1, respectively;

[0120] The optimal grid connection command acquisition model includes the optimal grid connection current command acquisition model and the optimal output voltage command acquisition model;

[0121] The current command generation stage is equivalent to power outer loop control, which can achieve power point tracking. Furthermore, the grid-connected current will change according to the given reference command. Therefore, the optimal grid-connected current command acquisition model is as follows:

[0122]

[0123]

[0124] in, and These are the optimal grid-connected current commands along the α and β axes at time k+1, respectively; P * (k+1) represents the optimal active power prediction value at time k+1; Q * (k+1) represents the optimal reactive power prediction value at time k+1; u α (k) and u β (k) represents the measured grid voltage values ​​along the α and β axes at time k;

[0125] The voltage command is derived from the discrete mathematical model. At time k+1, if the system achieves good power point tracking, the actual power of the system should be approximately equal to the input reference power. Therefore, the predicted output current should be approximately equal to the current reference command value, i.e.:

[0126]

[0127] By processing formula (16) according to formula (5), the optimal output voltage command acquisition model is obtained as shown below:

[0128]

[0129] in, and These are the optimal output voltage commands for the α-axis and β-axis at time k+1, respectively. and These are the optimal grid-connected current commands along the α and β axes at time k+1, respectively; P * (k+1) represents the optimal active power prediction value at time k+1; Q * (k+1) represents the optimal reactive power prediction value at time k+1; u α (k) and u β (k) represents the measured grid voltage values ​​along the α and β axes at time k; T S The sampling period;

[0130] (II) Optimized control under grid voltage imbalance

[0131] When the grid voltage is unbalanced, the optimized control of the decommissioned power battery energy storage system, combining LSTM-MPC control and carrier phase-shift modulation, is shown in the attached figure. Figure 7 As shown.

[0132] When an imbalance fault occurs in the power grid, the instantaneous power of the system will exhibit a second harmonic component, expressed as follows:

[0133]

[0134] Where P0 and Q0 are the average values ​​of P and Q, respectively; P C P S These are the average active harmonic values ​​of the second-order residual and sinusoidal terms, respectively; Q C Q S These are the average values ​​of the reactive harmonics of the second and sine terms, respectively.

[0135]

[0136] Among them, u acα+ u acα- These represent the positive and negative sequence components of the grid voltage on the α-axis, respectively; u acβ+ u acβ- These represent the positive and negative sequence components of the grid voltage on the β axis, respectively; i α+ i α- These represent the positive and negative sequence components of the grid-connected current on the α-axis, respectively; i β+ i β- These are the positive and negative sequence components of the grid-connected current on the α-axis, respectively.

[0137] When an imbalance occurs, the presence of a negative-sequence component in the current will cause output waveform distortion and power fluctuations. Therefore, it is necessary to suppress negative-sequence current and reactive and active power fluctuations. Under this condition, the cost function is:

[0138]

[0139] Among them, i α- (k+1),i β- (k+1) represent the negative sequence components of the grid-connected current prediction value at time k+1 on the α and β axes, respectively; λ i , λ P , λ Q , λ SOC These are negative sequence current suppression, constant active power operation, constant reactive power operation, and constant SOC operation, respectively, and the sum of the weights of the four control objectives in the cost function is equal to 1.

[0140] The optimal grid connection command acquisition model includes the optimal grid connection current command acquisition model and the optimal output voltage command acquisition model;

[0141] The construction process of the optimal grid-connected current command acquisition model includes:

[0142] (1) Current balance operation

[0143] When the three-phase voltage of the system is unbalanced, the grid-connected current contains negative sequence current. To obtain a balanced and symmetrical three-phase current, it is necessary to suppress the fluctuation component generated by the negative sequence current, so that i α- =0, i β- =0, thus the current command obtained is:

[0144]

[0145] (2) Constant active power operation

[0146] To eliminate the second harmonic ripple in active power, Q needs to be... C =0, Q S =0, P0, Q0 are given reference values. At this time, negative sequence current components are allowed in the system, and the current command equation is:

[0147]

[0148] (3) Constant reactive power operation

[0149] To suppress reactive power fluctuations, let Q C =0, Q S =0 is a given reference value, resulting in the current command equation:

[0150]

[0151] Under the same control objective, positive and negative current commands can be added together to obtain the optimal current commands for both current-balanced operation and constant power operation:

[0152]

[0153] Substituting the optimal output current command of the above three control objectives into the multi-objective control function, we obtain the optimal grid-connected current command acquisition model at time k+1, as shown below.

[0154]

[0155] in, The predicted values ​​of the current commands on the α and β axes at time k+1; i α负 (k+1),i β负 (k+1) represents the predicted current command values ​​on the α and β axes at time k+1 under the current balance operation target; i αP (k+1),i βP(k+1) represents the predicted current command values ​​on the α and β axes at time k+1 under the target of constant active power operation; i αQ (k+1),i βQ (k+1) represents the predicted current command values ​​on the α and β axes at time k+1 under the target of constant reactive power operation.

[0156] Optimal output voltage command acquisition model:

[0157]

[0158] in, and These are the optimal output voltage commands for the α-axis and β-axis at time k+1, respectively. and These are the optimal grid-connected current commands along the α and β axes at time k+1, respectively; P * (k+1) represents the optimal active power prediction value at time k+1; Q * (k+1) represents the optimal reactive power prediction value at time k+1; u α (k) and u β (k) represents the measured grid voltage values ​​along the α and β axes at time k; T S The sampling period.

[0159] Step S223: After coordinate transformation, calculate the duty cycle of each single-phase submodule based on the optimal grid-connected current command and the optimal output voltage command of the α-axis and β-axis at the next moment. The duty cycle is the ratio of the optimal output voltage command at the next moment to the predicted total output voltage of the single-phase submodule at the next moment.

[0160] For system power control, the converter arm output voltage must be greater than the grid-connected peak voltage of 311V. The number of single-phase sub-modules and the total voltage of each phase in the grid-connected retired power battery energy storage system are as follows:

[0161]

[0162] Among them, u j (k) represents the total output voltage value of the single-phase submodule; n j Indicates the number of single-phase submodules put into operation for each phase; V bati These are the output voltage values ​​of each single-phase submodule.

[0163] The duty cycle is the ratio of the optimal output voltage command for the next time step calculated by the LSTM-MPC control strategy to the predicted total output voltage of the single-phase submodule for the next time step.

[0164]

[0165] Among them, Dj (k) is the duty cycle for each single-phase submodule; u j (k+1) is the predicted total output voltage of the single-phase submodule at the next moment; This is the optimal output voltage command for the next moment.

[0166] Step S230: The duty cycle of each single-phase submodule is imported into the carrier phase-shift modulation algorithm to generate the single-phase submodule switching signal.

[0167] Step S240: Iterate the above steps within the set scheduling period to complete feedback correction and achieve optimal control.

[0168] In step S240, the above steps S210 to S230 are looped within the set scheduling period to perform an iterative process, realize feedback correction, minimize the cost function within one period, and thus achieve optimal control.

[0169] This invention, based on the topology of a retired power battery energy storage system, combines LSTM (Long Short-Term Memory) neural network, Model Predictive Control (MPC), and duty cycle modulation to disclose an optimized control method for a grid-connected retired power battery energy storage system capable of power point tracking (PPT) and State of Charge (SOC) tracking. The optimal output voltage is obtained by combining a discrete mathematical model of the retired power battery energy storage system with a cost function, and then the switching on and off of the corresponding single-phase sub-modules is controlled using a carrier phase-shift modulation algorithm. Addressing the issue that traditional ampere-hour integration methods cannot achieve high-precision estimation of SOC under nonlinear changes due to battery aging in cost function construction, this paper uses real-time acquired voltage and current measurements and LSTM network and discretized mathematical models to predict the SOC, output voltage, grid current, and output power of each single-phase submodule (i.e., each phase) of the grid-connected converter under different switching states at the next moment. By combining LSTM, MPC, and duty cycle modulation, it achieves SOC and power tracking for the battery pack under balanced grid voltage conditions. Under unbalanced grid voltage conditions, it considers the three-phase unbalanced operation characteristics of the system to suppress negative sequence current and power fluctuations, accurately obtaining the optimal grid current command and the optimal output voltage command for the α and β axes at the next moment, providing accurate data support for obtaining the switching signals of the single-phase submodule.

[0170] Example 3: To verify the effectiveness of the present invention, this example uses the Matlab / Simulink platform to build a simulation model of a retired power battery energy storage system to verify the effectiveness of the proposed algorithm. The parameter settings are shown in Table 1, the system simulation time is 1.5s, and the control step size is 100μs.

[0171] Table 1 Simulation Model Parameters

[0172]

[0173]

[0174] (1) Analysis of the optimal SOC prediction results of retired power batteries

[0175] This embodiment uses the root mean square error (RMSE), mean absolute percentage error (MAPE), and mean absolute error (MAE), as shown in the following formulas:

[0176]

[0177] Among them, C t For predicted value, C t This is the actual value.

[0178] This embodiment utilizes the 18650 battery dataset, employs an LSTM neural network to estimate the State of Charge (SOC) of the 18650 battery at different cycle counts, compares it with the true value, and contrasts it with the ampere-hour integration algorithm. Figure 8 As shown in Table 2.

[0179] Table 2 Algorithm Error Quantization Results

[0180]

[0181] As shown in the chart, compared with the traditional ampere-hour integration algorithm, the RMSE, MAPE and MAE values ​​of the LSTM neural network algorithm are significantly reduced. Therefore, the LSTM network has high accuracy in predicting the SOC of retired power batteries.

[0182] (2) Simulation verification during grid balance

[0183] (a) Power tracking simulation verification

[0184] Appendix Figure 9 The figure shows the simulation waveform of the system power point tracking. The reference values ​​for the system's active and reactive power are set to 72kW and 0kvar, respectively. At 0.3s, the system's active power changes from 72kW to 36kW, while the reactive power remains unchanged. As can be seen from the figure, the proposed strategy enables the system's output power to track the reference power well, and the dynamic response speed is fast.

[0185] (b) Simulation verification of charge-discharge switching

[0186] The simulation waveform of the system charging and discharging switching is attached. Figure 10As shown in the figure, the charging / discharging switch occurs at 0.3s, changing the direction of the system current from P = 72kW to P = -72kW. The figure demonstrates that the grid-connected current and reactive power fluctuate minimally and stabilize quickly.

[0187] (c) SOC tracking simulation verification

[0188] The simulation waveform of single-phase SOC is attached. Figure 11 As shown in the figure, the strategies proposed in this paper can enable the total SOC of a single-phase bridge arm to track the reference SOC well.

[0189] (3) Simulation verification under grid voltage imbalance

[0190] The system power, grid voltage, output voltage, and grid-connected current curves under grid-side voltage imbalance in the simulation model are shown in the attached figure. Figure 12 Appendix Figure 13 As shown, a single-phase voltage imbalance fault occurs in the grid voltage at 0.8s, causing the C-phase voltage to drop to 200V, resulting in distortion of the output current; the system's active and reactive power fluctuate at twice the frequency.

[0191] (a) Negative sequence current suppression

[0192] The simulation of negative sequence current suppression is attached. Figure 14 As shown. From the appendix Figure 14 It can be seen that the AC side current waveform is significantly improved after the negative sequence current suppression control strategy is added.

[0193] (b) Constant active power operation

[0194] The simulation of active power fluctuation suppression is attached. Figure 15 As shown in the figure, after the active power fluctuation suppression strategy was implemented, the active power fluctuation decreased significantly and basically returned to constant operation. However, due to the presence of negative sequence components, reactive power fluctuations still exist.

[0195] (c) Constant reactive power operation

[0196] The simulation of reactive power fluctuation suppression is attached. Figure 16 As shown in the figure, after implementing reactive power fluctuation suppression measures, reactive power fluctuations are eliminated, but active power fluctuations still exist.

[0197] (d) Simultaneous suppression of both active and reactive power

[0198] According to the appendix Figure 17 The waveform shown illustrates that power fluctuations occur in the system under three-phase voltage imbalance. However, once a fluctuation suppression control method is introduced, fluctuations in both active and reactive power are effectively suppressed.

[0199] (e) SOC tracking simulation verification

[0200] The simulation waveform of the single-phase bridge arm SOC is attached. Figure 18 As shown. From the appendix Figure 18 As can be seen from this paper, the proposed strategy has a certain degree of robustness and reliability. Even under the condition of unbalanced grid voltage, it can still make the total SOC of a single-phase bridge arm track the reference SOC well.

[0201] Example 4: This embodiment of the invention applies the above embodiments and discloses an optimized control device for a grid-connected retired power battery energy storage system, comprising:

[0202] The data acquisition unit acquires the voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment. The voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment include: the grid-connected current measurement values ​​of the α-axis and β-axis at the current moment, the grid voltage measurement values ​​of the α-axis and β-axis at the current moment, the grid voltage measurement values ​​of the α-axis and β-axis at the current moment, the total output voltage measurement values ​​of the single-phase sub-modules of the α-axis and β-axis at the current moment, and the total output voltage measurement values ​​of the single-phase sub-modules at the current moment.

[0203] The prediction and optimization unit, based on the voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment, combines the LSTM-MPC control strategy to obtain the duty cycle of each single-phase sub-module at the next moment. The LSTM-MPC control strategy includes a voltage and current prediction strategy based on a discretized mathematical model, a SOC prediction strategy based on an LSTM network model, and a current and voltage command optimization strategy based on a cost function.

[0204] The control signal generation unit imports the duty cycle of each single-phase submodule into the carrier phase-shift modulation algorithm to generate the single-phase submodule switching signal;

[0205] The iterative control unit iterates through the above steps within a set scheduling cycle to complete feedback correction and achieve optimal control.

[0206] The aforementioned prediction optimization unit includes:

[0207] The prediction module, based on the voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment, uses an LSTM network model and a discretized mathematical model to obtain the SOC prediction value, output voltage prediction value, grid current prediction value, and output power prediction value of each single-phase sub-module of the grid-connected converter under different switching states at the next moment.

[0208] The predictive data optimization module takes the minimum cost function as the optimization control objective in both grid voltage balance and grid voltage imbalance cases. It selects the converter switching state with the minimum cost function value as the input state for the next moment. Based on the optimal grid connection command acquisition model, it obtains the optimal grid connection current command and the optimal output voltage command for the α and β axes for the next moment.

[0209] The duty cycle acquisition module calculates the duty cycle of each single-phase submodule after transforming the optimal grid-connected current command and the optimal output voltage command of the α-axis and β-axis at the next moment. The duty cycle is the ratio of the optimal output voltage command at the next moment to the predicted total output voltage of the single-phase submodule at the next moment.

[0210] The aforementioned prediction data optimization module includes:

[0211] The first optimization submodule, when the grid voltage is balanced, takes minimizing the cost function as the optimization control objective, selects the converter switching state with the minimum cost function value as the input state for the next moment, and obtains the optimal grid-connected current command and the optimal output voltage command for the α and β axes for the next moment based on the optimal grid-connected command acquisition model. The cost function and the corresponding optimal grid-connected command acquisition model are shown below:

[0212] Cost function:

[0213]

[0214] Among them, P ref P(k+1) and Q are the given reference active power and the predicted active power at time k+1, respectively; ref Q(k+1) represents the given reference reactive power and the predicted reactive power at time k+1, respectively. S SOC (k+1) represents the given reference SOC value and the predicted SOC value at time k+1, respectively;

[0215] The optimal grid connection command acquisition model includes:

[0216] Optimal grid-connected current command acquisition model:

[0217]

[0218]

[0219] in, and These are the optimal grid-connected current commands along the α and β axes at time k+1, respectively; P * (k+1) represents the optimal active power prediction value at time k+1; Q* (k+1) represents the optimal reactive power prediction value at time k+1; u α (k) and u β (k) represents the measured grid voltage values ​​along the α and β axes at time k;

[0220] Optimal output voltage command acquisition model:

[0221]

[0222] in, and These are the optimal output voltage commands for the α-axis and β-axis at time k+1, respectively. and These are the optimal grid-connected current commands along the α and β axes at time k+1, respectively; P * (k+1) represents the optimal active power prediction value at time k+1; Q * (k+1) represents the optimal reactive power prediction value at time k+1; u α (k) and u β (k) represents the measured grid voltage values ​​along the α and β axes at time k; T S The sampling period.

[0223] The second optimization submodule, when the grid voltage is unbalanced, takes minimizing the cost function as the optimization control objective, selects the converter switching state with the minimum cost function value as the input state for the next moment, and obtains the optimal grid-connected current command and the optimal output voltage command for the α and β axes for the next moment based on the optimal grid-connected command acquisition model. The cost function and the corresponding optimal grid-connected command acquisition model are shown below:

[0224] The cost function is shown below:

[0225]

[0226] Among them, i α- (k+1),i β- (k+1) represent the negative sequence components of the grid-connected current prediction value at time k+1 on the α and β axes, respectively; λ i , λ P , λ Q , λ SOC These are negative sequence current suppression, constant active power operation, constant reactive power operation, and constant SOC operation, respectively, and the sum of the weights of the four control objectives in the cost function is equal to 1.

[0227] The optimal grid connection command acquisition model includes:

[0228] Optimal grid-connected current command acquisition model:

[0229]

[0230] in, The predicted values ​​of the current commands on the α and β axes at time k+1; i α负 (k+1),i β负 (k+1) represents the predicted current command values ​​on the α and β axes at time k+1 under the current balance operation target; i αP (k+1),i βP (k+1) represents the predicted current command values ​​on the α and β axes at time k+1 under the target of constant active power operation; i αQ (k+1),i βQ (k+1) represents the predicted current command values ​​on the α and β axes at time k+1 under the target of constant reactive power operation.

[0231] Optimal output voltage command acquisition model:

[0232]

[0233] in, and These are the optimal output voltage commands for the α-axis and β-axis at time k+1, respectively. and These are the optimal grid-connected current commands along the α and β axes at time k+1, respectively; P * (k+1) represents the optimal active power prediction value at time k+1; Q * (k+1) represents the optimal reactive power prediction value at time k+1; u α (k) and u β (k) represents the measured grid voltage values ​​along the α and β axes at time k; T S The sampling period.

[0234] The above technical features constitute the preferred embodiment of the present invention, which has strong adaptability and optimal implementation effect. Unnecessary technical features can be added or removed according to actual needs to meet the requirements of different situations.

Claims

1. A method for optimizing control of a grid-connected decommissioned power battery energy storage system, characterized in that, include: The system collects the voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment. The voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment include those at the current moment. shaft and The measured value of the grid-connected current of the shaft, the measured value of the grid-connected current at the current moment, and the current value at the current moment. shaft and Grid-connected output voltage measurement of the shaft, current time shaft and The total output voltage measurement value of the single-phase submodule of the shaft, and the total output voltage measurement value of the single-phase submodule at the current moment; Based on the voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment, and combined with the LSTM-MPC control strategy, the duty cycle of each phase at the next moment is obtained, including: Based on the voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment, the predicted SOC, grid-connected output voltage, grid-connected current, and output power of each phase of the grid-connected converter under different switching states at the next moment are obtained using the LSTM network model and the discretized mathematical model. In both cases of grid voltage balance and grid voltage imbalance, the goal is to minimize the cost function. The converter switching state with the minimum cost function value is selected as the input state for the next time step. The next time step is obtained based on a discretized mathematical model. shaft and The optimal grid-connected current command for the shaft, the next moment shaft and The optimal grid-connected output voltage command for the shaft; The next moment shaft and Grid current command for the shaft and the next moment shaft and The duty cycle of the grid-connected output voltage command of the axis is calculated after coordinate transformation. The duty cycle is the ratio of the optimal grid-connected output voltage command at the next moment to the predicted total output voltage of the single-phase submodule at the next moment. The cost function for grid voltage balance is shown below: Among them, P ref P(k+1) represents the given reference active power, Predicted active power at time Q; ref Q(k+1) are the given reference reactive power, Predicted reactive power at any given time; Given a reference SOC value, SOC prediction value at time; The cost function for grid voltage imbalance is shown below: in, They are respectively The predicted value of grid-connected current at any time shaft and Negative-order components on the axis; These are negative sequence current suppression, constant active power operation, constant reactive power operation, and constant SOC operation, respectively, and the sum of the weights of the four control objectives in the cost function is equal to 1. The duty cycle of each phase is imported into the carrier phase-shift modulation algorithm to generate the single-phase submodule switching signal; The above steps are iterated within the set scheduling period to complete feedback correction and achieve optimal control.

2. The optimized control method for a grid-connected retired power battery energy storage system according to claim 1, characterized in that, Based on the voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment, the predicted SOC, voltage, current, and output power of each phase of the grid-connected converter under different switching states at the next moment are obtained using an LSTM network model and a discretized mathematical model, including: The voltage and current measurements of the grid-connected retired power battery energy storage system at the current moment are input into the discretized mathematical model set. The discretized mathematical models under all different switching states in the set are traversed to obtain the predicted grid-connected output voltage and current values ​​for each phase of the grid-connected converter under different switching states at the next moment. The predicted output power values ​​for each phase of the grid-connected converter under different switching states at the next moment are also calculated. The discretized mathematical models include: Grid-connected current prediction model: in, They are respectively time shaft and Predicted grid-connected current value for the shaft; Predicted grid-connected current at any given time; They are time k respectively shaft and Measured value of the mains output voltage of the shaft; The measured value of the grid output voltage at any given time; They are time k respectively shaft and The total output voltage measurement of the single-phase submodule of the shaft; The measured total output voltage of the single-phase submodule at any given moment; They are time k respectively shaft and Grid-connected current measurement value of the shaft; Grid-connected output voltage prediction model: in, Predicted grid-connected output voltage at any given time; The predicted grid-connected output voltage and grid-connected current of each phase under different switching states of the grid-connected converter at the next moment are input into the LSTM network model to obtain the predicted SOC of each phase under different switching states of the grid-connected converter at the next moment. The LSTM network model is obtained by training the LSTM neural network with several samples, which are the charging and discharging voltage and current data of retired power batteries.

3. The optimized control method for a grid-connected retired power battery energy storage system according to claim 1 or 2, characterized in that, When the grid voltage is balanced, the goal of optimization control is to minimize the cost function. The converter switching state with the minimum cost function value is selected as the input state for the next time step. The next time step is obtained based on a discretized mathematical model. shaft and The optimal grid-connected current command for the shaft, the next moment shaft and The optimal grid-connected output voltage command for the shaft, wherein the corresponding discretized mathematical model includes: Optimal grid-connected current command acquisition model: in, They are respectively time shaft and The optimal current command for grid connection of the shaft; To give The optimal active power prediction value at time t; The optimal reactive power prediction value at time t; They are time k respectively shaft and Measured value of the mains output voltage of the shaft; Optimal grid-connected output voltage command acquisition model: in, They are respectively time shaft and The optimal grid-connected output voltage command for the shaft; They are respectively time shaft and The optimal current command for grid connection of the shaft; To give The optimal active power prediction value at time t; The optimal reactive power prediction value at time t; They are time k respectively shaft and Measured value of the mains output voltage of the shaft; The sampling period.

4. The optimized control method for a grid-connected retired power battery energy storage system according to claim 1 or 2, characterized in that, When the grid voltage is unbalanced, the goal of optimization control is to minimize the cost function. The converter switching state with the minimum cost function value is selected as the input state for the next time step. The next time step is obtained based on a discretized mathematical model. shaft and The optimal grid-connected current command for the shaft, the next moment shaft and The optimal grid-connected output voltage command for the shaft, wherein the discretized mathematical model includes: Optimal grid-connected current command acquisition model: in, time Current command prediction on the axis; Under the target of constant current balance operation Current command prediction on the axis; Under the objective of constant active power operation Current command prediction on the axis; Under the objective of constant reactive power operation Current command prediction on the axis; Optimal grid-connected output voltage command acquisition model: in, They are respectively time shaft and The optimal grid-connected output voltage command for the shaft; They are respectively time shaft and The optimal current command for grid connection of the shaft; To give The optimal active power prediction value at time t; The optimal reactive power prediction value at time t; They are time k respectively shaft and Measured value of the mains output voltage of the shaft; The sampling period.

Images