A method for verifying the voltage inflection point (SOC) of flow batteries based on GBDT machine learning prediction.

Abstract

The method for verifying the SOC of a flow battery based on GBDT machine learning prediction belongs to the field of application basic technology combining power electronics and long-term energy storage control. This invention addresses the current lack of mature and accurate SOC measurement technology and evaluation algorithms for ZBFB (Zero-Brain Battery). It applies the GBDT algorithm from machine learning to the real-time calculation and verification of ZBFB's SOC, effectively improving the accuracy of SOC calculation at the voltage inflection point of ZBFB and providing a reference for developing a general flow battery SOC algorithm. Compared with existing technologies, it has the following advantages: (1) Based on several sets of representative feature data, a GBDT verification algorithm matching the SOC measurement at the ZBFB discharge voltage inflection point can be used to achieve high-precision evaluation of the SOC at the ZBFB inflection point; (2) By applying the GBDT algorithm verification and timeliness judgment criteria based on limit cycle analysis, the problem of the long processing time of the GBDT algorithm is solved, providing a computational basis for improving the timeliness of GBDT from an algorithmic perspective.

Description

Technical Field

[0001] This invention belongs to the field of applied basic technology combining power electronics and long-term energy storage control, and relates to a high-precision evaluation and verification method for SOC (State of Charge) at the inflection point of battery discharge voltage based on GBDT (Gradient Boosting Decision Tree). Background Technology

[0002] Accurate measurement of the State of Charge (SOC) of zinc-bromine flow batteries (ZBFBs) is crucial for improving battery reliability, achieving series voltage equalization, suppressing parallel circulating currents, and advancing their application in long-term energy storage. However, due to the inherent electrochemical reaction characteristics of the battery and the influence of self-discharge, coulombic efficiency, electrolyte temperature, flow rate, and discharge rate, an unavoidable voltage inflection point occurs in the later stages of discharge. Limited by the sampling frequency of the Battery Management System (BMS), the sudden voltage drop at the inflection point easily leads to distortion of the sampled voltage and current, thus affecting the accuracy of SOC calculation. This is a problem that classical SOC calculation methods and current measurement techniques cannot completely solve. Therefore, it is urgent to investigate a feasible method for identifying the SOC at the discharge voltage inflection point of ZBFBs.

[0003] GBDT is a high-precision prediction algorithm based on time series data. Its advantages lie in its prediction accuracy, robustness, and computational capability with small sample sizes. However, the long training time due to the real-time adjustment of parameters such as the number and depth of trees and the learning rate affects the real-time performance of battery SOC monitoring, which is one of the bottlenecks in its application to SOC identification at ZBFB voltage inflection points. Previous research has shown that not all ZBFB discharge voltage inflection points exhibit SOC distortion; only some inflection points show this distortion. Since ZBFB discharge voltage inflection point data has a small sample size, GBDT identification and correction can be performed only on inflection points with SOC distortion, thus significantly reducing the computational cost and running time of GBDT. Summary of the Invention

[0004] The purpose of this invention is to address the bottleneck problem of classical SOC calculation methods being unable to achieve high-precision prediction due to the influence of the sampling frequency of the Battery Management System (BMS). This invention provides a method for identifying the SOC at the voltage inflection point of a flow battery based on limit cycle-assisted GBDT analysis, which can serve as a reference for developing a general high-precision SOC calculation algorithm for flow batteries. This method focuses on solving the following problems: ① Developing an efficient GBDT solution algorithm for battery SOC identification based on ZBFB discharge voltage and current; ② Investigating a solution method for determining SOC distortion at the ZBFB discharge voltage inflection point based on limit cycle analysis.

[0005] The objective of this invention is achieved through the following technical solution:

[0006] A method for verifying the state of charge (SOC) of flow batteries based on GBDT machine learning prediction, the method comprising the following steps:

[0007] Step 1: Establish a comprehensive experimental platform for zinc-bromine flow batteries, including the zinc-bromine flow battery body, BMS, EMS, PCS, and load AC motor; among them, BMS is used for monitoring the charge and discharge data of the battery body, EMS is used for powering the positive and negative electrode circulation pumps of the zinc-bromine flow battery, PCS is used to realize DC / AC power conversion, and load AC motor is used to simulate inductive load.

[0008] Step 2: Clarify the measured parameters and the identified parameters. The measured parameters include voltage, current, circulating pump speed, charge and discharge time, and measured SOC. The identified parameter is the SOC at the voltage inflection point. Use a battery charge and discharge comprehensive tester and BMS to monitor the experimental platform in real time for 24 hours with a sampling period of 1 minute to obtain the curves of measured SOC, battery voltage, and battery current data changing over time.

[0009] Step 3: Select limited voltage and current sampling data within multiple cycles of ZBFB discharge to identify discharge variation characteristics; extract limited feature samples at regular intervals within a whole discharge cycle to avoid sensitive outliers;

[0010] Step 4: Using the extracted limited feature samples as the dataset, and providing the loss function and learning rate of GBDT in real time, the optimal SOC value of ZBFB is obtained by GBDT, and a SOC solution paradigm that can be verified in real time is developed.

[0011] Step 5: Apply limit cycle theory to derive the judgment and verification criteria for calculating ZBFB high-precision SOC using GBDT. By reducing the number of times GBDT is applied, its computational efficiency within a complete charge-discharge cycle is improved, thereby enhancing the real-time performance of the proposed GBDT algorithm.

[0012] Furthermore, in step two, the sampling period is 1 minute, and the sampled measured data includes the voltage, current, circulation pump speed, charge and discharge time, and measured SOC of the zinc-bromine flow battery experimental platform.

[0013] Furthermore, in step three, extraction is performed at regular intervals within the displayed discharge cycle. There are N representative small sample feature data.

[0014] Considering that an excessive number and depth of trees would lead to a long training time, thus affecting the real-time performance of battery SOC monitoring, the number of small sample feature data after feature extraction is N=6. Based on this, corresponding voltage and current sample datasets are obtained and sorted from low to high according to the collected voltage values:

[0015] .

[0016] Furthermore, in step four, the SOC of ZBFB in the time domain after the voltage inflection point is evaluated in real time. The implementation process of the constructed GBDT model is as follows:

[0017] (1) Input feature sample data Input loss function Where N is the number of samples, Representing the Each feature sample data, , , Represents the input space; , , Represents the output space; and These represent the output data and the learner function, respectively.

[0018] (2) Initialize the weak learner:

[0019] ,

[0020] And initialize the iteration count m=1; where To initialize the weak learner function, The input features are denoted as c, which is the constant taken to minimize the loss function.

[0021] (3) For any sampled data The residual is calculated as follows:

[0022]

[0023] in, The pseudo residual of the k-th feature sample data in the m-th iteration is, i.e. The negative gradient of the loss function;

[0024] (4) The calculated residual value is used as the true value for the next training, and the fitted data is used as the true value. The sample data is used for the next training iteration to obtain a new regression tree. Its corresponding leaf node region is ,in Let be the number of leaf nodes in the m-th regression tree;

[0025] (5) Applying regression trees Calculate the best-fit value for each leaf node:

[0026]

[0027] in, This represents the best-fit value for the leaf region. The sample data given by the learner in the (m-1)th iteration. The predicted value;

[0028] (6) Refresh the strong learner:

[0029]

[0030] in, Let I be the learner for the m-th iteration, and let I be an indicator symbol with a value of 1, which is used to fit the classifier in the node region.

[0031] (7) After completing M rounds of iterations, the final boosting regression tree is obtained:

[0032] ,

[0033] in, This represents the number of iterations, i.e., the number of trees.

[0034] (8) Obtain and analyze the optimal decision tree ;

[0035] (9) Verification by comparison with the basic sample ;

[0036] (10) If If the predicted value is reasonable, output Conversely, m = m + 1, and repeat the above calculation process starting from (3).

[0037] Following the GBDT algorithm's solution process, the initialization of the first weak learner should minimize the loss function. The residual of the measured SOC relative to the first weak learner is calculated, and the optimal splitting node for the sample data is found based on the traversal value of each feature data. This program begins with voltage... Termination by current ;

[0038] To evaluate the state of charge (SOC) in the time domain after the voltage inflection point, the selection of left and right residuals at the segmentation node is aimed at optimizing the selection of the segmentation node. and As a sample partitioning node, it is about to As the optimal partitioning node; further, based on the principle of minimum variance and the traversal value analysis of the left and right leaf nodes, a tree diagram corresponding to the time domain after the voltage inflection point is obtained; based on the discharge current at the SOC that needs to be evaluated, the best path for SOC evaluation is selected.

[0039] Furthermore, in step four, considering that the coulombic efficiency of the battery varies with the discharge current over time, the learning rate of GBDT is ultimately selected. The optimal decision tree for ZBFB at the voltage inflection point can be calculated as shown in the following formula:

[0040] .

[0041] Furthermore, in step five, within a complete charge-discharge cycle, the necessity of GBDT for SOC verification at different discharge periods is determined through discharge voltage limit loop and discharge current limit loop analysis. The computational efficiency of GBDT within a complete charge-discharge cycle is improved by reducing the number of times GBDT is applied.

[0042] This invention addresses the current lack of mature and accurate SOC measurement technology and evaluation algorithms for ZBFB (Zero-Batch Flow Battery). It applies the GBDT (Guided Flow Data Determination) algorithm from machine learning to the real-time calculation and verification of ZBFB SOC. This work effectively improves the accuracy of SOC calculation at the voltage inflection point of ZBFB and provides a reference for developing a general SOC algorithm for flow batteries. Compared with existing technologies, it has the following advantages: (1) Based on several sets of representative feature data, a GBDT verification algorithm matching the SOC measurement at the ZBFB discharge voltage inflection point can be used to achieve high-precision evaluation of the SOC at the ZBFB inflection point; (2) By applying the GBDT algorithm verification and timeliness judgment criteria based on limit cycle analysis, the problem of the long processing time of the GBDT algorithm is solved, providing a computational basis for improving the timeliness of GBDT from an algorithmic perspective. Attached Figure Description

[0043] Figure 1 This is a flowchart illustrating the method implemented in this invention;

[0044] Figure 2 A diagram of the integrated experimental platform developed for the laboratory, including the ZBFB body, control circuit, and AC load;

[0045] Figure 3 The diagram shows the measured SOC and the corresponding ZBFB discharge voltage.

[0046] Figure 4 The measured SOC and corresponding ZBFB discharge current diagram are shown.

[0047] Figure 5The first regression tree plot of GBDT used to evaluate SOC in the [3520,3690] time domain (where the optimal branch is...). (As indicated by the red line)

[0048] Figure 6 The second regression tree plot for evaluating the GBDT of SOC in the [3520,3690] time domain (where the optimal branch is...) (As indicated by the red line)

[0049] Figure 7 The third regression tree plot for evaluating the GBDT of SOC in the [3520,3690] time domain (where the optimal branch is...) (As indicated by the red line)

[0050] Figure 8 The fourth regression tree plot of GBDT used to evaluate SOC in the [3520, 3690] time domain (where the optimal branch is...). (As indicated by the red line)

[0051] Figure 9 The fifth regression tree plot of GBDT used to evaluate SOC in the [3520, 3690] time domain (where the optimal branch is...). (As indicated by the red line)

[0052] Figure 10 The ZBFB discharge voltage phase diagram in the time domain [625, 2176] (relative to) Figure 2 The measured SOC shown is correct, indicating the existence of a stable limit cycle.

[0053] Figure 11 The phase diagram of ZBFB discharge voltage in the time domain [2965, 3960] (corresponding to the distorted SOC period, there is no limit cycle);

[0054] Figure 12 The ZBFB discharge voltage phase diagram in the time domain [3985, 4298] (relative to) Figure 2 The measured SOC shown is correct, indicating the existence of a stable limit cycle.

[0055] Figure 13 The phase diagram of ZBFB discharge current in the time domain [625, 2176] (relative to) Figure 2 The measured SOC shown is correct, indicating the existence of a stable limit cycle.

[0056] Figure 14 The phase diagram of ZBFB discharge current in the time domain [2965,3960] (corresponding to the distorted SOC period, there is no limit cycle);

[0057] Figure 15 The phase diagram of ZBFB discharge current in the time domain [3985, 4298] (relative to) Figure 2 The measured SOC shown is correct, indicating the existence of a stable limit cycle. Detailed Implementation

[0058] The technical solution of the present invention will be further described below with reference to the accompanying drawings, but it is not limited thereto. Any modifications or equivalent substitutions to the technical solution of the present invention that do not depart from the spirit and scope of the technical solution of the present invention should be covered within the protection scope of the present invention.

[0059] This invention, based on a 10kWh zinc-bromine flow battery experimental platform developed in the laboratory, addresses the limitation of current methods in accurately measuring the State of Charge (SOC) of zinc-bromine flow batteries. It applies Gradient Boosting Decision Tree (GBDT) to achieve high-precision SOC evaluation and verification at the voltage inflection point of the battery discharge. This provides a reference for developing a general algorithm for high-precision SOC calculation of flow batteries. The core work of this invention is: ① Constructing a GBDT modeling, solving, and evaluation system for efficient battery SOC evaluation based on finite voltage and current samples, emphasizing high-precision SOC identification at the voltage inflection point; ② To effectively reduce the computational cost and improve the efficiency of GBDT, establishing an SOC verification system based on discharge voltage and current limiting cycle analysis, improving the response time within a charge-discharge cycle by reducing the calculation frequency of GBDT. The following is a detailed explanation... Figure 1 The specific embodiments of the present invention are described below, including the following steps:

[0060] Step 1: Establish a comprehensive experimental platform for zinc-bromine redox flow batteries (e.g., Figure 2 Detailed parameter specifications are shown in Table 1, which includes the zinc-bromine flow battery body, BMS, EMS, PCS, and load AC motor. The BMS is used for monitoring the charge and discharge data of the battery body; the EMS supplies power to the positive and negative electrode circulation pumps of the zinc-bromine flow battery; the PCS is used for DC / AC power conversion; and the load AC motor is used to simulate inductive load. This establishes a model foundation for subsequent extraction of limited data samples and identification of the battery's SOC.

[0061] Table 1

[0062]

[0063] Step 2: Clarify the measured parameters and the identified parameters. The measured parameters include voltage, current, circulating pump speed, charge / discharge time, and measured SOC. The identified parameter is the SOC at the voltage inflection point. A battery charge / discharge comprehensive tester and BMS are used to monitor the experimental platform in real time for 24 hours, with a sampling period of 1 minute. Curves of the measured SOC, battery voltage, and battery current data changing over time are obtained (e.g., ...). Figures 3 to 4 );

[0064] Step 3: Within a complete discharge cycle, select a limited amount of voltage and current sampling data within one cycle of the ZBFB discharge process at regular intervals, extract a limited number of feature samples, and avoid sensitive outliers.

[0065] This invention focuses on solving the problem of poor accuracy in identifying the state of charge (SOC) of flow batteries. Figure 3 and Figure 4 This indicates that when both the discharge voltage and current of ZBFB are 0 (3250min-3690min), the SOC measured by the tester is not 0, therefore, the SOC at this point needs to be corrected. Feature samples extracted from the data obtained from the tester are shown in Table 2 below. The first six sets of data are used to improve the SOC identification accuracy at the voltage inflection point using GBDT. The seventh set of data falls within the interval after the voltage inflection point (3250min-3690min), and the measuring instrument displays inaccurate SOC values.

[0066] Table 2

[0067]

[0068] To improve the SOC identification accuracy at the voltage inflection point, six representative feature data points (625, 99.95), (1965, 90.29), (2176, 81.41), (2282, 99.87), (3985, 49.04), and (4298, 0.91) were extracted from the displayed charge / discharge cycles. Based on these, the corresponding voltage and current sample datasets are as follows:

[0069] Formula 1:

[0070] Step 4: Using the extracted limited feature samples as the dataset, and providing the loss function and learning rate of GBDT in real time, develop a SOC solution paradigm that can achieve real-time verification.

[0071] The GBDT algorithm, proposed by Jerome H. Friedman in 1999, belongs to the regression prediction method in the field of machine learning. Its core ideas are: ① performing multiple rounds of iterative training through residual basis analysis of weak classifiers; ② fitting the negative gradient of the loss function under the current model in each iteration; ③ controlling the loss function to descent along the gradient direction to converge as quickly as possible to reach a local optimum or global optimum. The implementation process of the GBDT model constructed in step 4 is as follows:

[0072] Step 4.1 Input feature sample data Input loss function Where N is the number of samples, Representing the Each feature sample data, , , Represents the input space. , , Represents the output space. and These represent the output data and the learner function, respectively.

[0073] Step 4.2 Initialize the weak learner:

[0074] ,

[0075] And initialize the iteration count m=1. To initialize the weak learner function, is the input feature, and c is the constant taken to minimize the loss function.

[0076] Step 4.3 For any sampled data The residual is calculated as follows:

[0077]

[0078] in, The pseudo residual of the k-th feature sample data in the m-th iteration is, i.e. This represents the negative gradient of the loss function.

[0079] The residual value calculated in step 4.4 is used as the true value for the next training iteration, and the fitted data is then used. The sample data is used for the next training iteration to obtain a new regression tree. Its corresponding leaf node region is ,in Let be the number of leaf nodes in the m-th regression tree.

[0080] Step 4.5 Applying Regression Trees Calculate the best-fit value for each leaf node:

[0081]

[0082] in, This represents the best-fit value for the leaf region. The sample data given by the learner in the (m-1)th iteration. The predicted value.

[0083] Step 4.6 Refresh the strong learner:

[0084]

[0085] in, Let I be the learner for the m-th iteration, and let I be an indicator symbol with a value of 1, which is used to fit the classifier to the node region.

[0086] Step 4.7 After completing M rounds of iterations, the final boosting regression tree is obtained:

[0087] ,

[0088] in, This represents the number of iterations, i.e., the number of trees.

[0089] Step 4.8 Obtain and analyze the optimal decision tree .

[0090] Step 4.9 Validation by comparison with the baseline sample .

[0091] Step 4.10 If If the predicted value is reasonable, output Conversely, m = m + 1, and the above calculation process is repeated starting from step 4.3.

[0092] Given the complexity of SOC calculation for ZBFB, the following related issues need to be addressed when using GBDT to evaluate the SOC at the discharge voltage inflection point: (1) Limited voltage and current sampling data within one cycle of the ZBFB discharge process need to be selected to identify discharge variation characteristics. (2) Given that the coulombic efficiency of ZBFB is time-varying relative to the discharge current, parameters such as the loss function and learning rate of GBDT need to be provided in real time. (3) The inherent data sensitivity of GBDT caused by the voltage drop at the ZBFB discharge voltage inflection point needs to be overcome. (4) To avoid overcompensation, improve the accuracy of SOC prediction and the real-time performance of the calculation, and provide the optimal SOC value for ZBFB, a SOC solution paradigm that can be verified in real time needs to be developed.

[0093] To evaluate the SOC of ZBFB in the [3520, 3690] time domain in real time, based on Table 2 and following the GBDT algorithm solution process, the first weak learner should be initialized to minimize the loss function. The first weak learner can be written in the following form:

[0094] Formula 2:

[0095] The measured residuals of SOC relative to the first weak learner are shown in Table 3. It is important to emphasize that the optimal splitting node for the sample data needs to be found based on the traversal values ​​of each feature data point. This program begins with voltage. Termination by current The core task is to calculate the variance of the data from the two sampled voltages or currents in the ZBFB algorithm.

[0096] Table 3

[0097]

[0098] Based on the above, choose and As a sample partitioning node, it is about to This is used to evaluate the optimal partitioning node for SOC within the [3520, 3690] time domain. Furthermore, based on the principle of minimum variance and the traversal value analysis of the left and right leaf nodes, the tree diagram corresponding to the [3520, 3690] time domain can be obtained as follows: Figures 5-9 As shown. This work evaluates the optimal SOC calculation path, which essentially involves training based on sample feature data to improve the SOC calculation accuracy at the ZBFB discharge voltage inflection point.

[0099] Compare Figures 5-9 As shown in Table 3, since the discharge current at the SOC to be evaluated is 2.6A, this value is at the treetop level. So choose As the optimal path for SOC evaluation, node 3 is the last strong learner computation path that needs to be considered.

[0100] Furthermore, considering that the coulombic efficiency of the battery varies with the discharge current, the learning rate of GBDT was ultimately chosen. Therefore, the SOC value of ZBFB in the time domain [3520, 3690] can be calculated as shown in the following formula:

[0101] Formula 3:

[0102] Comparing Formula 3 with Table 2 shows that, relative to the actual measured SOC value of 13.41%, the SOC evaluated using GBDT can reach 4.10%. This value is very close to the actual SOC value of 0, because at this point, the actual values ​​of the ZBFB discharge voltage and discharge current are both 0. Based on the above analysis, it can be seen that the proposed GBDT is applicable to the evaluation and verification of distorted SOC, thus providing a reference basis for correcting the SOC calculation error caused by sampling distortion at the ZBFB discharge voltage inflection point.

[0103] Step 5: Apply limit cycle theory to derive the judgment and verification criteria for calculating ZBFB high-precision SOC using GBDT. By reducing the number of times GBDT is applied, its computational efficiency within a complete charge-discharge cycle is improved, thereby enhancing the real-time performance of the proposed GBDT algorithm.

[0104] During a complete charge-discharge cycle, through the discharge voltage limit loop ( Figures 10-12 ), discharge current limiting cycle ( Figures 13-15 Analysis (see Table 4 for corresponding statistical results) determines whether GBDT is necessary for SOC verification during different discharge periods. The calculation efficiency of GBDT within a complete charge-discharge cycle can be improved by reducing the number of GBDT applications. The challenge lies in providing GBDT application criteria based on limit cycles.

[0105] Table 4

[0106]

[0107] Compare Figures 10-12 , Figures 13-15 And Table 4 shows that, Figure 10 , Figure 12 and Figure 13 , Figure 15 Correspondingly, the measured SOC at this time matches the ZBFB discharge voltage and discharge current in this domain, indicating the existence of a stable limiting cycle; such as Figure 11 , Figure 14 As shown, in the time domain [2965, 3960], the measured SOC is 13.41%, while the corresponding discharge voltage and discharge current are 0. That is, the SOC does not match the voltage and current. At this time, the voltage and voltage distortion rate, and the current and current distortion rate cannot form a stable limit cycle.

[0108] Conversely, such as Figures 10-12 and Figures 13-15As shown, the following conclusions can be drawn: (1) When the discharge voltage / current conversion rate of ZBFB can form a stable limiting cycle relative to voltage / current, there is no SOC distortion, and GBDT is not required to verify the SOC. (2) When the discharge voltage / current conversion rate of ZBFB cannot form a stable limiting cycle relative to voltage / current, or when the limiting cycle is unstable, there may be SOC distortion, and GBDT is required to verify the SOC.

[0109] In summary (1) and (2), since the present invention does not require GBDT verification of SOC at each voltage inflection point of ZBFB, it can greatly reduce the computational cost of GBDT and improve the real-time performance of this algorithm in the field of ZBFB.

[0110] This invention addresses the current lack of mature and accurate SOC measurement technology and evaluation algorithms for ZBFB (Zero-Batch Flow Battery). It applies the GBDT (Guided Flow Data Transformation) algorithm from machine learning to the real-time calculation and verification of ZBFB SOC, highlighting the following two research achievements: (1) A GBDT verification algorithm matching the SOC measurement at the ZBFB discharge voltage inflection point is provided, offering an algorithmic foundation for high-precision SOC evaluation at this point; (2) A GBDT algorithm verification and timeliness judgment criterion based on limit cycle analysis is developed, providing computational support for improving the timeliness of GBDT from an algorithmic perspective. Analysis of experimental results shows that the work effectively improves the SOC calculation accuracy at the ZBFB voltage inflection point and can provide a reference for developing a general flow battery SOC algorithm.

Claims

1. A method for verifying the state of charge (SOC) of flow batteries based on GBDT machine learning prediction, characterized in that: The method includes the following steps: Step 1: Establish a comprehensive experimental platform for zinc-bromine flow batteries, including the zinc-bromine flow battery body, BMS, EMS, PCS, and load AC motor; among them, BMS is used for monitoring the charge and discharge data of the battery body, EMS is used for powering the positive and negative electrode circulation pumps of the zinc-bromine flow battery, PCS is used to realize DC / AC power conversion, and load AC motor is used to simulate inductive load. Step 2: Clarify the measured parameters and the identified parameters. The measured parameters include voltage, current, circulating pump speed, charge and discharge time, and measured SOC. The identified parameter is the SOC at the voltage inflection point. Use a battery charge and discharge comprehensive tester and BMS to monitor the experimental platform in real time for 24 hours with a sampling period of 1 minute to obtain the curves of measured SOC, battery voltage, and battery current data changing over time. Step 3: Select limited voltage and current sampling data within multiple cycles of ZBFB discharge to identify discharge variation characteristics; extract limited feature samples at regular intervals within a whole discharge cycle to avoid sensitive outliers; Step 4: Using the extracted limited feature samples as the dataset, and providing the loss function and learning rate of GBDT in real time, the optimal SOC value of ZBFB is obtained by GBDT, and a SOC solution paradigm that can be verified in real time is developed; in Step 4, the SOC of ZBFB in the time domain after the voltage inflection point is evaluated in real time. The implementation process of the constructed GBDT model is as follows: (1) Input feature sample data Input loss function Where N is the number of samples, Representing the Each feature sample data, , , Represents the input space; , , Represents the output space; and These represent the output data and the learner function, respectively. (2) Initialize the weak learner: ， And initialize the iteration count m=1; where To initialize the weak learner function, The input features are denoted as c, which is the constant taken to minimize the loss function. (3) For any sampled data The residual is calculated as follows: in, The pseudo residual of the k-th feature sample data in the m-th iteration is, i.e. The negative gradient of the loss function; (4) The calculated residual value is used as the true value for the next training, and the fitted data is used as the true value. The sample data is used for the next training iteration to obtain a new regression tree. Its corresponding leaf node region is ,in Let be the number of leaf nodes in the m-th regression tree; (5) Applying regression trees Calculate the best-fit value for each leaf node: in, This represents the best-fit value for the leaf region. The sample data given by the learner in the (m-1)th iteration. The predicted value; (6) Refresh the strong learner: in, Let I be the learner for the m-th iteration, and let I be an indicator symbol with a value of 1, which is used to fit the classifier in the node region. (7) After completing M rounds of iterations, the final boosting regression tree is obtained: ， in, This represents the number of iterations, i.e., the number of trees. (8) Obtain and analyze the optimal decision tree ; (9) Verification by comparison with the basic sample ; (10) If If the predicted value is reasonable, output Conversely, m = m + 1, and repeat the above calculation process starting from (3); Step 5: Apply limit cycle theory to derive the judgment and verification criteria for calculating the high-precision SOC of ZBFB using GBDT. By reducing the number of times GBDT is applied, its computational efficiency within a complete charge-discharge cycle is improved, thereby enhancing the real-time performance of the proposed GBDT algorithm. In Step 5, within a complete charge-discharge cycle, through discharge voltage limit cycle and discharge current limit cycle analysis, the necessity of GBDT for SOC verification at different discharge stages is determined. By reducing the number of times GBDT is applied, its computational efficiency within a complete charge-discharge cycle is improved.

2. The method for verifying the state of charge (SOC) of a flow battery based on GBDT machine learning prediction according to claim 1, characterized in that: In step two, the sampling period is 1 minute, and the sampled measured data include the voltage, current, circulation pump speed, charge and discharge time, and measured SOC of the zinc-bromine flow battery experimental platform.

3. The method for verifying the state of charge (SOC) of a flow battery based on GBDT machine learning prediction according to claim 1, characterized in that: In step three, extraction is performed at regular intervals within the displayed discharge cycle. There are N representative small sample feature data.

4. The method for verifying the state of charge (SOC) of a flow battery based on GBDT machine learning prediction according to claim 1, characterized in that: In step four, the optimal decision tree for ZBFB at the voltage inflection point is shown in the following equation: 。

Images