Sharing-compensating efficient health assessment method for large energy storage battery system

Abstract

The invention discloses a sharing-compensation type efficient health assessment method for a large-scale energy storage battery system, and provides a state estimation modeling strategy based on sharing and difference compensation for the characteristics that the batteries in a power station are numerous and have health differences. Firstly, battery cluster degradation knowledge representation is mined, and a health state estimation basic model which can be shared by multiple batteries is constructed; then designing a residual error guided battery clustering strategy, revealing a battery health characteristic difference mode, and implementing iterative differentiation model precision addition for batteries in different modes; and finally, developing a self-adaptive mask model selection strategy, and realizing selection and efficient multiplexing of the health state estimation model for any battery. In one specific embodiment, 10 <-5 > magnitude mean square error performance is obtained in the aspect of health state estimation of a plurality of batteries; according to the method, a solution can be provided for efficient and accurate health state estimation of large-scale battery equipment.

Description

Technical Field

[0001] The present invention belongs to the technical field of energy storage power station health management, and particularly relates to a sharing-compensation type high-efficiency health assessment method for a large-scale energy storage battery system. Background Art

[0002] Lithium-ion batteries (LIBs) have gradually become the key energy storage units in the new generation of energy storage systems (ESS) due to their advantages of high energy density, long life, and low pollution. With the popularization, deployment, and application of lithium-ion batteries in the energy storage field, the management of the state of health (SOH) during their operation has gradually become a research hotspot in the industrial and academic fields.

[0003] During the long-term use process, the performance degradation of lithium-ion batteries inevitably occurs, and SOH provides an effective way to evaluate the degree of battery degradation. Usually, SOH is defined as the attenuation ratio of the current actual capacity of the battery to the nominal capacity. By accurately obtaining the SOH index of the battery, batteries with serious degradation can be detected in time, and targeted maintenance and replacement can be carried out to ensure the reliability of the energy storage system. However, in actual applications, the actual SOH of the battery is often difficult to measure and needs to be indirectly estimated using other parameters. Therefore, ensuring the accuracy of SOH is crucial for realizing reliable lithium battery health management.

[0004] Machine learning-based methods have been widely used in the SOH estimation task and have made remarkable progress. However, in a real energy storage power station, thousands of batteries are arranged in series / parallel stacks. In a large-scale battery stack, the physical characteristics, chemical characteristics, and working conditions of each battery may vary, resulting in different changes in the health status of each battery, and it is impossible to accurately describe them with a unified SOH estimation model. In order to achieve refined management of the health status of all batteries in the battery stack, accurate SOH estimation of each battery in the battery stack is required. However, most of the current existing methods only model individual lithium-ion batteries and then obtain their independent estimation results. Although this one-by-one modeling method performs well in small-scale scenarios, when faced with a large number of batteries, it will face heavy computational resource requirements and storage burdens, thus limiting the practical application of these methods in large-scale energy storage systems. Therefore, how to develop a lightweight SOH estimation modeling method while ensuring the model accuracy, optimize the computational efficiency and resource allocation of model training, and achieve efficient and feasible SOH estimation for a large-scale battery cluster is an urgent problem to be solved. Summary of the Invention

[0005] The object of the present invention is to provide a shared-compensated high-efficiency health assessment method for large-scale energy storage battery systems in view of the deficiencies of the prior art.

[0006] The object of the present invention is achieved by the following technical solutions: A shared-compensated high-efficiency health assessment method for large-scale energy storage battery systems, comprising the following steps:

[0007] (1) Obtain the total set of battery sampling data and the corresponding total set of health status indicators of the energy storage battery system, and construct the total set of expert knowledge representations of battery cluster degradation; (2) Randomly select the data of B0 batteries from the total set of health status indicators and the total set of expert knowledge representations of battery cluster degradation, perform standardized processing, and then train a long short-term memory network model to obtain a local shared model for health status estimation; (3) Based on the local shared model for health status estimation, retain B1 batteries greater than or equal to the deviation level threshold and perform residual spectrum clustering operations to obtain P clustering clusters; (4) Use the P clustering clusters to train multiple health status estimation compensation models respectively, and after removing duplicates, obtain the final set of health status estimation models; (5) Use the standardized data corresponding to the remaining batteries and the final set of health status estimation models to train a multi-layer perceptron model to obtain an adaptive mask selection network; (6) Use the adaptive mask selection network and the final set of health status estimation models to estimate the final health status estimation indicators corresponding to the batteries to be estimated.

[0008] Further, the obtaining of the total set of battery sampling data and the corresponding total set of health status indicators of the energy storage battery system specifically includes: Obtain the measurement point data set at all sampling moments in the discharge stage of all charge-discharge cycles during the full life cycle of B batteries in the energy storage battery system to form the total set of battery sampling data; wherein, the measurement point data includes voltage values and current values; Obtain the health status indicators of B batteries in the energy storage battery system during all charge-discharge cycles in the full life cycle to form the total set of health status indicators.

[0009] Further, the construction of the total set of expert knowledge representations of battery cluster degradation specifically includes: Obtain the corresponding capacity increment index vector according to the measurement point data vector of any one charge-discharge cycle of any one battery in the total set of battery sampling data, and perform regularization operations on it to obtain the corresponding regularized sample vector; Perform a segmentation operation on the regularized sample vectors, and obtain multiple local profile indicators based on the capacity increment indicators and the corresponding voltage values in each subset after segmentation, so as to obtain the corresponding degradation expert knowledge representation vectors of the battery cluster; Repeat the above steps for all measurement point data vectors of each charge-discharge cycle of each battery in the total set of battery sampling data to obtain the total set of degradation expert knowledge representations of the battery cluster.

[0010] Further, the regularization operation specifically includes: Set the curve truncation index to N, and compare it with the total number of sampling times in the discharge stage of any charge-discharge cycle of any battery If , sort the voltage values of all sampling times in this stage from small to large, and use the first N voltage values after sorting and the corresponding capacity increment indicators as the corresponding regularized sample vectors; if , sort the voltage values of all sampling times in this stage from small to large, and use the first voltage values and the th voltage value and the corresponding capacity increment indicators as the corresponding regularized sample vectors; where represents the total number of sampling times in the discharge stage of the cth charge-discharge cycle of the bth battery.

[0011] Further, the step (2) specifically includes: Randomly select the data corresponding to B0 batteries from the total set of degradation expert knowledge representations of the battery cluster and the total set of health status indicators, and perform standardization processing using the Z-score standardization method. Then randomly select batteries from the standardized set of degradation expert knowledge representations of the battery cluster corresponding to the batteries and input them into the long short-term memory network model to obtain the first total set of health status prediction indicators, and calculate the mean square error with the standardized health status indicator vectors corresponding to the batteries selected, to obtain the first loss function; train the long short-term memory network model based on the first loss function to obtain the trained long short-term memory network model as the local shared model for estimating the health status of B0 batteries.

[0012] Further, the step (3) specifically includes the following sub-steps: (3.1) The set of battery cluster degradation expert knowledge representations after standardization corresponding to the selected B0 batteries is input into the corresponding local shared model for health state estimation to obtain the total set of second health state prediction indicators, and the total set of first health state estimation residuals is obtained by calculating the health state estimation residuals between the total set of second health state prediction indicators and the corresponding health state indicator vector after standardization; (3.2) Obtain the set of average estimation deviation levels based on the total set of first health state estimation residuals, and retain the batteries with an average estimation deviation level greater than or equal to the deviation level threshold to obtain B1 retained batteries; (3.3) For the B1 retained batteries, obtain the residual embedding vectors according to their respective health state estimation residual vectors and the natural number S c Cluster the B1 retained batteries using spectral clustering with the residual embedding vectors as features to obtain P clusters.

[0013] Further, step (4) specifically includes the following sub-steps: (4.1) For all the clusters obtained in step (3.3), input the set of battery cluster degradation expert knowledge representations after standardization corresponding to all the batteries in each cluster into the support vector regression model to obtain the total set of third health state prediction indicators, and calculate the mean square error between the total set of third health state prediction indicators and the health state estimation residual vectors corresponding to all the batteries in this cluster to obtain the second loss function; train the support vector regression model based on the second loss function, and update the local shared model for health state estimation corresponding to the batteries in this cluster to the sum of the current local shared model for health state estimation corresponding to the batteries in this cluster and the trained support vector regression model, and determine whether the number of summation times experienced by the currently updated local shared model for health state estimation is greater than the threshold M p If not, execute step (4.2); otherwise, record the currently updated local shared model for health state estimation as a health state estimation compensation model and end the operation on this cluster; (4.2) Then, use the set of battery cluster degradation expert knowledge representations after standardization corresponding to all the batteries in this cluster as the input, repeat steps (3.1) - (3.2) to obtain the number of retained batteries in this cluster; if the number of retained batteries is 0, record the local shared model for health state estimation corresponding to this cluster as a health state estimation compensation model and end the operation on this cluster; otherwise, repeat steps (3.3) and (4.1) - (4.2) for the set of battery cluster degradation expert knowledge representations after standardization corresponding to the retained batteries until all clusters are traversed and the operation is completed; (4.3) Collect all the health state estimation compensation models obtained in steps (4.1) and (4.2), and remove the duplicate health state estimation compensation models, so as to form the final health state estimation model set.

[0014] Further, step (5) specifically includes: Normalize the data corresponding to the remaining batteries in the total set of battery cluster degradation expert knowledge representations and the total set of health state indicators by using the Z-score normalization method, and then input the normalized battery cluster degradation expert knowledge representation sets corresponding to the remaining batteries into each final health state estimation model in the final health state estimation model set respectively to obtain the fourth total set of health state prediction indicators, and calculate the absolute difference with the normalized health state indicator vectors corresponding to the remaining batteries to obtain the total set of model performance indicators and perform masking calculation to obtain the masked total set of model performance indicators; Input the normalized battery cluster degradation expert knowledge representation sets corresponding to the remaining batteries into the multi-layer perceptron model to obtain the total set of health state adaptive prediction indicators and perform masking calculation to obtain the masked total set of health state adaptive prediction indicators; Calculate the cross entropy between the masked total set of model performance indicators and the masked total set of health state adaptive prediction indicators to obtain the third loss function; and train the multi-layer perceptron model based on the third loss function to obtain the trained multi-layer perceptron model as the adaptive mask selection network.

[0015] Further, step (6) specifically includes: Construct the corresponding battery cluster degradation expert knowledge representation vector according to the measured point data at all sampling moments in the discharge stage of any charge-discharge cycle of the battery to be estimated, and perform normalization processing, and then input it into the adaptive mask selection network to obtain the corresponding model performance estimation indicator vector and perform masking calculation to obtain the masked model performance estimation indicator vector; When the h-th masked model performance estimation indicator in the masked model performance estimation indicator vector is 1, select the h-th final health state estimation model from the final health state estimation model set as the estimation prediction model; input the normalized battery cluster degradation expert knowledge representation vector of the battery to be estimated into the estimation prediction model to obtain the model output result, and perform anti-normalization processing to obtain the final health state estimation indicator corresponding to this charge-discharge cycle of the battery to be estimated.

[0016] The beneficial effects of the present invention are as follows: In view of the characteristics of battery stacks arranged in piles in the energy storage power station, with a large number and different degrees of differences among them, the present invention proposes a model sharing and differential compensation mechanism to achieve accurate and efficient estimation of the health state of large-scale battery systems. First, it excavates the degradation knowledge representation of the battery cluster and constructs a basic health state estimation model that can be shared by multiple batteries. Then, it designs a residual-guided battery clustering strategy to reveal the differential pattern of battery health characteristics, and implements iterative differential model accuracy compensation for different batteries. Finally, it develops an adaptive mask model selection strategy to achieve the selection and efficient reuse of the health state estimation model for any battery. Compared with the existing battery system health assessment methods, the health state estimation modeling method of resource sharing and differential training compensation proposed by this method avoids unnecessary resource waste and computational burden, greatly optimizes the modeling efficiency of large-scale energy storage battery stacks, and provides important support for the health state estimation of refined battery devices. Description of the Drawings

[0017] Figure 1 It is a flowchart of the shared-compensation type efficient health assessment method for the large-scale energy storage battery system of the present invention; Figure 2 It is a schematic diagram of the residual-guided model compensation strategy corresponding to steps (3)-(4) of the present invention; Figure 3 It is a result diagram of the final health state index estimation of Battery No. 1 in the LSTM method; Figure 4 It is a result diagram of the final health state index estimation of Battery No. 1 in the GRU method; Figure 5 It is a result diagram of the final health state index estimation of Battery No. 1 in the method described in the present invention; Figure 6 It is a result diagram of the final health state index estimation of Battery No. 2 in the LSTM method; Figure 7 It is a result diagram of the final health state index estimation of Battery No. 2 in the GRU method; Figure 8 It is a result diagram of the final health state index estimation of Battery No. 2 in the method described in the present invention; Figure 9 It is a result diagram of the final health state index estimation of Battery No. 3 in the LSTM method; Figure 10 It is a result diagram of the final health state index estimation of Battery No. 3 in the GRU method; Figure 11 It is a result diagram of the final health state index estimation of Battery No. 3 in the method described in the present invention. Detailed Embodiment

[0018] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the protection scope of the present invention.

[0019] See Figure 1 , the shared-compensated high-efficiency health assessment method for large-scale energy storage battery systems of the present invention specifically includes the following steps:

[0020] During the long-term use of lithium batteries, the phenomenon of performance degradation is inevitably faced. The degradation characteristics of different battery monomers are different, and an efficient state of health estimation strategy is required to accurately perceive the state of health of multiple batteries. In this embodiment, a set of measured point data at all sampling moments in the discharge stage of all charge-discharge cycles in the entire life cycle of 14 batteries in any energy storage battery system and the corresponding state of health index vector are collected as a training set to establish an SOH estimation model, and the set of measured point data at all sampling moments in the discharge stage of all charge-discharge cycles in the entire life cycle of another 14 batteries and the corresponding state of health index vector are used for state of health estimation testing. Among them, the batteries involved in the dataset are NCA ternary lithium batteries, the nominal capacity of the battery is 3500 mA, the charge-discharge mode of the battery is constant current-constant voltage (CC-CV) charging, the charging current rate in the constant current charging stage is 0.5C, the cut-off voltage is 4.2V, the voltage is maintained at 4.2V in the constant voltage charging stage, and the charging cut-off current rate is 0.05C; the discharge mode is constant current (CC) discharge, the discharge current rate is 1C, and the cut-off voltage is 2.65V; the experimental temperature is 45°C, and the measurement signals of the battery operation data include voltage signals and current signals.

[0021] (1) Obtain the total set of battery sampling data of the energy storage battery system and the corresponding total set of state of health (SOH) indicators, and construct the total set of expert knowledge representations of battery cluster degradation.

[0022] In this embodiment, obtaining the total set of battery sampling data of the energy storage battery system and the corresponding total set of state of health indicators specifically includes: obtaining a set of measured point data at all sampling moments in the discharge stage of all charge-discharge cycles in the entire life cycle of B = 14 batteries in the energy storage battery system to form the total set of battery sampling data. The set of measured point data of each battery is composed of the measured point data at all sampling moments in the discharge stage of all charge-discharge cycles in the entire life cycle of each battery, and each measured point data is composed of the current value and voltage value at the sampling moment. The total set of battery sampling data is expressed as , where represents the set of measured point data at all sampling moments in the discharge stage of all charge-discharge cycles of the b-th battery during its entire life cycle. , represents the vector of measured point data at all sampling moments in the discharge stage of the c-th charge-discharge cycle of the b-th battery. represents the total number of charge-discharge cycles experienced by the b-th battery, and the superscript T represents the transpose of a matrix or vector. , represents the measured point data at the k-th sampling moment in the discharge stage of the c-th charge-discharge cycle of the b-th battery. represents the total number of sampling moments in the discharge stage of the c-th charge-discharge cycle of the b-th battery. , represents the current value at the k-th sampling moment in the discharge stage of the c-th charge-discharge cycle of the b-th battery. represents the voltage value at the k-th sampling moment in the discharge stage of the c-th charge-discharge cycle of the b-th battery. Obtain the health state index of B = 14 batteries in the energy storage battery system during all charge-discharge cycles in their entire life cycle to form the total set of health state indices; among them, the health state indices corresponding to all charge-discharge cycles of each battery during its entire life cycle form the health state index vector corresponding to that battery, and the health state index vectors of B = 14 batteries in the energy storage battery system form the total set of health state indices. The total set of health state indices is denoted as , where represents the health state index vector corresponding to the b-th battery. , represents the health state index corresponding to the c-th charge-discharge cycle of the b-th battery during its entire life cycle.

[0023] In this embodiment, a total set of expert knowledge representations of battery cluster degradation is constructed, which specifically includes the following sub-steps:

[0024] (1.1) According to the vector of measured point data at all sampling moments in the discharge stage of the c-th charge-discharge cycle of the b-th battery in the total set of battery sampling data E calculate the capacity increment index vector corresponding to the c-th charge-discharge cycle of the b-th battery, where

[0025] represents the capacity increment index corresponding to the k-th sampling moment in the c-th charge-discharge cycle of the b-th battery, and its calculation formula is: Represents the time value at the k-th sampling moment in the c-th charge-discharge cycle of the b-th battery.

[0026] (1.2) Perform a regularization operation on the capacity increment index vector corresponding to the c-th charge-discharge cycle of the b-th battery obtained in step (1.1) to obtain the corresponding regularized sample vector.

[0027] Further, the regularization operation specifically includes: setting the curve truncation index to N, and comparing the total number of sampling moments in the discharge stage of the c-th charge-discharge cycle of the b-th battery with the set curve truncation index N: If , then sort all the voltage values at all sampling moments in the discharge stage of the c-th charge-discharge cycle of the b-th battery from smallest to largest, and the sorted voltage values are denoted as , where represents the d-th voltage value sorted from smallest to largest, and take the first N sorted voltage values and the corresponding capacity increment indicators as the regularized sample vector , where , represents the capacity increment indicator corresponding to the voltage value ; if , then sort all the voltage values at all sampling moments in the discharge stage of the c-th charge-discharge cycle of the b-th battery from smallest to largest, and the sorted voltage values are denoted as , where represents the f-th voltage value sorted from smallest to largest, , and take the first voltage values and the -th voltage value and the corresponding capacity increment indicators as the regularized sample vector , where , represents the capacity increment indicator corresponding to the voltage value . In this embodiment, the value of N is 250.

[0028] (1.3) Perform a splitting operation on the regularized sample vector, and obtain multiple local profile indicators according to the capacity increment indicators and the corresponding voltage values in each subset after splitting, so as to obtain the corresponding battery cluster degradation expert knowledge representation vector.

[0029] Specifically, perform a splitting operation on the regularized sample vector to split the regularized sample vector Equally divide it into L subsets at equal intervals, where L is a factor of N; then calculate the maximum value of the capacity increment index, the voltage value corresponding to the maximum value of the capacity increment index, the minimum value of the capacity increment index, the voltage value corresponding to the minimum value of the capacity increment index, and the average value of the capacity increment index in each subset, and respectively use them as local profile indicators to obtain the battery cluster degradation expert knowledge representation vector corresponding to the c-th charge-discharge cycle of the b-th battery , and this battery cluster degradation expert knowledge representation vector consists of local profile indicators, denotes the battery cluster degradation expert knowledge representation vector the j-th local profile indicator in, and J represents the total number of local profile indicators in the battery cluster degradation expert knowledge representation vector . In this embodiment, the value of L is 5.

[0030] (1.4) Repeat steps (1.1)-(1.3) for the measurement point data vectors at all sampling moments in the discharge stage of each charge-discharge cycle of each battery in the total set E of battery sampling data, to obtain the total set of battery cluster degradation expert knowledge representations composed of the battery cluster degradation expert knowledge representation sets of B batteries , where represents the battery cluster degradation expert knowledge representation set of the b-th battery, and the battery cluster degradation expert knowledge representation set of the b-th battery is composed of the battery cluster degradation expert knowledge representation vectors corresponding to all charge-discharge cycles experienced by this battery and is denoted as .

[0031] It should be noted that by constructing the total set of battery cluster degradation expert knowledge representations, a cluster degradation representation scheme that can be executed for all batteries in the energy storage battery system can be provided.

[0032] (2) Randomly select the data of B0 batteries from the total set of health state indicators and the total set of battery cluster degradation expert knowledge representations, perform standardization processing, and then train a long short-term memory (LSTM) model to obtain a local shared model for health state estimation.

[0033] (2.1) Randomly select from the total set of battery cluster degradation expert knowledge representations and the total set of health state indicators the battery cluster degradation expert knowledge representation sets and health state indicator vectors corresponding to the batteries as the initial health state estimation data set , where , denotes the selected The initial total set of battery cluster degradation expert knowledge representations composed of the battery cluster degradation expert knowledge representation sets corresponding to each battery Denote the selected The initial total set of health state index vectors composed of the health state index vectors corresponding to each battery; the initial total set of battery cluster degradation expert knowledge representations Denote as , Denote the th battery cluster degradation expert knowledge representation set corresponding to the selected , Denote the th battery cluster degradation expert knowledge representation vector corresponding to the selected th charge-discharge cycle of the battery Denote the total number of charge-discharge cycles experienced by the selected th battery , Denote the jth battery cluster degradation expert knowledge representation in the battery cluster degradation expert knowledge representation vector ; the initial total set of health state index vectors Denote as , Denote the health state index vector corresponding to the selected th battery , Denote the health state index corresponding to the selected th battery at the th charge-discharge cycle in its entire life cycle. In this embodiment, takes the value of 9, that is, randomly select the data corresponding to 9 batteries out of 14 batteries.

[0034] (2.2) For any battery cluster degradation expert knowledge representation in the initial total set of battery cluster degradation expert knowledge representations , perform normalization processing. The normalization processing is carried out by the Z-score normalization method, and finally the normalized initial battery cluster degradation expert knowledge representation can be obtained, and its calculation formula is as follows: ; ; ;

[0035] In the formula, denotes the average value of the battery cluster degradation expert knowledge representation, denotes the standard deviation of the battery cluster degradation expert knowledge representation.

[0036] (2.3) For any one of the health state indicators in the initial total set of health state indicators perform standardization processing. The standardization processing is carried out by the Z - score standardization method, and finally the initial health state indicators after standardization processing can be obtained , and the calculation formula is as follows: ; ; ; ;

[0037] In the formula, represents the average value of the health state indicator, represents the standard deviation of the health state indicator.

[0038] (2.4) Repeat step (2.2) for each battery cluster degradation expert knowledge representation in the initial total set of battery cluster degradation expert knowledge representations to obtain the initial total set of battery cluster degradation expert knowledge representations after standardization processing , denoted as , where represents the set of initial battery cluster degradation expert knowledge representations after standardization processing corresponding to the selected th battery, , represents the vector of initial battery cluster degradation expert knowledge representations after standardization processing for the th charge - discharge cycle of the selected th battery, , represents the vector of initial battery cluster degradation expert knowledge representations after standardization processing for the th charge - discharge cycle of the selected th battery and the jth battery cluster degradation expert knowledge representation in it.

[0039] Repeat step (2.3) for each health state indicator in the initial total set of health state indicators to obtain the initial total set of health state indicators after standardization processing , denoted as , where represents the vector of initial health state indicators after standardization processing corresponding to the selected th battery, , represents the th battery's The health state index after standardization corresponding to the first charge-discharge cycle.

[0040] Subsequently, according to the total set of expert knowledge representations of the initial battery cluster degradation after standardization and the total set of initial health state indices after standardization , the initial health state estimation data set after standardization is obtained , denoted as .

[0041] (2.5) From the initial health state estimation data set after standardization randomly select sets of expert knowledge representations of the initial battery cluster degradation after standardization corresponding to the batteries and the initial health state index vectors after standardization as the local health state estimation data set , denoted as , where , represents the total set of local battery cluster degradation expert knowledge representations composed of the sets of expert knowledge representations of the initial battery cluster degradation after standardization corresponding to the selected batteries, represents the total set of local health state indices composed of the initial health state index vectors after standardization corresponding to the selected batteries. The total set of local battery cluster degradation expert knowledge representations is denoted as , represents the set of expert knowledge representations of the initial battery cluster degradation after standardization corresponding to the th selected battery, , represents the th charge-discharge cycle of the th selected battery, the expert knowledge representation vector of the initial battery cluster degradation after standardization, represents the total number of charge-discharge cycles experienced by the th selected battery; , represents the th charge-discharge cycle of the th selected battery, the expert knowledge representation vector of the initial battery cluster degradation after standardization in the represents the jth battery cluster degradation expert knowledge representation in the total set of local health state indices. The total set of local health state indices , represents the initial health state index vector after standardization corresponding to the th selected battery, , indicates the health state index after standardized processing corresponding to the th charge-discharge cycle of the th extracted battery. In this embodiment, takes the value of 3, that is, 3 sets of data corresponding to 3 randomly selected batteries from the data of 9 batteries are taken.

[0042] (2.6) Subsequently, the total set of local battery cluster degradation expert knowledge representations is input into the long short-term memory network model to obtain the total set of first health state prediction indicators , denoted as , where represents the first health state prediction indicator vector corresponding to the th extracted battery, , where represents the first health state prediction indicator corresponding to the th charge-discharge cycle of the th extracted battery, , represents the mapping function of the long short-term memory network model. Subsequently, the total set of first health state prediction indicators and the total set of local health state indicators are used to calculate the mean square error to obtain the first loss function ; and based on the first loss function , the long short-term memory network model is trained with the goal of minimizing the first loss function to adjust the parameters of the long short-term memory network model until the preset number of training rounds is reached, so as to obtain the trained long short-term memory network model as the local shared model for estimating the health state of all B0 batteries . In this training process of this embodiment, the number of training rounds is , and its value is 20; the learning rate is , and its value is 1×10 -3 , and the training batch size is , and its value is 32.

[0043] Furthermore, the calculation formula of the first loss function is:

[0044] It should be noted that the local shared model for estimating the health state obtained through the current step (2) can be used as the health state estimation model that can be shared by all batteries, and there is no need to construct models for all batteries separately, which can effectively reduce the overhead of the number of models.

[0045] (3) Based on the health state estimation local sharing model, retain B1 batteries greater than or equal to the deviation level threshold and perform residual spectrum clustering operation to obtain P clustering clusters. The schematic diagram of the steps is as Figure 2 shown.

[0046] Specifically, obtain the total set of second health state prediction indicators through the health state estimation local sharing model, and then calculate the health state estimation residual. Calculate the retained average estimation deviation level of B0 selected batteries, retain B1 batteries greater than or equal to the deviation level threshold and perform residual spectrum clustering operation to obtain P clustering clusters. Specifically, it includes the following sub-steps:

[0047] (3.1) First, input the total set of standardized initial battery cluster degradation expert knowledge representations corresponding to the B0 selected batteries in the standardized initial health state estimation data set into the health state estimation local sharing model to obtain the total set of second health state prediction indicators , denoted as , where represents the second health state prediction indicator vector corresponding to the selected th battery, , represents the second health state prediction indicator corresponding to the th extracted selected battery for the th charge-discharge cycle, , , represents the mapping function of the health state estimation local sharing model .

[0048] Then, based on the total set of second health state prediction indicators and the total set of standardized initial health state indicators, calculate the health state estimation residual for all charge-discharge cycles corresponding to the batteries between the second health state prediction indicators and the corresponding standardized initial health state indicators to obtain the total set of first health state estimation residuals , denoted as , where represents the health state estimation residual vector corresponding to the selected th battery, , represents the health state estimation residual corresponding to the th extracted selected battery for the th charge-discharge cycle, , .

[0049]

[0049] (3.2) According to the total set of first health state estimation residuals Calculate the average estimated deviation level set , denoted as , where represents the average estimated deviation level corresponding to the selected th battery, .

[0050] And according to the average estimated deviation level set Compare the average estimated deviation level corresponding to each battery with the deviation level threshold : If , then the standardized initial battery cluster degradation expert knowledge representation set and the standardized initial health state index vector corresponding to this battery are removed from the total set of standardized initial battery cluster degradation expert knowledge representations , and the health state estimation residual vector corresponding to this battery is removed from the first health state estimation residual total set , and the average estimated deviation level corresponding to this battery is removed from the average estimated deviation level set ; If , then the standardized initial battery cluster degradation expert knowledge representation set and the standardized initial health state index vector corresponding to this battery are retained, and the health state estimation residual vector and the average estimated deviation level corresponding to this battery are retained; After completing the level threshold comparison for batteries, the number of retained batteries is . In this embodiment, the value of the deviation level threshold is 0.03.

[0051] (3.3) Set the natural number S c , for retained batteries, divide the length of the respective corresponding health state estimation residual vector by S c , obtain the quotient as S w , intercept the first S c * S w elements of the respective corresponding health state estimation residual vector to form the second residual vector; Starting from the first element of the second residual vector, at intervals of S wExtract the element values one by one to form a residual embedding vector. Using the residual embedding vector as a feature, perform spectral clustering on the retained batteries to obtain P clustering clusters. Denote that the p-th clustering cluster contains retained batteries respectively, where . In this embodiment, S c is selected as 50, and the number of clustering clusters is taken as 2, that is, P = 2.

[0052] It should be noted that the obtained P clustering clusters indicate the differences in the health characteristics of different batteries. At the same time, this clustering operation directly reflects the gap between the health state estimation results of the constructed local shared model of the health state and the true health state estimation results on the selected batteries, providing a direct basis for subsequent customized precision compensation for different batteries.

[0053] (4) Use the P clustering clusters to train multiple health state estimation compensation models respectively, and obtain the final set of health state estimation models after removing duplicates. The schematic diagram of the steps is as Figure 2 shown. Specifically, use the standardized initial battery cluster degradation expert knowledge representation sets corresponding to all batteries in each of the P clustering clusters to train the corresponding health state estimation models, and obtain the final set of health state estimation models after removing duplicates. After the above B0 batteries are clustered, for the batteries in each clustering cluster, using their respective health state estimation residuals as the training target, train the corresponding local shared health state estimation models. The local shared health state estimation models can use only simple machine learning models. By adding a customized model according to the residual superposition on the original basis, further implement the improvement of the health state estimation accuracy with high efficiency and customization.

[0054] (4.1) Based on all the clustering clusters obtained in step (3.3), for each clustering cluster, perform the following steps in sequence: Take the standardized initial battery cluster degradation expert knowledge representation set corresponding to the retained batteries in the p-th clustering cluster as the total standardized initial battery cluster degradation expert knowledge representation set of the p-th clustering cluster , and take the health state estimation residual vector corresponding to the retained batteries as the total second health state estimation residuals set .

[0055] Subsequently, input the total standardized initial battery cluster degradation expert knowledge representation set of the p-th clustering cluster into a support vector regression (SVR) model to obtain the total third health state prediction index set ; Subsequently, the total set of the third health status prediction indicators and the total set of the second health status estimation residuals are used to calculate the mean square error to obtain the second loss function; and the support vector regression model is trained based on the second loss function. With minimizing the second loss function as the optimization objective, the parameters of the support vector regression model are adjusted until the model reaches the tolerance condition for stopping training, so as to obtain a trained support vector regression model , where represents the trained support vector regression model corresponding to the p-th clustering cluster; and the local shared model for estimating the health status of the batteries in the p-th clustering cluster is updated to the current local shared model for estimating the health status of the batteries in this clustering cluster which is the sum of the current local shared model for estimating the health status of the batteries in the p-th clustering cluster and the trained support vector regression model , that is , where represents the updated local shared model for estimating the health status of the batteries corresponding to the p-th clustering cluster. In this embodiment, the kernel function of the support vector regression model is selected as the rbf function, and the tolerance for stopping training is -3 , and its value is 1×10 , the regularization coefficient is , and its value is 1, and the error tolerance bandwidth is

[0056] Subsequently, it is judged whether the number of summation times experienced by the local shared model for estimating the health status in the updated local shared model for estimating the health status corresponding to the p-th clustering cluster is greater than the threshold . If it is not greater than the threshold , then step (4.2) is executed. Otherwise, the updated local shared model for estimating the health status corresponding to the p-th clustering cluster is used as the health status estimation compensation model , and the operation on this clustering cluster is ended. In this embodiment, the value of the threshold is 10.

[0057] (4.2) Then, the total set of the battery cluster degradation expert knowledge representations after standardization processing of the p-th clustering cluster is input into the updated local shared model for estimating the health status , and steps (3.1) - step (3.2) are repeated to obtain that the number of batteries retained in the p-th clustering cluster is ; if then the updated local shared model for estimating the health status is used as the health status estimation model , and the operation on this clustering cluster is ended; otherwise, the retained Repeat step (3.3) and steps (4.1)-(4.2) for the set of initial battery cluster degradation expert knowledge representations after standardization corresponding to each battery until all clusters are traversed and the operation is completed.

[0058] (4.3) Repeat steps (4.1)-(4.2) for each cluster, collect the health state estimation compensation models corresponding to all clusters, and remove duplicate health state estimation compensation models to form the final set of health state estimation models , denoted as , where represents the m-th final health state estimation model, and M represents the total number of health state estimation models in the set of final health state estimation models.

[0059] (5) Use the standardized data corresponding to the remaining batteries and the set of final health state estimation models to train the multi-layer perceptron model to obtain an adaptive mask selection network. This step is based on the battery data that did not participate in the model training in steps (1)-(4). The constructed adaptive mask selection network can take the cluster degradation expert knowledge representation of any battery as input and output the corresponding optimal health state estimation model selection scheme. Therefore, the purpose of this step is to enable the models in the model library constructed in step (4) to be used for the health state estimation tasks of more unseen batteries. It specifically includes the following sub-steps:

[0060] (5.1) Use the total set of battery cluster degradation expert knowledge representations and the remaining battery cluster degradation expert knowledge representation sets and health state index vectors corresponding to the batteries as the adaptive mask training data set , where represents the total set of adaptive battery cluster degradation expert knowledge representations composed of the battery cluster degradation expert knowledge representation sets corresponding to the remaining batteries, represents the total set of adaptive health state indexes composed of the health state index vectors corresponding to the remaining batteries; the total set of adaptive battery cluster degradation expert knowledge representations is denoted as , represents the battery cluster degradation expert knowledge representation set corresponding to the -th unselected battery, , represents the battery cluster degradation expert knowledge representation vector of the -th charge-discharge cycle corresponding to the -th unselected battery, represents the The total number of charge-discharge cycles experienced by a battery , represents the degradation expert knowledge representation vector of the j-th battery cluster in the battery cluster degradation expert knowledge representation vector set ; The total set of adaptive health state indicators is denoted as , represents the health state indicator vector corresponding to the i-th battery that has not been selected , , represents the health state indicator corresponding to the i-th battery that has not been selected during the k-th charge-discharge cycle in its entire life cycle , .

[0061] (5.2) For each battery cluster degradation expert knowledge representation in the total set of adaptive battery cluster degradation expert knowledge representations , perform standardization processing using the Z-score standardization method, that is, repeat step (2.2) to obtain the total set of standardized adaptive battery cluster degradation expert knowledge representations , denoted as , where represents the set of standardized adaptive battery cluster degradation expert knowledge representations corresponding to the i-th battery that has not been selected , , represents the standardized adaptive battery cluster degradation expert knowledge representation vector of the i-th battery that has not been selected during the k-th charge-discharge cycle , , , represents the standardized adaptive battery cluster degradation expert knowledge representation vector of the i-th battery that has not been selected during the k-th charge-discharge cycle , , represents the degradation expert knowledge representation of the j-th battery cluster in the battery cluster degradation expert knowledge representation vector set ; represents the total number of charge-discharge cycles experienced by the i-th battery that has not been selected

[0062] For each health state indicator in the total set of adaptive health state indicators , perform standardization processing using the Z-score standardization method, that is, repeat step (2.3) to obtain the total set of standardized adaptive health state indicators , denoted as , where represents the vector of standardized adaptive health state indicators corresponding to the i-th battery that has not been selected , , represents the normalized health state indicator corresponding to the th charge-discharge cycle of the unselected

[0063] Subsequently, based on the total set of normalized adaptive battery cluster degradation expert knowledge representations and the total set of normalized adaptive health state indicators , the normalized adaptive health state estimation data set is obtained, denoted as .

[0064] (5.3) Input the total set of normalized adaptive battery cluster degradation expert knowledge representations into each final health state estimation model in the final health state estimation model set respectively, and the total set of fourth health state prediction indicators is obtained, denoted as , where represents the set of fourth health state prediction indicators corresponding to the unselected th battery, represents the vector of fourth health state prediction indicators corresponding to the th charge-discharge cycle of the unselected th battery, represents the fourth health state prediction indicator corresponding to the th charge-discharge cycle of the selected th battery in the th final health state estimation model, represents the mapping function of the th final health state estimation model

[0065] (5.4) Subsequently, based on the total set of fourth health state prediction indicators and the total set of normalized adaptive health state indicators , calculate the absolute difference between the fourth health state prediction indicators corresponding to all charge-discharge cycles of the th battery and the normalized adaptive health state indicators, and the total set of model performance indicators is obtained, denoted as , where represents the set of model performance indicators corresponding to the selected th battery, Indicates the vector of model performance metrics corresponding to the th charge-discharge cycle of the th unselected battery, , Indicates the model performance metric corresponding to the th charge-discharge cycle of the th selected battery in the mth final health state estimation model, .

[0066] (5.5) Subsequently, a masking calculation is performed on the total set of model performance metrics to obtain the masked total set of model performance metrics; where the masked model performance metric at the i-th row and j-th column in the masked total set of model performance metrics is calculated as follows:

[0067] In the formula, represents any model performance metric in the i-th row of the total set of model performance metrics.

[0068] (5.6) Input the total set of adaptively represented expert knowledge on battery cluster degradation after standardization into a Multilayer Perceptron (MLP) model to obtain the total set of adaptively predicted health state metrics, denoted as , where represents the set of adaptively predicted health state metrics corresponding to the th unselected battery, , represents the vector of adaptively predicted health state metrics corresponding to the th charge-discharge cycle of the th unselected battery, , represents the mth adaptively predicted health state metric output by the MLP model corresponding to the th charge-discharge cycle of the th selected battery. Subsequently, a masking calculation is performed on the total set of adaptively predicted health state metrics to obtain the masked total set .

[0069] (5.7) Subsequently, the masked total set The total set of adaptive prediction indicators for the masked health status Perform cross-entropy calculation to obtain the third loss function; and train the multi-layer perceptron model based on the third loss function, with the goal of minimizing the third loss function, adjust the parameters of the multi-layer perceptron model until the preset number of training rounds is reached, so as to obtain the trained multi-layer perceptron model as the adaptive mask selection network. In this training process of this embodiment, the number of training rounds is , and its value is 100; the learning rate is , and its value is 0.001; the training batch size is , and its value is 64.

[0070] (6) Use the adaptive mask selection network and the final health status estimation model set to estimate the final health status estimation index corresponding to the battery to be estimated.

[0071] (6.1) Construct the battery cluster degradation expert knowledge representation vector of the battery to be estimated according to the measurement point data at all sampling moments in the discharge stage of any charge-discharge cycle of the battery to be estimated ; then perform standardization processing on the battery cluster degradation expert knowledge representation vector of the battery to be estimated to obtain the standardized battery cluster degradation expert knowledge representation vector of the battery to be estimated ; then input the standardized battery cluster degradation expert knowledge representation vector of the battery to be estimated into the adaptive mask selection network to obtain the model performance estimation index vector corresponding to the battery to be estimated ; and perform mask calculation on the model performance estimation index vector to obtain the masked model performance estimation index vector .

[0072] (6.2) When the h-th masked model performance estimation index in the masked model performance estimation index vector is 1, select the h-th final health status estimation model from the final health status estimation model set as the estimation prediction model ; then input the standardized battery cluster degradation expert knowledge representation vector of the battery to be estimated into the estimation prediction model to obtain the model output result , denoted as , where represents the mapping function of the estimation prediction model ; finally, the model output result Perform denormalization to obtain the estimated final state of health index corresponding to any charge-discharge cycle of the battery to be estimated. : .

[0073] It should be understood that, as can be seen from the description of step (6) above, in the online application stage, the method of the present invention can be applied to any battery in the system that has not participated in the modeling of steps (1) to (5), and directly output the estimated result of the final state of health index without any model training.

[0074] The present invention predicts the estimated final state of health index of lithium batteries in the test set. It can be found that the present invention shows good prediction effects of the estimated final state of health index on 14 batteries. To more clearly demonstrate the superiority of the present invention in the prediction task of the estimated final state of health index of lithium batteries, the LSTM method, the GRU method and the method provided by the present invention are selected for comparison here. Among them, the training schemes adopted by the LSTM method and the GRU method are: using all battery data in the training set to train the model. This method learns with a large amount of battery data and can demonstrate certain general capabilities in the health estimation task for large-scale batteries, but it has a large amount of training data and a time-consuming training process. The estimated result graph of the final state of health index of battery No. 1 in the test set in the LSTM method is as Figure 3 shown, the estimated result graph of the final state of health index of battery No. 1 in the test set in the GRU method is as Figure 4 shown, and the estimated result graph of the final state of health index of battery No. 1 in the test set in the method provided by the present invention is as Figure 5 shown. The estimated result graph of the final state of health index of battery No. 2 in the test set in the LSTM method is as Figure 6 shown, the estimated result graph of the final state of health index of battery No. 2 in the test set in the GRU method is as Figure 7 shown, and the estimated result graph of the final state of health index of battery No. 2 in the test set in the method provided by the present invention is as Figure 8 shown. The estimated result graph of the final state of health index of battery No. 3 in the test set in the LSTM method is as Figure 9 shown, the estimated result graph of the final state of health index of battery No. 3 in the test set in the GRU method is as Figure 10 shown, and the estimated result graph of the final state of health index of battery No. 3 in the test set in the method provided by the present invention is as Figure 11 shown. The estimated accuracy of the final state of health of all 14 batteries used for testing corresponding to the LSTM method, the GRU method and the method used in the present invention is shown in Table 1, and the model training consumption time of the LSTM method, the GRU method and the method used in the present invention is shown in Table 2.

[0075] Table 1: Prediction Accuracy of Final Health Status Estimation Metrics

[0076] Table 2: Comparison Results of Modeling Time Used by Each Method

[0077] As can be seen from Table 1, in most cases, the method provided by the present invention is superior to the LSTM method and the GRU method in terms of the MSE and R 2 metrics. Moreover, the average level of the mean squared error (MSE) metric of the prediction results of the final health status estimation metrics of the method provided by the present invention reaches 3.274×10 -5 , and the average level of the R 2 metric reaches 0.9892, which proves the superiority of the present method in terms of accuracy. In addition, as can be seen from Table 2, compared with the existing LSTM method and GRU method, the modeling time of the method of the present invention is shortened by more than 2 times, and the modeling efficiency has advantages, further proving the effectiveness of the method provided by the present invention.

[0078] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments or perform equivalent replacements on some of the technical features. However, these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A shared-compensated efficient health assessment method for large-scale energy storage battery systems, characterized in that It includes the following steps: (1) Obtain the total set of battery sampling data and the corresponding total set of health state indicators of the energy storage battery system, and construct the total set of battery cluster degradation expert knowledge representations; (2) Randomly select the data of B0 batteries from the total set of health state indicators and the total set of battery cluster degradation expert knowledge representations, perform standardization processing, and then train a long short-term memory network model to obtain a local shared model for health state estimation; (3) Based on the local shared model for health state estimation, retain B1 batteries greater than or equal to the deviation level threshold and perform residual spectrum clustering operations to obtain P clustering clusters; (4) Use the P clustering clusters to train multiple health state estimation compensation models respectively, and after removing duplicates, obtain the final set of health state estimation models; (5) Use the standardized data corresponding to the remaining batteries and the final set of health state estimation models to train a multi-layer perceptron model to obtain an adaptive mask selection network; (6) Use the adaptive mask selection network and the final set of health state estimation models to estimate the final health state estimation indicators corresponding to the battery to be estimated.

2. The shared-compensated high-efficiency health assessment method for large-scale energy storage battery systems according to claim 1, characterized in that The obtaining of the total set of battery sampling data and the corresponding total set of health state indicators of the energy storage battery system specifically includes: Obtain the set of measurement point data at all sampling moments during the discharge stage of all charge-discharge cycles in the full life cycle of B batteries in the energy storage battery system to form the total set of battery sampling data; among them, the measurement point data includes voltage values and current values; Obtain the health state indicators of B batteries in the energy storage battery system during all charge-discharge cycles in the full life cycle to form the total set of health state indicators.

3. The shared-compensated high-efficiency health assessment method for large-scale energy storage battery systems according to claim 1, characterized in that The construction of the total set of battery cluster degradation expert knowledge representations specifically includes: Obtain the corresponding capacity increment index vector according to the measurement point data vector of any charge-discharge cycle of any battery in the total set of battery sampling data, and perform regularization operations on it to obtain the corresponding regularized sample vector; Perform segmentation operations on the regularized sample vector, and obtain multiple local profile indicators according to the capacity increment indicators and the corresponding voltage values in each subset after segmentation to obtain the corresponding battery cluster degradation expert knowledge representation vector; Repeat the above steps for all measurement point data vectors of each charge-discharge cycle of each battery in the total set of battery sampling data to obtain the total set of battery cluster degradation expert knowledge representations.

4. The shared-compensated high-efficiency health assessment method for large-scale energy storage battery systems according to claim 3, wherein The regularization operation specifically includes: Set the curve truncation index to N and compare it with the total number of sampling times in the discharge stage of any charge-discharge cycle of any battery as follows: If , then sort the voltage values of all sampling times in this stage from smallest to largest, and use the first N voltage values after sorting and the corresponding capacity increment index as the corresponding normalized sample vector; if , then sort the voltage values of all sampling times in this stage from smallest to largest, and use the first voltage values and the th voltage value and the corresponding capacity increment index as the corresponding normalized sample vector; where represents the total number of sampling times in the discharge stage of the cth charge-discharge cycle of the bth battery.

5. The shared-compensated high-efficiency health assessment method for large-scale energy storage battery systems according to claim 1, wherein The step (2) specifically includes: Randomly select the data corresponding to B0 batteries from the total set of battery cluster degradation expert knowledge representations and the total set of health state indicators, and perform standardization processing using the Z-score standardization method. Then, randomly extract the standardized battery cluster degradation expert knowledge representation set corresponding to the batteries and input it into the long short-term memory network model to obtain the first total set of health state prediction indicators, and calculate the mean square error with the standardized health state indicator vectors corresponding to the extracted batteries to obtain the first loss function; train the long short-term memory network model based on the first loss function to obtain the trained long short-term memory network model as the local shared model for estimating the health state of B0 batteries.

6. The shared-compensated high-efficiency health assessment method for large-scale energy storage battery systems according to claim 1, wherein The step (3) specifically includes the following sub-steps: (3.1) Input the standardized battery cluster degradation expert knowledge representation set corresponding to the selected B0 batteries into the corresponding local shared model for health state estimation to obtain the second total set of health state prediction indicators, and perform health state estimation residual calculation with its corresponding standardized health state indicator vector to obtain the first total set of health state estimation residuals; (3.2) Obtain the average estimation deviation level set according to the first total set of health state estimation residuals, and retain the batteries with an average estimation deviation level greater than or equal to the deviation level threshold to obtain B1 retained batteries; For B1 reserved batteries, according to the respective health state estimation residual vectors and natural number S c Obtain the residual embedding vectors; using the residual embedding vectors as features, perform spectral clustering on the B1 reserved batteries to obtain P clustering clusters.

7. The shared-compensated high-efficiency health assessment method for large-scale energy storage battery systems according to claim 6, wherein The specific steps of step (4) include the following sub-steps: (4.1)For all the clustering clusters obtained in step (3.3), input the normalized battery cluster degradation expert knowledge representation sets corresponding to all the batteries in each clustering cluster into the support vector regression model to obtain the total set of the third health state prediction indicators, and calculate the mean square error with the health state estimation residual vectors corresponding to all the batteries in this clustering cluster to obtain the second loss function; train the support vector regression model based on the second loss function, and update the local shared model of the health state estimation corresponding to the batteries in this clustering cluster to the sum of the current local shared model of the health state estimation corresponding to the batteries in this clustering cluster and the trained support vector regression model, and determine whether the number of summation times experienced by the currently updated local shared model of the health state estimation is greater than the threshold M p If not, execute step (4.2); otherwise, record the currently local shared model of the health state estimation as a health state estimation compensation model, and end the operation on this clustering cluster; (4.2) Then, take the set of standardized battery cluster degradation expert knowledge representations corresponding to all batteries in this cluster as the input, and repeat steps (3.1)-(3.2) to obtain the number of batteries retained in this cluster; if the number of retained batteries is 0, record the local shared health state estimation model corresponding to this cluster as a health state estimation compensation model, and end the operation on this cluster; otherwise, repeat steps (3.3) and (4.1)-(4.2) for the set of standardized battery cluster degradation expert knowledge representations corresponding to the retained batteries until all clusters are traversed and the operation is completed; (4.3) Collect all the health state estimation compensation models obtained in steps (4.1) and (4.2), and remove the duplicate health state estimation compensation models to form the final health state estimation model set.

8. The shared-compensated high-efficiency health assessment method for large-scale energy storage battery systems according to claim 1, characterized in that The specific steps of step (5) include: The data corresponding to the remaining batteries in the total set of battery cluster degradation expert knowledge representations and the total set of health state indicators is standardized using the Z-score normalization method, and then the standardized battery cluster degradation expert knowledge representation sets corresponding to the remaining batteries are respectively input into each final health state estimation model in the final health state estimation model set to obtain the fourth total set of health state prediction indicators, and the absolute difference is calculated with the standardized health state indicator vectors corresponding to the remaining batteries to obtain the total set of model performance indicators and perform masking calculation to obtain the masked total set of model performance indicators; ​ The remaining The set of expert knowledge representations of the degradation of the standardized battery cluster corresponding to the Calculate the cross-entropy between the total set of masked model performance metrics and the total set of masked health state adaptive prediction metrics to obtain the third loss function; and train the multi-layer perceptron model based on the third loss function to obtain the trained multi-layer perceptron model as the adaptive mask selection network.

9. The shared-compensated high-efficiency health assessment method for large-scale energy storage battery systems according to claim 1, wherein The specific steps of step (6) include: Construct the corresponding battery cluster degradation expert knowledge representation vector based on the measurement point data at all sampling moments in the discharge stage of any charge-discharge cycle of the battery to be estimated, and perform standardization processing. Then, input it into the adaptive mask selection network to obtain the corresponding model performance estimation metric vector and perform mask calculation to obtain the masked model performance estimation metric vector; When the h-th masked model performance estimation metric in the masked model performance estimation metric vector is 1, select the h-th final health state estimation model from the final health state estimation model set as the estimation prediction model; input the standardized battery cluster degradation expert knowledge representation vector of the battery to be estimated into the estimation prediction model to obtain the model output result, and perform inverse standardization processing to obtain the final health state estimation metric corresponding to this charge-discharge cycle of the battery to be estimated.

Images