Shared-compensation type efficient health evaluation method for large energy storage battery system

Abstract

The present invention discloses a shared-compensation efficient health assessment method for large-scale energy storage battery systems. This method aims at the characteristics of a large number of batteries in a power station and the differences in health between them, and proposes a state estimation modeling strategy based on sharing and difference compensation. First, the knowledge representation of battery cluster degradation is mined to construct a basic health state estimation model that can be shared by multiple batteries; then, a residual-guided battery clustering strategy is designed to reveal the difference pattern of battery health characteristics, and iterative differentiation model accuracy is implemented for batteries with different modes; finally, an adaptive mask model selection strategy is developed to realize the selection and efficient reuse of health state estimation models for any battery. In a specific embodiment, the present invention has achieved 10 ‑5 The present invention can provide a solution for efficient and accurate health status 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 in particular relates to a shared-compensation efficient health assessment method for large energy storage battery systems. Background Art

[0002] Lithium-ion batteries (LIBs), with their high energy density, long lifespan, and low pollution, are becoming a key energy storage unit in next-generation energy storage systems (ESS). With the widespread deployment and application of LIBs in energy storage, the management of their state of health (SOH) during operation has become a research hotspot in both industry and academia.

[0003] Over extended use, lithium-ion batteries inevitably experience performance degradation, and SOH provides an effective way to assess the extent of battery degradation. Typically, SOH is defined as the ratio of a battery's current actual capacity to its nominal capacity. Accurately measuring a battery's SOH allows for timely identification of severely degraded batteries, enabling targeted maintenance and replacement to ensure energy storage system reliability. However, in practice, a battery's actual SOH is often difficult to measure and must be indirectly estimated using other parameters. Therefore, ensuring the accuracy of SOH is crucial for reliable lithium battery health management.

[0004] Machine learning-based methods have been widely used in state-of-health (SOH) estimation and have achieved significant progress. However, in real-world energy storage power plants, thousands of batteries are stacked in series or parallel. In large-scale battery stacks, each battery may have different physical, chemical, and operating conditions, resulting in varying state-of-health (SOH) profiles that cannot be accurately described using a unified SOH estimation model. To achieve refined state-of-health management for all batteries in a battery stack, accurate SOH estimation for each battery is necessary. However, most existing methods only model individual lithium-ion batteries, generating independent SOH estimates. While this modeling approach performs well in small-scale scenarios, it faces heavy computational and storage requirements when faced with a large number of batteries, limiting its practical application in large-scale energy storage systems. Therefore, developing lightweight SOH estimation methods while ensuring model accuracy, optimizing computational efficiency and resource allocation for model training, and achieving efficient and feasible SOH estimation for large battery clusters remains an urgent challenge. Summary of the Invention

[0005] The purpose of the present invention is to address the deficiencies of the existing technology and provide a shared-compensation efficient health assessment method for large energy storage battery systems.

[0006] The objective of the present invention is achieved through the following technical solution: a shared-compensation efficient health assessment method for a large energy storage battery system, 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 representation of battery cluster degradation;

[0008] (2) Randomly select B0 battery data from the total set of health status indicators and the total set of battery cluster degradation expert knowledge representation, perform standardization processing, and then train a long short-term memory network model to obtain a local shared model for health status estimation;

[0009] (3) Based on the local shared model for health state estimation, B1 batteries with a deviation level threshold are retained and residual spectrum clustering is performed to obtain P clusters;

[0010] (4) Using P clusters to train multiple health state estimation compensation models, and after removing duplicates, the final health state estimation model set is obtained;

[0011] (5) Using the standardized data corresponding to the remaining batteries and the final health state estimation model set to train the multi-layer perceptron model, an adaptive mask selection network is obtained;

[0012] (6) The final health state estimation index corresponding to the battery to be estimated is estimated using the adaptive mask selection network and the final health state estimation model set.

[0013] Furthermore, the acquisition 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:

[0014] Obtain a set of measurement point data at all sampling moments of the discharge phase of all charge-discharge cycles of B batteries in the energy storage battery system throughout their entire life cycle to form a total set of battery sampling data; wherein the measurement point data includes voltage values ​​and current values;

[0015] Obtain health status indicators of all charge-discharge cycles of B batteries in the energy storage battery system throughout their life cycle to form a total set of health status indicators.

[0016] Furthermore, the construction of a total set of expert knowledge representations of battery cluster degradation specifically includes:

[0017] According to the measurement point data vector of any one charging-discharging cycle of any one battery in the total set of battery sampling data, a corresponding capacity increment index vector is obtained, and a normalization operation is performed thereon to obtain a corresponding normalized sample vector;

[0018] The normalized sample vector is segmented, and a plurality of local profile indicators are obtained according to the capacity increment indicators and the corresponding voltage values in each subset after segmentation, so as to obtain a corresponding battery cluster degradation expert knowledge representation vector.

[0019] The above steps are repeated for all measurement point data vectors of each charging-discharging cycle of each battery in the total set of battery sampling data to obtain a total set of battery cluster degradation expert knowledge representations.

[0020] Further, the normalization operation specifically includes:

[0021] The curve truncation index N is set, and the total number of sampling time points in the discharge stage of any one charging-discharging cycle of any one battery is compared: if , the voltage values of all sampling time points in the stage are sorted from small to large, and the first N voltage values and the corresponding capacity increment indicators are taken as the corresponding normalized sample vector; if , the voltage values of all sampling time points in the stage are sorted from small to large, and the first voltage values and the corresponding capacity increment indicators are taken as the corresponding normalized sample vector; wherein represents the total number of sampling time points in the discharge stage of the cth charging-discharging cycle of the bth battery. Further, the step (2) specifically includes:

[0022] From the total set of battery cluster degradation expert knowledge representations and the total set of health state indicators, the data corresponding to B0 batteries are randomly selected and standardized by Z-score standardization, and then battery cluster degradation expert knowledge representation sets after standardization corresponding to the B0 batteries are randomly selected and input into the long short-term memory network model to obtain a first total set of health state prediction indicators, and the mean square error calculation is performed with the health state indicator vectors corresponding to the B0 batteries after standardization to obtain a first loss function; the long short-term memory network model is trained based on the first loss function to obtain a trained long short-term memory network model as a health state estimation local shared model of the B0 batteries.

[0023]

[0024] ​​​Further, the step (3) specifically comprises the following sub-steps:

[0025] (3.1) the standardized processed battery cluster degradation expert knowledge representation set corresponding to the selected B0 batteries is input into the corresponding health state estimation local shared model, to obtain a second health state prediction index total set, and the health state estimation residual calculation is performed on the second health state prediction index total set and the corresponding standardized processed health state index vector, to obtain a first health state estimation residual total set;

[0026] (3.2) the average estimation bias level set is obtained according to the first health state estimation residual total set, to retain the batteries with the average estimation bias level greater than or equal to the bias level threshold, to obtain B1 retained batteries;

[0027] (3.3) for the B1 retained batteries, the residual embedding vector is obtained according to the respective corresponding health state estimation residual vector and the natural number S c ; the residual embedding vector is used as a feature, and spectral clustering is used to cluster the B1 retained batteries, to obtain P clusters.

[0028] Further, the step (4) specifically comprises the following sub-steps:

[0029] (4.1) for all the cluster clusters obtained in step (3.3), the standardized processed battery cluster degradation expert knowledge representation set corresponding to all the batteries in each cluster cluster is input into the support vector regression model, to obtain a third health state prediction index total set, and the mean square error calculation is performed on the third health state prediction index total set and the health state estimation residual vector corresponding to all the batteries in the cluster cluster, to obtain a second loss function; the support vector regression model is trained based on the second loss function, and the health state estimation local shared model corresponding to the batteries in the cluster cluster is updated to the sum of the current health state estimation local shared model of the batteries in the cluster cluster and the trained support vector regression model, and it is judged whether the number of times of summation experienced by the current updated health state estimation local shared model is greater than a threshold M p , if not, step (4.2) is performed, otherwise the current health state estimation local shared model is recorded as a health state estimation compensation model, and the operation on the cluster cluster is ended;

[0030] (4.2) Then, the standardized battery cluster degradation expert knowledge representation set corresponding to all batteries in the cluster is used as input, and steps (3.1)-step (3.2) are repeated to obtain the number of batteries retained in the cluster; if the number of retained batteries is 0, the health state estimation local shared model corresponding to the cluster is recorded as a health state estimation compensation model, and the operation on the cluster is ended; otherwise, the standardized battery cluster degradation expert knowledge representation set corresponding to the retained batteries is repeated with steps (3.3) and steps (4.1)-step (4.2) until all clusters are traversed and the operation is completed;

[0031] (4.3) Collect all the health state estimation compensation models obtained in step (4.1) and step (4.2), and remove duplicate health state estimation compensation models to form the final health state estimation model set.

[0032] Furthermore, the step (5) specifically includes:

[0033] The remaining The data corresponding to each battery are standardized using the Z-score standardization method, and the remaining The standardized battery cluster degradation expert knowledge representation set corresponding to each battery is input into each final health state estimation model in the final health state estimation model set to obtain the fourth health state prediction index total set, and then combined with the remaining The absolute difference calculation is performed on the standardized health status indicator vectors corresponding to each battery to obtain the total set of model performance indicators and perform mask calculation to obtain the total set of masked model performance indicators;

[0034] The remaining The standardized battery cluster degradation expert knowledge representation set corresponding to each battery is input into the multi-layer perceptron model to obtain the total set of health state adaptive prediction indicators and perform mask calculation to obtain the total set of masked health state adaptive prediction indicators;

[0035] A cross entropy calculation is performed on the total set of masked model performance indicators and the total set of masked health status adaptive prediction indicators to obtain a third loss function; and the multilayer perceptron model is trained based on the third loss function to obtain the trained multilayer perceptron model as the adaptive mask selection network.

[0036] Furthermore, the step (6) specifically includes:

[0037] Based on the measurement point data at all sampling moments in the discharge phase of any charge-discharge cycle of the battery to be estimated, the corresponding battery cluster degradation expert knowledge representation vector is constructed and standardized. The vector is then input into the adaptive mask selection network to obtain the corresponding model performance estimation index vector and perform mask calculation to obtain the masked model performance estimation index vector.

[0038] When the h-th masked model performance estimation index in the masked model performance estimation index vector is 1, the h-th final health state estimation model is selected from the final health state estimation model set as the estimation prediction model; the standardized battery cluster degradation expert knowledge representation vector of the battery to be estimated is input into the estimation prediction model to obtain the model output result, and denormalization is performed to obtain the final health state estimation index corresponding to the charge-discharge cycle of the battery to be estimated.

[0039] The beneficial effects of the present invention are as follows: In view of the characteristics of energy storage power stations where batteries are arranged in stacks, are numerous in number, and have varying degrees of differences between each other, the present invention proposes a model sharing and differentiated compensation mechanism to achieve accurate and efficient estimation of the health status of large-scale battery systems; first, the knowledge representation of battery cluster degradation is mined to construct a basic health status estimation model that can be shared by multiple batteries; then, a residual-guided battery clustering strategy is designed to reveal the difference pattern of battery health characteristics, and iterative differentiated model accuracy compensation is implemented for different batteries; finally, an adaptive mask model selection strategy is developed to achieve health status estimation model selection and efficient reuse for any battery. Compared with existing battery system health assessment methods, the health status estimation modeling method of resource sharing and differentiated training compensation proposed in 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 refined health status estimation of battery devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a flow chart of the shared-compensation efficient health assessment method for a large energy storage battery system of the present invention;

[0041] Figure 2 Schematic diagram of the residual-guided model compensation strategy corresponding to steps (3) to (4) of the present invention;

[0042] Figure 3 This is the result diagram of the final health status indicator estimation of battery No. 1 in the LSTM method;

[0043] Figure 4 This is the result diagram of the final health status indicator estimation of battery No. 1 in the GRU method;

[0044] Figure 5This is a result diagram of the final health status index estimation of battery No. 1 in the method of the present invention;

[0045] Figure 6 This is the result diagram of the final health status indicator estimation of battery No. 2 in the LSTM method;

[0046] Figure 7 This is the result diagram of the final health status indicator estimation of battery No. 2 in the GRU method;

[0047] Figure 8 This is a result diagram of the final health status index estimation of battery No. 2 in the method of the present invention;

[0048] Figure 9 This is the result diagram of the final health status indicator estimation of battery No. 3 in the LSTM method;

[0049] Figure 10 This is the result diagram of the final health status indicator estimation of battery No. 3 in the GRU method;

[0050] Figure 11 This is a result diagram of the final health status indicator estimation of the No. 3 battery in the method described in the present invention. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to illustrate the present invention, rather than to represent all embodiments. All other embodiments obtained by persons of ordinary skill in the art based on the embodiments of the present invention without creative work are within the scope of protection of the present invention.

[0052] See also Figure 1 The shared-compensation efficient health assessment method for a large energy storage battery system of the present invention specifically includes the following steps:

[0053] During the long-term use of lithium batteries, performance degradation is inevitable. Different battery cells have different degradation characteristics, requiring an efficient health status estimation strategy to accurately perceive the health status of multiple batteries. In this embodiment, a set of measurement point data at all sampling moments in the discharge phase of all charge-discharge cycles in the entire life cycle of 14 batteries in any energy storage battery system and the corresponding health status indicator vectors are collected as a training set to establish an SOH estimation model. The health status estimation test is then performed using a set of measurement point data at all sampling moments in the discharge phase of all charge-discharge cycles in the entire life cycle of another 14 batteries and the corresponding health status indicator vectors. Among them, the battery involved in the data set is an NCA ternary lithium battery with a nominal capacity of 3500mA. The battery's charge and discharge mode 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, and the constant voltage charging stage maintains the voltage at 4.2V, 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.

[0054] (1) Obtain the total set of battery sampling data and the corresponding total set of State of Health (SOH) indicators of the energy storage battery system, and construct the total set of expert knowledge representation of battery cluster degradation.

[0055] In this embodiment, obtaining a total set of battery sampling data and a corresponding total set of health status indicators of the energy storage battery system specifically includes: obtaining a set of measurement point data at all sampling moments in the discharge phase of all charge-discharge cycles of B=14 batteries in the energy storage battery system throughout their life cycle to form a total set of battery sampling data. The measurement point data set of each battery is composed of measurement point data at all sampling moments in the discharge phase of all charge-discharge cycles of each battery throughout its life cycle, and each measurement point data is composed of current and voltage values ​​at the sampling moment. The total set of battery sampling data is represented as ,in, represents the set of measurement point data at all sampling moments in the discharge phase of all charge-discharge cycles of the b-th battery in its entire life cycle, , represents the measurement point data vector of all sampling moments in the discharge phase of the cth charge-discharge cycle of the bth battery, represents the total number of charge-discharge cycles experienced by the b-th battery, and the superscript T represents the transpose of the matrix or vector. , represents the measurement point data at the kth sampling moment of the discharge phase in the cth charge-discharge cycle of the bth battery, represents the total number of sampling moments in the discharge phase of the cth charge-discharge cycle of the bth battery, , represents the current value at the kth sampling moment of the discharge phase in the cth charge-discharge cycle of the bth battery, Represents the voltage value at the kth sampling moment of the discharge phase in the cth charge-discharge cycle of the bth battery. Obtain the health status indicators of all charge-discharge cycles of B=14 batteries in the energy storage battery system throughout their life cycle to form a total set of health status indicators; among which, the health status indicators corresponding to all charge-discharge cycles of each battery in its life cycle constitute the health status indicator vector corresponding to the battery, and the health status indicator vectors of B=14 batteries in the energy storage battery system constitute the total set of health status indicators. The total set of health status indicators is expressed as ,in, represents the health status indicator vector corresponding to the b-th battery, , Represents the health status indicator corresponding to the cth charge-discharge cycle of the bth battery in its entire life cycle.

[0056] In this embodiment, constructing a total set of expert knowledge representations of battery cluster degradation specifically includes the following sub-steps:

[0057] (1.1) According to the measurement point data vector of all sampling moments in the discharge phase of the cth charge-discharge cycle of the bth battery in the total battery sampling data set E, Calculate the capacity increment index vector corresponding to the cth charge-discharge cycle of the bth battery , It represents the capacity increment index corresponding to the kth sampling moment in the cth charge-discharge cycle of the bth battery. Its calculation formula is:

[0058]

[0059] Where, Represents the time value of the kth sampling moment in the cth charge-discharge cycle of the bth battery.

[0060] (1.2) The capacity increment index vector corresponding to the cth charge-discharge cycle of the bth battery obtained in step (1.1) is Perform a normalization operation to obtain the corresponding normalized sample vector.

[0061] Furthermore, the normalization operation specifically includes: setting the curve truncation index to N, and the total number of sampling moments of the discharge phase in the cth charge-discharge cycle of the bth battery Compare with the set curve truncation index N: If , then sort all the voltage values ​​at all sampling moments in the discharge phase of the cth charge-discharge cycle of the bth battery from small to large, and get the sorted voltage value, which is expressed as ,in, Represents the dth voltage value after sorting from small to large, and takes the first N voltage values ​​after sorting and the corresponding capacity increment index as the normalized sample vector ,in, , Indicates voltage value The corresponding capacity increment indicator; if , then sort all the voltage values ​​at all sampling moments in the discharge phase of the cth charge-discharge cycle of the bth battery from small to large, and get the sorted voltage value, which is expressed as ,in, Indicates the fth voltage value after sorting from small to large, , and the sorted front voltage values ​​and No. Voltage value And the corresponding capacity increment index as the normalized sample vector ,in, , Indicates voltage value The corresponding capacity increment indicator. In this embodiment, the value of N is 250.

[0062] (1.3) The normalized sample vector is segmented, and multiple local profile indicators are obtained based on the capacity increment indicator and the corresponding voltage value in each subset after segmentation to obtain the corresponding battery cluster degradation expert knowledge representation vector.

[0063] Specifically, for the normalized sample vector Perform a split operation to normalize the sample vector The system is divided into L subsets at equal intervals, where L is a factor of N. 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 are calculated and used as local profile indicators to obtain the battery cluster degradation expert knowledge representation vector corresponding to the cth charge-discharge cycle of the bth battery. , the battery cluster degradation expert knowledge representation vector is composed of The local contour index is composed of Represents the expert knowledge representation vector of battery cluster degradation The jth local profile index in , J represents the battery cluster degradation expert knowledge representation vector The total number of local contour indices in . In this embodiment, the value of L is 5.

[0064] (1.4) Repeat steps (1.1) to (1.3) for all sampling points of the discharge phase of each charge-discharge cycle of each battery in the total battery sampling data set E, and obtain the total battery cluster degradation expert knowledge representation set consisting of the battery cluster degradation expert knowledge representation set of B batteries. ,in, The expert knowledge representation set of battery cluster degradation for the b-th battery is represented by The expert knowledge representation vector of battery cluster degradation corresponding to all charge-discharge cycles experienced by the battery Composition, expressed as .

[0065] It should be noted that by constructing a total set of expert knowledge representations of battery cluster degradation, a cluster degradation representation solution that can be executed for all batteries in the energy storage battery system can be provided.

[0066] (2) Randomly select B0 battery data from the total set of health status indicators and the total set of battery cluster degradation expert knowledge representation, perform standardization processing, and then train a long short-term memory network (LSTM) model to obtain a local shared model for health status estimation.

[0067] (2.1) Randomly select from the total set of battery cluster degradation expert knowledge representation and the total set of health status indicators The battery cluster degradation expert knowledge representation set and health status indicator vector corresponding to each battery are used as the initial health status estimation data set ,in, , Indicates selected The total set of initial battery cluster degradation expert knowledge representations is composed of the battery cluster degradation expert knowledge representation set corresponding to each battery. Indicates selected The total set of initial health status indicators consisting of the health status indicator vectors corresponding to the batteries; the total set of initial battery cluster degradation expert knowledge representation Expressed as , Indicates the selected The battery cluster degradation expert knowledge representation set corresponding to each battery, , Indicates the selected The battery corresponding to Expert knowledge representation vector of battery cluster degradation over charge-discharge cycles, Indicates the selected The total number of charge-discharge cycles that a battery undergoes, , Represents the expert knowledge representation vector of battery cluster degradation Expert knowledge representation of degradation of the j-th battery cluster; the total set of initial health status indicators Expressed as , Indicates the selected The health status indicator vector corresponding to each battery, , Shows the selected The first time a battery is used in its entire life cycle The health status indicator corresponding to the charge-discharge cycle. In this embodiment, The value of is 9, which means that the data corresponding to 9 batteries are randomly selected from 14 batteries.

[0068] (2.2) Total set of expert knowledge representations for initial battery cluster degradation Expert knowledge representation of degradation of any battery cluster in Standardization is performed by Z-score standardization, and finally the initial battery cluster degradation expert knowledge representation after standardization can be obtained. , which is calculated as follows:

[0069] ;

[0070] ;

[0071] ;

[0072] Where, represents the average value of expert knowledge representation of battery cluster degradation, represents the standard deviation of the expert knowledge representation of battery cluster degradation.

[0073] (2.3) For the total set of initial health status indicators Any health status indicator Standardization is performed by Z-score standardization, and the initial health status index after standardization can be obtained. , the calculation formula is as follows:

[0074] ;

[0075] ;

[0076] ;

[0077] Where, represents the average value of health status indicators, Represents the standard deviation of the health status indicator.

[0078] (2.4) Total set of expert knowledge representations for initial battery cluster degradation Repeat step (2.2) for each battery cluster degradation expert knowledge representation in to obtain the total set of standardized initial battery cluster degradation expert knowledge representations , expressed as ,in, Indicates the selected The normalized initial battery cluster degradation expert knowledge representation set corresponding to each battery, , Indicates the selected The first battery The expert knowledge representation vector of the initial battery cluster degradation after normalization of charge-discharge cycles, , Indicates the selected The first battery Initial battery cluster degradation expert knowledge representation vector after normalization of charge-discharge cycles Expert knowledge representation of the degradation of the j-th battery cluster in .

[0079] For the total set of initial health status indicators Repeat step (2.3) for each health status indicator in to obtain the total set of initial health status indicators after standardization , expressed as ,in, Indicates the selected The normalized initial health status indicator vector corresponding to each battery is: , Indicates the selected Battery No. The normalized health status index corresponding to the charge-discharge cycle.

[0080] Then, the total set of expert knowledge representations of the initial battery cluster degradation after standardization is and the total set of initial health status indicators after standardization , get the initial health status estimation data set after standardization , expressed as .

[0081] (2.5) Estimating the dataset from the normalized initial health state It is a random draw The standardized initial battery cluster degradation expert knowledge representation set corresponding to the battery and the standardized initial health state indicator vector are used as the local health state estimation data set , expressed as ,in, , Indicates that the extracted The total set of local battery cluster degradation expert knowledge representations is composed of the standardized initial battery cluster degradation expert knowledge representation set corresponding to each battery. Indicates the extracted The total set of local health status indicators consisting of the standardized initial health status indicator vectors corresponding to the batteries, and the total set of local battery cluster degradation expert knowledge representation Expressed as , Indicates the extracted The normalized initial battery cluster degradation expert knowledge representation set corresponding to each battery, , Indicates the extracted The first battery The expert knowledge representation vector of the initial battery cluster degradation after normalization of charge-discharge cycles, Indicates the extracted The total number of charge-discharge cycles that the battery undergoes; , Indicates the extracted The first battery Initial battery cluster degradation expert knowledge representation vector after normalization of charge-discharge cycles The degradation expert knowledge representation of the j-th battery cluster in the total set of local health status indicators Expressed as , Indicates the extracted The normalized initial health status indicator vector corresponding to each battery is: , Indicates the extracted The first battery The health status index after normalization corresponding to the charge-discharge cycle. In this embodiment, The value of is 3, which means that the data corresponding to 3 batteries are randomly selected from the data of 9 batteries.

[0082] (2.6) Then the total set of local battery cluster degradation expert knowledge representation is Input into the long short-term memory network model to obtain the total set of the first health status prediction indicators , expressed as ,in, Indicates the extracted The first health status prediction indicator vector corresponding to the battery, ,in, Indicates the extracted The first battery The first health status prediction indicator corresponding to the charge-discharge cycle, , Represents the mapping function of the long short-term memory network model. Then the first health status prediction index is totaled and the total set of local health status indicators Calculate the mean square error and get the first loss function ; and based on the first loss function The long short-term memory network model is trained to minimize the first loss function To optimize the target, adjust the parameters of the LSTM network model until the preset number of training rounds is reached to obtain the trained LSTM network model as the local shared model for estimating the health status of all B0 batteries. In the training process of this embodiment, the number of training rounds is , its value is 20; the learning rate is , whose value is 1×10 -3 , the training batch size is , its value is 32.

[0083] Furthermore, the first loss function The calculation formula is:

[0084]

[0085] It should be noted that the health state estimation local shared model obtained through the current step (2) is used as a health state estimation model that can be shared by all batteries. There is no need to build models for all batteries separately, which can effectively reduce the cost of the number of models.

[0086] (3) Based on the local shared model for health state estimation, retain B1 batteries that are greater than or equal to the deviation level threshold and perform residual spectrum clustering operations to obtain P clusters. The steps are shown in the following diagram: Figure 2 shown.

[0087] Specifically, the second health state prediction indicator total set is obtained by the health state estimation local sharing model, and then the health state estimation residual calculation is performed, and the reserved average estimation deviation level of the selected B0 batteries is calculated, the B1 batteries with a reserved average estimation deviation level greater than or equal to the deviation level threshold are reserved, and the residual spectrum clustering operation is performed to obtain P clustering clusters. Specifically, the following sub-steps are included:

[0088] (3.1) First, the standardized initial health state estimation data set is input into the health state estimation local sharing model , and the standardized initial battery cluster degradation expert knowledge representation total set corresponding to the selected B0 batteries is obtained. , is expressed as , wherein represents the second health state prediction indicator vector corresponding to the selected i-th battery, , represents the second health state prediction indicator corresponding to the j-th charge-discharge cycle of the selected i-th battery, , represents the mapping function of the health state estimation local sharing model . Then, according to the second health state prediction indicator total set and the standardized initial health state indicator total set , the second health state prediction indicator corresponding to all charge-discharge cycles of the B1 batteries and the corresponding standardized initial health state indicator are calculated, and the first health state estimation residual total set

[0089] is obtained, which is expressed as , wherein represents the health state estimation residual vector corresponding to the selected i-th battery, , represents the health state estimation residual corresponding to the j-th charge-discharge cycle of the selected i-th battery. (3.2) According to the first health state estimation residual total set , the average estimation deviation level set is calculated, which is expressed as , wherein . (3.3) The B1 batteries with a reserved average estimation deviation level greater than or equal to the deviation level threshold are reserved, and the residual spectrum clustering operation is performed to obtain P clustering clusters.

[0090] (3.4) The health state estimation residual spectrum clustering operation is performed on the B1 batteries with a reserved average estimation deviation level greater than or equal to the deviation level threshold, and the P clustering clusters are obtained. (3.5) The health state estimation residual spectrum clustering operation is performed on the B1 batteries with a reserved average estimation deviation level greater than or equal to the deviation level threshold, and the P clustering clusters are obtained. (3.6) The health state estimation residual spectrum clustering operation is performed on the B1 batteries with a reserved average estimation deviation level greater than or equal to the deviation level threshold, and the P clustering clusters are obtained.​​​Indicates the selected The average estimated deviation level corresponding to the batteries, .

[0091] And according to the average estimated deviation level set The average estimated deviation level corresponding to each battery Deviation level threshold Perform level threshold comparison: If , then the initial battery cluster degradation expert knowledge representation set corresponding to the standardized processing of the battery is and the initial health status indicator vector after normalization The total set of expert knowledge representations of initial battery cluster degradation after standardization Remove it and estimate the health status residual vector corresponding to the battery Estimating the total set of residuals from the first healthy state Remove the average estimated deviation level corresponding to the battery Deviations from the mean estimated level set Remove from , then the initial battery cluster degradation expert knowledge representation set corresponding to the standardized processing of the battery is and the initial health status indicator vector after normalization Retain and estimate the residual vector of the health status corresponding to the battery and the average estimated bias level to reserve; to complete After comparing the level thresholds of the batteries, the number of batteries retained is In this embodiment, the deviation level threshold The value of is 0.03.

[0092] (3.3) Set a natural number S c ,for The retained batteries are divided by the length of the corresponding health state estimation residual vector S c , and the quotient is S w , intercept the corresponding health state estimation residual vector S c *S w elements form the residual second vector; starting from the first element of the residual second vector, the interval S w Take out the element values ​​one by one to form the residual embedding vector. Using the residual embedding vector as a feature, use spectral clustering to The retained batteries are clustered to obtain P clusters. The p-th cluster contains retained batteries, of which In this embodiment, Sc The value is 50, and the number of clusters is 2, that is, P=2.

[0093] It should be noted that the obtained P clusters indicate the differences in health characteristics of different batteries. At the same time, the clustering operation directly reflects the constructed health status estimation local sharing model in The gap between the health status estimation results of the selected batteries and the actual health status estimation results provides a direct basis for the subsequent implementation of customized accuracy compensation for different batteries.

[0094] (4) Use P clusters to train multiple health state estimation compensation models, and after deduplication, obtain the final health state estimation model set. The steps are shown in the following diagram: Figure 2 As shown. Specifically, the standardized initial battery cluster degradation expert knowledge representation set corresponding to all batteries in each cluster of P clusters is used to train the corresponding health state estimation model, and the final health state estimation model set is obtained after deduplication. After the above-mentioned B0 batteries are clustered, for the batteries in each cluster, the respective health state estimation residuals are used as training targets to train the corresponding local shared health state estimation model. The local shared health state estimation model only uses a simple machine learning model. By superimposing a customized model based on the residuals on the original basis, the accuracy of the health state estimation can be further improved in an efficient and customized manner.

[0095] (4.1) Based on all clusters obtained in step (3.3), for each cluster, perform the following steps in sequence: The initial battery cluster degradation expert knowledge representation set after standardization corresponding to the batteries is used as the total set of initial battery cluster degradation expert knowledge representation after standardization of the p-th cluster , and the The health state estimation residual vectors corresponding to the retained batteries are used as the second health state estimation residual total set .

[0096] Then the total set of expert knowledge representation of initial battery cluster degradation after normalization of the p-th cluster is Input into the support vector regression (SVR) model to obtain the total set of third health status prediction indicators ; Then the third health status prediction index is combined and the total set of the second health state estimated residuals The second loss function is calculated by using the mean square error, and the support vector regression model is trained based on the second loss function, so as to minimize the second loss function as an optimization objective, and adjust the parameters of the support vector regression model until the model reaches a tolerance condition of stopping training, so as to obtain the trained support vector regression model , wherein represents the trained support vector regression model corresponding to the pth clustering cluster, and the health state estimation local sharing model corresponding to the battery in the pth clustering cluster is updated to the current health state estimation local sharing model of the battery in the clustering cluster and the trained support vector regression model , that is, , wherein represents the updated health state estimation local sharing model corresponding to the battery in the pth clustering cluster. In the embodiment, the kernel function of the support vector regression model is selected as the rbf function, the tolerance of stopping training is , and the value of the tolerance is 1x10 -3 , the regularization coefficient is , the value of the regularization coefficient is 1, and the error tolerance bandwidth is , and the value of the error tolerance bandwidth is 0.1.

[0097] Then, it is judged whether the number of times of summation of the updated health state estimation local sharing model corresponding to the pth clustering cluster exceeds a threshold value , if the number of times of summation does not exceed the threshold value , step (4.2) is performed, otherwise, the updated health state estimation local sharing model corresponding to the pth clustering cluster is taken as a health state estimation compensation model , and the operation on the clustering cluster is ended. In the embodiment, the value of the threshold value is 10.

[0098] (4.2) Then, the standardized battery cluster degradation expert knowledge representation total set of the pth clustering cluster is input into the updated health state estimation local sharing model , steps (3.1)-(3.2) are repeated, and the number of the retained batteries of the pth clustering cluster is obtained as , if , the updated health state estimation local sharing model is taken as a health state estimation model , and the operation on the clustering cluster is ended; otherwise, the retained Repeat steps (3.3) and (4.1)-(4.2) for the standardized initial battery cluster degradation expert knowledge representation set corresponding to each battery until all clusters are traversed and the operation is completed.

[0099] (4.3) Repeat steps (4.1) to (4.2) for each cluster, collect the health state estimation compensation models corresponding to all clusters, and remove the duplicate health state estimation compensation models to form the final health state estimation model set , expressed as ,in, represents the mth final health state estimation model, and M represents the total number of health state estimation models in the final health state estimation model set.

[0100] (5) The multi-layer perceptron model is trained using the standardized data corresponding to the remaining batteries and the final health state estimation model set 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) to (4). The constructed adaptive mask selection network can take the cluster degradation expert knowledge representation of any battery as input and output its corresponding optimal health state estimation model selection scheme. Therefore, this step aims to enable the model in the model library constructed in step (4) to be used for health state estimation tasks for more unseen batteries. It specifically includes the following sub-steps:

[0101] (5.1) The remaining The battery cluster degradation expert knowledge representation set and health status indicator vector corresponding to each battery are used as the adaptive mask training dataset ,in, Indicates the remainder The total set of adaptive battery cluster degradation expert knowledge representations is composed of the battery cluster degradation expert knowledge representation set corresponding to each battery. Indicates the remainder The total set of adaptive health status indicators consisting of the health status indicator vectors corresponding to the batteries; the total set of adaptive battery cluster degradation expert knowledge representation Expressed as , Indicates the unselected The battery cluster degradation expert knowledge representation set corresponding to each battery, , Indicates the unselected The corresponding battery Expert knowledge representation vector of battery cluster degradation over charge-discharge cycles, Indicates the unselected The total number of charge-discharge cycles that a battery undergoes, , Represents the expert knowledge representation vector of battery cluster degradation Expert knowledge representation of degradation of the j-th battery cluster; total set of adaptive health status indicators Expressed as , Indicates the unselected The health status indicator vector corresponding to each battery, , Indicates the unselected The first time a battery is used in its entire life cycle Health status indicator corresponding to the charge-discharge cycle.

[0102] (5.2) Total set of expert knowledge representation for adaptive battery cluster degradation Each battery cluster degradation expert knowledge representation in is standardized using the Z-score standardization method, that is, repeating step (2.2) to obtain the total set of standardized adaptive battery cluster degradation expert knowledge representations , expressed as ,in, Indicates the unselected The standardized adaptive battery cluster degradation expert knowledge representation set corresponding to each battery, , Indicates the unselected The first battery The adaptive battery cluster degradation expert knowledge representation vector after normalization of charge-discharge cycles, , Indicates the unselected The first battery Adaptive battery cluster degradation expert knowledge representation vector after normalization of charge-discharge cycles The expert knowledge representation of the degradation of the j-th battery cluster in Indicates the unselected The total number of charge-discharge cycles a battery undergoes.

[0103] For the total set of adaptive health status indicators Each health status indicator in is standardized using the Z-score standardization method, that is, repeating step (2.3) to obtain the total set of standardized adaptive health status indicators , expressed as ,in, Indicates the unselected The normalized adaptive health status indicator vector corresponding to each battery is: , Indicates the unselected The first battery The normalized health status index corresponding to the charge-discharge cycle.

[0104] Then, the total set of adaptive battery cluster degradation expert knowledge representation after standardization is and the total set of standardized adaptive health status indicators , and obtain the standardized adaptive health status estimation dataset , expressed as .

[0105] (5.3) The standardized adaptive battery cluster degradation expert knowledge representation total set Input into the final health state estimation model set In each final health state estimation model, the fourth health state prediction index total set is obtained , expressed as ,in, Indicates the unselected The fourth health status prediction indicator set corresponding to each battery, , Indicates the unselected The first battery The fourth health status prediction index vector corresponding to the charge-discharge cycle, , Indicates the selected The first battery The first charge-discharge cycle The corresponding fourth health status prediction indicator in the final health status estimation model is , Represents the mth final health state estimation model The mapping function.

[0106] (5.4) Then according to the fourth health status prediction indicator set and the total set of standardized adaptive health status indicators Will The absolute difference between the fourth health status prediction index corresponding to all charge-discharge cycles of the battery and the standardized adaptive health status index is calculated to obtain the total set of model performance indicators. , expressed as ,in, Indicates the selected The set of model performance indicators corresponding to each battery, , Indicates the unselected The first battery The model performance index vector corresponding to the charge-discharge cycle, , Indicates the selected The first battery The model performance index corresponding to the mth charge-discharge cycle in the final health state estimation model is, .

[0107] (5.5) Then the total set of model performance indicators Perform mask calculation to obtain the total set of masked model performance indicators ; Among them, the total set of model performance indicators after masking The performance index of the masked model in row i and column j The calculation formula is as follows:

[0108]

[0109] Where, Represents the total set of model performance indicators Any model performance indicator in the i-th row.

[0110] (5.6) The standardized adaptive battery cluster degradation expert knowledge representation total set Input into the Multilayer Perceptron (MLP) model to obtain the total set of health status adaptive prediction indicators , expressed as ,in, Indicates the unselected A set of adaptive health status prediction indicators corresponding to each battery, , Indicates the unselected The first battery The health status adaptive prediction indicator vector corresponding to the charge-discharge cycle, , Indicates that the multilayer perceptron model corresponds to the selected The first battery The mth health state adaptive prediction index is output by the charge-discharge cycle. Perform mask calculation to obtain the total set of masked health status adaptive prediction indicators .

[0111] (5.7) Then the total set of masked model performance indicators and the total set of adaptive health status prediction indicators after masking Perform cross entropy calculation to obtain a third loss function; and train the multilayer perceptron model based on the third loss function, with minimizing the third loss function as the optimization goal, and adjust the parameters of the multilayer perceptron model until a preset number of training rounds is reached to obtain a trained multilayer perceptron model as an adaptive mask selection network. In this training process of this embodiment, the number of training rounds is , its value is 100; the learning rate is , its value is 0.001; the training batch size is , its value is 64.

[0112] (6) The final health state estimation index corresponding to the battery to be estimated is estimated using the adaptive mask selection network and the final health state estimation model set.

[0113] (6.1) The battery cluster degradation expert knowledge representation vector of the battery to be estimated is constructed based on the measurement point data of all sampling moments in the discharge phase of any charge-discharge cycle of the battery to be estimated. ; Then the battery cluster degradation expert knowledge representation vector of the battery to be estimated Perform standardization to obtain the battery cluster degradation expert knowledge representation vector after standardization of the battery to be estimated ; Then the battery cluster degradation expert knowledge representation vector after the standardized processing of the battery to be estimated is Input into the adaptive mask selection network to obtain the model performance estimation index vector corresponding to the battery to be estimated ; and estimate the model performance indicator vector Perform mask calculation to obtain the masked model performance estimation index vector .

[0114] (6.2) When the masked model performance estimation indicator vector The model performance estimation index after the hth mask is 1, and the model set is estimated from the final health state Select the hth final health state estimation model as the estimation prediction model ; Then the battery cluster degradation expert knowledge representation vector after the standardized processing of the battery to be estimated is Input to the estimated forecast model The model output is obtained , expressed as ,in, Represents the estimated prediction model The mapping function; finally, the model outputs the result Perform denormalization to obtain the final health status estimation index corresponding to any charge-discharge cycle of the battery to be estimated : .

[0115] It should be understood that, as described in the above step (6), in the online application stage, the method of the present invention can directly output the final health status estimation index results for any remaining batteries in the system that have not participated in the modeling of steps (1) to (5) without the need for any model training.

[0116] The present invention predicts the final health status estimation indicators of lithium batteries in the test set. It can be found that the present invention shows good prediction effects of final health status estimation indicators on 14 batteries. In order to more clearly reflect the superiority of the present invention in the prediction task of the final health status estimation indicators of lithium batteries, the LSTM method and the GRU method are selected here for comparison with the method provided by the present invention. Among them, the training schemes adopted by the LSTM method and the GRU method are: using all the battery data in the training set to train the model. This method uses a large amount of battery data for learning, and can reflect certain general capabilities in the health estimation tasks for large-scale batteries, but the training data is large and the training process is time-consuming. The estimation results of the final health status indicators of battery No. 1 in the test set using the LSTM method are shown in the figure below. Figure 3 As shown in the figure, the estimated results of the final health status index of battery No. 1 in the test set using the GRU method are as follows Figure 4 As shown in the figure, the estimated results of the final health status index of battery No. 1 in the test set in the method provided by the present invention are as follows Figure 5 As shown in the figure below. The estimated results of the final health status index of battery No. 2 in the test set using the LSTM method are shown in the figure below. Figure 6 As shown in the figure, the estimated results of the final health status index of battery No. 2 in the test set in the GRU method are as follows Figure 7 As shown in the figure, the estimated results of the final health status index of the No. 2 battery in the test set in the method provided by the present invention are as follows Figure 8 As shown in the figure below. The estimated results of the final health status index of battery No. 3 in the test set using the LSTM method are shown in the figure below. Figure 9 As shown in the figure, the estimated results of the final health status index of battery No. 3 in the test set in the GRU method are as follows Figure 10 As shown in the figure, the estimated results of the final health status index of the No. 3 battery in the test set in the method provided by the present invention are as follows Figure 11 The final health state estimation accuracy of the LSTM method, the GRU method, and the method used in the present invention for all 14 batteries used in the test is shown in Table 1. The model training time of the LSTM method, the GRU method, and the method used in the present invention is shown in Table 2.

[0117] Table 1: Prediction accuracy of final health status estimation indicators

[0118]

[0119] Table 2: Comparison of modeling time for each method

[0120]

[0121] As shown in Table 1, in most cases, the method provided by the present invention is more effective in MSE and R 2 The indicators are better than those of the LSTM method and the GRU method, and the average level of the mean square error (MSE) of the final health status estimation index prediction results of the method provided by the present invention reaches 3.274×10 -5 , R 2 The average level of the indicator reached 0.9892, proving the superiority of the method in terms of accuracy. In addition, as shown in 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 is superior, further proving the effectiveness of the method provided by the present invention.

[0122] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A shared-compensation efficient health assessment method for large energy storage battery systems, characterized in that: The following steps are involved: (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 representation of battery cluster degradation; The construction of the total set of expert knowledge representations of battery cluster degradation specifically includes: Obtaining a corresponding capacity increment index vector according to a measurement point data vector of any charge-discharge cycle of any battery in the total battery sampling data set, and performing a normalization operation on the vector to obtain a corresponding normalized sample vector; The normalized sample vector is segmented, and multiple local profile indicators are obtained based on the capacity increment indicator and the corresponding voltage value in each segmented subset 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 representation; (2) Randomly select B0 battery data from the total set of health status indicators and the total set of battery cluster degradation expert knowledge representation, perform standardization 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 health state estimation local sharing model, retain B1 batteries that are greater than or equal to the deviation level threshold and perform residual spectrum clustering operation to obtain P clusters; the step (3) specifically includes the following sub-steps: (3.1) The standardized battery cluster degradation expert knowledge representation set corresponding to the selected B0 batteries is input into the corresponding health state estimation local shared model to obtain the second health state prediction index total set, and the health state estimation residual is calculated by combining the second health state prediction index and the corresponding standardized health state indicator vector to obtain the first health state estimation residual total set; (3.2) obtaining an average estimated deviation level set based on the total set of the first health state estimated residuals, and retaining batteries whose average estimated deviation levels are greater than or equal to the deviation level threshold, to obtain B1 retained batteries; (3.3) For B1 retained batteries, estimate the residual vector and the natural number S according to their corresponding health status c Obtain the residual embedding vector; use spectral clustering to cluster the B1 retained batteries using the residual embedding vector as a feature, and obtain P clusters; (4) Using P clusters to train multiple health state estimation compensation models, and after removing duplicates, obtaining a final health state estimation model set; the step (4) specifically includes the following sub-steps: (4.1) For all clusters obtained in step (3.3), the standardized battery cluster degradation expert knowledge representation set corresponding to all batteries in each cluster is input into the support vector regression model to obtain the third health status prediction index total set, and the mean square error is calculated between it and the health status estimation residual vector corresponding to all batteries in the cluster to obtain the second loss function; the support vector regression model is trained based on the second loss function, and the health status estimation local shared model corresponding to the batteries in the cluster is updated to the sum of the current health status estimation local shared model of the batteries in the cluster and the trained support vector regression model, and it is judged whether the number of summations experienced by the current updated health status estimation local shared model is greater than the threshold M p If not, execute step (4.2); otherwise, record the current health state estimation local shared model as a health state estimation compensation model and end the operation on the cluster; (4.2) Then, the standardized battery cluster degradation expert knowledge representation set corresponding to all batteries in the cluster is used as input, and steps (3.1) to (3.2) are repeated to obtain the number of batteries retained in the cluster; if the number of retained batteries is 0, the health state estimation local shared model corresponding to the cluster is recorded as a health state estimation compensation model, and the operation on the cluster is ended; otherwise, the standardized battery cluster degradation expert knowledge representation set corresponding to the retained batteries is repeated with steps (3.3) and steps (4.1) to (4.2) until all clusters are traversed and the operation is completed; (4.3) Collect all health state estimation compensation models obtained in step (4.1) and step (4.2), and remove duplicate health state estimation compensation models to form a final health state estimation model set; (5) using the standardized data corresponding to the remaining batteries and the final health state estimation model set to train the multi-layer perceptron model to obtain an adaptive mask selection network; (6) The final health state estimation index corresponding to the battery to be estimated is estimated using the adaptive mask selection network and the final health state estimation model set.

2. The shared-compensation efficient health assessment method for large energy storage battery systems according to claim 1 is characterized in that: 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 a set of measurement point data at all sampling moments of the discharge phase of all charge-discharge cycles of B batteries in the energy storage battery system throughout their entire life cycle to form a total set of battery sampling data; wherein the measurement point data includes voltage values ​​and current values; Obtain health status indicators of all charge-discharge cycles of B batteries in the energy storage battery system throughout their life cycle to form a total set of health status indicators.

3. The shared-compensation efficient health assessment method for large energy storage battery systems according to claim 1 is characterized in that: The regularization operation specifically includes: Set the curve cutoff index to N and compare it with the total number of sampling moments N in the discharge phase of any charge-discharge cycle of any battery. b,c For comparison: If N b,c ≥N, then sort the voltage values ​​of all sampling moments in this stage from small to large, and take the first N voltage values ​​after sorting and the corresponding capacity increment index as the corresponding normalized sample vector; if N b,c <N, then sort the voltage values ​​of all sampling moments in this stage from small to large, and sort the first N b,c voltage values ​​and NN b,c Nth b,c voltage values ​​and the corresponding capacity increment indicators are used as the corresponding normalized sample vectors; where N b,c represents the total number of sampling moments in the discharge phase of the cth charge-discharge cycle of the bth battery.

4. The shared-compensation efficient health assessment method for large energy storage battery systems according to claim 1 is characterized in that: The step (2) specifically includes: Data corresponding to B0 batteries are randomly selected from the total set of battery cluster degradation expert knowledge representations and the total set of health status indicators and standardized using the Z-score standardization method. Then, the standardized battery cluster degradation expert knowledge representation set corresponding to B′0 batteries is randomly extracted and input into the long short-term memory network model to obtain a first total set of health status prediction indicators. The mean square error is calculated with the standardized health status indicator vector corresponding to the extracted B′0 batteries to obtain a first loss function; the long short-term memory network model is trained based on the first loss function to obtain the trained long short-term memory network model as a local shared model for health status estimation of B0 batteries.

5. The shared-compensation efficient health assessment method for large energy storage battery systems according to claim 1 is characterized in that: The step (5) specifically includes: The data corresponding to the remaining B-B0 batteries in the total set of expert knowledge representations of battery cluster degradation and the total set of health status indicators are standardized using the Z-score standardization method. The standardized expert knowledge representation sets of battery cluster degradation corresponding to the remaining B-B0 batteries are then input into each final health status estimation model in the final health status estimation model set to obtain a fourth total set of health status prediction indicators. The fourth total set of health status prediction indicators is then subjected to absolute difference calculation with the standardized health status indicator vectors corresponding to the remaining B-B0 batteries to obtain a total set of model performance indicators and perform mask calculation to obtain a masked total set of model performance indicators. The standardized battery cluster degradation expert knowledge representation set corresponding to the remaining B-B0 batteries is input into the multi-layer perceptron model to obtain the total set of health state adaptive prediction indicators and perform mask calculation to obtain the total set of masked health state adaptive prediction indicators; A cross entropy calculation is performed on the total set of masked model performance indicators and the total set of masked health status adaptive prediction indicators to obtain a third loss function; and the multilayer perceptron model is trained based on the third loss function to obtain the trained multilayer perceptron model as the adaptive mask selection network.

6. The shared-compensation efficient health assessment method for large energy storage battery systems according to claim 1 is characterized in that: The step (6) specifically includes: Based on the measurement point data at all sampling moments in the discharge phase of any charge-discharge cycle of the battery to be estimated, the corresponding battery cluster degradation expert knowledge representation vector is constructed and standardized. The vector is then input into the adaptive mask selection network to obtain the corresponding model performance estimation index vector and perform mask calculation to obtain the masked model performance estimation index vector. When the h-th masked model performance estimation index in the masked model performance estimation index vector is 1, the h-th final health state estimation model is selected from the final health state estimation model set as the estimation prediction model; the standardized battery cluster degradation expert knowledge representation vector of the battery to be estimated is input into the estimation prediction model to obtain the model output result, and denormalization is performed to obtain the final health state estimation index corresponding to the charge-discharge cycle of the battery to be estimated.

Images