Energy storage battery fault early warning method adopting generative model

Abstract

The invention provides an energy storage battery fault early warning method adopting a generative model, and the method comprises the steps: collecting the historical data of an energy storage battery, and constructing a feature data set based on a time sequence; constructing an unsupervised learning data generation model based on a variational auto-encoder, and obtaining a reconstructed electrical parameter data model through training; and outputting the risk level of the energy storage battery based on the electrical parameter reconstruction result and the actually measured electrical parameter deviation result. According to the invention, based on a variational auto-encoder technology, energy storage battery fault detection is carried out by comprehensively considering charging, discharging and standing phase processes, and new data is generated by combining data analysis of charging and discharging segments and standing segments and utilizing existing data, so that the problems of less training data and low training efficiency are effectively solved. The problem that data in the standing stage characterizes the health state of the battery is not considered, the fault detection precision of the energy storage battery is remarkably improved, the abnormal energy storage battery lacking characterization in the charging and discharging stage can be effectively distinguished, and the method is of great significance to safe and stable operation of an energy storage system.

Description

Technical Field

[0001] This invention relates to the field of energy storage technology, and more specifically to a method for early warning of energy storage battery faults using a generative model. Background Technology

[0002] With the rapid growth of installed capacity of renewable energy sources such as photovoltaics and wind power, large-capacity electrochemical energy storage systems, as a key technology for solving grid fluctuation and intermittency problems, have been deployed on a large scale globally. Among them, lithium-ion batteries have become the mainstream technology due to their high energy density, long cycle life, and fast response characteristics; in recent years, lithium-ion batteries have accounted for a very high proportion of new electrochemical energy storage installations.

[0003] However, with the large-scale deployment of electrochemical energy storage systems, the safety issues of battery energy storage systems are becoming increasingly prominent. Typical failures include overcharging / over-discharging, internal and external short circuits, thermal runaway, and insulation failure. These failures are often caused by inconsistencies in battery performance parameters and evolve into safety accidents through multi-factor coupling mechanisms under complex operating conditions.

[0004] Currently, traditional battery management systems (BMS) mainly rely on threshold judgment methods and equivalent circuit models for fault detection, which has the following significant limitations:

[0005] 1) Early warning lag: Thermal runaway usually only has a few minutes to react from the accumulation of hidden dangers to the trigger threshold, and traditional methods cannot achieve early intervention.

[0006] 2) Strong model dependence: Diagnostic methods based on Kalman filtering or electrochemical models have an error increase of 5-10% under scenarios such as changes in ambient temperature and battery aging.

[0007] 3) Difficulty in identifying multiple fault couplings: Microscopic failure mechanisms such as short circuits, lithium plating, and dendrite growth are difficult to accurately capture using statistical models of macroscopic parameters such as voltage and current.

[0008] 4) High data annotation cost: Supervised learning methods require a large number of fault samples, but the actual fault occurrence rate is less than 0.01%.

[0009] The aforementioned problems result in low accuracy of fault detection for existing energy storage batteries, posing a significant threat to the safe operation of electrochemical energy storage systems. Summary of the Invention

[0010] This application addresses the problems existing in the prior art by providing a data processing method for energy storage batteries that uses a generative model for data reconstruction. Through multi-stage data combination analysis, it can discover potential anomalies in energy storage batteries and improve fault detection accuracy.

[0011] To achieve the above objectives, the technical solution adopted in this application is as follows:

[0012] This application provides a method for early warning of energy storage battery faults using a generative model, which mainly includes the following steps:

[0013] Collect historical data from energy storage batteries;

[0014] Based on the historical data, filter the charging / discharging and static segment data;

[0015] A time-series-based feature dataset is constructed using the charging / discharging and static segment data.

[0016] An unsupervised learning data generation model is constructed based on a variational autoencoder; the data generation model is trained using the feature dataset to obtain a reconstructed electrical parameter data model;

[0017] Real-time charging and discharging data of the energy storage battery are collected and substituted into the reconstructed electrical parameter data model to obtain the electrical parameter reconstruction result;

[0018] By comparing the reconstructed electrical parameters with the corresponding measured electrical parameters, the deviation results are obtained.

[0019] The risk level of the energy storage battery is output based on the deviation results.

[0020] Optionally, the historical data includes charging data, discharging data, and static data; the charging / discharging and static data segments include charging sample segment data and discharging sample segment data.

[0021] The method for filtering historical data includes the following steps:

[0022] Select a charging segment with a continuous charging data length greater than or equal to a frames, and a first static segment with a static data length greater than or equal to b frames after charging; store the charging segment and the corresponding first static segment as the charging sample segment data with a total length greater than or equal to a+b frames; a and b are both positive integers, and a is greater than b; preferably a is 30 and b is 20.

[0023] Select a discharge segment with a continuous discharge data length greater than or equal to c frames, and a second static segment with a static data length greater than or equal to d frames after discharge; store the discharge segment and the second static segment as the discharge sample segment data with a total length greater than or equal to c+d frames; c and d are both positive integers, and c is greater than d; preferably c is 30 and d is 20.

[0024] Optionally, in the charge / discharge and resting segment data, the feature data of each frame includes at least time, voltage, current, state of charge, temperature and total running time.

[0025] Optionally, the method for constructing the feature dataset includes the following steps:

[0026] The charging / discharging and static data segments are preprocessed to obtain preprocessed data;

[0027] The preprocessed data is normalized to obtain the feature dataset.

[0028] Optionally, the method for obtaining the reconstructed electrical parameter data model includes the following steps:

[0029] The encoder and decoder of the data generation model are constructed based on the Transformer architecture.

[0030] The input sample dataset is generated using the aforementioned feature dataset;

[0031] The data generation model is trained using the input sample data, and the reconstructed electrical parameter data model is generated.

[0032] Optionally, the method for constructing the reconstructed electrical parameter data model further includes the following steps:

[0033] The input sample dataset is divided into charging segments or discharging segments, and corresponding static segments;

[0034] The stationary segment is masked by a masking operation, and the charging or discharging segment is input into the encoder to obtain an encoding vector;

[0035] The encoded vector is reparameterized to obtain a first data sequence; the encoded vector is computed through a linear network to obtain a second data sequence.

[0036] Combine the first data sequence and the second data sequence to obtain a combined feature vector;

[0037] The combined feature vector is input into the decoder to obtain reconstructed data;

[0038] A loss function for model training is constructed based on the input sample data and the reconstructed data.

[0039] Optionally, the feature dataset may also be subjected to anomaly data filtering;

[0040] In the abnormal data filtering, sample data corresponding to energy storage batteries in abnormal states are removed.

[0041] Optionally, the process of solving for the deviation result includes the following steps:

[0042] Calculate the cumulative mean square error of the reconstructed electrical parameters and the corresponding measured electrical parameters, and record the cumulative mean square error as the deviation result;

[0043] The cumulative mean square error also includes the tracking error during the data acquisition period.

[0044] Optionally, the method for obtaining the risk level of the energy storage battery includes the following steps:

[0045] Statistical data-based risk assessment methods yield multiple risk assessment thresholds;

[0046] Based on the risk assessment threshold, the range of the deviation result is determined by the threshold comparison method, and the corresponding risk level is obtained.

[0047] Optionally, the following steps may also be included:

[0048] Select a maintenance strategy based on the risk level;

[0049] The maintenance strategy is executed, and the measured status of the energy storage battery is fed back based on the maintenance results;

[0050] The risk assessment threshold is dynamically updated based on the actual measured status of the energy storage battery.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] This invention proposes a novel generative method for detecting faults in energy storage batteries based on variational autoencoder technology, which comprehensively considers the charging, discharging, and resting phases. By combining data analysis of the charging / discharging and resting phases, it generates new data using existing data, effectively solving the problems of insufficient training data and the lack of consideration for the representation of battery health status by data from the resting phase. This significantly improves the accuracy of fault detection in energy storage batteries and can effectively identify abnormal energy storage batteries that lack representation during the charging and discharging phases. It is of great significance for the safe and stable operation of energy storage systems. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a flowchart of a method in a specific embodiment of the present invention;

[0055] Figure 2 This is a structural diagram of the data generation model in a specific embodiment of the present invention;

[0056] Figure 3 This is a comparison chart of voltage curves of normal batteries in a specific embodiment of the present invention;

[0057] Figure 4 This is a comparison chart of voltage curves of abnormal batteries in a specific embodiment of the present invention; Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0060] It is worth noting that, unless otherwise specified, the methods used in this invention are all conventional methods; and the raw materials and equipment used are all conventional commercially available products, and their sources are not specifically limited.

[0061] This embodiment provides a method for early warning of energy storage battery faults using a generative model, such as... Figure 1 As shown, it mainly includes the following steps:

[0062] Historical data of the energy storage battery is collected. First, the daily operating data of the energy storage battery for nearly 90 days is read from the cloud storage server. The data is arranged according to the sampling time sequence, and each time point contains various parameter data of the energy storage battery. Specifically, the operating status of the energy storage battery is divided into stages according to time periods, including charging data, discharging data, and resting data.

[0063] The historical data is used to filter charge / discharge and static storage segment data, which includes both charging and discharging sample data. Specifically, the historical data filtering method includes the following steps:

[0064] The stored content in the cloud storage server is saved in the form of data frames. A data frame (hereinafter referred to as "frame") is a protocol data unit of the data link layer. It consists of three parts: frame header, data part, and frame trailer. Its data part records the various parameter data of the energy storage battery within a sampling interval.

[0065] A charging segment with a continuous charging data length greater than or equal to 30 frames is selected, and a first static segment with a static data length greater than or equal to 20 frames following this charging phase is selected. The charging segment and the corresponding first static segment are stored as charging sample segment data with a total length greater than or equal to 50 frames; specifically, in this embodiment, the number of frames in the charging segment is 30, the number of frames in the first static segment is 20, and the total number of frames in the charging sample segment data is 50.

[0066] A discharge segment with a continuous discharge data length greater than or equal to 30 frames is selected, and a second static segment with a static data length greater than or equal to 20 frames is selected after the discharge stage; the discharge segment and the second static segment are stored as discharge sample segment data with a total length greater than or equal to 50 frames; specifically, in this embodiment, the number of frames of the discharge segment is 30, the number of frames of the second static segment is 20, and the total number of frames of the discharge sample segment data is 50.

[0067] Therefore, after filtering, the historical data can be divided into two types: charging sample fragments containing data of the resting phase and discharging sample fragments. The data within the samples are continuous fragments in time, namely charging + resting and discharging + resting.

[0068] Furthermore, in the charge / discharge and resting segment data, the feature data of each frame includes at least six feature data: time, voltage, current, state of charge (SOC), temperature, and total running time.

[0069] A time-series-based feature dataset is constructed using charge / discharge and resting segment data. The method for constructing the feature dataset includes the following steps:

[0070] Data preprocessing is performed on the charging, discharging, and static segment data, that is, the values ​​of each feature data in the selected charging and discharging segments are removed, and samples with abnormal values ​​and missing data are eliminated, thereby obtaining preprocessed data;

[0071] The preprocessed data is normalized to obtain a feature dataset. Furthermore, in order to obtain a training model of the normal state of the energy storage battery in subsequent training, the feature dataset is also subjected to abnormal data screening. That is, in the abnormal data screening, the sample data corresponding to the abnormal state of the energy storage battery is removed, thereby retaining the data of the healthy state of the energy storage battery.

[0072] An unsupervised learning data generation model is constructed based on a variational autoencoder; the data generation model is trained using a feature dataset to obtain a reconstructed electrical parameter data model; specifically, the method for obtaining the reconstructed electrical parameter data model includes the following steps:

[0073] Combination Figure 2 As shown, the encoder and decoder of the data generation model are first constructed based on the Transformer architecture;

[0074] The input sample dataset is generated through the feature dataset, namely, the aforementioned input sample dataset composed of data from known fault-free and normally operating energy storage batteries. Further, based on the operating state of the energy storage batteries, the continuous segmented data can be divided into charging segments + resting segments, or discharging segments + resting segments. Optionally, to ensure the model's generalization ability, the ratio of charging samples to discharging samples should be kept as close as possible, and a portion of the data should be extracted as a validation dataset.

[0075] A data generation model is trained using input sample data to generate a reconstructed electrical parameter data model. To allow the generation model to focus more on learning the feature representation of the resting phase, the resting segment is first masked using a masking operation. The charging or discharging segment is then input into an encoder for encoding. The encoder maps the input data to a latent space z and outputs distribution parameters to obtain an encoded vector. Next, the encoded vector is reparameterized by z = μ + ε·σ to obtain a first data sequence. Here, ε ~ N(0,1) is external standard normal noise independent of the model, introducing randomness into the sampling process, while μ and σ are deterministic learnable parameters output by the encoder. Furthermore, a second data sequence is obtained by computing the encoded vector using a linear network. The input tensor dimension of the linear network is the length of the charging or discharging stage data multiplied by the number of features, and the output tensor dimension is the length of the resting stage data multiplied by the number of features. The first and second data sequences are combined sequentially to form a combined feature vector. This combined feature vector is then input into a decoder to obtain the reconstructed data. During this period, since the model training only uses known normal samples, the decoder can only generate reconstructed data from the distribution learned from the normal data. Therefore, regardless of whether there are any anomalies in the input data of the energy storage battery during the subsequent model testing process, as long as the latent variables fall within the range of the normal distribution, the reconstruction result still conforms to the normal pattern, that is, the operating result of the energy storage battery in a healthy state is obtained.

[0076] Based on the aforementioned model structure and training method, a loss function for model training is constructed using the input sample data and reconstructed data. In this embodiment, voltage is selected as the representative quantity; therefore, the training loss function comprises two parts: the KL divergence loss function and the mean square error of the reconstructed voltage. Thus, the total loss function (Loss) is calculated as follows:

[0077] Loss = N1 × KL loss +N2×Volt loss ;

[0078] In the formula, N1 and N2 are the weights corresponding to the two loss terms. The initial values ​​of N1 are 4 and N2 are 10000. Their specific values ​​are adjusted according to the training data. KL loss Let Volt be the loss function value of the KL divergence. lossThe loss function value of the reconstructed voltage.

[0079] When the number of training iterations reaches the set quantity and the training loss is lower than the set target value, the training is completed, and the trained model is saved as the reconstructed electrical parameter data model for actual measurement.

[0080] Load the trained reconstructed electrical parameter data model for inference. In the inference mode, collect the real-time charge and discharge data of the energy storage battery, perform the aforementioned data preprocessing, and substitute the preprocessed data into the reconstructed electrical parameter data model to obtain the voltage reconstruction result.

[0081] Compare the voltage reconstruction result with the corresponding measured voltage to obtain the deviation result; the solution process of the deviation result includes the following steps:

[0082] Calculate the cumulative mean square error of the voltage reconstruction result and the corresponding measured voltage. The error includes the tracking error of the charging / discharging part and the static part of multiple collected segments, and record the cumulative mean square error as the deviation result. Calculating the error based on a complete period including the charging / discharging and static stages can not only detect energy storage batteries with abnormal conditions in the charging / discharging stage, but also detect energy storage batteries with potential problems, whose abnormal performance in the charging / discharging stage is not obvious and cannot be discovered by the conventional threshold judgment method or the artificial intelligence feature discrimination method, while this embodiment can discover its abnormality through the voltage change in the static stage.

[0083] Output the risk level of the energy storage battery based on the deviation result; the specific method includes the following steps:

[0084] Obtain multiple risk assessment thresholds based on the risk assessment method of statistical data. A feasible way is to use a large number of existing samples (regardless of normal or abnormal), and respectively count the 25%, 50%, and 75% quantiles of the cumulative mean square error. Set the medium risk threshold T2 = 75% quantile + k1*(75% - 25% quantile), and the high risk threshold T1 = 75% quantile + k2*(75% - 25% quantile), where k1 < k2, preferably k1 = 1.5, k2 = 3; the actual value can be adjusted according to the allowed maximum alarm rate.

[0085] Based on risk assessment thresholds T1 and T2, the range of deviation results is determined using a threshold comparison method, and the corresponding risk level is obtained. A maintenance strategy is then selected based on the risk level. Example: When the battery's cumulative mean square error is greater than or equal to threshold T1, the battery is classified as high-risk, and a red alert is issued; maintenance personnel immediately proceed to the site for inspection. When the battery's cumulative mean square error is greater than threshold T2 but less than T1, the battery's risk level is classified as medium, and a yellow alert is issued; the battery is added to the observation list. When the battery's cumulative mean square error is less than or equal to threshold T2, the battery's risk level is classified as low, and no alert is issued.

[0086] When implementing maintenance strategies, maintenance personnel report the actual measured status of the energy storage battery based on the inspection results, and dynamically update the risk assessment threshold based on the actual measured status of the energy storage battery. The specific method is as follows:

[0087] 1) Set the initial parameters Y = 0 and N = 0;

[0088] 2) When the model determines that the battery is high-risk and the maintenance feedback does indeed show an anomaly, then Y = Y + 1, N = 0; when the model determines that the battery is high-risk but the maintenance feedback does not show an anomaly, then N = N + 1, Y = 0.

[0089] 3) When Y = 3, k1 = k1 * 0.9 and k2 = k2 * 0.9, and Y = 0 is reset; when N = 3, k1 = k1 * 1.1 and k2 = k2 * 1.1, and N = 0 is reset.

[0090] Comprehensive analysis shows that data on the resting phase is lacking in real-world engineering projects, and it is generally believed that data on the resting phase of energy storage batteries can only assist in the threshold judgment of the charging and discharging phases to increase the reference amount. Furthermore, without clear theoretical research, it is impossible to develop a highly accurate method for judging abnormal conditions simply by increasing the monitoring data of the resting phase. Therefore, this application addresses the limitations of existing energy storage battery fault detection methods, particularly their focus on charging and discharging data without considering the representation of battery health status by data from the resting phase. This application first constructs an unsupervised learning voltage data generation model based on a variational autoencoder (VAE). This model learns from a large amount of normal battery sample data, inputting characteristic data from the charging or discharging process, and reconstructing the output voltage data for the complete charging plus resting or discharging plus resting phases, thus expanding the sample data. Furthermore, by calculating the tracking error between the reconstructed voltage data and the actual voltage data, and then determining the battery's risk level based on a threshold, it issues warnings for high-risk batteries. This not only effectively solves a series of limitations in existing energy storage battery fault detection methods, comprehensively considering all processes in the operation of the energy storage battery and improving the accuracy of fault detection, but also, through joint analysis of data from different stages based on time-series conditions, especially the accumulation of tracking errors, it can quickly and concisely identify potential faults in the energy storage battery. Furthermore, the cumulative mean square error of the reconstructed data and measured data over time in this application not only focuses on the voltage change at a certain point in time, but also reflects the overall trend of change, thus providing a fault early warning method that combines static voltage data.

[0091] Comparative Example 1;

[0092] Two sets of batteries were selected: a normal energy storage battery and an abnormal energy storage battery. Each set of energy storage batteries was discharged to a standby state, and the collected data was substituted into the model of the above embodiment to reconstruct the voltage data.

[0093] like Figure 3 and Figure 4 The figure shows a comparison of the partially reconstructed voltage curves. The red curve is the measured voltage curve, and the blue curve is the reconstructed voltage curve. The horizontal axis represents the number of samples, and the vertical axis represents the battery voltage value.

[0094] in, Figure 3 For a known normal energy storage battery, it can be seen from the figure that the model-reconstructed voltage curve 1# and the measured voltage curve 1# have good tracking accuracy, the curve positions are basically consistent, and the corresponding cumulative mean square error is 0.04, which is less than the set threshold T2 (0.05), so the risk level is judged to be low.

[0095] further, Figure 4For the known anomalous energy storage battery, it can be seen from the figure that there is a significant difference between the model-reconstructed voltage curve of #2 and the measured voltage curve of #2, especially... Figure 4 The trend of the stationary voltage curve differs significantly from the trend of the actual voltage curve. The cumulative mean square error of the two samples is 0.23, which is greater than the set threshold T1 (0.15), and the risk level is determined to be high.

[0096] The above comparative examples clearly demonstrate that even if the abnormal battery behaves normally during the data collection period and does not exceed the conventional threshold during the charging phase according to existing technology, the method of this application embodiment can still clearly warn of its risks.

[0097] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, and is not intended to limit the scope of protection of the present invention. Simple modifications or equivalent substitutions made by those skilled in the art to the technical solution of the present invention do not depart from the essence and scope of the technical solution of the present invention.

Claims

1. A method for early warning of energy storage battery faults using a generative model, characterized in that, Includes the following steps: Collect historical data from energy storage batteries; Based on the historical data, filter the charging / discharging and static segment data; A time-series-based feature dataset is constructed using the charging / discharging and static segment data. An unsupervised learning data generation model is constructed based on a variational autoencoder; the data generation model is trained using the feature dataset to obtain a reconstructed electrical parameter data model; Real-time charging and discharging data of the energy storage battery are collected and substituted into the reconstructed electrical parameter data model to obtain the electrical parameter reconstruction result; By comparing the reconstructed electrical parameters with the corresponding measured electrical parameters, the deviation results are obtained. The risk level of the energy storage battery is output based on the deviation results.

2. The energy storage battery fault early warning method using a generative model according to claim 1, characterized in that: The historical data includes charging data, discharging data, and static data; the charging, discharging, and static data segments include charging sample segment data and discharging sample segment data. The method for filtering historical data includes the following steps: Select a charging segment with a continuous charging data length greater than or equal to a frames, and a first static segment with a static data length greater than or equal to b frames after charging; store the charging segment and the corresponding first static segment as the charging sample segment data with a total length greater than or equal to a+b frames; a and b are both positive integers, and a is greater than b; Select a discharge segment with a continuous discharge data length greater than or equal to c frames, and a second static segment with a static data length greater than or equal to d frames after discharge; store the discharge segment and the second static segment as the discharge sample segment data with a total length greater than or equal to c+d frames; c and d are both positive integers, and c is greater than d.

3. The energy storage battery fault early warning method using a generative model according to claim 1 or 2, characterized in that: In the charge / discharge and resting segment data, the characteristic data of each frame includes at least time, voltage, current, state of charge, temperature and total running time.

4. The energy storage battery fault early warning method using a generative model according to claim 1, characterized in that: The method for constructing the feature dataset includes the following steps: The charging / discharging and static data segments are preprocessed to obtain preprocessed data; The preprocessed data is normalized to obtain the feature dataset.

5. The energy storage battery fault early warning method using a generative model according to claim 1, characterized in that: The method for obtaining the reconstructed electrical parameter data model includes the following steps: The encoder and decoder of the data generation model are constructed based on the Transformer architecture. The input sample dataset is generated using the aforementioned feature dataset; The data generation model is trained using the input sample data, and the reconstructed electrical parameter data model is generated.

6. The energy storage battery fault early warning method using a generative model according to claim 5, characterized in that: The method for constructing the reconstructed electrical parameter data model further includes the following steps: The input sample dataset is divided into charging segments or discharging segments, and corresponding static segments; The stationary segment is masked by a masking operation, and the charging or discharging segment is input into the encoder to obtain an encoding vector; The encoded vector is reparameterized to obtain a first data sequence; the encoded vector is computed through a linear network to obtain a second data sequence. Combine the first data sequence and the second data sequence to obtain a combined feature vector; The combined feature vector is input into the decoder to obtain reconstructed data; A loss function for model training is constructed based on the input sample data and the reconstructed data.

7. The energy storage battery fault early warning method using a generative model according to any one of claims 4-6, characterized in that: The feature dataset also undergoes anomaly data filtering; In the abnormal data filtering, sample data corresponding to energy storage batteries in abnormal states are removed.

8. The energy storage battery fault early warning method using a generative model according to claim 1, characterized in that: The process of solving the deviation result includes the following steps: Calculate the cumulative mean square error of the reconstructed electrical parameters and the corresponding measured electrical parameters, and record the cumulative mean square error as the deviation result; The cumulative mean square error also includes the tracking error during the data acquisition period.

9. The energy storage battery fault early warning method using a generative model according to claim 1, characterized in that: The method for obtaining the risk level of the energy storage battery includes the following steps: Statistical data-based risk assessment methods yield multiple risk assessment thresholds; Based on the risk assessment threshold, the range of the deviation result is determined by the threshold comparison method, and the corresponding risk level is obtained.

10. The energy storage battery fault early warning method using a generative model according to claim 9, characterized in that: It also includes the following steps: Select a maintenance strategy based on the risk level; The maintenance strategy is executed, and the measured status of the energy storage battery is fed back based on the maintenance results; The risk assessment threshold is dynamically updated based on the actual measured status of the energy storage battery.

Images