Battery fault prediction method, model training method, equipment, system and medium

Abstract

The embodiment of the invention relates to the technical field of battery management, in particular to a battery fault prediction method, a model training method, equipment, a system and a medium. The method mainly comprises the steps of obtaining a training data set and a fault data set which carry multiple battery state parameters from historical operation data of a battery, performing data processing on the training data set and the fault data set, extracting feature data corresponding to the battery state parameters, obtaining a training feature set and a fault feature set, and performing data processing on the training feature set and the fault feature set; carrying out feature correlation analysis on each feature in the fault feature set to obtain feature correlation of each feature corresponding to the fault type, taking the training feature set as training sample data, inputting the training sample data into an initial model constructed based on a recurrent neural network and Transform to obtain an initial prediction model, and carrying out prediction on the initial prediction model; and adjusting parameters of the initial prediction model according to the feature relevancy to obtain a battery fault prediction model. By adopting the method, the accuracy of predicting and identifying the battery fault by the battery fault prediction model can be improved.

Description

Technical Field

[0001] The present disclosure relates to the field of battery management technology, and in particular to a battery fault prediction method, model training method, device, system and medium. Background Art

[0002] With the rapid development of electric vehicles, energy storage systems, and other fields, the safety and reliability of batteries, as core energy components, have attracted significant attention. However, batteries can fail during use due to factors such as overcharging, overdischarging, and temperature fluctuations, which can even lead to safety accidents. Therefore, accurately predicting battery failures is crucial to ensuring the safe operation of battery systems.

[0003] Currently, battery failure prediction technology primarily relies on traditional statistical methods and simple machine learning models. For example, threshold detection methods based on operating data such as battery voltage and current, or predictions based on shallow machine learning models, while simple to implement, struggle to fully capture the complex characteristics of battery failures, resulting in inaccurate battery failure prediction results. Summary of the Invention

[0004] Based on this, it is necessary to provide a battery fault prediction method, model training method, device, system and medium to improve the accuracy of battery fault prediction in response to at least one of the above technical problems.

[0005] In a first aspect, an embodiment of the present disclosure provides a battery fault prediction model training method, which may include the following steps: obtaining a training data set and a fault data set carrying multiple battery status parameters from the battery historical operation data, performing data processing on the training data set and the fault data set, extracting feature data corresponding to the battery status parameters, obtaining a training feature set and a fault feature set, performing feature correlation analysis on each feature in the fault feature set, obtaining the feature correlation of each feature corresponding to the fault type, using the training feature set as training sample data, inputting it into an initial model constructed based on a recurrent neural network and a Transformer, obtaining an initial prediction model, adjusting the parameters of the initial prediction model according to the feature correlation, and obtaining a battery fault prediction model.

[0006] Among them, the training data set includes multiple normal data recording the normal operation status of the battery and multiple fault data recording the fault status of the battery. Any fault data includes a fault type label used to indicate the fault type, and the feature correlation is used to indicate the frequency or intensity of abnormal battery status parameters corresponding to the feature.

[0007] By using recurrent neural networks and Transformers to build an initial model and inputting the training dataset into the initial model for training, the initial prediction model is obtained. This combines the advantages of recurrent neural networks and Transformers to capture short-term and long-term dependencies in the sequence data of the training dataset. A comprehensive analysis of the time series data is performed to better capture the dynamic changes in the feature data corresponding to the battery state parameters, improving the accuracy of the prediction results. Furthermore, by adjusting the parameters of the initial prediction model based on feature correlation, a battery fault prediction model is obtained, which can further optimize the model training effect and improve model performance.

[0008] In some embodiments, the characteristic data includes at least voltage characteristic data, current characteristic data, temperature characteristic data, and internal resistance characteristic data.

[0009] When training a model, considering multiple feature data can improve the accuracy of model training.

[0010] In some embodiments, performing feature correlation analysis on each feature in the fault feature set may include the steps of: calculating a Pearson correlation coefficient or performing a heat map analysis on each feature in the fault feature set.

[0011] The Pearson correlation coefficient can be used to quantify the correlation between features. Heat map analysis can intuitively display the correlation between features through color depth, which has the effect of intuitively displaying feature relationships. By performing feature correlation analysis, redundant features can be removed, the model complexity can be reduced, and the computational cost and time of model training can be reduced.

[0012] In some embodiments, performing feature correlation analysis on each feature in the fault feature set may include the following steps: performing Pearson correlation coefficient calculation or heat map analysis on each feature in the fault feature set to obtain preliminary feature correlation of each feature, and optimizing the preliminary feature correlation according to preset fault mechanism data.

[0013] Among them, the fault mechanism data is used to represent the change process of various battery status parameters corresponding to various types of battery faults.

[0014] By further utilizing the fault mechanism data to refine the results of the preliminary feature correlation analysis, the feature correlation situation is made more consistent with the actual situation, thereby improving the accuracy and generalization ability of the model.

[0015] In some embodiments, the structure of the initial model may include an input layer, an intermediate layer, a mapping layer, and an output layer.

[0016] Among them, the middle layer includes the Transformer sublayer and the LSTM sublayer, and the mapping layer includes the pooling layer or the fully connected layer.

[0017] The input layer receives the training sample data and converts it into vector data to meet the model's data requirements. The middle layer, comprising an LSTM sublayer and a Transformer sublayer, employs a cascaded LSTM and Transformer architecture. The LSTM sublayer processes the time series data within the vector data, extracting local temporal and data features. The Transformer sublayer uses a self-attention mechanism to capture global dependencies between the local temporal and data features to obtain perceptual sequence features. The mapping layer maps the perceptual sequence features to obtain predicted data. The output layer processes the predicted data and outputs the corresponding prediction results.

[0018] By adopting a multi-layer LSTM and Transformer cascade structure, the dynamic changes of battery status parameters in the battery capture training sample data, that is, the battery historical operation data, can be better captured, thereby improving the ability to accurately predict battery failures.

[0019] In some embodiments, the battery fault prediction model training method may further include the following steps: obtaining a test data set from the battery historical operation data, performing data processing on the test data set to obtain a test feature set, inputting the test feature set into the battery fault prediction model to obtain test results, obtaining evaluation indicators of the battery fault prediction model based on the test results, and adjusting the number of training times, learning rate and / or the number of network layers within the model based on the evaluation indicators to optimize the model quality.

[0020] By using test data to test the accuracy of the trained battery failure prediction model, the model effect can be verified, thereby improving the accuracy of the model prediction results.

[0021] In some embodiments, the training feature set is used as training sample data and input into an initial model constructed by a recurrent neural network and a Transformer to obtain an initial prediction model, which can include the steps of iteratively training the training sample data through the initial model, calculating the error using a cross-entropy loss function, and updating the model parameter weights through an Adam optimizer to obtain an initial prediction model.

[0022] Calculating error using the cross-entropy loss function effectively measures the difference between the predicted probability distribution and the true distribution. The gradient calculation of the cross-entropy loss function is simple and stable, which helps the model converge quickly. The Adam optimizer's momentum mechanism and adaptive learning rate improve the stability of model training, thereby enhancing model performance.

[0023] In some embodiments, the battery fault prediction model training method may further include the steps of: performing data cleaning and data conversion on the training data set and the fault data set.

[0024] By performing data cleaning and data conversion on the operating data in the training dataset and the fault dataset, the data quality of the training dataset and the fault dataset can be improved, the consistency of the data can be ensured, and the speed of model training can be increased.

[0025] In a second aspect, an embodiment of the present disclosure provides a battery fault prediction method, which may include the following steps: acquiring operating data of a battery to be tested, inputting the operating data of the battery to be tested into a battery fault prediction model, and obtaining a fault prediction result of the battery to be tested.

[0026] The battery fault prediction model is obtained by training the battery fault prediction model training method provided in any embodiment of the first aspect of the present disclosure.

[0027] By inputting the operating data of the battery to be tested into a pre-trained battery fault prediction model for fault prediction, a highly accurate battery fault prediction result can be obtained, battery faults can be discovered and managed in a timely manner, and the safety of battery operation can be improved.

[0028] In some embodiments, the fault prediction result includes predicted fault type information, and the fault prediction method may further include: sending the fault prediction result to a BMS unit, so that the BMS unit controls the operating state of the battery to be tested according to the predicted fault type information.

[0029] Sending the fault prediction results to the BMS unit can enable the BMS unit to promptly control the battery operating status according to the battery fault, thereby improving the safety of battery operation and the timeliness of battery fault processing.

[0030] In a third aspect, an embodiment of the present disclosure provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the battery fault prediction model training method provided in any embodiment of the present disclosure in the first aspect are implemented.

[0031] In a fourth aspect, an embodiment of the present disclosure provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the battery fault prediction method provided in any embodiment of the second aspect of the present disclosure are implemented.

[0032] In a fifth aspect, an embodiment of the present disclosure provides a battery failure prediction system, which includes a BMS unit and a model operation unit.

[0033] The battery fault prediction model deployed in the model operation unit is trained by the battery fault prediction model training method provided in any embodiment of the first aspect of the present disclosure.

[0034] The BMS unit is communicatively connected to the model operation unit, and is used to send the operating data of the battery to be tested to the model operation unit, and control the operating state of the battery to be tested according to the fault prediction result fed back by the model operation unit.

[0035] The model operation unit is used to obtain the operation data of the battery to be tested, input the operation data of the battery to be tested into the battery fault prediction model, and obtain the fault prediction result.

[0036] The battery fault prediction model is deployed in the model operation unit and does not occupy the system resources of the BMS unit. It can improve the response speed and real-time performance of battery fault prediction.

[0037] In some embodiments, the battery fault prediction system further includes a remote upgrade unit, which is communicatively connected to the model operation unit and is used to remotely upgrade the battery fault prediction model deployed in the model operation unit.

[0038] The remote upgrade unit can realize remote upgrade and update of the battery fault prediction model, which can make the model upgrade more convenient, improve the efficiency of fault prediction model update, and avoid poor model performance due to untimely model update.

[0039] In a sixth aspect, an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the battery fault prediction model training method provided in any embodiment of the first aspect of the present disclosure.

[0040] In a seventh aspect, an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the battery fault prediction method provided in any embodiment of the second aspect of the present disclosure.

[0041] The battery fault prediction method, model training method, device, system, and medium described above utilize a recurrent neural network (RNN) and a Transformer to construct an initial model. A training dataset consisting of both normal and faulty data is input into the initial model for training to obtain an initial prediction model. This method combines the advantages of RNN and Transformer to capture short-term and long-term dependencies in the sequence data of the training dataset, and comprehensively analyzes the time series data therein. This allows for better capture of dynamic changes in the characteristic data corresponding to battery state parameters, thereby improving the accuracy of the prediction results. Furthermore, by adjusting the parameters of the initial prediction model based on feature correlation, a battery fault prediction model is obtained, which can further optimize the model training effect and improve model performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A diagram illustrating an application environment of a battery failure prediction model training method in some embodiments;

[0043] Figure 2 A flowchart of a battery failure prediction model training method in some embodiments;

[0044] Figure 3 is a flowchart of a feature correlation analysis step in some embodiments;

[0045] Figure 4 is a schematic diagram of a flow chart involving model testing steps in some embodiments;

[0046] Figure 5 is a flowchart of a battery failure prediction method in some embodiments;

[0047] Figure 6 is a diagram of the internal structure of a computer device in some embodiments;

[0048] Figure 7 FIG. 1 is a schematic diagram of a battery failure prediction system in some embodiments. DETAILED DESCRIPTION

[0049] In order to make the technical solutions and advantages of the present disclosure more clearly understood, the following embodiments of the present disclosure and related technical contents are further described in detail in conjunction with the accompanying drawings and text descriptions. It should be understood that the embodiments described below are only used to explain the technical solutions of the embodiments of the present disclosure and are not intended to limit the further possible implementations of the present disclosure.

[0050] It should be noted that relational terms such as "first" and "second" appearing herein are used solely to distinguish between things, states, or actions and do not necessarily indicate or imply relative importance or a sequential relationship. The terms "include," "comprising," or any other variations thereof are used to indicate non-exclusive inclusion, and the objects included are not limited to the objects listed herein. The term "plurality" or any other variations thereof is used to indicate that the number of objects is two or more.

[0051] In a first aspect, an embodiment of the present disclosure provides a method for training a battery fault prediction model. The method can be applied to Figure 1 In the application environment shown. The application environment includes a first device 101 and a second device 102, wherein the first device 101 is used to perform battery fault prediction model training, and can be a local computer device, a server, or other devices capable of performing battery fault prediction model training. The second device 102 can be a device that runs a battery management unit (BMS), a memory that stores battery historical operation data, or other devices for collecting or recording battery historical operation data. The first device 101 can communicate with the second device 102 through a network or other means to receive battery historical operation data from the second device 102.

[0052] The battery failure prediction model training method is applied to Figure 1 The first device 101 in FIG. 1 is taken as an example to illustrate, in some embodiments, such as Figure 2 As shown, the battery fault prediction model training method includes steps S201 to S205 that can be executed by the first device 101. Each step is described in detail below.

[0053] Step S201: obtaining a training data set and a fault data set containing multiple battery status parameters from historical battery operation data.

[0054] Among them, the training data set includes multiple normal data recording the normal operating status of the battery and multiple fault data recording the fault status of the battery. The battery historical operation data in the training data set and the fault data set both carry multiple battery status parameters, and any fault data includes a fault type label for indicating the fault type. The multiple battery status parameters may include voltage, current, temperature, internal resistance and other parameters used to indicate the health status of the battery, and may also include battery state of charge, health status, pressure difference, gas, etc. The fault type label is used to indicate the fault type corresponding to the current fault data, and may include faults such as internal short circuit, overcharge, over-discharge, thermal runaway, etc. of the battery. The battery historical operation data may be the historical operation data of the battery in the battery compartment of a completed energy storage project, for example, it may be the historical operation data of energy storage equipment used for industrial, commercial and grid-level applications.

[0055] In some specific examples, the training data set may be operating data that accounts for 70% of all historical operating data of the battery, and the fault data set may be all or part of the fault data in the historical operating data of the battery.

[0056] Specifically, the first device 101 obtains normal operation data and fault operation data from the battery historical operation data as a training data set, and obtains fault data from the battery historical operation data as a fault data set. The battery historical operation data can be obtained directly from the BMS unit that records the operation data, or from the battery database or memory that stores the battery historical operation data. The source of the operation data is not particularly limited here.

[0057] In some specific embodiments, the battery fault prediction model training method may further include grouping the operating data in the fault dataset according to fault type labels to obtain multiple fault data groups. The grouped fault data groups may include an overcharge fault data group, an over-discharge fault data group, a short-circuit fault data group, and a thermal runaway fault data group.

[0058] In some specific embodiments, after obtaining the training data set and the fault data set of the battery, a training database and a fault database can be constructed respectively so that they can be directly called when there is a need for model training of other batteries in the future.

[0059] Step S202: Process the training data set and the fault data set to extract feature data corresponding to the battery status parameters to obtain a training feature set and a fault feature set.

[0060] The characteristic data may include voltage characteristic data, current characteristic data, temperature characteristic data and internal resistance characteristic data corresponding to various battery state parameters.

[0061] Specifically, preliminary feature engineering is performed on the operating data in the training data set and the fault data set to extract various features corresponding to various battery status parameters and obtain feature data.

[0062] Step S203: performing feature correlation analysis on each feature data in the fault feature set to obtain the feature correlation of each feature corresponding to the fault type.

[0063] The feature correlation is used to indicate the frequency or intensity of abnormality of the battery status parameter corresponding to the feature.

[0064] Abnormal battery status parameters refer to abnormalities in the parameters used to represent the battery's health status, such as abnormal voltage increase or decrease, voltage outside the normal range, voltage returning to zero, abnormal current fluctuation, current returning to zero, temperature increase, internal resistance decrease, internal resistance increase, or internal resistance failure. Different abnormalities or combinations of abnormalities correspond to different fault types. For example, when the characteristic data includes a sudden voltage drop, abnormal current fluctuation, temperature increase, and internal resistance decrease, the corresponding battery fault type is an internal short circuit. For another example, when the characteristic data shows voltage returning to zero, current returning to zero, and a sharp temperature increase, the corresponding battery fault type is thermal runaway.

[0065] In some specific embodiments, feature correlation analysis is performed on the feature data in the fault feature set. This can be performed by statistically analyzing the probability or intensity of occurrence of features corresponding to each type of fault in the fault data. The feature correlation of each feature can be expressed as a percentage. Based on the feature correlation of each feature corresponding to different fault types, the strength of the correlation between each feature and the corresponding fault type can be determined.

[0066] Step S204: The training feature set is used as a training sample and input into the initial model constructed by the recurrent neural network and the Transformer for training to obtain an initial prediction model.

[0067] In the model training phase, the training feature set is input into the initial model constructed by recurrent neural networks (RNN) and Transformer (a sequence model based on attention mechanism) for training to obtain the initial prediction model.

[0068] Among them, the initial model includes a Transformer layer and a recurrent neural network layer, and the recurrent neural network layer can adopt a long short-term memory network (LSTM, Long Short-Term Memory).

[0069] Step S205: adjusting the parameters of the initial prediction model according to the feature correlation to obtain a battery failure prediction model.

[0070] The parameters of the initial prediction model adjusted according to the feature relevance may be feature weight coefficients corresponding to each feature, or may be judgment thresholds corresponding to classification results in the initial prediction model.

[0071] In some specific embodiments, the feature weight coefficient corresponding to each feature in the initial prediction model is adjusted according to the feature correlation. When the feature correlation is high, the weight coefficient corresponding to the feature is increased to enhance the influence of the feature on the model prediction result.

[0072] In some specific embodiments, the judgment threshold of the initial prediction model is adjusted according to the feature relevance, and the interval of the judgment threshold is adjusted according to the feature relevance to improve the accuracy of the model prediction result.

[0073] By using a recurrent neural network (RNN) and Transformer to build an initial model, a training dataset consisting of normal data and fault data is input into the initial model for training to obtain an initial prediction model. This combines the advantages of RNN and Transformer to capture the short-term and long-term dependencies of the sequence data in the training dataset. A comprehensive analysis of the time series data is performed to better capture the dynamic changes in the feature data corresponding to the battery status parameters, improving the accuracy of the prediction results. Furthermore, by adjusting the parameters of the initial prediction model based on feature correlation, a battery fault prediction model is obtained, which can further optimize the model training effect and improve model performance.

[0074] In some embodiments, the battery fault prediction method may further include inputting a training data set into an initial model for training, and adjusting or training a judgment threshold in the initial model using feature correlation to obtain a battery fault prediction model.

[0075] In some embodiments, the characteristic data includes at least voltage characteristic data, current characteristic data, temperature characteristic data, and internal resistance characteristic data. The characteristic data may also include pressure difference characteristic data, gas characteristic data, and insulation failure characteristic data. Considering multiple characteristic data during model training can improve the accuracy of model training.

[0076] In some embodiments, performing feature correlation analysis on each feature in the fault feature set may include: performing Pearson correlation coefficient calculation or heat map analysis on each feature data in the fault feature set.

[0077] The Pearson correlation coefficient can be used to quantify the correlation between features. Heat map analysis can intuitively display the correlation between features through color depth, which has the effect of intuitively displaying feature relationships. In addition, by performing feature correlation analysis, highly correlated feature pairs can be found, redundant features can be removed, model complexity can be reduced, and the computational cost and time of model training can be reduced.

[0078] In some embodiments, performing feature correlation analysis on each feature in the fault feature set may include step S301 and step S302.

[0079] Step S301: Calculate the Pearson correlation coefficient or perform heat map analysis on each feature in the fault feature set to obtain preliminary feature correlation.

[0080] Step S302: Optimize the preliminary feature correlation according to the preset fault mechanism data.

[0081] The fault mechanism data represents the changes in various battery status parameters corresponding to various battery faults. The fault mechanism data may include characteristic data corresponding to various battery faults. The feature correlations of the initial feature correlations in the feature data that do not match the fault mechanism data are adjusted to obtain optimized feature correlations.

[0082] By further utilizing the fault mechanism data to refine the results of the preliminary feature correlation analysis, the feature correlation situation is made more consistent with the actual situation, thereby improving the accuracy and generalization ability of the model.

[0083] In some embodiments, feature correlation may include high correlation, medium correlation and low correlation, and the feature correlation may be divided into high correlation, medium correlation and low correlation according to the numerical value of the calculated Pearson correlation number or the color depth of the heat map.

[0084] In some embodiments, the feature correlation can be expressed as a percentage, and the feature correlation of each feature corresponding to each fault type can be as shown in Table 1.

[0085] Table 1: Correlation of various fault characteristics

[0086]

[0087] Table 1 shows the characteristic data and characteristic correlation of each characteristic corresponding to each type of battery failure. The battery failure types in the table include short circuit, overcharge, overdischarge and thermal runaway. The percentage in the table is the characteristic correlation of each characteristic data.

[0088] In some embodiments, the structure of the initial model may include an input layer, an intermediate layer, a mapping layer, and an output layer.

[0089] Among them, the middle layer includes the Transformer sublayer and the LSTM sublayer, and the mapping layer includes the pooling layer or the fully connected layer.

[0090] The input layer receives the training sample data and converts it into vector data to meet the model's data requirements. The middle layer, comprising an LSTM sublayer and a Transformer sublayer, employs a cascaded LSTM and Transformer architecture. The LSTM sublayer processes the time series data within the vector data, extracting local temporal and data features. The Transformer sublayer uses a self-attention mechanism to capture global dependencies between these features and obtain perceptual sequence features. The mapping layer maps the perceptual sequence features to obtain predicted data. The output layer processes the predicted data and outputs the corresponding prediction results.

[0091] The Transformer and LSTM sublayers can each have multiple layers. The training sample data includes multi-dimensional features such as voltage, current, and temperature, which represent the battery's health status. By stacking multiple layers of Transformers and LSTMs, the dynamic changes in battery status parameters in the training sample data—that is, historical battery operating data—can be better captured, improving the ability to accurately predict battery failures. The perceived sequence features are context-aware feature representations generated by the Transformer sublayer through a self-attention mechanism, capturing global dependencies from local time series features and data features.

[0092] In some embodiments, as Figure 4 As shown, the battery fault prediction model training method further includes steps S401 to S404.

[0093] Step S401: Acquire a test data set from historical battery operation data.

[0094] Step S402: Process the test data set to obtain a test feature set.

[0095] Step S403: input the test feature set into the battery failure prediction model to obtain the test results.

[0096] Step S404: obtaining evaluation indicators of the battery fault prediction model according to the test results, and adjusting the number of training times, learning rate and / or the number of network layers within the model according to the evaluation indicators to optimize the model quality.

[0097] The test dataset can be 30% of the battery's historical operating data. This data can be recorded and stored by the BMS unit or stored in a specific database. Evaluation metrics can include accuracy and recall, as well as classification task evaluation metrics such as precision and F1 score.

[0098] The data processing operations performed on the running data in the test data set may include data cleaning, data conversion, and preliminary feature engineering to extract features from the test data set and obtain a test feature set.

[0099] When the obtained evaluation index is lower than the preset qualification standard, the model's hyperparameters are tuned. For example, the number of training times, learning rate, and the number of Transformer layers can be adjusted accordingly to improve the accuracy of the model's prediction results.

[0100] By using test data to test the accuracy of the trained battery failure prediction model, the model effect can be verified, thereby improving the accuracy of the model prediction results.

[0101] In some embodiments, step S204 may include: iteratively training the training sample data through the initial model, calculating the error using the cross entropy loss function, updating the model parameter weights through the Adam optimizer (Adaptive Moment Est i mat ion, an optimization algorithm based on gradient descent), and obtaining the initial prediction model.

[0102] Calculating error using the cross-entropy loss function effectively measures the difference between the predicted probability distribution and the true distribution. The gradient calculation of the cross-entropy loss function is simple and stable, which helps the model converge quickly. The Adam optimizer's momentum mechanism and adaptive learning rate improve the stability of model training, thereby enhancing model performance.

[0103] In some embodiments, the battery fault prediction model training method may further include performing data cleaning and data conversion on the operating data in the training data set and the fault data set.

[0104] Data cleaning can include processing missing values in the data, for example, supplementing or deleting missing data. Data transformation can be standardizing or normalizing continuous values of data so that different data features have the same dimension.

[0105] By performing data cleaning and data conversion on the operating data in the training dataset and the fault dataset, the data quality of the training dataset and the fault dataset can be improved, the consistency of the data can be ensured, and the speed of model training can be increased.

[0106] It should be understood that although Figures 2 to 4 The steps in the flowchart are displayed in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Figures 2 to 4 Unless otherwise specified herein, the steps shown and the steps involved in other embodiments are not strictly limited in order of execution and can be executed in other orders. Moreover, at least a portion of the steps in the aforementioned embodiments may include multiple sub-steps or multiple stages, which are not necessarily executed at the same time but can be executed at different times. The order of execution of these sub-steps or stages is not necessarily sequential but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0107] In a second aspect, the present disclosure provides a method for predicting battery failure. Figure 5 As shown, the battery failure prediction method includes step S501 and step S502.

[0108] Step S501: Acquire operating data of the battery to be tested.

[0109] Step S502: inputting the operating data of the battery to be tested into the battery fault prediction model to obtain the battery fault prediction result.

[0110] The battery fault prediction model is obtained by training the battery fault prediction model training method provided in any embodiment of the first aspect of the present disclosure.

[0111] The battery to be tested can be a large energy storage device battery, such as a battery stack or battery cell in a power station, or a power battery pack in a new energy vehicle.

[0112] Specifically, the operating data may be battery status data collected in real time by a battery status monitoring unit, including battery voltage, current, temperature, internal resistance, etc. Obtaining the operating data of the battery to be tested may be that the first device 101 obtains the operating data recorded by the BMS unit.

[0113] By inputting the operating data of the battery to be tested into a pre-trained battery fault prediction model for fault prediction, a highly accurate battery fault prediction result can be obtained, battery faults can be discovered and managed in a timely manner, and the safety of battery operation can be improved.

[0114] In some embodiments, the fault prediction result may include predicted fault type information, where the predicted fault type information is used to indicate the predicted fault type. The predicted fault type may include short circuit, overcharge, overdischarge, and thermal runaway. The predicted fault type may be set based on actual conditions and is not particularly limited herein. The battery fault prediction method may further include transmitting the fault prediction result to a battery management system (BMS) unit, so that the BMS unit can control the operating state of the battery under test based on the predicted fault type information.

[0115] Sending the fault prediction results to the BMS unit can enable the BMS unit to promptly control the battery operating status according to the battery fault, thereby improving the safety of battery operation and the timeliness of battery fault processing.

[0116] Specifically, the BMS unit will obtain the operating data of the battery under test and input it into the battery fault prediction model to perform fault prediction and feedback the fault prediction results to the BMS unit. The BMS unit will determine and execute the corresponding fault response operation based on the fault type in the fault prediction results and the preset fault response strategy. In some specific examples, the fault response strategy can include the following three categories:

[0117] (1) Level 1 fault: charging and discharging are prohibited, and the battery system is shut down as a whole;

[0118] (2) Level 2 fault: charging and discharging are prohibited, and the battery system is partially shut down;

[0119] (3) Level 3 fault: an alarm will be issued.

[0120] In some embodiments, the fault prediction result may include fault level information, which is used to indicate the fault level. The battery fault prediction method may further include sending the fault prediction result to the BMS unit so that the BMS unit can respond to the fault according to the fault level.

[0121] Among them, the fault level can be set according to the actual situation. For example, thermal runaway and short circuit faults are level one faults, and voltage overcharge and over-discharge faults can be set to level one, level two or level three faults according to the specific voltage threshold. Specifically, different fault levels can be determined by setting the percentage interval range of the voltage overcharge value or over-discharge value exceeding the voltage standard value. Different levels correspond to different fault response strategies. For example, when the fault level is level one, the fault response strategy of the BMS unit may include issuing an early warning notification, controlling the entire battery system to shut down, and starting the fire protection system. When the fault level is level two, the fault response strategy may include issuing an early warning notification and controlling the partial shutdown of the battery system. When the fault level is level three, the fault response strategy may include issuing an early warning notification. The early warning notification may take the form of text notification, sound alarm notification, etc., and the notification may be sent in the form of a message prompt, pop-up window reminder or other methods on the supervision platform.

[0122] In a third aspect, an embodiment of the present disclosure provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the battery fault prediction model training method provided in any embodiment of the present disclosure in the first aspect are implemented.

[0123] In a fourth aspect, an embodiment of the present disclosure provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the battery fault prediction method provided in any embodiment of the second aspect of the present disclosure are implemented.

[0124] In some embodiments, the computer device may be a server, and its internal structure may be as follows: Figure 6As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store historical battery operation data. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, it implements the battery fault prediction model training method in any embodiment of the present invention, or, when the computer program is executed by the processor, it implements the battery fault prediction method in any embodiment of the present invention.

[0125] Those skilled in the art will understand that Figure 6 The structure shown in the figure is merely a block diagram of a portion of the structure related to the embodiment scheme of the present disclosure, and does not constitute a limitation on the computer device to which the embodiment scheme of the present disclosure is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0126] In a fifth aspect, an embodiment of the present disclosure provides a battery failure prediction system, such as Figure 7 As shown, the battery fault prediction system 700 includes a BMS unit 710 and a model operation unit 720. The battery fault prediction model deployed in the model operation unit is trained by the battery fault prediction model training method provided in any embodiment of the first aspect of the present disclosure.

[0127] The BMS unit 710 is in communication with the model operation unit 720 and is configured to send the operation data of the battery to be tested to the model operation unit and control the operation state of the battery to be tested according to the fault prediction result fed back by the model operation unit.

[0128] The model operation unit 720 is used to obtain the operation data of the battery to be tested, input the operation data of the battery to be tested into the battery fault prediction model, and obtain a fault prediction result.

[0129] The battery fault prediction model is independently deployed in the model operation unit and does not occupy the system resources of the BMS unit, which can improve the response speed and real-time performance of battery fault prediction.

[0130] In some embodiments, the system further includes a remote upgrade unit, which is in communication with the model operation unit and is configured to remotely upgrade the battery fault prediction model deployed in the model operation unit 720 .

[0131] The remote upgrade unit can realize remote upgrade and update of the battery fault prediction model, which can make the model upgrade more convenient, improve the efficiency of fault prediction model update, and avoid poor model performance due to untimely model update.

[0132] In a sixth aspect, an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the battery fault prediction model training method provided in any embodiment of the first aspect of the present disclosure.

[0133] The computer readable storage medium may be Figure 6 A computer-readable storage medium in the computer device shown.

[0134] In a seventh aspect, an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the battery fault prediction method provided in any embodiment of the second aspect of the present disclosure.

[0135] The computer readable storage medium may be Figure 6 A computer-readable storage medium in the computer device shown.

[0136] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program. The aforementioned computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments of the present disclosure may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0137] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this disclosure.

[0138] The above embodiments merely represent several implementation methods of the present disclosure. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of protection of the present disclosure. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present disclosure, all of which fall within the scope of protection of the present disclosure. Therefore, the scope of protection of the present disclosure shall be determined by the appended claims.

Claims

1. A battery failure prediction model training method, characterized in that: The method comprises: A training dataset and a fault dataset containing multiple battery status parameters are obtained from historical battery operation data, where each piece of fault data includes a fault type label indicating the fault type. Processing the training data set and the fault data set to extract feature data corresponding to the battery state parameters to obtain a training feature set and a fault feature set; Performing feature correlation analysis on each feature in the fault feature set to obtain a feature correlation of each feature corresponding to the fault type, wherein the feature correlation is used to indicate the frequency or intensity of abnormality of the battery status parameter corresponding to the feature; The training feature set is used as training sample data and input into an initial model constructed by a recurrent neural network and a Transformer to obtain an initial prediction model; The parameters of the initial prediction model are adjusted according to the feature correlation to obtain a battery failure prediction model.

2. The method according to claim 1, characterized in that The characteristic data at least includes voltage characteristic data, current characteristic data, temperature characteristic data and internal resistance characteristic data.

3. The method according to claim 1, characterized in that The performing feature correlation analysis on each feature in the fault feature set includes: performing Pearson correlation coefficient calculation or heat map analysis on each feature in the fault feature set.

4. The method according to claim 1, wherein The performing feature correlation analysis on each feature in the fault feature set includes: Performing Pearson correlation coefficient calculation or heat map analysis on each feature in the fault feature set to obtain preliminary feature correlation of each feature; The preliminary feature correlation is optimized according to preset fault mechanism data, wherein the fault mechanism data is used to represent the change process of various battery status parameters corresponding to various types of battery faults.

5. The method according to claim 1, wherein The initial model includes: An input layer, configured to receive the training sample data and convert the data format of the training sample data into vector data; The middle layer includes an LSTM sublayer and a Transformer sublayer, wherein the LSTM sublayer is used to process the time series data in the vector data and extract local time series features and data features, and the Transformer sublayer is used to capture global dependencies between the local time series features and the data features through a self-attention mechanism to obtain perceptual sequence features; A mapping layer, including a pooling layer or a fully connected layer, is used to map the perception sequence features to obtain prediction data; The output layer is used to process the prediction data and output the prediction results.

6. The method according to claim 1, characterized in that The method further comprises, Acquire a test data set from the battery historical operating data; Performing data processing on the test data set to obtain a test feature set; Inputting the test feature set into the battery failure prediction model to obtain a test result; An evaluation index of the battery fault prediction model is obtained according to the test results, and the number of training times, the learning rate and / or the number of network layers within the model are adjusted according to the evaluation index to perform model quality tuning.

7. The method according to claim 1, characterized in that The training feature set is used as training sample data and input into an initial model built based on a recurrent neural network and a Transformer to obtain an initial prediction model, including: Iteratively training the training sample data using the initial model; The error is calculated using the cross entropy loss function; The model parameter weights are updated using the Adam optimizer to obtain the initial prediction model.

8. The method according to claim 1, characterized in that The method further comprises: Data cleaning and data conversion are performed on the training data set and the fault data set.

9. A battery failure prediction method, characterized in that: The method comprises: Obtaining the operating data of the battery under test; The operating data of the battery to be tested is input into the battery fault prediction model to obtain a fault prediction result of the battery to be tested, wherein the battery fault prediction model is trained by the training method according to any one of claims 1 to 8.

10. The method according to claim 9, characterized in that The fault prediction result includes predicted fault type information, and the method further includes: The fault prediction result is sent to the BMS unit, so that the BMS unit can control the operating state of the battery to be tested according to the predicted fault type information.

11. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 10 are implemented.

12. A battery failure prediction system, characterized in that: The system includes a BMS unit and a model operation unit, wherein the battery fault prediction model deployed in the model operation unit is trained by the training method according to any one of claims 1 to 8; The BMS unit is communicatively connected to the model operation unit, and is used to send the operating data of the battery to be tested to the model operation unit, and control the operating state of the battery to be tested according to the fault prediction result fed back by the model operation unit; The model operation unit is used to obtain the operation data of the battery to be tested, input the operation data of the battery to be tested into the battery fault prediction model, and obtain a fault prediction result.

13. The system according to claim 12, wherein: The system also includes a remote upgrade unit, which is in communication with the model operation unit and is used to remotely upgrade the battery failure prediction model deployed in the model operation unit.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 10 are implemented.

Images