A method and system for predicting abnormal state of lithium battery based on deep learning

Abstract

The present invention relates to the technical field of lithium battery state prediction, and in particular to a method and system for predicting abnormal states of lithium batteries based on deep learning. The method comprises the following steps: acquiring and processing multi-source parameters of lithium batteries, including missing value interpolation and feature mining, to form a multi-dimensional feature matrix; performing working situation analysis and behavior pattern mining on the matrix to generate a lithium battery behavior knowledge graph; combining the multi-dimensional feature matrix with the knowledge graph, performing simulation-driven processing, and creating a simulated lithium battery model; using environmental data to simulate the behavior of the model to obtain simulated lithium battery working data; performing abnormal anchor point positioning and state mapping on the simulated data based on the knowledge graph to generate anchor-state abnormal prediction data; finally, using these prediction data to predict abnormal states of multi-source parameters of lithium batteries to obtain lithium battery abnormal state data. The present invention can effectively improve the accuracy and timeliness of abnormal state prediction of lithium batteries.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery state prediction, and in particular to a method and system for predicting abnormal state of a lithium battery based on deep learning. Background Art

[0002] With the rapid development of electric vehicles and energy storage systems, the safety and reliability of lithium batteries have received increasing attention. In the actual application process, the working environment of lithium batteries is complex and changeable, and will be affected by a variety of external factors, such as temperature, humidity, and charge and discharge rate. These factors will cause abnormal conditions in lithium batteries, thereby affecting their life and performance. Therefore, how to accurately predict the abnormal state of lithium batteries in complex environments has become an important topic of current research. However, most of the current lithium battery abnormal state prediction methods rely on traditional statistical models, rules and methods, which often fail to achieve ideal results when processing multi-source heterogeneous data and complex behavior patterns. In addition, existing studies usually only focus on a single parameter or a single type of abnormal state of lithium batteries, lacking a comprehensive and integrated analysis of multi-source parameters of lithium batteries, resulting in insufficient accuracy and robustness of the prediction results. At the same time, in actual applications, the abnormal state of lithium batteries often presents concealment and suddenness, and the existing methods have weak early warning capabilities for these abnormal states. Summary of the invention

[0003] Based on this, it is necessary to provide a lithium battery abnormal state prediction method and system based on deep learning to solve at least one of the above technical problems.

[0004] To achieve the above purpose, a lithium battery abnormal state prediction method based on deep learning includes the following steps:

[0005] Step S1: obtaining multi-source parameters of lithium batteries; performing multiple interpolation processing on missing values ​​of multi-source parameters of lithium batteries, and performing multi-source feature mining to obtain a multi-dimensional feature matrix of lithium batteries;

[0006] Step S2: Analyze the working status of the lithium battery multi-dimensional feature matrix, and mine the behavior patterns to generate a lithium battery behavior knowledge graph;

[0007] Step S3: Perform fusion simulation drive processing on the lithium battery multi-dimensional feature matrix and the lithium battery behavior knowledge graph to generate a simulated lithium battery model;

[0008] Step S4: obtaining lithium battery use environment data; simulating the working behavior of the simulated lithium battery model based on the lithium battery use environment data to obtain simulated lithium battery working data;

[0009] Step S5: locating abnormal change anchor points of the simulated lithium battery working data according to the lithium battery behavior knowledge graph, and performing predicted state mapping to generate anchor point-state abnormality prediction data;

[0010] Step S6: Based on the anchor point-state abnormality prediction data, abnormal state prediction is performed on the lithium battery multi-source parameters to obtain lithium battery abnormal state data.

[0011] The present invention can efficiently integrate and process data from different sources by performing multiple interpolation processing of missing values ​​and multi-source feature mining on multi-source parameters of lithium batteries. The integrity and consistency of the data are ensured, and comprehensive and accurate data support is provided for subsequent model training, thereby improving the prediction effect of the deep learning model. By performing battery working situation analysis and behavior pattern mining on the multi-dimensional feature matrix, a lithium battery behavior knowledge map is generated, which can reveal the complex behavior patterns of the battery under different working conditions, enhance the ability to identify abnormal states, and provide rich background knowledge and prior information for the deep learning model. By fusing the multi-dimensional feature matrix with the behavior knowledge map for simulation drive processing, a simulated lithium battery model is generated, which can simulate the behavior of the battery in the actual working environment, and generate real simulated lithium battery working data, ensuring the effectiveness and accuracy of the model in various practical application scenarios. The lithium battery behavior knowledge map is used to locate the abnormal change anchor point of the simulated lithium battery working data, and predict the state mapping, which can accurately identify potential abnormal points, and associate these abnormal states with the predicted data in the model, thereby improving the accuracy and timeliness of abnormal state prediction.

[0012] The present invention also provides a lithium battery abnormal state prediction system based on deep learning, which is used to execute the lithium battery abnormal state prediction method based on deep learning as described above. The lithium battery abnormal state prediction system based on deep learning includes:

[0013] The data processing module is used to obtain multi-source parameters of lithium batteries; perform multiple interpolation processing on missing values ​​of multi-source parameters of lithium batteries, and perform multi-source feature mining to obtain a multi-dimensional feature matrix of lithium batteries;

[0014] Pattern mining module, used to analyze the working status of lithium batteries and conduct behavior pattern mining on the multi-dimensional feature matrix of lithium batteries to generate a knowledge graph of lithium battery behavior;

[0015] The simulation driving module is used to perform fusion simulation driving processing on the lithium battery multi-dimensional feature matrix and the lithium battery behavior knowledge graph to generate a simulated lithium battery model;

[0016] The behavior simulation module is used to obtain the lithium battery use environment data; based on the lithium battery use environment data, the working behavior of the simulated lithium battery model is simulated to obtain the simulated lithium battery working data;

[0017] The state mapping module is used to locate the abnormal change anchor point of the simulated lithium battery working data according to the lithium battery behavior knowledge graph, and perform predicted state mapping to generate anchor point-state abnormality prediction data;

[0018] The state prediction module is used to predict the abnormal state of the lithium battery multi-source parameters based on the anchor point-state abnormality prediction data to obtain the abnormal state data of the lithium battery.

[0019] The present invention can effectively fill in the missing data and extract useful features by performing multiple interpolation processing on the missing values ​​of the multi-source parameters of the lithium battery and performing multi-source feature mining. The integrity and reliability of the data are ensured, high-quality input data is provided for subsequent model training, and the prediction accuracy of the model is improved. The multidimensional feature matrix is ​​analyzed for the battery working status by deep learning technology, and the lithium battery behavior knowledge map is generated by behavior pattern mining. The complex working mode and potential abnormal behavior of the lithium battery are revealed, providing a comprehensive knowledge basis for the identification of abnormal states and improving the model's perception of abnormal patterns. The simulated lithium battery model is generated by fusion simulation drive processing of the multidimensional feature matrix and the behavior knowledge map of the lithium battery. The working behavior of the model is simulated based on the actual use environment data, and accurate simulated lithium battery working data can be generated. It is helpful to verify the effectiveness of the model in a real environment and ensure its reliability and robustness under various working conditions. Combined with the lithium battery behavior knowledge map, the abnormal change anchor point of the simulated lithium battery working data is located, and the predicted state mapping is performed to generate anchor point-state abnormal prediction data. By identifying and mapping abnormal conditions of anchor points, abnormal conditions can be accurately predicted at an early stage, improving the early warning capability and response speed to abnormal conditions of lithium batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic diagram of the steps of a method for predicting abnormal state of a lithium battery based on deep learning;

[0021] Figure 2 for Figure 1 Detailed implementation steps of step S2 in the flowchart;

[0022] Figure 3 for Figure 1 Detailed implementation steps of step S3 in FIG.

[0023] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0024] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0025] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities are implemented in software form, or in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0026] It should be understood that although the terms "first", "second", etc. are used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit is referred to as a second unit, and similarly, a second unit is referred to as a first unit, and the term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0027] To achieve this, please refer to Figures 1 to 3 , a method for predicting abnormal state of lithium battery based on deep learning, comprising the following steps:

[0028] Step S1: obtaining multi-source parameters of lithium batteries; performing multiple interpolation processing on missing values ​​of multi-source parameters of lithium batteries, and performing multi-source feature mining to obtain a multi-dimensional feature matrix of lithium batteries;

[0029] Step S2: Analyze the working status of the lithium battery multi-dimensional feature matrix, and mine the behavior patterns to generate a lithium battery behavior knowledge graph;

[0030] Step S3: Perform fusion simulation drive processing on the lithium battery multi-dimensional feature matrix and the lithium battery behavior knowledge graph to generate a simulated lithium battery model;

[0031] Step S4: obtaining lithium battery use environment data; simulating the working behavior of the simulated lithium battery model based on the lithium battery use environment data to obtain simulated lithium battery working data;

[0032] Step S5: locating abnormal change anchor points of the simulated lithium battery working data according to the lithium battery behavior knowledge graph, and performing predicted state mapping to generate anchor point-state abnormality prediction data;

[0033] Step S6: Based on the anchor point-state abnormality prediction data, abnormal state prediction is performed on the lithium battery multi-source parameters to obtain lithium battery abnormal state data.

[0034] The present invention can efficiently integrate and process data from different sources by performing multiple interpolation processing of missing values ​​and multi-source feature mining on multi-source parameters of lithium batteries. The integrity and consistency of the data are ensured, and comprehensive and accurate data support is provided for subsequent model training, thereby improving the prediction effect of the deep learning model. By performing battery working situation analysis and behavior pattern mining on the multi-dimensional feature matrix, a lithium battery behavior knowledge map is generated, which can reveal the complex behavior patterns of the battery under different working conditions, enhance the ability to identify abnormal states, and provide rich background knowledge and prior information for the deep learning model. By fusing the multi-dimensional feature matrix with the behavior knowledge map for simulation drive processing, a simulated lithium battery model is generated, which can simulate the behavior of the battery in the actual working environment, and generate real simulated lithium battery working data, ensuring the effectiveness and accuracy of the model in various practical application scenarios. The lithium battery behavior knowledge map is used to locate the abnormal change anchor point of the simulated lithium battery working data, and predict the state mapping, which can accurately identify potential abnormal points, and associate these abnormal states with the predicted data in the model, thereby improving the accuracy and timeliness of abnormal state prediction.

[0035] In this embodiment, reference Figure 1 The above is a schematic diagram of the steps of the method for predicting abnormal state of a lithium battery based on deep learning of the present invention. In this example, the method for predicting abnormal state of a lithium battery based on deep learning includes the following steps:

[0036] Step S1: obtaining multi-source parameters of lithium batteries; performing multiple interpolation processing on missing values ​​of multi-source parameters of lithium batteries, and performing multi-source feature mining to obtain a multi-dimensional feature matrix of lithium batteries;

[0037] In this embodiment, multi-source parameters of the lithium battery are obtained, including voltage, current, temperature, number of charge and discharge cycles, and internal resistance. Data cleaning tools are used to process missing values, and multiple interpolation methods are used to complete data, such as using the K nearest neighbor algorithm (KNN) or the multiple interpolation Bayesian algorithm (MICE). After processing the missing values, feature mining is performed on the data, and important features are extracted using principal component analysis (PCA) and feature selection techniques, such as the L1 regularization method. Finally, a multidimensional feature matrix containing each parameter and its derived features is generated to ensure that the matrix can accurately reflect the status information of the lithium battery.

[0038] Step S2: Analyze the working status of the lithium battery multi-dimensional feature matrix, and mine the behavior patterns to generate a lithium battery behavior knowledge graph;

[0039] In this embodiment, the multidimensional feature matrix is ​​used to analyze the battery working status, and the state space model (SSM) or regression analysis model (such as linear regression and support vector regression) is used to analyze the working mode of the lithium battery. During the analysis process, time series analysis technology is used to identify the performance changes of the battery under different working conditions. By mining the behavior pattern, clustering algorithms (such as K-means or DBSCAN) are used to divide the data into different working modes, and generate a lithium battery behavior knowledge graph. The knowledge graph should include the battery status, abnormal mode and its corresponding feature relationship diagram for subsequent analysis.

[0040] Step S3: Perform fusion simulation drive processing on the lithium battery multi-dimensional feature matrix and the lithium battery behavior knowledge graph to generate a simulated lithium battery model;

[0041] In this embodiment, the multidimensional feature matrix is ​​fused with the behavioral knowledge graph, and a simulated lithium battery model is generated through simulation-driven processing. Multi-physics simulation software, such as COMSOL Multiphysics or ANSYS, is used to input the multidimensional feature data of the battery and the behavioral pattern information in the knowledge graph to establish a simulation model of the battery. The simulation process includes the thermal, electrical and mechanical behaviors of the battery, and the simulation is based on the working state of the battery in a real environment. The generated simulation model can simulate the actual behavior of the battery under different operating conditions, including the charging and discharging process of the battery and temperature changes, and provide accurate simulation data.

[0042] Step S4: obtaining lithium battery use environment data; simulating the working behavior of the simulated lithium battery model based on the lithium battery use environment data to obtain simulated lithium battery working data;

[0043] In this embodiment, the use environment data of the lithium battery is obtained, including the ambient temperature, humidity, and use load, and these data are collected by environmental sensors. These environmental data are input into the simulated lithium battery model that has been generated to simulate the working behavior. During the simulation process, dynamic simulation technology (such as real-time simulation or offline simulation) is used to simulate the performance of the battery under different environmental conditions to generate simulated lithium battery working data. The data should include the voltage, current, and temperature parameters of the battery under different environmental conditions, reflecting the performance of the battery in the actual use environment.

[0044] Step S5: locating abnormal change anchor points of the simulated lithium battery working data according to the lithium battery behavior knowledge graph, and performing predicted state mapping to generate anchor point-state abnormality prediction data;

[0045] In this embodiment, based on the knowledge graph of lithium battery behavior, the simulated lithium battery working data is analyzed to identify the abnormal change anchor points appearing in the data. The anchor points are specifically the turning points of a local minimum and a local maximum of the system state curve. At this point, the trajectory in the state space will spread outward, showing unstable characteristics. Use anomaly detection algorithms, such as Isolation Forest or the statistical Z-score method, to determine the abnormal patterns and anchor point positions in the data. Compare the identified abnormal changes with the abnormal states in the knowledge graph to perform predicted state mapping. The generated anchor-state anomaly prediction data contains the location information of the anchor point and its corresponding abnormal state prediction results, providing detailed information on potential anomalies of the lithium battery.

[0046] Step S6: Based on the anchor point-state abnormality prediction data, abnormal state prediction is performed on the lithium battery multi-source parameters to obtain lithium battery abnormal state data.

[0047] In this embodiment, the generated anchor-state abnormality prediction data is used to predict the abnormal state of multi-source data of lithium batteries. The predicted data is input into a deep learning model (such as a convolutional neural network (CNN) or a long short-term memory network (LSTM)) for abnormal state recognition and prediction. During the model training process, the labeled abnormal state data is used as a training set, and the model parameters are optimized by the back propagation algorithm. After the training is completed, the model is applied to the actual multi-source data to obtain the predicted abnormal state data, including the specific type of abnormal state and the time point of occurrence, to provide accurate abnormality detection results.

[0048] Preferably, step S1 comprises the following steps:

[0049] Step S11: Obtaining multi-source parameters of lithium battery;

[0050] Step S12: performing outlier interval detection on the multi-source parameters of the lithium battery to obtain parameter outlier intervals;

[0051] Step S13: performing multiple interpolation processing on parameter outlier intervals based on lithium battery multi-source parameters to generate an interpolated lithium battery parameter set;

[0052] Step S14: performing independent component analysis on the interpolated lithium battery parameter set to generate a dimension-reduced lithium battery parameter set;

[0053] Step S15: Perform multi-source feature mining on the dimension-reduced lithium battery parameter set to obtain a multi-dimensional feature matrix of the lithium battery.

[0054] The present invention ensures the comprehensiveness of data through multi-source parameter acquisition, so that the abnormal state prediction is based on rich information sources, thereby improving the accuracy and reliability of abnormal detection. Outlier interval detection helps to quickly discover potential abnormal behaviors by identifying abnormal fluctuations in parameters, thereby improving the timeliness of early warning. Multiple interpolation processing effectively handles missing values ​​or outliers in the data, generates a more complete and consistent set of lithium battery parameters, and improves the effectiveness and accuracy of subsequent analysis. The interpolated data set provides more reliable input, enhancing the data quality and prediction ability in the model training process. Independent component analysis reduces the complexity of data through dimensionality reduction processing, removes redundant information, and makes feature analysis more focused on actual effective parameters, thereby improving computational efficiency. Multi-source feature mining extracts useful features from dimensionality reduction data to form a multidimensional feature matrix, which provides more representative input data for the deep learning model, and helps the model to more accurately identify and predict abnormal states. The comprehensive use of the generated high-quality data (interpolation data set, dimensionality reduction data set, and multidimensional feature matrix) can significantly improve the training effect and prediction performance of the deep learning model, making the prediction of abnormal states of lithium batteries more accurate and reliable. Through the systematic data preprocessing and feature extraction process, the model training time is significantly reduced, the prediction efficiency is improved, and real-time monitoring and early warning can be provided in complex battery operating environments. Through the structured data processing process, it is ensured that the deep learning model can effectively extract useful information from multi-source data, thereby improving the accuracy and reliability of lithium battery abnormal state prediction.

[0055] In this embodiment, multi-source parameters of the lithium battery are obtained, including voltage, current, temperature, internal resistance, and charge and discharge cycle. Real-time monitoring and recording are performed through a sensor network and a data acquisition system. The data acquisition system should have high-precision sensors to ensure that the collected data has high accuracy and reliability. The sensor transmits the collected data to the data storage system to form an original parameter data set for subsequent processing. The acquired multi-source parameters of the lithium battery are detected for outlier intervals. Statistical methods such as box plot analysis and Z-score analysis are used to identify outliers in parameter data. Box plot analysis determines the outlier interval by calculating the quartiles and inner and outer limits of the parameters, while Z-score analysis identifies outliers by calculating the standard deviation of each parameter value from the mean. Multiple interpolation processing is performed on the determined parameter outlier intervals, such as the multiple interpolation Bayesian algorithm (MICE) or the K-nearest neighbor interpolation (KNN Imputation) method. Use existing normal data to fill in the missing data in the outlier interval. The MICE algorithm predicts each outlier by building a regression model, while KNN interpolation interpolates by finding the most similar data points. Generate a complete interpolated lithium battery parameter set to replace the original outlier data. Perform independent component analysis (ICA) on the interpolated lithium battery parameter set to achieve data dimensionality reduction. ICA technology can separate mixed signals and extract independent components, thereby reducing the dimension of the data. The data components are separated by calculating the covariance matrix of the data and then applying whitening transformation and rotation algorithms (such as FastICA algorithm). Generate a reduced-dimensional lithium battery parameter set that retains the main information in the data while reducing redundant features. Perform multi-source feature mining on the reduced-dimensional lithium battery parameter set, using feature selection techniques such as principal component analysis (PCA) or feature selection algorithms (such as L1 regularization method). Identify and extract important features through methods to comprehensively analyze battery performance. PCA projects the reduced-dimensional data onto the main components, reducing the feature dimensions without losing information. The final generated lithium battery multidimensional feature matrix contains key feature information for subsequent modeling and analysis.

[0056] Preferably, step S2 comprises the following steps:

[0057] Step S21: performing time series analysis on the multi-dimensional characteristic matrix of the lithium battery to obtain a lithium battery working state sequence;

[0058] Step S22: performing working status analysis on the lithium battery working status sequence to generate lithium battery status status data;

[0059] Step S23: mining the behavior pattern of the lithium battery status data to generate status-behavior pattern data;

[0060] Step S24: construct a behavior knowledge graph for the state-behavior pattern data to obtain a lithium battery behavior knowledge graph.

[0061] The present invention enables the multi-dimensional feature matrix of lithium batteries to reflect the changing trend in the time dimension through time series analysis, helps capture the working status of lithium batteries at different time points, and improves the understanding and prediction ability of dynamic changes in battery status. Working status analysis generates detailed status status data by analyzing the working status sequence, thereby providing a comprehensive evaluation of the overall working status of the lithium battery and helping to identify potential abnormal modes or potential problems. Behavior pattern mining extracts behavior patterns from the status status data and generates state-behavior pattern data. Identifying the behavior patterns of lithium batteries under different working conditions helps to discover the potential root causes of abnormal behavior. Behavior knowledge graph construction converts state-behavior pattern data into behavior knowledge graphs, integrates various behaviors and status information of lithium batteries, and forms a comprehensive knowledge system that can systematically describe and analyze the working behavior and abnormal modes of lithium batteries. By comprehensively analyzing time series data, working status, behavior patterns, and constructed behavior knowledge graphs, it is possible to provide a deep insight into the abnormal state of lithium batteries. The prediction accuracy and foresight of the model for complex abnormal states are improved, and effective warnings can be provided before potential battery failures occur. The generated behavioral knowledge graph, as the input of the deep learning model, can significantly enrich and optimize the model's training data, so that the model can better understand the battery's behavior pattern, thereby improving the accuracy and reliability of abnormal state prediction. Through a systematic analysis process, from time series to the construction of behavioral knowledge graphs, it is ensured that the deep learning model can use rich battery behavior data to provide more accurate and reliable abnormal state predictions.

[0062] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes:

[0063] Step S21: performing time series analysis on the multi-dimensional characteristic matrix of the lithium battery to obtain a lithium battery working state sequence;

[0064] In this embodiment, various parameter data in the multidimensional feature matrix of the lithium battery are organized into a time series format to reflect the state of the battery at different time points. Use time series analysis tools such as ARIMA (autoregressive integrated moving average model) or LSTM (long short-term memory network) to model the time series data. By analyzing the trends, seasonality and periodicity in historical data, future state changes can be predicted. Through analysis, a working state sequence of the lithium battery is generated, including a detailed record of the changes of various parameters of the battery over time, revealing the dynamic characteristics and working mode of the battery.

[0065] Step S22: performing working status analysis on the lithium battery working status sequence to generate lithium battery status status data;

[0066] In this embodiment, the generated lithium battery working state sequence is subjected to working state analysis, using statistical analysis methods such as sliding window technology and Fourier transform. The sliding window technology applies a fixed-length window to the time series and calculates the data features in each window, such as the mean, variance, and maximum value, to analyze the changing trend of the battery state. The Fourier transform is used to identify periodic fluctuations in the sequence. The lithium battery state state data is generated to describe the state of the battery in each time period and its changing trend, providing basic data for subsequent behavior pattern mining.

[0067] Step S23: mining the behavior pattern of the lithium battery status data to generate status-behavior pattern data;

[0068] In this embodiment, the behavior pattern mining of lithium battery state situation data is carried out, using pattern recognition techniques such as cluster analysis (K-means or DBSCAN) and association rule mining (Apriori algorithm). Cluster analysis divides the state situation data into different groups and identifies state patterns with similar characteristics. Association rule mining discovers the potential relationship between states and reveals which state changes will lead to the emergence of specific behavior patterns. Generate state-behavior pattern data, describe the typical behavior patterns of batteries in various states and their relationships, and provide information for building a behavioral knowledge graph.

[0069] Step S24: construct a behavior knowledge graph for the state-behavior pattern data to obtain a lithium battery behavior knowledge graph.

[0070] In this embodiment, the generated state-behavior pattern data is used to construct a behavioral knowledge graph. A graph database technology such as Neo4j is used to create a graph structure model, in which nodes represent different states and behavior patterns, and edges represent the relationship between them. By mapping the state-behavior pattern data into a graph database, a behavioral knowledge graph of lithium batteries is constructed. The complex relationship between battery state and behavior is displayed, providing a full picture of battery operation and potential behavior patterns, which facilitates subsequent anomaly detection and state prediction analysis.

[0071] Preferably, step S3 comprises the following steps:

[0072] Step S31: performing association rule analysis on the lithium battery multidimensional feature matrix and the lithium battery behavior knowledge graph to obtain an association feature matrix;

[0073] Step S32: backtracking the behavior pattern of the lithium battery behavior knowledge graph according to the associated feature matrix to generate a behavior pattern matching matrix;

[0074] Step S33: performing multi-dimensional topological fusion processing on the behavior pattern matching matrix and the lithium battery multi-dimensional feature matrix to obtain a dynamic behavior fusion matrix;

[0075] Step S34: performing fusion simulation drive processing on the dynamic behavior fusion matrix to generate a simulated lithium battery model.

[0076] The present invention associates the multidimensional feature matrix of lithium batteries with the lithium battery behavior knowledge graph through association rule analysis to generate an association feature matrix. The potential relationship between different features and behaviors is excavated, which improves the understanding of complex data relationships and helps to identify the potential causes and influencing factors of abnormal states. Behavior pattern backtracking is based on the association feature matrix to perform backtracking analysis on the behavior pattern and generate a behavior pattern matching matrix. By comparing historical behavior patterns with current data, the behavior patterns related to abnormal states are identified, thereby improving the precision and accuracy of abnormal prediction. Multidimensional topological fusion processing fuses the behavior pattern matching matrix and the multidimensional feature matrix to generate a dynamic behavior fusion matrix. By fusing data of different dimensions, a more comprehensive behavior data representation can be created, which improves the model's ability to analyze complex behavior patterns. Fusion simulation driven processing uses the dynamic behavior fusion matrix to generate a simulated lithium battery model. By simulating the performance of lithium batteries under different states, the prediction ability of the model can be verified and optimized, and the reliable prediction of abnormal states can be improved. Through systematic association rule analysis, behavioral pattern backtracking, multi-dimensional fusion, and simulation-driven processing, the deep learning model can process and understand complex battery behavior data at a higher level, optimize the model's predictive ability and reliability, and thus provide more accurate and effective abnormal state predictions. The generated simulated lithium battery model can monitor and warn the dynamic behavior of lithium batteries in real time, improve the ability to respond to potential abnormal states in a timely manner, and reduce safety hazards and failure risks. Through in-depth analysis and fusion processing of multi-dimensional data, it can provide more accurate and comprehensive predictions of lithium battery abnormal states, improving the effectiveness and reliability of the entire prediction system.

[0077] As an example of the present invention, refer to Figure 3 As shown, in this example, step S3 includes:

[0078] Step S31: performing association rule analysis on the lithium battery multidimensional feature matrix and the lithium battery behavior knowledge graph to obtain an association feature matrix;

[0079] In this embodiment, association rule analysis is performed on the multi-dimensional feature matrix of lithium batteries and the knowledge graph of lithium battery behavior, using association rule mining algorithms such as Apriori or FP-Growth. First, the Apriori algorithm is used to identify frequent item sets, and feature combinations with significant associations are found by setting minimum support and confidence thresholds. Subsequently, the FP-Growth algorithm is applied to further optimize the discovery process of frequent item sets and reduce computational complexity. The extracted association rules are applied to the feature matrix and knowledge graph to generate an association feature matrix to reveal the potential relationships and patterns between features and behaviors.

[0080] Step S32: backtracking the behavior pattern of the lithium battery behavior knowledge graph according to the associated feature matrix to generate a behavior pattern matching matrix;

[0081] In this embodiment, the behavior pattern backtracking of the lithium battery behavior knowledge graph is performed based on the generated association feature matrix. First, the important features in the association feature matrix are mapped back to the behavior pattern nodes in the knowledge graph. Through backtracking analysis, the behavior pattern corresponding to each feature is determined, and their matching degree is calculated. Using backtracking algorithms such as dynamic time warping (DTW) or similarity matching methods, the behavior patterns are compared to generate a behavior pattern matching matrix that describes the matching between the features and the actual behavior patterns and their strength.

[0082] Step S33: performing multi-dimensional topological fusion processing on the behavior pattern matching matrix and the lithium battery multi-dimensional feature matrix to obtain a dynamic behavior fusion matrix;

[0083] In this embodiment, the behavior pattern matching matrix and the lithium battery multi-dimensional feature matrix are subjected to multi-dimensional topological fusion processing, using graph theory and matrix decomposition technology. First, the behavior pattern matching matrix and the feature matrix are constructed as a graph structure, in which nodes represent features or behavior patterns, and edges represent the associations between them. A graph convolutional network (GCN) is applied to perform convolution operations on the graph to fuse feature and pattern information. Next, matrix decomposition techniques such as singular value decomposition (SVD) or non-negative matrix decomposition (NMF) are used to reduce the dimension of the graph structure matrix and generate a dynamic behavior fusion matrix.

[0084] Step S34: performing fusion simulation drive processing on the dynamic behavior fusion matrix to generate a simulated lithium battery model.

[0085] In this embodiment, the dynamic behavior fusion matrix is ​​subjected to fusion simulation drive processing, and a simulation-driven modeling method is adopted. First, a suitable simulation platform, such as MATLAB / Simulink or ANSYS, is selected to create the model. The dynamic behavior fusion matrix is ​​used as input data to set simulation parameters and boundary conditions. A simulated lithium battery model is constructed through the modeling tools of the simulation platform, integrating multidimensional features and dynamic behavior patterns. Run the simulation to verify the performance of the model under different working conditions, and adjust the model parameters to optimize performance. Finally, a high-precision simulated lithium battery model is generated, which can accurately simulate the behavior and state of the battery in actual operation.

[0086] Preferably, step S4 comprises the following steps:

[0087] Step S41: Acquire lithium battery usage environment data;

[0088] Step S42: extracting multi-dimensional variables from the lithium battery usage environment data to obtain an environment variable matrix;

[0089] Step S43: constructing a simulated environment field for the environmental variable matrix to generate a changing simulated environment field;

[0090] Step S44: mapping the behavior characteristics of the simulated lithium battery model based on the changed simulated environment field and the environment variable matrix to obtain a simulated behavior characteristic matrix;

[0091] Step S45: performing multi-scale time series analysis on the simulation behavior feature matrix to generate a behavior pattern sequence matrix;

[0092] Step S46: performing simulation behavior fitting on the simulation lithium battery model according to the behavior pattern sequence matrix to obtain simulated lithium battery operating data.

[0093] The present invention ensures that the model can be analyzed based on the actual use environment by obtaining the lithium battery use environment data. A comprehensive understanding of lithium batteries under different environmental conditions is provided, and the adaptability of the prediction model to actual use scenarios is enhanced. Multidimensional variable extraction extracts environmental variables of multiple dimensions from environmental data and generates an environmental variable matrix. The model can comprehensively consider multiple environmental factors, thereby improving the understanding of the impact of the environment on the behavior of lithium batteries. The simulation environment field construction provides an environmental model that is closer to the actual use conditions for simulation by generating a changing simulation environment field. The working performance of lithium batteries under different environmental conditions is more accurately simulated, which improves the realism and effectiveness of the simulation results. Behavior feature mapping maps the simulation lithium battery model based on the changing simulation environment field and the environmental variable matrix to generate a simulation behavior feature matrix. The environmental factors are combined with the model behavior features to provide a detailed description of the battery behavior and improve the simulation accuracy. Multiscale time series analysis analyzes the simulation behavior feature matrix and generates a behavior pattern sequence matrix. Identifying and parsing the behavior pattern changes on different time scales provides an in-depth understanding of the dynamic behavior of lithium batteries. Simulation behavior fitting fits the simulation lithium battery model through the behavior pattern sequence matrix to generate simulated lithium battery working data. Compare the simulation results with the actual data to optimize the model's prediction ability, thereby improving the accuracy of abnormal state prediction. The comprehensive use of environmental data, simulated environmental fields, behavioral feature mapping, and time series analysis can make the deep learning model highly adaptable under different environmental conditions and be able to predict and warn of abnormal states of lithium batteries in real time. Through comprehensive analysis from environmental data to simulated behavior, it is ensured that the deep learning model can provide more accurate and reliable abnormal state predictions in actual use scenarios, enhancing the effectiveness and breadth of application of the prediction system.

[0094] In this embodiment, various data of the lithium battery in the actual use environment are collected, and these data include temperature, humidity, charge and discharge frequency, and workload environmental variables. Environmental monitoring equipment, such as temperature and humidity sensors, load testers, and data acquisition systems are used to record the working status of the lithium battery under different environmental conditions in real time. The data is centrally stored through the data acquisition system to ensure the integrity and accuracy of the data and provide reliable basic data for subsequent analysis and simulation. Multidimensional variables are extracted from the collected environmental data, and data preprocessing techniques such as standardization and normalization are used to ensure the consistency and comparability of the data. Using feature extraction methods such as principal component analysis (PCA) or linear discriminant analysis (LDA), important environmental variable features are extracted from the original data, including temperature change trends, humidity fluctuations, and charge and discharge cycles. Finally, an environmental variable matrix is ​​generated, which contains all key variables and their values ​​under different environmental conditions, providing a data basis for the construction of a simulated environmental field. The environmental variable matrix is ​​used to build a changing simulated environmental field, and environmental modeling tools such as OpenFOAM or COMSOL Multiphysics are applied to perform three-dimensional modeling and simulation of the environmental field. According to the data in the environmental variable matrix, the boundary conditions and environmental parameters of the simulated environment field are set, such as temperature gradient, humidity distribution and battery load conditions. Through numerical simulation methods, the scene changes under different environmental conditions are simulated to generate a changing simulated environment field. According to the changing simulated environment field and the environmental variable matrix, the behavior characteristics of the simulated lithium battery model are mapped. The environmental conditions in the simulated environment field are used as input data, and the environmental data are mapped to the simulated lithium battery model using the behavior mapping tool in the simulation software. Through this process, the behavior characteristics of lithium batteries under different environmental conditions are extracted, such as battery temperature changes, internal resistance changes and charge and discharge efficiency. Finally, a simulation behavior characteristic matrix is ​​generated to describe the behavior characteristics of lithium batteries under specific environmental conditions. Multi-scale time series analysis is performed on the simulation behavior characteristic matrix. Time series analysis methods such as wavelet transform or multi-scale analysis (MSA) are used to decompose the time series data in the simulation behavior characteristic matrix, which can capture the changes and periodic patterns of different scales in the behavior characteristics. By multi-scale decomposition of the time series data, the key features of the behavior pattern are identified, and the behavior pattern sequence matrix is ​​generated. The behavior pattern sequence matrix is ​​used to fit the simulation behavior of the simulated lithium battery model, and the fitting algorithm such as least squares method or Bayesian optimization is used to adjust and optimize the simulation model. The behavior pattern in the behavior pattern sequence matrix is ​​used as input to adjust the parameters of the simulation model so that the model output is consistent with the actual behavior pattern. Through simulation drive, the optimized model is run to generate simulated lithium battery working data.

[0095] Preferably, step S5 comprises the following steps:

[0096] Step S51: Perform dynamic graph embedding processing on the lithium battery behavior knowledge graph to obtain low-dimensional graph embedding representation data;

[0097] Step S52: performing abnormal change detection on the simulated lithium battery working data based on the low-dimensional embedding representation data of the graph to generate an abnormal candidate sequence;

[0098] Step S53: positioning abnormal change anchor points of the abnormal candidate sequence to obtain an abnormal change anchor point set;

[0099] Step S54: performing predicted state mapping on the abnormal change anchor point set according to the lithium battery behavior knowledge graph to generate anchor point-state abnormality prediction data.

[0100] The present invention converts the lithium battery behavior knowledge graph into low-dimensional embedding representation data through dynamic graph embedding processing. By reducing the data dimension, the graph data is easier to process and analyze, while maintaining the core information in the graph, improving the computational efficiency and effect of subsequent processing. Abnormal change detection performs abnormal detection on the simulated lithium battery working data based on low-dimensional embedded data to generate an abnormal candidate sequence. By combining the embedded representation of the graph, the abnormal pattern can be more accurately identified, and the detection sensitivity and accuracy of the abnormal state are improved. Abnormal change anchor point positioning processes the abnormal candidate sequence to obtain an abnormal change anchor point set. By accurately locating the key points of abnormal changes, an accurate basis is provided for further analysis and prediction, and the accuracy and reliability of abnormal state positioning are improved. Prediction state mapping maps the abnormal change anchor point set to the lithium battery behavior knowledge graph to generate anchor-state abnormal prediction data. Combining the knowledge and anchor information in the graph, a more accurate prediction of the abnormal state can be made, providing a comprehensive view and prediction results of the abnormal state. By combining the behavioral knowledge graph with the anomaly detection and prediction process, the deep learning model can use the rich information in the graph to enhance the accuracy of the prediction results. The ability to identify and predict complex abnormal conditions has been improved, making the prediction more consistent with actual battery behavior and operating environment. The comprehensive use of dynamic graph embedding, anomaly detection, anchor point positioning and state mapping can provide real-time and accurate abnormal state warnings and detect potential problems in a timely manner, thereby improving the safety and reliability of lithium battery systems. Through the combination of dynamic graph embedding and abnormal change detection, the abnormal state of lithium batteries can be predicted efficiently and accurately, improving the overall performance and application effect of the prediction system.

[0101] In this embodiment, the lithium battery behavior knowledge graph is processed using dynamic graph embedding technology. First, a graph embedding algorithm such as node embedding (Node2Vec) or graph convolutional network (GCN) is used to represent and learn the nodes and edges in the knowledge graph, which can convert high-dimensional graph data into vector representations in a low-dimensional space, retaining the structural information of the graph and the relationship between nodes. The embedding process includes mapping each node and edge in the knowledge graph into a low-dimensional vector space, and optimizing the model parameters through training to minimize the reconstruction error. Based on the low-dimensional embedding representation data of the graph, abnormal changes are detected for the simulated lithium battery working data. The low-dimensional embedded data is compared with the simulated lithium battery working data using anomaly detection algorithms such as isolation forest or local outlier factor (LOF). First, the simulated lithium battery working data is mapped to the low-dimensional embedding space of the graph, and then the anomaly score of the data point is calculated. The anomaly detection algorithm identifies the anomaly by evaluating the distance between the data point and its neighborhood data point. The generated abnormal candidate sequence includes all detected abnormal data points and marks the time series with abnormal behavior. The abnormal candidate sequence is processed using nonlinear transformation (HHT). First, the abnormal candidate sequence is decomposed into intrinsic mode functions (IMFs) and trend terms. Each IMF is processed using the Hilbert transform to calculate its instantaneous frequency and instantaneous amplitude. A set of abnormal candidate curves is obtained by comprehensive analysis of the instantaneous frequency and amplitude of these IMFs. It reflects the instantaneous characteristic changes of the signal in the abnormal candidate sequence and lays the foundation for subsequent anomaly detection. The tool for HHT processing uses the HHT toolbox in MATLAB or the relevant library in Python. Saddle point detection is performed on the obtained abnormal candidate curve set. Each curve is analyzed using the second-order derivative method or optimization algorithm to determine the position of the saddle point on the curve. The saddle point is usually manifested as a local extreme point of the curve, characterized by the second-order derivative of the curve at this point being zero. By calculating the second-order derivative of each curve and locating the point where it is zero, the saddle point set of the abnormal curve is obtained. The second-order derivative calculation and extreme value detection are performed using the SciPy library in Python, or completed using the numerical analysis tools in MATLAB. The detected saddle point set is used to locate the abnormal change anchor points of the abnormal candidate curve. First, the distribution of saddle points in the abnormal candidate curve is checked and used as potential abnormal anchor points. According to the saddle point characteristics of the curve, the anchor points related to the actual abnormal behavior are identified and extracted from the abnormal curve. The obtained initial abnormal anchor point set marks the locations where significant changes occur in the abnormal candidate curve. A relationship network between the abnormal anchor points is constructed to determine the upstream and downstream influence relationship of each anchor point in the time series. Use causal relationship analysis methods, such as Granger causality test or time series correlation analysis, to evaluate the upstream and downstream impact of abnormal anchor points.The analysis results will reveal how the abnormal anchor points affect the behavior of the entire system and associate the abnormal anchor points with the actual abnormal change events. Finally, the abnormal change anchor point set obtained includes all abnormal locations confirmed by analysis, which can effectively indicate abnormal states. All behavior patterns in the lithium battery behavior knowledge graph are clustered and analyzed to identify abnormal patterns. The graph embedding technology is used to convert the nodes and edges in the graph into low-dimensional vector representations. These vectors are classified by clustering algorithms (such as k-means and DBSCAN) to extract patterns related to abnormal behaviors. The extraction process is carried out using Python's graphics library (such as NetworkX) combined with a machine learning library (such as Scikit-Learn). The abnormal behavior patterns obtained by clustering results can be used for subsequent anchor point matching. The abnormal behavior patterns are compared with the anchor point features in the abnormal change anchor point set to calculate the similarity between them. Similarity measurement methods (such as cosine similarity and Euclidean distance) are used to evaluate the degree of match between each abnormal change anchor point and the abnormal behavior pattern. By threshold screening these measurement results, matching abnormal anchor point patterns are obtained. An abnormal state prediction model is constructed based on the matching abnormal anchor point patterns. Use machine learning models (such as support vector machines and random forests) to train the abnormal anchor point pattern to learn the relationship between the pattern and the actual abnormal state. Then, use the trained model to predict the abnormal change anchor point set and generate the corresponding abnormal state data.

[0102] Preferably, locating abnormal change anchor points for abnormal candidate sequences includes the following steps:

[0103] Perform nonlinear transformation on the abnormal candidate sequence to obtain an abnormal candidate curve set;

[0104] Perform saddle point detection on the abnormal candidate curve set to obtain the abnormal curve saddle point set;

[0105] Based on the abnormal curve saddle point set, the abnormal change anchor point of the abnormal candidate curve set is located to obtain the initial abnormal anchor point set;

[0106] The upstream and downstream impact chain analysis is performed on the initial abnormal anchor point set to obtain the abnormal change anchor point set.

[0107] The present invention converts the data into a set of abnormal candidate curves by performing nonlinear transformation processing on the abnormal candidate sequence. The instantaneous frequency and amplitude information in the time series data are effectively extracted, providing accurate curve features for subsequent abnormal detection. The key feature points of the abnormal curve are determined by identifying the saddle points of the curve. The key positions of abnormal changes are accurately located, providing important reference data for further analysis. The preliminary abnormal anchor point set is identified by analyzing the saddle point set of the abnormal curve. The preliminary identification of the abnormal change point is provided, which helps to accurately locate the occurrence location of the abnormal state. By analyzing the upstream and downstream influence relationship of the abnormal point, the real anchor point of the abnormal change can be more accurately determined. The impact of the abnormal change on the overall system is considered, and a comprehensive understanding of the abnormal point is provided, which improves the accuracy and reliability of the abnormal anchor point positioning. The comprehensive use of nonlinear transformation, saddle point detection, initial anchor point positioning and influence chain analysis can provide a comprehensive and in-depth understanding of the abnormal change. The abnormal change is efficiently identified and located, and the prediction ability of the deep learning model for the abnormal state is enhanced. Through the refined abnormal change anchor point positioning process, the deep learning model can handle more complex abnormal states, provide more accurate prediction results, and improve the ability to cope with complex battery behaviors. Through the systematic abnormal change anchor point positioning method, the accuracy of abnormal detection and the timeliness of early warning can be significantly improved, the potential risk of failure can be reduced, and the safety and reliability of the lithium battery system can be improved. By performing detailed processing and analysis on the abnormal candidate sequence, the accurate positioning of the abnormal change anchor point is ensured, thereby improving the accuracy and reliability of the abnormal state prediction of the lithium battery.

[0108] In this embodiment, the abnormal candidate sequence is processed using nonlinear transformation (HHT). First, the abnormal candidate sequence is decomposed into intrinsic mode functions (IMFs) and trend terms. Each IMF is processed using the Hilbert transform to calculate its instantaneous frequency and instantaneous amplitude. By comprehensively analyzing the instantaneous frequency and amplitude of these IMFs, a set of abnormal candidate curves is obtained. It reflects the instantaneous characteristic changes of the signal in the abnormal candidate sequence, laying the foundation for subsequent abnormal detection. The tool for HHT processing uses the HHT toolbox in MATLAB or the relevant library in Python. Saddle point detection is performed on the obtained abnormal candidate curve set, and the saddle point is specifically that the saddle point is a fixed point in a state space, which is locally manifested as a combination of a stable point and an unstable point. It is neither the highest point (peak) nor the lowest point (valley) on the curve or signal, but it is still a key point in the local morphology of the curve, providing trend information of the curve and helping to identify potential abnormal positions (i.e., anchor points). In abnormal detection, saddle points are often used to identify turning points in trends or patterns. Each curve is analyzed using a second-order derivative method or an optimization algorithm to determine the position of the saddle point on the curve. Saddle points are usually manifested as local extreme points of a curve, characterized by the fact that the second-order derivative of the curve at this point is zero. By calculating the second-order derivative of each curve and locating the point where it is zero, the saddle point set of the abnormal curve is obtained. The second-order derivative calculation and extreme value detection are performed through the SciPy library in Python, or completed using the numerical analysis tools in MATLAB. The detected saddle point set is used to locate the abnormal change anchor points of the abnormal candidate curve. First, the distribution of saddle points in the abnormal candidate curve is checked and used as potential abnormal anchor points. According to the saddle point characteristics of the curve, the anchor points related to the actual abnormal behavior are identified and extracted from the abnormal curve. The obtained initial abnormal anchor point set marks the locations where significant changes occur in the abnormal candidate curve. The relationship network between the abnormal anchor points is constructed to determine the upstream and downstream influence relationship of each anchor point in the time series. Use causal relationship analysis methods, such as Granger causality test or time series correlation analysis, to evaluate the upstream and downstream impact of the abnormal anchor points. The analysis results will reveal how the abnormal anchor points affect the behavior of the entire system and associate the abnormal anchor points with the actual abnormal change events. Finally, the obtained abnormal change anchor point set includes all abnormal locations confirmed by analysis, which can effectively indicate the abnormal state.

[0109] Preferably, predicting the state mapping of the abnormal change anchor point set according to the lithium battery behavior knowledge graph includes the following steps:

[0110] Extract abnormal patterns from the lithium battery behavior knowledge graph and generate abnormal behavior patterns;

[0111] According to the abnormal behavior pattern, the abnormal change anchor point set is matched with the change anchor point to obtain the matching abnormal anchor point pattern;

[0112] Based on the matching abnormal anchor point pattern, the abnormal state of the abnormal change anchor point set is predicted to generate anchor point-state abnormal prediction data.

[0113] The present invention can identify key abnormal behavior patterns from the spectrum through abnormal pattern extraction, provide a detailed abnormal pattern description based on historical and existing data for prediction, and enhance the ability to understand and identify potential abnormalities. Change anchor point matching matches the abnormal change anchor point set with the extracted abnormal behavior pattern. By comparing the actual abnormal anchor point with the known abnormal pattern, the specific pattern of the abnormal point can be more accurately identified, thereby improving the accuracy of abnormal detection. Abnormal state prediction predicts the abnormal change anchor point set and generates anchor-state abnormal prediction data. By using the information of the abnormal behavior pattern, the nature and properties of the abnormal state can be accurately predicted, thereby providing a reliable prediction result for the early warning system. The combination of abnormal pattern extraction, anchor point matching and abnormal state prediction can enable the deep learning model to have a more comprehensive understanding and prediction of the abnormal state of the lithium battery, thereby improving the model's prediction of complex abnormalities. The processing capability of the state improves the accuracy of the prediction. By matching the abnormal behavior pattern with the actual anchor point, the information of the behavioral knowledge graph can be fully utilized in the abnormal state prediction, improving the accuracy and reliability of the prediction, reducing false alarms and missed alarms, and the generated anchor point-state abnormal prediction data can provide detailed prediction information for the early warning system, helping the system to more effectively identify and respond to potential abnormal states, thereby improving the safety and reliability of the lithium battery system. Through the systematic abnormal pattern extraction and prediction process, it can better cope with complex abnormal states, optimize the performance of the deep learning model in practical applications, and improve the effect and application scope of the overall prediction system. Through detailed analysis of the behavioral knowledge graph and abnormal state prediction, it ensures that the deep learning model can accurately identify and predict the abnormal state of the lithium battery, thereby improving the overall performance and reliability of the prediction system.

[0114] In this embodiment, all the behavior patterns in the lithium battery behavior knowledge graph are clustered and analyzed to identify the abnormal patterns. The graph embedding technology is used to convert the nodes and edges in the graph into low-dimensional vector representations, and these vectors are classified by clustering algorithms (such as k-means, DBSCAN) to extract patterns related to abnormal behavior. The extraction process is carried out using Python's graphics library (such as NetworkX) combined with a machine learning library (such as Scikit-Learn), and the abnormal behavior patterns obtained by clustering results can be used for subsequent anchor point matching. The abnormal behavior pattern is compared with the anchor point features in the abnormal change anchor point set to calculate the similarity between them. A similarity measurement method (such as cosine similarity and Euclidean distance) is used to evaluate the degree of match between each abnormal change anchor point and the abnormal behavior pattern. By threshold screening these measurement results, a matching abnormal anchor point pattern is obtained. An abnormal state prediction model is constructed based on the matching abnormal anchor point pattern. The abnormal anchor point pattern is trained using a machine learning model (such as a support vector machine and a random forest) to learn the relationship between the pattern and the actual abnormal state. Then, the trained model is used to predict the abnormal change anchor point set and generate the corresponding abnormal state data.

[0115] Preferably, step S6 comprises the following steps:

[0116] Step S61: clustering anchor point potential categories on the anchor point-state anomaly prediction data to obtain clustered anchor point anomaly prediction data;

[0117] Step S62: performing deep learning model training on the cluster anchor point abnormality prediction data to obtain an abnormal state prediction model;

[0118] Step S63: using the abnormal state prediction model to predict the abnormal state of the lithium battery multi-source parameters to obtain the abnormal state data of the lithium battery.

[0119] The present invention generates clustered anchor abnormal prediction data by clustering anchor-state abnormal prediction data. The abnormal prediction data is grouped to identify potential abnormal categories, thereby improving the classification and analysis capabilities of abnormal states. Model training is performed based on clustered abnormal prediction data. The structured data of the clustering results is used to improve the learning effect and training efficiency of the model on abnormal states, so that the model can better capture and predict abnormal patterns. The abnormal state prediction model obtained by training predicts multi-source parameters of lithium batteries. The abnormal state of lithium batteries under different operating conditions is accurately identified, and lithium battery abnormal state data is generated, thereby providing effective early warning and monitoring information. By training the clustered anchor abnormal prediction data, the deep learning model can better handle diverse abnormal patterns, and improve the robustness and accuracy of the model for various abnormal states. By combining clustering processing and deep learning training, the detection of abnormal states is made more efficient and accurate. Clustering helps to refine abnormal data, and the deep learning model uses these data to make more accurate abnormal state predictions. Through the generated abnormal state data, the early warning system can monitor the state of the lithium battery in real time, discover and respond to potential abnormalities in a timely manner, enhance the real-time performance and reliability of the system, and reduce safety hazards. The clustered abnormal prediction data provides structured abnormal information, which facilitates deeper data analysis and decision support, helps maintenance personnel and system administrators make more accurate decisions, and optimizes battery usage and maintenance strategies. Through clustering processing of abnormal prediction data and deep learning model training, accurate prediction of lithium battery abnormal status is ensured, improving the effect and application performance of the entire prediction system.

[0120] In this embodiment, feature information such as abnormal type and frequency is extracted from the anchor-state abnormal prediction data. Then, a clustering algorithm (such as k-means, DBSCAN) is applied to classify these data. In the specific operation, the abnormal prediction data is converted into a feature vector, and the Scikit-Learn library in Python is used for clustering. When k-means is selected, the preset number of categories k is defined, and the data is grouped using the k-means algorithm; when DBSCAN is selected, the clustering parameters are determined according to the density distribution of the data. The result obtained after clustering divides the data into multiple categories, each category represents a potential abnormal type, and generates clustered anchor abnormal prediction data. Prepare a training data set, and organize the clustered anchor abnormal prediction data and its corresponding label (abnormal state) into training samples. Then, design and build a deep learning model architecture, such as using a convolutional neural network (CNN) or a long short-term memory network (LSTM), and select a suitable network structure according to the characteristics of the data. Use TensorFlow or PyTorch deep learning framework to compile, train and verify the model. During the training process, the cross entropy loss function and optimizer (such as Adam optimizer) are used to optimize the model parameters. After multiple iterations of training, the hyperparameters are adjusted to improve the accuracy of the model. After the training is completed, the trained abnormal state prediction model is saved for subsequent data prediction tasks. The multi-source parameters of the lithium battery are converted into the input format required by the model. Then, the converted data is input into the trained abnormal state prediction model for forward propagation calculation. Using the model prediction function in the deep learning framework (such as TensorFlow or PyTorch), the input data is predicted by the model and the prediction results of the abnormal state are output. According to the prediction results of the model, the abnormal state data of the lithium battery is generated, including the predicted abnormal type and its probability of occurrence.

[0121] The present invention also provides a lithium battery abnormal state prediction system based on deep learning, which is used to execute the lithium battery abnormal state prediction method based on deep learning as described above. The lithium battery abnormal state prediction system based on deep learning includes:

[0122] The data processing module is used to obtain multi-source parameters of lithium batteries; perform multiple interpolation processing on missing values ​​of multi-source parameters of lithium batteries, and perform multi-source feature mining to obtain a multi-dimensional feature matrix of lithium batteries;

[0123] Pattern mining module, used to analyze the working status of lithium batteries and conduct behavior pattern mining on the multi-dimensional feature matrix of lithium batteries to generate a knowledge graph of lithium battery behavior;

[0124] The simulation driving module is used to perform fusion simulation driving processing on the lithium battery multi-dimensional feature matrix and the lithium battery behavior knowledge graph to generate a simulated lithium battery model;

[0125] The behavior simulation module is used to obtain the lithium battery use environment data; based on the lithium battery use environment data, the working behavior of the simulated lithium battery model is simulated to obtain the simulated lithium battery working data;

[0126] The state mapping module is used to locate the abnormal change anchor point of the simulated lithium battery working data according to the lithium battery behavior knowledge graph, and perform predicted state mapping to generate anchor point-state abnormality prediction data;

[0127] The state prediction module is used to predict the abnormal state of the lithium battery multi-source parameters based on the anchor point-state abnormality prediction data to obtain the abnormal state data of the lithium battery.

[0128] The present invention can effectively fill in the missing data and extract useful features by performing multiple interpolation processing on the missing values ​​of the multi-source parameters of the lithium battery and performing multi-source feature mining. The integrity and reliability of the data are ensured, high-quality input data is provided for subsequent model training, and the prediction accuracy of the model is improved. The multidimensional feature matrix is ​​analyzed for the battery working status by deep learning technology, and the lithium battery behavior knowledge map is generated by behavior pattern mining. The complex working mode and potential abnormal behavior of the lithium battery are revealed, providing a comprehensive knowledge basis for the identification of abnormal states and improving the model's perception of abnormal patterns. The simulated lithium battery model is generated by fusion simulation drive processing of the multidimensional feature matrix and the behavior knowledge map of the lithium battery. The working behavior of the model is simulated based on the actual use environment data, and accurate simulated lithium battery working data can be generated. It is helpful to verify the effectiveness of the model in a real environment and ensure its reliability and robustness under various working conditions. Combined with the lithium battery behavior knowledge map, the abnormal change anchor point of the simulated lithium battery working data is located, and the predicted state mapping is performed to generate anchor point-state abnormal prediction data. By identifying and mapping abnormal conditions of anchor points, abnormal conditions can be accurately predicted at an early stage, improving the early warning capability and response speed to abnormal conditions of lithium batteries.

[0129] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0130] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for predicting abnormal state of lithium battery based on deep learning, characterized in that: The following steps are involved: Step S1: obtaining multi-source parameters of lithium batteries; performing multiple interpolation processing on missing values ​​of multi-source parameters of lithium batteries, and performing multi-source feature mining to obtain a multi-dimensional feature matrix of lithium batteries; Step S2: Analyze the working status of the lithium battery multi-dimensional feature matrix, and mine the behavior patterns to generate a lithium battery behavior knowledge graph, wherein step S2 includes the following steps: Step S21: performing time series analysis on the multi-dimensional characteristic matrix of the lithium battery to obtain a lithium battery working state sequence; Step S22: Performing a working state analysis on the lithium battery working state sequence to generate lithium battery state state data. The working state analysis specifically includes: applying a statistical analysis method and a sliding window technology to a time series by applying a fixed-length window to calculate the data features in each window, thereby analyzing the change trend of the battery state; Step S23: Performing behavior pattern mining on the lithium battery status data to generate status-behavior pattern data. The behavior pattern mining specifically includes: dividing the status data into different groups, identifying status patterns with similar characteristics, and discovering potential relationships between statuses; Step S24: constructing a behavior knowledge graph for the state-behavior pattern data to obtain a lithium battery behavior knowledge graph; Step S3: Perform fusion simulation drive processing on the lithium battery multi-dimensional feature matrix and the lithium battery behavior knowledge graph to generate a simulated lithium battery model, wherein step S3 includes the following steps: Step S31: performing association rule analysis on the lithium battery multidimensional feature matrix and the lithium battery behavior knowledge graph to obtain an association feature matrix; Step S32: backtracking the behavior pattern of the lithium battery behavior knowledge graph according to the associated feature matrix to generate a behavior pattern matching matrix; Step S33: performing multi-dimensional topological fusion processing on the behavior pattern matching matrix and the lithium battery multi-dimensional feature matrix to obtain a dynamic behavior fusion matrix; Step S34: performing fusion simulation drive processing on the dynamic behavior fusion matrix to generate a simulated lithium battery model; Step S4: obtaining lithium battery use environment data; simulating the working behavior of the simulated lithium battery model based on the lithium battery use environment data to obtain simulated lithium battery working data; Step S5: positioning the abnormal change anchor point of the simulated lithium battery working data according to the lithium battery behavior knowledge graph, and performing predicted state mapping to generate anchor point-state abnormality prediction data, wherein step S5 includes the following steps: Step S51: Perform dynamic graph embedding processing on the lithium battery behavior knowledge graph to obtain low-dimensional graph embedding representation data; Step S52: performing abnormal change detection on the simulated lithium battery working data based on the low-dimensional embedding representation data of the graph to generate an abnormal candidate sequence; Step S53: positioning abnormal change anchor points of the abnormal candidate sequence to obtain an abnormal change anchor point set, wherein step S53 includes the following steps: Perform nonlinear transformation on the abnormal candidate sequence to obtain an abnormal candidate curve set; Perform saddle point detection on the abnormal candidate curve set to obtain the abnormal curve saddle point set; Based on the abnormal curve saddle point set, the abnormal change anchor point of the abnormal candidate curve set is located to obtain the initial abnormal anchor point set; Perform upstream and downstream impact chain analysis on the initial abnormal anchor point set to obtain the abnormal change anchor point set; Step S54: performing a predicted state mapping on the abnormal change anchor point set according to the lithium battery behavior knowledge graph to generate anchor point-state abnormality prediction data, wherein step S54 includes the following steps: Extract abnormal patterns from the lithium battery behavior knowledge graph and generate abnormal behavior patterns; According to the abnormal behavior pattern, the abnormal change anchor point set is matched with the change anchor point to obtain the matching abnormal anchor point pattern; Based on the matching abnormal anchor point pattern, the abnormal state of the abnormal change anchor point set is predicted to generate anchor point-state abnormal prediction data; Step S6: Based on the anchor point-state abnormality prediction data, abnormal state prediction is performed on the lithium battery multi-source parameters to obtain abnormal state data of the lithium battery, wherein step S6 includes the following steps: Step S61: clustering anchor point potential categories on the anchor point-state anomaly prediction data to obtain clustered anchor point anomaly prediction data; Step S62: performing deep learning model training on the cluster anchor point abnormality prediction data to obtain an abnormal state prediction model; Step S63: using the abnormal state prediction model to predict the abnormal state of the lithium battery multi-source parameters to obtain the abnormal state data of the lithium battery.

2. The method for predicting abnormal state of lithium battery based on deep learning according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: Obtaining lithium battery multi-source parameters; Step S12: performing outlier interval detection on the multi-source parameters of the lithium battery to obtain parameter outlier intervals; Step S13: performing multiple interpolation processing on parameter outlier intervals based on lithium battery multi-source parameters to generate an interpolated lithium battery parameter set; Step S14: performing independent component analysis on the interpolated lithium battery parameter set to generate a dimension-reduced lithium battery parameter set; Step S15: Perform multi-source feature mining on the dimension-reduced lithium battery parameter set to obtain a multi-dimensional feature matrix of the lithium battery.

3. The method for predicting abnormal state of lithium battery based on deep learning according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: Acquire lithium battery usage environment data; Step S42: extracting multi-dimensional variables from the lithium battery usage environment data to obtain an environment variable matrix; Step S43: constructing a simulated environment field for the environmental variable matrix to generate a changing simulated environment field; Step S44: mapping the behavior characteristics of the simulated lithium battery model based on the changed simulated environment field and the environment variable matrix to obtain a simulated behavior characteristic matrix; Step S45: performing multi-scale time series analysis on the simulation behavior feature matrix to generate a behavior pattern sequence matrix; Step S46: performing simulation behavior fitting on the simulation lithium battery model according to the behavior pattern sequence matrix to obtain simulated lithium battery operating data.

4. A lithium battery abnormal state prediction system based on deep learning, characterized in that: Used to execute the lithium battery abnormal state prediction method based on deep learning as claimed in claim 1, the lithium battery abnormal state prediction system based on deep learning comprises: The data processing module is used to obtain multi-source parameters of lithium batteries; perform multiple interpolation processing on missing values ​​of multi-source parameters of lithium batteries, and perform multi-source feature mining to obtain a multi-dimensional feature matrix of lithium batteries; Pattern mining module, used to analyze the working status of lithium batteries and conduct behavior pattern mining on the multi-dimensional feature matrix of lithium batteries to generate a knowledge graph of lithium battery behavior; The simulation driving module is used to perform fusion simulation driving processing on the lithium battery multi-dimensional feature matrix and the lithium battery behavior knowledge graph to generate a simulated lithium battery model; The behavior simulation module is used to obtain the lithium battery use environment data; based on the lithium battery use environment data, the working behavior of the simulated lithium battery model is simulated to obtain the simulated lithium battery working data; The state mapping module is used to locate the abnormal change anchor point of the simulated lithium battery working data according to the lithium battery behavior knowledge graph, and perform predicted state mapping to generate anchor point-state abnormality prediction data; The state prediction module is used to predict the abnormal state of the lithium battery multi-source parameters based on the anchor point-state abnormality prediction data to obtain the abnormal state data of the lithium battery.

Images