Battery fast charging safety control method, device, equipment and storage medium

Abstract

The present invention relates to a method, device, equipment and storage medium for fast charging safety control of a storage battery. The method: modeling the equivalent circuit and thermal equivalent circuit of the storage battery to obtain a dynamic decoupling matrix; obtaining battery safety status data through parallel calculation; performing abnormal pattern recognition to obtain abnormal pattern recognition results; inputting the abnormal pattern recognition results into a heterogeneous graph gated recurrent unit, extracting spatiotemporal features through heterogeneous graph convolution and GRU structure to obtain deep fault features; calculating the charging safety capacity, and performing multiple constraint control on battery thermal management and charge state to obtain a charging power allocation scheme; inputting the charging power allocation scheme into a two-stage model predictive controller for distributed power redistribution to obtain a charging control instruction, which can monitor the battery status in real time, accurately identify abnormal patterns, and perform adaptive control based on multiple constraints to achieve a safe and efficient fast charging process.

Description

Technical Field

[0001] The present invention relates to the technical field of battery charging, and in particular to a method, device, equipment and storage medium for safe control of rapid charging of a battery. Background Art

[0002] Fast charging can not only significantly shorten charging time and improve user experience, but also improve the operational efficiency of charging stations. However, the high current and high power during fast charging will bring huge thermal and electrochemical stress to the battery, which may lead to battery performance degradation, shortened life, and even cause safety accidents.

[0003] Traditional charging control methods often have difficulty striking a good balance between charging efficiency and safety. On the one hand, overly conservative control strategies can lead to long charging times, affecting user experience; on the other hand, overly aggressive control strategies may cause safety hazards such as battery overheating and overcharging. In addition, existing charging control methods usually have difficulty accurately capturing the complex dynamic characteristics of batteries during fast charging, especially thermal-electric coupling effects and various abnormal modes, which poses challenges to control accuracy and reliability. Summary of the invention

[0004] The main purpose of the present invention is to provide a battery fast charging safety control method, device, equipment and storage medium, which can monitor the battery status in real time, accurately identify abnormal modes, and perform adaptive control based on multiple constraints to achieve a safe and efficient fast charging process.

[0005] To achieve the above object, the present invention provides a battery fast charging safety control method, comprising the following steps:

[0006] Modeling an equivalent circuit and a thermal equivalent circuit of the battery to obtain a virtual circuit model, and decoupling the thermal state and the electrochemical state of the virtual circuit model to obtain a dynamic decoupling matrix;

[0007] Inputting charging current and voltage sampling data into the virtual circuit model, performing state separation using the dynamic decoupling matrix, and obtaining battery safety state data through parallel calculation;

[0008] Perform abnormal pattern recognition based on the battery safety status data to obtain an abnormal pattern recognition result;

[0009] The abnormal pattern recognition result is input into the heterogeneous graph gated recurrent unit, and the spatiotemporal features are extracted through the heterogeneous graph convolution and GRU structure to obtain the deep fault features;

[0010] Calculate the safe charging capacity according to the deep fault characteristics, and perform multiple constraint controls on battery thermal management and charge state to obtain a charging power allocation plan;

[0011] The charging power allocation scheme is input into a two-stage model predictive controller for distributed power redistribution to obtain a charging control instruction.

[0012] The present invention also provides a battery fast charging safety control device, comprising:

[0013] A decoupling module is used to model the equivalent circuit and thermal equivalent circuit of the battery to obtain a virtual circuit model, and decouple the thermal state and electrochemical state of the virtual circuit model to obtain a dynamic decoupling matrix;

[0014] A separation module, used to input charging current and voltage sampling data into the virtual circuit model, perform state separation using the dynamic decoupling matrix, and obtain battery safety state data through parallel calculation;

[0015] an identification module, configured to perform abnormal pattern identification based on the battery safety status data to obtain an abnormal pattern identification result;

[0016] An extraction module, used for inputting the abnormal pattern recognition result into a heterogeneous graph gated recurrent unit, extracting spatiotemporal features through heterogeneous graph convolution and GRU structure, and obtaining deep fault features;

[0017] A control module, used to calculate the charging safety capacity according to the deep fault characteristics, and perform multiple constraint control on battery thermal management and charge state to obtain a charging power allocation plan;

[0018] The allocation module is used to input the charging power allocation plan into the two-stage model predictive controller to perform distributed power redistribution and obtain a charging control instruction.

[0019] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0020] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0021] In summary, the technical solution provided by the present invention realizes accurate modeling of complex dynamic characteristics of batteries and improves the accuracy of state estimation by constructing a virtual circuit model and decoupling thermal state and electrochemical state. The state separation and parallel calculation are performed using a dynamic decoupling matrix, which significantly improves the efficiency of obtaining battery safety state data. The improved abnormal pattern recognition algorithm is adopted, combined with the sensor topology structure with internal and external measurement points separated, to enhance the detection ability of various abnormal patterns. The heterogeneous graph gate recurrent unit is introduced, and the spatiotemporal features are extracted through heterogeneous graph convolution and GRU structure, so as to effectively capture the deep fault features. The charging safety capacity is calculated based on the deep fault features, and multiple constraint control is performed to maximize the charging efficiency while ensuring safety. The two-stage model predictive controller is used for distributed power redistribution to achieve coordinated optimization of the overall performance of the charging station. The present invention can monitor the battery state in real time, accurately identify abnormal modes, and perform adaptive control based on multiple constraints to achieve a safe and efficient fast charging process, and achieve a better balance between charging efficiency, safety and battery life. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 1 is a schematic diagram of the steps of a method for controlling the rapid charging safety of a battery in one embodiment of the present invention;

[0023] Figure 2 This is a structural block diagram of a battery fast charging safety control device in one embodiment of the present invention;

[0024] Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0025] 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

[0026] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0027] Reference Figure 1 This embodiment provides a battery fast charging safety control method, comprising the following steps:

[0028] S1, modeling the equivalent circuit and thermal equivalent circuit of the battery to obtain a virtual circuit model, and decoupling the thermal state and electrochemical state of the virtual circuit model to obtain a dynamic decoupling matrix;

[0029] Among them, the open circuit voltage, internal resistance and polarization characteristics of the battery are parameterized. By measuring the voltage response curves under different charge states and combining the least squares fitting method, the open circuit voltage function, internal resistance function and polarization impedance function of the battery are obtained, and an electrochemical equivalent circuit model is constructed. This model can reflect the electrochemical behavior characteristics inside the battery, including the relationship between its open circuit voltage and charge state, the change of internal resistance with the state, and the voltage hysteresis characteristics caused by the polarization effect. At the same time, the thermal capacity and thermal resistance of the battery are parameterized. Through constant current charge and discharge tests and temperature response curves, the specific heat capacity of the battery and the thermal resistance of each layer structure are calculated using the thermal balance equation and combined with finite element analysis. The thermal equivalent circuit model is used to describe the thermal behavior of the battery during charging and discharging, and can reveal the thermal impedance characteristics of the battery during temperature changes, as well as the response of each layer structure during heat conduction. The electrochemical equivalent circuit model and the thermal equivalent circuit model are connected in series to establish the coupling relationship between the internal current, voltage and temperature of the battery, and obtain a unified virtual circuit model. In this coupling relationship, the Joule heat generated by the electrochemical equivalent circuit model is used as the input of the thermal equivalent circuit model, while the temperature of the battery affects the parameters (such as internal resistance, polarization impedance, etc.) in the electrochemical equivalent circuit model through reaction. The virtual circuit model can simultaneously describe the electrochemical and thermal processes of the battery and more comprehensively reflect the overall dynamic behavior of the battery. The virtual circuit model is represented in state space, and the state of charge, polarization voltage, core temperature and surface temperature are selected as state variables to construct the state equation and output equation. In this way, a state vector containing thermal and electrochemical states is obtained, and the thermal and electrochemical characteristics of the battery are jointly described. In order to achieve effective decoupling of the thermal state and the electrochemical state, an orthogonal transformation matrix is ​​constructed, and the state vector is linearly transformed to decompose the state space into thermal subspace and electrochemical subspace. According to the characteristic changes in the charging stage, time-varying parameters are introduced to construct an adaptive orthogonal transformation matrix. The adaptive orthogonal transformation matrix can be adjusted accordingly according to the changes in the thermal and electrochemical behaviors of the battery during the charging process, making the decoupling of the state space more accurate, thereby effectively separating the thermal state and the electrochemical state. Using the constructed adaptive orthogonal transformation matrix, the state vector of the virtual circuit model is projected into the thermal subspace and the electrochemical subspace to obtain the decoupled thermal state and electrochemical state. The decoupled thermal state is used to describe the temperature characteristics of the battery, while the electrochemical state reflects the characteristics of the battery's electrochemical process. The two decoupled states are combined to obtain a dynamic decoupling matrix. The dynamic decoupling matrix makes safety management and control during fast charging more efficient and accurate by accurately distinguishing the thermal and electrochemical behaviors of the battery.

[0030] S2, input charging current and voltage sampling data into the virtual circuit model, use the dynamic decoupling matrix to separate the states, and obtain the battery safety state data through parallel calculation;

[0031] Specifically, the charging current and voltage are sampled at high frequency. The sampling frequency is set to 10kHz, and the high sampling frequency can ensure that sufficiently detailed current and voltage change information is obtained. A sliding window median filter with a length of 100 data points is used to remove outliers and high-frequency noise in the sampled data to ensure the stability and credibility of the data. In order to improve the smoothness of the data, the Savitzky-Golay filter is used to smooth the data. The order of the filter is set to 3, and the window length is set to 51 to maintain the smoothness of the data and retain the characteristic changes in the current and voltage signals to obtain the current and voltage data sequence after noise reduction. The current and voltage data sequence after noise reduction is input into the virtual circuit model for numerical integration. In order to ensure the accuracy and stability of the integration process, the fourth-order Runge Kutta method is used to numerically solve the model to obtain the initial state estimate. The fourth-order Runge Kutta method is a numerical integration method with high accuracy and stability, which is suitable for the dynamic solution of multi-physics coupling systems such as batteries. The dynamic decoupling matrix is ​​applied to perform orthogonal transformation on the initial state estimate to achieve decoupling of the thermal state and the electrochemical state, and obtain the decoupled thermal state component and electrochemical state component. Through decoupling, different physical processes in the complex coupled system are effectively separated, which facilitates independent analysis and control of each state. In the calculation of the thermal state, the thermal state observation equation is constructed based on the decoupled thermal state component to describe the thermal behavior of the battery, and the thermal state is updated using the unscented Kalman filter algorithm. The unscented Kalman filter is a filtering method suitable for state estimation of nonlinear systems. Through iterative calculations of two steps, time update and measurement update, the predicted value of the state is gradually corrected to improve the accuracy of state estimation. The core temperature and surface temperature of the battery are finally obtained through the unscented Kalman filter algorithm. In the calculation of the electrochemical state, the electrochemical state observation equation is constructed based on the decoupled electrochemical state component, and the state is updated using the extended Kalman filter algorithm. The extended Kalman filter linearizes the nonlinear part of the system to generate a Jacobian matrix, and based on this Jacobian matrix, updates the system state and the prediction error covariance matrix, and calculates the Kalman gain at the same time. The battery state of charge and polarization voltage are obtained through iterative updates over time. The state of charge is an important indicator for measuring the remaining power of the battery, while the polarization voltage can reflect the polarization phenomenon generated by the battery during the charging process. The battery core temperature, surface temperature, state of charge and polarization voltage are combined to construct a complete battery state vector, which can comprehensively reflect the important state indicators of the battery during the fast charging process. Based on the battery state vector, the safety state of the battery is evaluated. The thermal runaway risk index is calculated using the thermal runaway probability model, which is used to quantify the thermal runaway risk of the battery during the charging process.At the same time, the Coulomb efficiency model is used to calculate the battery health index, which is used to reflect the overall health of the battery, including its performance degradation after multiple charge and discharge cycles. The thermal runaway risk index and the battery health index are weighted and fused to comprehensively consider the thermal safety and battery health status during the charging process to obtain comprehensive battery safety status data.

[0032] S3, performing abnormal pattern recognition based on the battery safety status data to obtain an abnormal pattern recognition result;

[0033] It should be noted that, first of all, the battery safety status data is segmented into time series. The data is processed using a sliding window with a length of 600 seconds and a step size of 300 seconds. This ensures that when analyzing the battery core temperature, surface temperature, state of charge, and polarization voltage, it can not only reflect the trend of changes in a relatively long period of time, but also capture abnormal behaviors in a relatively short period of time, and obtain a multidimensional time series data set. When extracting features from the multidimensional time series data set, a comprehensive statistical analysis is performed from multiple aspects. For each time window, the statistical features such as the mean, standard deviation, kurtosis, and skewness of the core temperature, surface temperature, state of charge, and polarization voltage are calculated to obtain the central trend, degree of dispersion, and distribution form of the data. Time domain features such as the maximum rate of increase and the maximum rate of decrease are calculated to capture the extreme cases of battery state changes. These features reflect the various state change modes of the battery during the charging process, and finally a feature vector containing 48-dimensional features is constructed. In order to optimize the features and reduce the redundancy between features, the 48-dimensional feature vector is subjected to correlation analysis. By calculating the Pearson correlation coefficient between features, the degree of correlation between features is obtained, and a correlation coefficient matrix is ​​constructed. Based on the correlation coefficient matrix, feature pairs with absolute values ​​of correlation coefficients greater than 0.8 are selected. In order to avoid information redundancy and multicollinearity problems, principal component analysis is performed on the selected feature pairs. By calculating the feature covariance matrix and solving the eigenvalues ​​and eigenvectors, the main components that mainly affect the system changes are found. In order to ensure that the feature data after dimensionality reduction can retain most of the information, the principal components with a cumulative contribution rate of 95% are selected as the new feature set. Based on the feature data after dimensionality reduction, a sensor topology structure with internal and external measurement points separated is constructed, and the features are divided into internal features and external features for anomaly detection. By analyzing the features from different sources separately, the abnormal behavior of the battery under the influence of internal electrochemical changes and external environment is captured. When detecting internal features and external features separately, the local anomaly factor algorithm is used to calculate the local reachable density and local anomaly factor of each data point to obtain the internal anomaly score and external anomaly score. The local anomaly factor algorithm is a density-based anomaly detection method that can effectively identify data points in the data set that appear abnormal due to local density differences, thereby identifying abnormal patterns in the battery state. The internal anomaly score and the external anomaly score are weighted and fused to obtain a comprehensive anomaly score. The comprehensive anomaly score reflects the overall abnormality of the battery during the charging process, and the battery state is evaluated by combining the anomaly scores of internal and external features. The Otsu threshold method is used to determine the optimal anomaly threshold. The Otsu threshold method is an adaptive threshold selection method that can automatically select the optimal threshold by maximizing the inter-class variance, thereby optimally distinguishing between normal and abnormal states. Based on the comprehensive anomaly score and the optimal anomaly threshold, the battery state is classified into abnormal patterns to obtain the final abnormal pattern recognition result.

[0034] S4, inputs the abnormal pattern recognition results into the heterogeneous graph gate recurrent unit, extracts spatiotemporal features through heterogeneous graph convolution and GRU structure, and obtains deep fault features;

[0035] Specifically, the results of abnormal pattern recognition are constructed into a heterogeneous graph structure. The nodes are used to represent different parameters of the battery, such as core temperature, surface temperature, state of charge, and polarization voltage, while the edges are used to represent the relationship between these parameters, thereby obtaining an initial heterogeneous graph. In order to improve the information interaction effect between nodes in the graph structure, the initial heterogeneous graph is processed by a multi-head attention mechanism. By calculating the attention weights between nodes of different types, an attention-enhanced heterogeneous graph is obtained. The multi-head attention mechanism comprehensively considers the influence between nodes of different types and depicts the degree of correlation between parameters, making the representation of nodes in the graph more accurate. The attention-enhanced heterogeneous graph is input into the heterogeneous graph convolution layer for feature extraction. The heterogeneous graph convolution layer contains three parallel graph convolution sublayers, which process the node features, the relationship between nodes of the same type, and the relationship between nodes of different types respectively. Through these three convolution sublayers, the local features of the node, the correlation features between itself and nodes of the same type, and the connection between nodes of different types are captured respectively, and the local spatial features are obtained. The local spatial features describe the interaction and influence between various battery parameters during the charging process, and reveal the dynamic change characteristics of the battery state in the microscopic space. The local spatial features are processed by skip connection. The original node features are concatenated with the features extracted after convolution, and the concatenated features are fused through the fully connected layer to obtain enhanced spatial features. The skip connection process helps to retain the information of the original input features, avoid information loss in the multi-layer convolution process, and enhance the expressiveness of the features. The enhanced spatial features are input into the gated graph neural network layer for processing. The gated graph neural network layer contains update gates and reset gates. Through these gating mechanisms, the node information is selectively updated and reset to better extract and retain important information. This process enables the obtained node representation to perceive timing information, that is, the timing-aware node representation. The timing-aware node representation is input into the bidirectional GRU structure for processing. The bidirectional GRU contains two GRU branches, the forward GRU and the reverse GRU. The forward GRU is used to capture the forward timing dependency, while the reverse GRU is used to capture the reverse timing dependency. Through the processing of the bidirectional GRU, a more comprehensive bidirectional timing feature can be obtained, which contains all the dynamic change information of the battery state in time. In order to improve the information interaction between features, the self-attention mechanism is applied to the bidirectional time series features for global information interaction, and the importance weights of different time steps are calculated. The self-attention mechanism can dynamically adjust the contribution of each node in the time series to the overall representation by assigning weights to each time step, and extract more representative features. Through multi-level feature extraction, including from spatial features to time series features, and then to global information interaction, the deep fault features of the battery system are obtained, which comprehensively reflect the state changes of the battery during the fast charging process and its potential abnormal modes.

[0036] S5, calculate the charging safety capacity according to the deep fault characteristics, and perform multiple constraint controls on the battery thermal management and charge state to obtain the charging power allocation plan;

[0037] Among them, the deep fault feature is converted into a 64-dimensional vector and input into a three-layer fully connected neural network for processing. In the three-layer fully connected neural network, the first layer contains 128 neurons, the second layer contains 64 neurons, and the third layer contains 32 neurons. Each layer is followed by a ReLU activation function to ensure that the network can effectively capture nonlinear features. At the same time, in order to prevent overfitting, a Dropout layer with a probability of 0.5 is added after each layer for regularization. The last layer uses a Sigmoid activation function to map the output between 0 and 1 to obtain the charging safety capacity coefficient, which reflects the safe charging capacity ratio of the battery in the current state. Through the charging safety capacity coefficient, combined with the rated capacity and current state of charge of the battery, the maximum allowable charging current and the maximum allowable charging power are calculated to construct the constraints of the charging safety capacity. The constraints can provide a safe boundary for the selection of charging power during fast charging to prevent battery damage or safety accidents caused by overcharging. Based on the thermal-electric coupling model, the temperature dynamic equation of the battery is established, and the temperature equation is discretized by the finite difference method to obtain a discrete temperature prediction model. Based on the model, the temperature change of the battery under different charging strategies is predicted. In order to ensure that the battery will not cause safety problems due to excessive temperature during the charging process, the maximum temperature rise rate, the maximum temperature limit and the temperature uniformity requirements are set to construct the temperature constraint conditions. By establishing the dynamic equation of the state of charge and combining it with the voltage dynamic equation, the charge change of the battery during the charging process is described. Based on these equations, the upper limits of the maximum charging voltage and the state of charge are set to construct the state of charge constraint conditions to ensure that the battery will not be damaged due to overcharging. The charging safety capacity constraint conditions, temperature constraint conditions and state of charge constraint conditions are combined to construct a multi-objective optimization problem. In the multi-objective optimization problem, the first objective function includes minimizing the charging time and maximizing the battery life, and the decision variable is the charging current. The non-dominated sorting genetic algorithm is used to solve the multi-objective optimization problem. The non-dominated sorting genetic algorithm is an evolutionary algorithm suitable for multi-objective optimization. In this algorithm, the population size is set to 100, the number of iterations is 200, the crossover probability is 0.9, and the mutation probability is 0.1. Through the congestion distance calculation, individuals in the population are selected to obtain a set of non-dominated solution sets, representing charging strategies with different trade-offs between charging time and battery life. The fuzzy comprehensive evaluation method is applied to the non-dominated solution set, and the membership function is designed for each objective function, and the weight vector of each objective is determined by the hierarchical analysis method. By calculating the comprehensive score, the solution with the highest score is selected from the non-dominated solution set as the optimal charging strategy. Based on the selected optimal charging strategy, the power allocation problem of the actual charging station is considered. Considering the total power limit of the charging station and the demand for load balancing, a quadratic programming problem is constructed and solved using the interior point method to obtain a specific charging power allocation scheme.Through quadratic programming solution, it is ensured that reasonable power allocation is achieved for the charging process of multiple batteries while meeting the total power limit of the charging station, thereby optimizing the power utilization efficiency of the charging station to the greatest extent and ensuring that the charging of multiple batteries can be carried out in a coordinated manner.

[0038] S6, input the charging power allocation plan into the two-stage model predictive controller for distributed power redistribution to obtain charging control instructions.

[0039] Among them, the charging power allocation scheme is converted into the initial charging power sequence. Based on the initial charging power sequence, a charging station-level model predictive control problem is constructed. In the charging station-level model predictive control, the goal is to minimize the total power deviation and power fluctuation of the charging station through the second objective function to ensure the smooth and efficient operation of the entire charging process. In order to achieve this goal, a series of constraints are considered, including the upper limit of the total power of the charging station and the limit of the power change rate, so as to ensure that the total power of the charging station does not exceed its rated range, and at the same time avoid the adverse effects of drastic changes in power on the charging equipment and batteries. In order to solve the charging station-level model predictive control problem, the augmented Lagrangian method is used. The augmented Lagrangian method is an effective optimization solution method. By introducing Lagrangian multipliers and penalty functions to deal with constraint problems, the objective function is optimized while ensuring that the constraints are met, and the optimized total power sequence of the charging station is obtained. The optimized power sequence can effectively reduce the power fluctuation and power deviation in the entire charging process, making the power allocation of all charging devices in the charging station more reasonable and stable. Based on the optimized total power sequence of the charging station, a distributed consensus algorithm is constructed to realize power redistribution between charging devices. The distributed consensus algorithm enables adjacent charging devices to exchange local information with each other by designing a communication topology matrix, and continuously updates their respective power allocation values ​​based on these local information. Through the information exchange and update process, each charging device works together without relying on the central controller, thereby reaching a consensus on power allocation, making the power load between charging devices more balanced. For each charging device, a charging device-level model predictive control problem is constructed based on the updated local power allocation value. In this level of model predictive control, the goal is to minimize the power tracking error and charging time. The thermal management constraints and state of charge constraints of the battery are considered to ensure that the thermal state and health state of the battery are not adversely affected while the charging speed is accelerated. The fast gradient method is used to solve the charging device-level model predictive control problem and obtain the optimized charging current sequence. Using the optimized charging current sequence, the battery state of the next N time steps is predicted, including voltage, temperature and state of charge. Based on the predicted battery state, the safety margin index is calculated. The safety margin index reflects the safety status of the battery under the current charging conditions. If the safety margin is low, it indicates that there is a certain risk. According to the safety margin index, the charging current is corrected to obtain a corrected charging current sequence. The correction process helps to adjust the charging strategy in time during the charging process to ensure that the battery always works within a safe range. The first time step of the corrected charging current sequence is used as the charging control instruction of the current control cycle, and this charging control instruction is sent to the corresponding charging device.

[0040] In one example, the equivalent circuit and thermal equivalent circuit of the battery are modeled to obtain a virtual circuit model, and the thermal state and electrochemical state of the virtual circuit model are decoupled to obtain a dynamic decoupling matrix, including: parameterizing the open circuit voltage, internal resistance and polarization characteristics of the battery, measuring the voltage response curve under different charge states, fitting the open circuit voltage function, internal resistance function and polarization impedance function using the least squares method, and constructing an electrochemical equivalent circuit model; parameterizing the thermal capacity and thermal resistance of the battery, and calculating the specific heat capacity of the battery and the thermal resistance of each layer structure through constant current charge and discharge tests and temperature response curves, using thermal balance equations and finite element analysis, and constructing a thermal equivalent circuit model; connecting the electrochemical equivalent circuit model and the thermal equivalent circuit model in series to establish a coupling relationship between current, voltage and temperature. , a virtual circuit model is obtained, in which the Joule heat of the electrochemical equivalent circuit model is used as the input of the thermal equivalent circuit model, and the temperature reacts on the parameters of the electrochemical equivalent circuit model; the virtual circuit model is represented in state space, and the state of charge, polarization voltage, core temperature and surface temperature are selected as state variables, and the state equation and output equation are constructed to obtain a state vector containing thermal state and electrochemical state; an orthogonal transformation matrix is ​​constructed, and the state vector is linearly transformed to decompose the state space into thermal subspace and electrochemical subspace, and according to the characteristic changes in the charging stage, time-varying parameters are introduced to construct an adaptive orthogonal transformation matrix; the thermal subspace and the electrochemical subspace are projected using the adaptive orthogonal transformation matrix to obtain the decoupled thermal state and electrochemical state, and the decoupled thermal state and electrochemical state are combined to construct a dynamic decoupling matrix.

[0041] In this example, by measuring the voltage response curves at different states of charge (SOC), the change of the battery voltage with SOC is obtained, reflecting the internal electrochemical reaction characteristics of the battery at different SOCs, including its open circuit voltage, internal resistance and polarization characteristics. The least squares method is used to fit the measured data to obtain the open circuit voltage function, internal resistance function and polarization impedance function. The least squares method obtains the optimal parameters by minimizing the error between the measured data and the fitting model, so that the model accurately describes the voltage response behavior of the battery. The open circuit voltage function is expressed as

[0042] ;

[0043] in, represents the open circuit voltage, are the fitting coefficients, which are obtained by fitting the actual measured data using the least square method. Similarly, the variation characteristics of the internal resistance and polarization impedance are obtained by a similar method, and the internal resistance function is obtained. and polarization impedance function , these functions reflect the changes in the internal resistance and polarization characteristics of the battery under different charging states. At the same time, in order to establish the thermal model of the battery, the thermal capacity and thermal resistance of the battery are parameterized. Through constant current charge and discharge tests and temperature response curve measurements, the thermal balance equation and finite element analysis are used to calculate the specific heat capacity of the battery and the thermal resistance of each layer structure. The thermal balance equation is expressed as

[0044] ;

[0045] in, is the quality of the battery, is the specific heat capacity of the battery, is the battery temperature, It is the heat generated by the battery during charging or discharging (mainly composed of Joule heat and polarization heat). Represents heat loss. Through constant current charge and discharge test, measure the temperature response curve of the battery, and use the above heat balance equation to infer the specific heat capacity And the thermal resistance of each layer structure . Finite element analysis is used to accurately describe the heat conduction process. Taking into account the thermal resistance between different layers of the battery (such as electrodes, diaphragms, electrolytes, etc.), the thermal resistance characteristics of each layer of the structure are determined through finite element simulation to construct a thermal equivalent circuit model. The electrochemical equivalent circuit model and the thermal equivalent circuit model are connected in series to establish a coupling relationship between current, voltage, and temperature to obtain a virtual circuit model. In the virtual circuit model, the Joule heat generated by the electrochemical equivalent circuit model is used as the input of the thermal equivalent circuit model, that is, the heat generated by the flow of current during the charging and discharging process of the battery will cause the battery temperature to rise, and the temperature in turn will affect the parameters in the electrochemical process, such as internal resistance and polarization impedance. Therefore, there is a coupling relationship between the electrochemical process and the thermal process. For example, when the battery temperature rises, the internal resistance usually decreases, but the polarization effect increases. These mutual influences need to be fully described by the virtual circuit model. In order to more effectively control and analyze the virtual circuit model, it is converted into the form of state space. Select the state of charge (SOC), polarization voltage ( ), core temperature( ) and surface temperature ( ) as the state variable, construct the state equation and output equation. The state equation is expressed as

[0046] ;

[0047] in, is the state vector, is the state matrix, which describes the relationship between the variables in the system. is the input matrix, To control the input (such as charging current). Through the representation of the state space, the thermal state and electrochemical state of the system are described in a unified mathematical framework. In order to decouple the thermal state and electrochemical state, an orthogonal transformation matrix is ​​constructed to linearly transform the state vector. Through this linear transformation, the state space is decomposed into thermal subspace and electrochemical subspace, so that the two processes can be analyzed and controlled independently. Specifically, according to the characteristic changes in the charging stage, time-varying parameters are introduced to construct an adaptive orthogonal transformation matrix, so that the decoupling process can be dynamically adjusted as the charging process changes to adapt to different changes in the system state. The construction method of the adaptive orthogonal transformation matrix takes into account the characteristic changes in the charging process. For example, in the early stage of charging, the temperature changes relatively slowly, and as the charging proceeds, the thermal effect begins to be significant, and the coupling strength of the thermal subspace and electrochemical subspace of the system changes, so it is adaptively adjusted through time-varying parameters. Using the constructed adaptive orthogonal transformation matrix, the state vector of the virtual circuit model is projected into the thermal subspace and electrochemical subspace to obtain the decoupled thermal state and electrochemical state. The decoupled thermal state mainly describes the temperature characteristics of the battery, including the changes in the core temperature and surface temperature, while the electrochemical state describes the changes in the battery's state of charge and polarization voltage. These two decoupled states are combined to construct a dynamic decoupling matrix to describe how the thermal state and electrochemical state of the battery dynamically evolve during the charging process.

[0048] In one example, the charging current and voltage sampling data are input into a virtual circuit model, and the state separation is performed using a dynamic decoupling matrix, and the battery safety state data is obtained through parallel calculation, including: high-frequency sampling of the charging current and voltage, with the sampling frequency set to 10kHz, median filtering using a sliding window with a length of 100 data points to remove outliers and high-frequency noise, and then applying a Savitzky-Golay filter for smoothing, with a filter order of 3 and a window length of 51 to obtain a denoised current and voltage data sequence; the denoised current and voltage data sequence is input into a virtual circuit model, and numerical integration is performed using a fourth-order Runge-Kutta method to obtain an initial state estimate, and a dynamic decoupling matrix is ​​applied to the initial state estimate to perform an orthogonal transformation to obtain a decoupled thermal state component and a decoupled electrochemical state component; based on the decoupling Based on the decoupled thermal state component, the thermal state observation equation is constructed, and the state is updated using the unscented Kalman filter algorithm. The battery core temperature and surface temperature are obtained through iterative calculation in two steps: time update and measurement update. Based on the decoupled electrochemical state component, the electrochemical state observation equation is constructed, and the state is updated using the extended Kalman filter algorithm. The Jacobian matrix is ​​obtained through linearization processing, and the prediction error covariance matrix and the Kalman gain are iteratively updated over time to obtain the battery state of charge and polarization voltage. The battery core temperature, surface temperature, state of charge and polarization voltage are combined to obtain the battery state vector. Based on the battery state vector, the thermal runaway risk index is calculated using the thermal runaway probability model, and the battery health state index is calculated using the Coulomb efficiency model. The thermal runaway risk index and the battery health state index are weighted and fused to obtain the comprehensive battery safety status data.

[0049] In this example, the charging current and voltage are sampled at high frequency, and the sampling frequency is set to 10kHz. High-frequency sampling reflects the electrochemical process inside the battery and its changes over time. The collected data is preprocessed, and a sliding window median filter with a length of 100 data points is applied to remove outliers and high-frequency noise. Median filtering is a commonly used denoising technique. By taking the median of the data in each sliding window, isolated outliers are effectively eliminated to obtain smoother data. The Savitzky-Golay filter is applied to smooth the data. The Savitzky-Golay filter is a smoothing filter based on polynomial fitting, which can reduce noise while retaining important features in the data. The filter order is selected as 3 and the window length is 51 to ensure that the key information in the charging current and voltage data is retained while removing noise, and the current and voltage data series after denoising are obtained. The denoised current and voltage data series are input into the virtual circuit model for numerical integration. The fourth-order Runge-Kutta method is used for numerical integration to calculate the state of the system with high accuracy. The fourth-order Runge-Kutta method estimates the state of the next time step by combining the values ​​of multiple intermediate points, thereby effectively improving the accuracy of the integration process. For a given battery state equation, the numerical integration process is expressed as:

[0050] ;

[0051] in, represents the current state vector, is the increment of different intermediate points in the fourth-order Runge-Kutta method, which is used to calculate the state vector of the next time step . Through this integration method, the initial state estimate of the battery is obtained. The initial state estimate is applied to the dynamic decoupling matrix for orthogonal transformation to obtain the decoupled thermal state component and electrochemical state component. The role of the dynamic decoupling matrix is ​​to separate the internal thermal effect and electrochemical effect of the battery so that the system can analyze these two states separately. The thermal state component describes the temperature change of the battery, including the core temperature and surface temperature; while the electrochemical state component describes the change of the battery's state of charge and polarization voltage. Based on the decoupled thermal state component, the observation equation of the thermal state is constructed, and the state is updated using the unscented Kalman filter algorithm. The unscented Kalman filter is a state estimation method suitable for nonlinear systems. It performs iterative calculations through two steps, time update and measurement update, and gradually corrects the predicted value of the temperature to obtain more accurate battery core temperature and surface temperature. In the time update stage, the model is used to predict the state of the next time step; in the measurement update stage, the predicted value is corrected in combination with the actual measured data. This process enables the system to obtain the temperature change of the battery during fast charging. Similarly, based on the decoupled electrochemical state components, the observation equation of the electrochemical state is constructed, and the state is updated using the extended Kalman filter algorithm. The extended Kalman filter linearizes the nonlinear part of the system, obtains the Jacobian matrix of the system, and uses the matrix to calculate the prediction error covariance matrix and Kalman gain. Through the time iterative update of these steps, the state of charge and polarization voltage of the battery are obtained. Specifically, the state update process of the extended Kalman filter is expressed as:

[0052] ;

[0053] in, is the updated state vector, is the Kalman gain, is the measured value, It is the measured prediction value estimated based on the prior state. Through continuous updating, more accurate state of charge and polarization voltage are obtained. The battery core temperature, surface temperature, state of charge and polarization voltage are combined to obtain the battery state vector. Based on the battery state vector, the thermal runaway risk index is calculated using the thermal runaway probability model. The thermal runaway risk index is used to assess the possibility of thermal runaway of the battery under the current charging conditions. Its calculation is usually based on parameters such as temperature change rate and core temperature. At the same time, the battery health status index is calculated using the Coulomb efficiency model. Coulomb efficiency is an important indicator of the efficiency of power conversion during battery charging and discharging. The Coulomb efficiency model is used to judge the health status of the battery, including its capacity attenuation and performance degradation. The thermal runaway risk index and the battery health status index are weighted and fused to obtain a comprehensive battery safety status data.

[0054] In one example, abnormal pattern recognition is performed based on battery safety status data to obtain abnormal pattern recognition results, including: segmenting the battery safety status data into time series, using a sliding window with a length of 600 seconds and a step size of 300 seconds to segment the battery core temperature, surface temperature, state of charge, and polarization voltage to obtain a multidimensional time series data set; extracting features from the multidimensional time series data set, calculating the statistical features of the mean, standard deviation, kurtosis, and skewness in each time window, and the time domain features of the maximum rising rate and the maximum falling rate, and constructing a 48-dimensional feature vector; performing correlation analysis on the 48-dimensional feature vector, calculating the Pearson correlation coefficient between features, constructing a correlation coefficient matrix, and selecting feature pairs with absolute values ​​of correlation coefficients greater than 0.8 based on the correlation coefficient matrix; The principal component analysis is performed on the selected feature pairs, the feature covariance matrix is ​​calculated, the eigenvalues ​​and eigenvectors are solved, and the principal components with a cumulative contribution rate of 95% are selected to obtain the feature data after dimensionality reduction; based on the feature data after dimensionality reduction, a sensor topology structure with internal and external measurement points separated is constructed, and the features are divided into internal features and external features, and anomaly detection is performed separately; the local anomaly factor algorithm is applied to the internal features and the external features, respectively, and the local reachable density and local anomaly factor of each data point are calculated to obtain the internal anomaly score and the external anomaly score; the internal anomaly score and the external anomaly score are weightedly fused to obtain a comprehensive anomaly score, and the Otsu threshold method is used to determine the optimal anomaly threshold, and the anomaly pattern classification is performed based on the comprehensive anomaly score and the optimal anomaly threshold to obtain the anomaly pattern recognition result.

[0055] In this example, the parameters such as battery core temperature, surface temperature, state of charge and polarization voltage are segmented. The state data is segmented using a sliding window with a length of 600 seconds and a step size of 300 seconds. The length of the sliding window of 600 seconds means that each analysis covers 10 minutes of data, and the step size of 300 seconds means that each sliding window moves forward for 5 minutes, ensuring that the data is fully covered and overlapped at different time points to capture the dynamic changes of the battery state. Through the sliding window segmentation, multiple overlapping multidimensional time series data sets are obtained. Feature extraction is performed on the multidimensional time series data set, and the statistical characteristics of the mean, standard deviation, kurtosis, and skewness in each time window are calculated to describe the central tendency, degree of dispersion, and distribution form of these parameters in this time period. At the same time, the time domain characteristics such as the maximum rate of increase and the maximum rate of decrease of each parameter are calculated to reflect the drastic changes of the state parameters. These features jointly construct a feature vector containing 48-dimensional features, of which the mean, standard deviation, kurtosis, and skewness each account for four dimensions, and the features of the maximum rate of increase and the maximum rate of decrease account for two dimensions for each parameter. By extracting features, the changing pattern of the battery status is described in different time windows. The 48-dimensional feature vector is subjected to correlation analysis, and the Pearson correlation coefficient between features is calculated to measure the linear correlation between each pair of features. The formula of the Pearson correlation coefficient is:

[0056] ;

[0057] in, Represents two features and The Pearson correlation coefficient between and Respectively represent characteristics and data points, and Characteristics and If The absolute value of is greater than 0.8, indicating that there is a strong linear correlation between the two features. Therefore, consider retaining only one of the features, or performing dimensionality reduction through principal component analysis. Perform principal component analysis on the selected feature pairs to reduce the dimension of the features and reduce redundancy. Through principal component analysis, calculate the feature covariance matrix. The elements of the covariance matrix represent the covariance between different features, thereby describing the correlation between each feature. The calculation formula of the covariance matrix is:

[0058] ;

[0059] in, Features and The covariance of is the sample size, and is the eigenvalue of the corresponding sample. A set of principal components is obtained by decomposing the covariance matrix by eigenvalues ​​and eigenvectors. The principal components with a cumulative contribution rate of 95% are selected to achieve dimensionality reduction of the features and obtain the reduced-dimensional feature data. Based on the reduced-dimensional feature data, a sensor topology with separated internal and external measurement points is constructed. By analyzing the source of the features, the reduced-dimensional features are divided into internal features and external features. The internal features are mainly related to the internal state of the battery, such as core temperature and polarization voltage; the external features are related to the external environment or macroscopic state of the battery, such as surface temperature and SOC. Through division, anomaly detection is performed on internal features and external features respectively to improve the accuracy of detection. In order to detect anomalies, the local anomaly factor algorithm is applied to internal features and external features respectively. The local anomaly factor is a density-based anomaly detection algorithm that determines the degree of anomaly of a data point by calculating the local reachable density of each data point. The local anomaly factor determines whether the point is abnormal by comparing the density difference between the data point and other points in its neighborhood. The specific calculation formula is:

[0060] ;

[0061] in, For data points The local anomaly factor, For data points of nearest neighbors, For data points The local reachable density of is the number of neighbors. By calculating the LOF value of each data point, the anomaly scores of internal features and external features are obtained respectively. The internal anomaly score and the external anomaly score are weighted and fused to obtain a comprehensive anomaly score. The comprehensive anomaly score reflects the overall abnormality of the battery in its current state. By combining the abnormal conditions of internal and external features, the battery state is more comprehensively evaluated. The Otsu threshold method is used to determine the optimal anomaly threshold. The Otsu threshold method automatically selects the optimal threshold by maximizing the inter-class variance to optimally distinguish between normal and abnormal states. Based on the comprehensive anomaly score and the optimal anomaly threshold, the battery state is classified into abnormal patterns to obtain the final abnormal pattern recognition result.

[0062] In one example, the abnormal pattern recognition result is input into the heterogeneous graph gated recurrent unit, and the spatiotemporal features are extracted through the heterogeneous graph convolution and GRU structure to obtain the deep fault features, including: constructing the abnormal pattern recognition result into a heterogeneous graph structure, in which the nodes represent the battery parameters and the edges represent the relationship between the parameters, to obtain the initial heterogeneous graph, and performing a multi-head attention mechanism on the initial heterogeneous graph to calculate the attention weights between nodes of different types to obtain the attention-enhanced heterogeneous graph; inputting the attention-enhanced heterogeneous graph into the heterogeneous graph convolution layer, which contains 3 parallel graph convolution sublayers, which respectively process the node characteristics, the relationship between nodes of the same type and the relationship between nodes of different types, to obtain the local space Features; Perform skip connection processing on local spatial features, concatenate the original node features with the post-convolution features, and fuse them through the fully connected layer to obtain enhanced spatial features, and input the enhanced spatial features into the gated graph neural network layer, which contains an update gate and a reset gate. The node information is selectively updated through the gating mechanism to obtain a timing-aware node representation; A bidirectional GRU structure is applied to the timing-aware node representation, the forward GRU captures the forward timing dependency, and the reverse GRU captures the reverse timing dependency to obtain a bidirectional timing feature, and the bidirectional timing feature is used for global information interaction through the self-attention mechanism, and the importance weights of different time steps are calculated to obtain deep fault features.

[0063] In this example, the abnormal pattern recognition results of the battery are represented in the form of a graph structure. In this graph structure, each node represents a parameter of the battery, such as core temperature, surface temperature, state of charge, polarization voltage, etc., and the edge represents the relationship between these parameters, such as the coupling relationship between temperature and state of charge. The constructed initial heterogeneous graph structure is used to represent the interaction between different physical quantities in the battery system. The initial heterogeneous graph is processed by a multi-head attention mechanism to calculate the attention weights between nodes of different types. The multi-head attention mechanism captures the relationships of different dimensions between nodes by introducing multiple attention heads, so that each node can perform weighted aggregation of its information according to the importance of neighboring nodes. For each node , its attention weight is calculated by the following formula:

[0064] ;

[0065] in, Representation Node and nodes The attention weights between is a linear transformation matrix used to map node features. and The nodes are and nodes Features, is the learning parameter in the attention mechanism, || represents the concatenation operation of the vector, and LeakyReLU is an activation function that can introduce nonlinear factors. Through the multi-head attention mechanism, the weights between nodes of different types are obtained, thereby constructing an attention-enhanced heterogeneous graph. The attention-enhanced heterogeneous graph is input into the heterogeneous graph convolution layer for feature extraction. The heterogeneous graph convolution layer contains three parallel graph convolution sublayers, which are used to process the node's own features, the relationship between nodes of the same type, and the relationship between nodes of different types. The multi-layer structure enables each node to treat neighbor nodes of the same type and different types differently when aggregating neighbor information, and obtain more detailed features. For example, the core temperature node of the battery updates its own state by aggregating other temperature-related node information, while the nodes related to electrochemical characteristics (such as SOC) capture the interaction with temperature through convolution. In this way, richer local spatial features are obtained to describe the local interaction information of nodes in the graph structure. After obtaining the local spatial features, jump connection processing is performed. Jump connection is an effective deep neural network structure design method that can avoid the gradient disappearance problem in deep networks. The original node features are concatenated with the features obtained after convolution, and then fused through the fully connected layer to obtain enhanced spatial features. The enhanced spatial features are input into the gated graph neural network layer to process the node information. The gated graph neural network layer contains update gates and reset gates. Through these gating mechanisms, the node information is selectively updated and reset. For example, the update gate determines which information needs to be retained, while the reset gate determines which information is ignored. Through the selective update mechanism, the node representation is fine-tuned to obtain a time-aware node representation. The update representation of the gating mechanism is:

[0066] ;

[0067] in, Indicates Step Node The hidden state of To update the output of the gate, control the weighted ratio of new information to old information. is the candidate state of the node, and © represents element-by-element multiplication. Through updating, the dynamic change characteristics of the node are captured so that its representation can adapt to the change of the battery state. After obtaining the timing-aware node representation, the bidirectional GRU structure is applied to process the node representation. The bidirectional GRU structure consists of two parts: the forward GRU and the reverse GRU. The forward GRU is used to capture the positive timing dependency, while the reverse GRU is used to capture the reverse timing dependency. Through the processing of the bidirectional GRU, a more comprehensive timing feature is obtained, thereby ensuring that the dependencies between nodes can be fully considered in the time dimension. The forward GRU captures the evolution of the node state from the past time step, while the reverse GRU captures the evolution path in reverse from the future time step, thereby obtaining forward and reverse information in the time series. In order to enhance the information interaction between nodes, the output of the bidirectional GRU will interact with global information through the self-attention mechanism. The self-attention mechanism enables the model to dynamically focus on the most important time points for the current state by calculating the importance weights of different time steps. Attention weight The calculation is expressed as:

[0068] ;

[0069] in, is the time step The attention weight, is the query vector, is the time step The key vector of each time step is queried to obtain the importance score of each time point. The self-attention mechanism enables the model to adaptively select the historical state that is most important to the final output based on the current node representation, thereby calculating the final deep fault feature.

[0070] In one example, the charging safety capacity is calculated according to the deep fault characteristics, and the battery thermal management and charge state are controlled by multiple constraints to obtain a charging power allocation plan, including: converting the deep fault characteristics into a 64-dimensional vector, processing it through a three-layer fully connected neural network, the first layer uses 128 neurons, the second layer uses 64 neurons, and the third layer uses 32 neurons. Each layer is followed by a ReLU activation function and a Dropout layer with a probability of 0.5. The last layer uses a Sigmoid activation function to output a charging safety capacity coefficient between 0 and 1; using the charging safety capacity coefficient, combined with the battery rated capacity and the current charge state, the maximum allowable charging current and the maximum allowable charging power are calculated to construct a charging safety capacity constraint condition; based on the thermal-electric coupling model, a battery temperature dynamic equation is established, discretized by the finite difference method, a temperature prediction model is obtained, and the maximum temperature rise rate, the maximum temperature limit and the temperature uniformity requirements are set to construct a temperature constraint condition; establishing a charging The dynamic equation of the electric state is combined with the voltage dynamic equation to set the maximum charging voltage and the upper limit of the state of charge to construct the state of charge constraint. The charging safety capacity constraint, temperature constraint and state of charge constraint are combined to construct a multi-objective optimization problem. The first objective function includes minimizing the charging time and maximizing the battery life, and the decision variable is the charging current. The non-dominated sorting genetic algorithm is used to solve the multi-objective optimization problem. The population size is set to 100, the number of iterations is 200, the crossover probability is 0.9, and the mutation probability is 0.1. Individual selection is performed through congestion distance calculation to obtain a set of non-dominated solution sets. The fuzzy comprehensive evaluation method is applied to the non-dominated solution set, and the membership function is designed for each objective function. The weight vector is determined by the hierarchical analysis method, the comprehensive score is calculated, and the solution with the highest score is selected as the optimal charging strategy. Based on the optimal charging strategy, considering the total power limit of the charging station and the load balancing demand, a quadratic programming problem is constructed, and the interior point method is used to solve it to obtain the charging power allocation plan.

[0071] In this example, deep fault features are input to the first layer of the neural network for processing. Deep fault features contain key state information of the battery during the charging process, such as temperature, state of charge, and polarization features. After converting these feature data into 64-dimensional vectors, they are used as input to the fully connected neural network. In the first layer of the neural network, 128 neurons are used to linearly transform the input data. Each neuron is connected to all nodes of the input, so that higher-level features can be extracted. The ReLU activation function is applied in this layer to introduce nonlinearity so that the network can represent complex patterns and features. A Dropout layer with a probability of 0.5 is then added to reduce the risk of overfitting. Dropout makes the network more robust during training by randomly masking some neurons and avoiding dependence on certain nodes. In the second layer, the number of neurons is reduced to 64. Key features are extracted through this layer, and the ReLU activation function and Dropout layer are also used to ensure the nonlinear representation and generalization capabilities of the features. Then in the third layer, 32 neurons are used to compress and streamline the features while continuing to maintain the validity of the features. In the last layer, the Sigmoid activation function is used to map the network output to between 0 and 1, and a charging safety capacity coefficient is obtained, which reflects whether the current battery state is suitable for high-power charging and the acceptable charging degree. The form of the Sigmoid function is:

[0072] ;

[0073] in, represents the output of the previous layer of the network, The value range of is between 0 and 1. Through this function, the output value is effectively limited to a suitable range as the coefficient of charging safety capacity. Using the charging safety capacity coefficient, combined with the rated capacity and current state of charge of the battery, the maximum allowable charging current ( ) and the maximum allowable charging power ( ). The maximum allowable charging current is calculated as:

[0074] ;

[0075] in, is the charging safety capacity factor, is the rated capacity of the battery, is the current state of charge. This formula ensures that the charging current is dynamically adjusted according to the current state of charge and safety capacity of the battery, thereby avoiding the risk of overcharging. At the same time, the maximum allowable charging power is calculated based on the maximum allowable charging current:

[0076] ;

[0077] in, is the current battery terminal voltage. Through this formula, the power constraint condition during the charging process is obtained. During the charging process, considering the safety constraints of current and power, it is also necessary to ensure the control of battery temperature. Based on the thermal-electric coupling model, the dynamic equation of battery temperature is established, and the temperature prediction model is obtained by discretization through the finite difference method. The temperature dynamic equation describes the temperature rise of the battery during the charging process. The finite difference method is used to convert continuous differential equations into discrete forms for calculation in digital systems. After the temperature prediction model is established, the maximum temperature rise rate ( ), maximum temperature limit ( ) and temperature uniformity requirements, and build temperature constraints to ensure that the battery temperature is always kept within a safe range. At the same time, a dynamic equation for the state of charge is established, and by combining it with the voltage dynamic equation, the maximum charging voltage ( ) and the upper limit of state of charge ( ), construct the state of charge constraint. Ensure that the battery can reach the full charge state as quickly as possible during the charging process, and will not be damaged due to overcharging. Combine the charging safety capacity constraint, temperature constraint and state of charge constraint to construct a multi-objective optimization problem. In the multi-objective optimization problem, the first objective function includes minimizing the charging time and maximizing the battery life. Minimizing the charging time is achieved by maximizing the charging current, while maximizing the battery life requires the charging process to be as gentle as possible to reduce the stress and loss of the battery material. The decision variable is the charging current ( ), which needs to be optimized under multiple constraints to find the optimal charging solution. In order to solve the multi-objective optimization problem, a non-dominated sorting genetic algorithm is used. In the algorithm, the population size is set to 100, the number of iterations is 200, the crossover probability is 0.9, and the mutation probability is 0.1. Through the congestion distance calculation, the individuals in the population are selected to obtain a set of non-dominated solution sets, which represent the optimal solutions with different trade-offs between charging time and battery life. The fuzzy comprehensive evaluation method is applied to the non-dominated solution set, and the membership function is designed for each objective function. The weight vector is determined by the hierarchical analysis method, the comprehensive score is calculated, and the solution with the highest score is selected as the optimal charging strategy. Based on the optimal charging strategy, the power limit and load balancing requirements of the actual charging station are considered. A quadratic programming problem is constructed and solved by the interior point method, and finally the charging power allocation scheme is obtained. The interior point method is an effective algorithm for solving constrained optimization problems, which can find the optimal solution globally. Through the quadratic programming solution, it is ensured that the charging power allocation of all devices in the charging station not only meets the total power limit, but also achieves load balancing between different charging devices, thereby improving the efficiency of the entire charging system.

[0078] In one example, the charging power allocation plan is input into a two-stage model predictive controller for distributed power redistribution to obtain a charging control instruction, including: converting the charging power allocation plan into an initial charging power sequence, constructing a charging station-level model predictive control problem, and the second objective function is to minimize the total power deviation and power fluctuation of the charging station. The constraints include the total power upper limit of the charging station and the power change rate limit; using the augmented Lagrangian method to solve the charging station-level model predictive control problem to obtain an optimized charging station total power sequence; based on the optimized charging station total power sequence, constructing a distributed consensus algorithm, designing a communication topology matrix, enabling adjacent charging devices to exchange local information and update local power allocation values; for each charging device, based on the updated local The power allocation value is calculated, and a charging device-level model predictive control problem is constructed. The third objective function is to minimize the power tracking error and charging time. The constraints include battery thermal management constraints and charge state constraints. The fast gradient method is used to solve the charging device-level model predictive control problem to obtain an optimized charging current sequence, and the optimized charging current sequence is used to predict the battery status in the next N time steps, including voltage, temperature and charge state. Based on the predicted battery status, the safety margin index is calculated, and the charging current is corrected according to the safety margin index to obtain a corrected charging current sequence. The first time step of the corrected charging current sequence is used as the charging control instruction of the current control cycle, and the charging control instruction is issued to the corresponding charging device.

[0079] In this example, the charging power allocation scheme is converted into an initial charging power sequence. The initial charging power sequence represents the charging power allocated by the charging station in each time period. When constructing the charging station-level model predictive control problem, the total power of the charging station is kept stable through optimization control, and the power deviation and power fluctuation of the charging station are minimized. In order to achieve this goal, a second objective function is defined to minimize the total power deviation and power fluctuation of the charging station. The objective function is expressed in the form of:

[0080] ;

[0081] in, is the objective function, which represents the minimization of power deviation and fluctuation over the entire time range; Indicates that at time step The actual total power of the charging station, is the reference power, usually the average power of the charging station target; is the time step The rate of change of power compared to the previous time step; is a weight coefficient used to balance the weight between power deviation and fluctuation. Constraints include the total power limit of the charging station ( ) and the power change rate are limited to ensure that the entire charging process operates within a controllable range and avoid overload and severe power fluctuations. In order to solve the charging station-level model predictive control problem, the augmented Lagrangian method is used to deal with the constraint problem by introducing Lagrangian multipliers and penalty terms. The goal of the augmented Lagrangian method is to combine the original objective function with the constraints and approximate the optimal solution by continuously updating the Lagrangian multipliers. The form of the augmented Lagrangian function is:

[0082] ;

[0083] in, is the augmented Lagrangian function, is the objective function, As constraints, is the Lagrange multiplier, is the penalty parameter, by adjusting and Approximate the optimal solution of the problem. Solve this optimization problem by augmented Lagrangian method and obtain the optimized total power sequence of charging stations. Based on the optimized total power sequence of charging stations, construct a distributed consensus algorithm so that the power can be reasonably distributed among various charging devices. Design the communication topology matrix so that adjacent charging devices can exchange local information and reach a global consensus through multiple iterations. Communication topology matrix Describes the information exchange relationship between various devices. The elements of the matrix Indicates the device and equipment If there is direct communication between the devices, then ,otherwise . In each iteration, the charging device updates its local power allocation value until the power allocation values ​​of all devices converge and reach a consensus. For each charging device, a charging device-level model predictive control problem is constructed based on the updated local power allocation value. In the charging device-level model predictive control, the third objective function is to minimize the power tracking error and charging time. The minimization of the power tracking error is to make the actual charging power of the device follow the target power as much as possible, while the minimization of the charging time is to complete the charging process as quickly as possible. The constraints include the thermal management constraints and the state of charge constraints of the battery to ensure that charging is carried out within a safe range. In order to solve the charging device-level model predictive control problem, the fast gradient method is used, which is an efficient algorithm suitable for convex optimization problems and can quickly find the optimal solution. The problem is solved by the fast gradient method to obtain the optimized charging current sequence of each charging device. Based on the optimized charging current sequence, the battery state in the next N time steps is predicted, including the battery voltage, temperature and state of charge. The predicted battery state helps to evaluate the safety hazards in the future charging process. Based on the predicted battery state, the safety margin index is calculated. The safety margin index is used to evaluate the safety of the battery under the current charging conditions, and is usually related to the battery's temperature, state of charge, and voltage parameters. If the safety margin is low, the charging current is corrected accordingly to avoid safety problems. Therefore, the charging current is corrected according to the safety margin index to obtain a corrected charging current sequence. The first time step in the corrected charging current sequence is used as the charging control instruction for the current control cycle, and this charging control instruction is sent to the corresponding charging device to execute the charging process.

[0084] Reference Figure 2 This embodiment provides a battery fast charging safety control device, including:

[0085] A decoupling module 1 is used to model the equivalent circuit and thermal equivalent circuit of the battery to obtain a virtual circuit model, and decouple the thermal state and electrochemical state of the virtual circuit model to obtain a dynamic decoupling matrix;

[0086] Separation module 2, used for inputting charging current and voltage sampling data into a virtual circuit model, performing state separation using a dynamic decoupling matrix, and obtaining battery safety state data through parallel calculation;

[0087] Identification module 3, used for performing abnormal pattern recognition based on battery safety status data to obtain abnormal pattern recognition results;

[0088] Extraction module 4, used to input the abnormal pattern recognition result into the heterogeneous graph gate recurrent unit, extract the spatiotemporal features through the heterogeneous graph convolution and GRU structure, and obtain the deep fault features;

[0089] Control module 5, used to calculate the charging safety capacity according to the deep fault characteristics, and perform multiple constraint control on the battery thermal management and charge state to obtain a charging power allocation plan;

[0090] The allocation module 6 is used to input the charging power allocation plan into the two-stage model predictive controller to perform distributed power redistribution and obtain charging control instructions.

[0091] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0092] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer 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 the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0093] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0094] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0095] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the 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 provided by the present invention and used in the embodiments 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. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0096] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0097] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A battery fast charging safety control method, characterized in that: The following steps are involved: Modeling an equivalent circuit and a thermal equivalent circuit of the battery to obtain a virtual circuit model, and decoupling the thermal state and the electrochemical state of the virtual circuit model to obtain a dynamic decoupling matrix; Inputting charging current and voltage sampling data into the virtual circuit model, performing state separation using the dynamic decoupling matrix, and obtaining battery safety state data through parallel calculation; The abnormal pattern recognition is performed based on the battery safety status data to obtain the abnormal pattern recognition result; specifically comprising: performing time series segmentation on the battery safety status data, using a sliding window with a length of 600 seconds and a step size of 300 seconds to segment the battery core temperature, surface temperature, state of charge and polarization voltage to obtain a multidimensional time series data set; performing feature extraction on the multidimensional time series data set, calculating the statistical characteristics of the mean, standard deviation, kurtosis, and skewness in each time window, and the time domain characteristics of the maximum rising rate and the maximum falling rate, and constructing a 48-dimensional feature vector; performing correlation analysis on the 48-dimensional feature vector, calculating the Pearson correlation coefficient between the features, constructing a correlation coefficient matrix, and selecting feature pairs with absolute values ​​of correlation coefficients greater than 0.8 based on the correlation coefficient matrix; for the selected The principal component analysis is performed on the feature pairs, the feature covariance matrix is ​​calculated, the eigenvalues ​​and eigenvectors are solved, and the principal components with a cumulative contribution rate of 95% are selected to obtain the feature data after dimensionality reduction; based on the feature data after dimensionality reduction, a sensor topology structure with internal and external measurement points separated is constructed, and the features are divided into internal features and external features, and anomaly detection is performed separately; the local anomaly factor algorithm is applied to the internal features and the external features, respectively, and the local reachable density and local anomaly factor of each data point are calculated to obtain the internal anomaly score and the external anomaly score; the internal anomaly score and the external anomaly score are weightedly fused to obtain a comprehensive anomaly score, and the optimal anomaly threshold is determined by the Otsu threshold method, and the anomaly pattern classification is performed based on the comprehensive anomaly score and the optimal anomaly threshold to obtain the anomaly pattern recognition result; The abnormal pattern recognition result is input into the heterogeneous graph gated recurrent unit, and the spatiotemporal features are extracted through the heterogeneous graph convolution and GRU structure to obtain the deep fault features; Calculate the safe charging capacity according to the deep fault characteristics, and perform multiple constraint controls on battery thermal management and charge state to obtain a charging power allocation plan; The charging power allocation scheme is input into a two-stage model predictive controller for distributed power redistribution to obtain a charging control instruction.

2. The battery fast charging safety control method according to claim 1, characterized in that: The equivalent circuit and thermal equivalent circuit of the battery are modeled to obtain a virtual circuit model, and the thermal state and electrochemical state of the virtual circuit model are decoupled to obtain a dynamic decoupling matrix, including: The open circuit voltage, internal resistance and polarization characteristics of the battery are parameterized. By measuring the voltage response curves under different states of charge, the open circuit voltage function, internal resistance function and polarization impedance function are fitted using the least squares method to construct an electrochemical equivalent circuit model. The thermal capacity and thermal resistance of the battery are parameterized. Through constant current charge and discharge tests and temperature response curves, the specific heat capacity of the battery and the thermal resistance of each layer structure are calculated using the thermal balance equation and finite element analysis, and a thermal equivalent circuit model is constructed. The electrochemical equivalent circuit model and the thermal equivalent circuit model are connected in series, and a coupling relationship among current, voltage and temperature is established to obtain a virtual circuit model, wherein the Joule heat of the electrochemical equivalent circuit model is used as an input of the thermal equivalent circuit model, and the temperature reacts on the parameters of the electrochemical equivalent circuit model; The virtual circuit model is represented in state space, the state of charge, polarization voltage, core temperature and surface temperature are selected as state variables, a state equation and an output equation are constructed, and a state vector including a thermal state and an electrochemical state is obtained; An orthogonal transformation matrix is ​​constructed to linearly transform the state vector, decompose the state space into a thermal subspace and an electrochemical subspace, and introduce a time-varying parameter according to the characteristic changes in the charging stage to construct an adaptive orthogonal transformation matrix; The thermal subspace and the electrochemical subspace are projected using the adaptive orthogonal transformation matrix to obtain a decoupled thermal state and an electrochemical state, and the decoupled thermal state and the electrochemical state are combined to construct a dynamic decoupling matrix.

3. The battery fast charging safety control method according to claim 2, characterized in that: The step of inputting the charging current and voltage sampling data into the virtual circuit model, performing state separation using the dynamic decoupling matrix, and obtaining the battery safety state data through parallel calculation includes: The charging current and voltage are sampled at high frequency, and the sampling frequency is set to 10kHz. A sliding window with a length of 100 data points is used for median filtering to remove outliers and high-frequency noise. Then, a Savitzky-Golay filter is applied for smoothing, with a filter order of 3 and a window length of 51 to obtain the current and voltage data sequence after noise reduction. Inputting the noise-reduced current and voltage data sequence into the virtual circuit model, performing numerical integration using the fourth-order Runge-Kutta method to obtain an initial state estimate, and applying the dynamic decoupling matrix to the initial state estimate to perform orthogonal transformation to obtain a decoupled thermal state component and a decoupled electrochemical state component; Based on the decoupled thermal state components, a thermal state observation equation is constructed, and the state is updated using an unscented Kalman filter algorithm. The core temperature and surface temperature of the battery are obtained by iterative calculation through two steps of time update and measurement update. Based on the decoupled electrochemical state components, an electrochemical state observation equation is constructed, and the state is updated using an extended Kalman filter algorithm. The Jacobian matrix is ​​obtained through linearization processing, and the prediction error covariance matrix and the Kalman gain are iteratively updated over time to obtain the battery state of charge and polarization voltage; Combining the battery core temperature, the surface temperature, the state of charge and the polarization voltage to obtain a battery state vector; Based on the battery state vector, a thermal runaway risk index is calculated using a thermal runaway probability model, a battery health state index is calculated using a Coulomb efficiency model, and the thermal runaway risk index and the battery health state index are weightedly fused to obtain comprehensive battery safety state data.

4. The battery fast charging safety control method according to claim 1, characterized in that: The abnormal pattern recognition result is input into the heterogeneous graph gate recurrent unit, and the spatiotemporal features are extracted through heterogeneous graph convolution and GRU structure to obtain deep fault features, including: The abnormal pattern recognition result is constructed into a heterogeneous graph structure, in which nodes represent battery parameters and edges represent relationships between parameters, to obtain an initial heterogeneous graph, and the initial heterogeneous graph is processed by a multi-head attention mechanism, and attention weights between nodes of different types are calculated to obtain an attention-enhanced heterogeneous graph; Inputting the attention-enhanced heterogeneous graph into a heterogeneous graph convolution layer, which includes three parallel graph convolution sublayers, respectively processing node features, relationships between nodes of the same type, and relationships between nodes of different types, to obtain local spatial features; The local spatial features are processed by skip connection, the original node features are concatenated with the post-convolution features, and the enhanced spatial features are obtained by fusing them through a fully connected layer. The enhanced spatial features are input into a gated graph neural network layer, which includes an update gate and a reset gate. The node information is selectively updated through a gating mechanism to obtain a time-aware node representation. A bidirectional GRU structure is applied to the timing-aware node representation. The forward GRU captures the positive timing dependency, and the reverse GRU captures the reverse timing dependency to obtain a bidirectional timing feature. The bidirectional timing feature is subjected to global information exchange through a self-attention mechanism, and the importance weights of different time steps are calculated to obtain a deep fault feature.

5. The battery fast charging safety control method according to claim 4, characterized in that: The method of calculating the charging safety capacity according to the deep fault characteristics and performing multiple constraint control on the battery thermal management and the state of charge to obtain a charging power allocation scheme includes: The deep fault features are converted into 64-dimensional vectors and processed through a three-layer fully connected neural network. The first layer uses 128 neurons, the second layer uses 64 neurons, and the third layer uses 32 neurons. Each layer is followed by a ReLU activation function and a Dropout layer with a probability of 0.

5. The last layer uses a Sigmoid activation function to output a charging safety capacity coefficient between 0 and 1; The maximum allowable charging current and the maximum allowable charging power are calculated by using the charging safety capacity coefficient in combination with the rated capacity and the current state of charge of the battery, and a charging safety capacity constraint condition is constructed; Based on the thermal-electric coupling model, the battery temperature dynamic equation is established, and the temperature prediction model is obtained by discretizing it through the finite difference method. The maximum temperature rise rate, maximum temperature limit and temperature uniformity requirements are set to construct temperature constraints. Establish the state of charge dynamic equation, combine it with the voltage dynamic equation, set the maximum charging voltage and the upper limit of the state of charge, and construct the state of charge constraint condition; The charging safety capacity constraint, temperature constraint and state of charge constraint are combined to construct a multi-objective optimization problem. The first objective function includes minimizing the charging time and maximizing the battery life, and the decision variable is the charging current. The non-dominated sorting genetic algorithm is used to solve the multi-objective optimization problem. The population size is set to 100, the number of iterations is 200, the crossover probability is 0.9, the mutation probability is 0.1, and the individual selection is performed through the crowding distance calculation to obtain a set of non-dominated solutions. Apply fuzzy comprehensive evaluation method to the non-dominated solution set, design membership function for each objective function, determine weight vector through analytic hierarchy process, calculate comprehensive score, and select the solution with the highest score as the optimal charging strategy; Based on the optimal charging strategy, considering the total power limit of the charging station and the load balancing requirement, a quadratic programming problem is constructed and solved using the interior point method to obtain the charging power allocation solution.

6. The battery fast charging safety control method according to claim 5, characterized in that: The step of inputting the charging power allocation scheme into a two-stage model predictive controller for distributed power redistribution to obtain a charging control instruction includes: The charging power allocation scheme is converted into an initial charging power sequence, and a charging station-level model predictive control problem is constructed. The second objective function is to minimize the total power deviation and power fluctuation of the charging station. The constraints include the total power upper limit and the power change rate limit of the charging station. The augmented Lagrangian method is used to solve the charging station level model predictive control problem to obtain an optimized total power sequence of the charging station; Based on the optimized total power sequence of the charging stations, a distributed consensus algorithm is constructed and a communication topology matrix is ​​designed to enable adjacent charging devices to exchange local information and update local power allocation values; For each charging device, a charging device-level model predictive control problem is constructed based on the updated local power allocation value. The third objective function is to minimize the power tracking error and charging time. The constraints include battery thermal management constraints and state of charge constraints. The fast gradient method is used to solve the charging device-level model predictive control problem to obtain an optimized charging current sequence, and the optimized charging current sequence is used to predict the battery state in the next N time steps, including voltage, temperature and state of charge; Based on the predicted battery state, a safety margin index is calculated, and the charging current is corrected according to the safety margin index to obtain a corrected charging current sequence; The first time step of the corrected charging current sequence is used as the charging control instruction of the current control cycle, and the charging control instruction is sent to the corresponding charging device.

7. A battery fast charging safety control device, characterized in that: For implementing the steps of the method according to any one of claims 1 to 6, the device comprises: A decoupling module is used to model the equivalent circuit and thermal equivalent circuit of the battery to obtain a virtual circuit model, and decouple the thermal state and electrochemical state of the virtual circuit model to obtain a dynamic decoupling matrix; A separation module, used to input charging current and voltage sampling data into the virtual circuit model, perform state separation using the dynamic decoupling matrix, and obtain battery safety state data through parallel calculation; an identification module, configured to perform abnormal pattern identification based on the battery safety status data to obtain an abnormal pattern identification result; An extraction module, used for inputting the abnormal pattern recognition result into a heterogeneous graph gated recurrent unit, extracting spatiotemporal features through heterogeneous graph convolution and GRU structure, and obtaining deep fault features; A control module, used to calculate the charging safety capacity according to the deep fault characteristics, and perform multiple constraint control on battery thermal management and charge state to obtain a charging power allocation plan; The allocation module is used to input the charging power allocation plan into the two-stage model predictive controller to perform distributed power redistribution and obtain a charging control instruction.

8. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. 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 6 are implemented.

Images