Lithium battery charging and discharging detection and management method, device and equipment and storage medium

Abstract

The invention relates to the technical field of lithium battery charging and discharging management, and discloses a lithium battery charging and discharging detection and management method, device and equipment and a storage medium, and the method comprises the steps: carrying out the data collection of the charging and discharging process of a lithium battery pack, applying consistent Gaussian noise, and obtaining an electrode-level expansion data set and a battery-level expansion data set; performing feature mapping and feature mixing by minimizing the maximum average difference loss function to obtain a mapping feature vector; estimating the state of charge and the state of health of the battery through a fusion model of a recurrent neural network and a physical information hybrid neural network to obtain a state of charge estimated value and a state of health estimated value; and constructing a battery pack electric heating model of the lithium battery pack, and generating a corresponding charging and discharging control instruction sequence. Self-adaptive adjustment can be carried out according to the real-time state of the battery, potential safety hazards such as over-charging and over-discharging and out-of-control temperature are effectively avoided, and the use safety of the battery pack is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery charge and discharge management, and in particular to a lithium battery charge and discharge detection and management method, device, equipment and storage medium. Background Art

[0002] The increasing size and complexity of lithium battery packs have put forward higher requirements for charge and discharge management. Traditional charge and discharge management methods mainly rely on simple voltage and current monitoring, which makes it difficult to accurately grasp the changes in the internal state of the battery, resulting in low charge and discharge efficiency and shortened battery life.

[0003] Existing data-driven methods often ignore the physical properties of batteries and have difficulty accurately capturing the dynamic behavior of batteries under different operating conditions. In addition, the acquisition cost of battery data is high and the cycle is long. How to effectively use limited data to build an accurate battery model is also an urgent problem to be solved. On the other hand, there are inconsistencies between the individual cells within the battery pack. How to achieve balanced charging and discharging of the battery pack while taking into account charging efficiency and safety is another challenge facing the battery management system. Most of the existing balancing strategies are based on simple voltage balancing, which makes it difficult to consider the influence of multi-dimensional factors such as temperature and health status, and easily lead to safety hazards such as overcharging, over-discharging, and temperature runaway. Summary of the invention

[0004] The present invention provides a lithium battery charge and discharge detection and management method, device, equipment and storage medium. The present invention can perform adaptive adjustment according to the real-time status of the battery, effectively avoiding safety hazards such as overcharging, overdischarging and temperature runaway, and significantly improving the safety of battery pack use.

[0005] In a first aspect, the present invention provides a lithium battery charge and discharge detection and management method, the lithium battery charge and discharge detection and management method comprising:

[0006] Collect data on the charge and discharge process of the lithium battery pack to obtain electrode-level charge and discharge data and battery-level charge and discharge data;

[0007] Applying consistent Gaussian noise to the same charge-discharge cycle sequence in the electrode-level charge-discharge data and the battery-level charge-discharge data by generating an adversarial network, respectively, to obtain an electrode-level expanded data set and a battery-level expanded data set;

[0008] Respectively extracting the electrode-level time-frequency domain features of the electrode-level extended data set and the battery-level time-frequency domain features of the battery-level extended data set, and performing feature mapping and feature mixing on the electrode-level time-frequency domain features and the battery-level time-frequency domain features by minimizing the maximum mean difference loss function to obtain a mapping feature vector;

[0009] Inputting the mapped feature vector into a fusion model of a recurrent neural network and a physical information hybrid neural network to estimate the battery state of charge and state of health, thereby obtaining an estimated value of the state of charge and an estimated value of the state of health;

[0010] A battery pack electrothermal model of the lithium battery pack is constructed based on the state of charge estimation value and the health state estimation value, and a charge and discharge current trajectory and a balance signal are calculated, and a corresponding charge and discharge control instruction sequence is generated according to the charge and discharge current trajectory and the balance signal.

[0011] In a second aspect, the present invention provides a lithium battery charge and discharge detection and management device, the lithium battery charge and discharge detection and management device comprising:

[0012] The data acquisition module is used to collect data on the charging and discharging process of the lithium battery pack to obtain electrode-level charging and discharging data and battery-level charging and discharging data;

[0013] A data expansion module, used to apply consistent Gaussian noise to the same charge and discharge cycle sequence in the electrode-level charge and discharge data and the battery-level charge and discharge data respectively through a generative adversarial network to obtain an electrode-level expanded data set and a battery-level expanded data set;

[0014] A feature mapping module, used to extract the electrode-level time-frequency domain features of the electrode-level extended data set and the battery-level time-frequency domain features of the battery-level extended data set, respectively, and perform feature mapping and feature mixing on the electrode-level time-frequency domain features and the battery-level time-frequency domain features by minimizing the maximum average difference loss function to obtain a mapping feature vector;

[0015] A state estimation module, used for inputting the mapping feature vector into a fusion model of a recurrent neural network and a physical information hybrid neural network to estimate the battery state of charge and state of health, and obtaining a state of charge estimation value and a state of health estimation value;

[0016] A generation module is used to construct a battery pack electrothermal model of the lithium battery pack based on the charge state estimation value and the health state estimation value, and calculate the charge and discharge current trajectory and the balance signal, and generate a corresponding charge and discharge control instruction sequence according to the charge and discharge current trajectory and the balance signal.

[0017] The third aspect of the present invention provides a computer device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the computer device executes the above-mentioned lithium battery charge and discharge detection and management method.

[0018] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned lithium battery charge and discharge detection and management method.

[0019] In the technical solution provided by the present invention, by combining electrode-level and battery-level data collection and using generative adversarial networks for data enhancement, the present invention significantly improves data utilization efficiency, solves the problem of difficulty in obtaining battery data, uses time-frequency domain feature extraction and maximum average difference loss function for feature mapping, effectively integrates electrode-level and battery-level information, enhances the model's perception of the internal state of the battery, and improves the accuracy of state estimation. Combining the recurrent neural network with the physical information hybrid neural network not only utilizes the powerful fitting ability of the deep learning model, but also incorporates the prior knowledge of the battery physical model, and realizes high-precision estimation of the battery state of charge and health state. Based on the battery pack electrothermal model and model predictive control algorithm, the present invention can simultaneously consider multiple objectives such as charging time, temperature balance and state of charge balance, realizes optimized charge and discharge management of the battery pack, improves charging efficiency, and prolongs battery life. By generating a dynamic sequence of charge and discharge control instructions, the method can be adaptively adjusted according to the real-time state of the battery, effectively avoiding safety hazards such as overcharging, overdischarging and temperature runaway, and significantly improving the safety of the battery pack. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0021] Figure 1 A schematic diagram of the steps of a lithium battery charge and discharge detection and management method according to an embodiment of the present invention;

[0022] Figure 2 It is a schematic diagram of the structure of a lithium battery charge and discharge detection and management device in an embodiment of the present invention;

[0023] Figure 3 It is a schematic block diagram of the structure of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] Embodiments of the present invention provide a lithium battery charge and discharge detection and management method, device, equipment and storage medium. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0025] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of the lithium battery charge and discharge detection and management method in the embodiment of the present invention includes:

[0026] Step S1, collecting data on the charge and discharge process of the lithium battery pack to obtain electrode-level charge and discharge data and battery-level charge and discharge data;

[0027] It is understandable that the execution subject of the present invention may be a lithium battery charge and discharge detection and management device, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0028] Specifically, reference electrodes are installed on the positive and negative electrodes of each single cell in the lithium battery pack to monitor the actual working state of the battery. The positive and negative electrode potentials are collected respectively by the high-precision voltage acquisition module, and these data are recorded at a sampling frequency of 1kHz to obtain the time series data of each electrode potential. At the same time, the total input and output current of the lithium battery pack is monitored in real time using a Hall current sensor, and the sampling frequency is also 1kHz to capture the current changes of the lithium battery pack during the charging and discharging process. At the same time, in order to monitor the terminal voltage of the battery pack, a high-precision voltage sensor is used for measurement with a sampling frequency of 1kHz to obtain detailed time series data of the battery pack current and voltage. Multiple temperature sensors are evenly arranged on the surface of each single cell, using thermocouples or thermistors, and the sampling frequency is set to 0.1Hz to ensure an appropriate temperature sampling frequency. Temperature monitoring helps to identify the thermal management of the battery during the charging and discharging process, and helps to predict temperature anomalies or thermal runaway risks. In terms of data processing, the charge state of the electrode level is calculated using the incremental capacity analysis method based on the electrode potential time series data. By analyzing the capacity changes of the electrode, the charge state of the electrode can be accurately evaluated. At the same time, the capacity of the electrode is calculated by the coulomb counting method, and the electrode-level internal resistance obtained by electrochemical impedance spectroscopy is combined to construct the electrode-level state parameter vector. For battery-level data processing, the extended Kalman filter algorithm is used to estimate the battery-level state of charge based on the collected time series data of battery pack current and voltage. Extended Kalman filtering is an estimation method for nonlinear systems, suitable for complex nonlinear systems such as lithium batteries. The battery-level capacity is obtained by constant current-constant voltage charge and discharge test, and the battery-level internal resistance obtained by DC internal resistance measurement is combined to construct a battery-level state parameter vector to reflect the electrochemical performance and state of the entire battery pack. In the feature extraction stage, time domain feature extraction is performed on the electrode-level and battery-level state parameter vectors to analyze the characteristics of the data in the time dimension. At the same time, the time domain data is fast Fourier transformed to obtain frequency domain features. The frequency domain features provide the frequency response information of the system and capture the dynamic change characteristics of the battery pack in terms of frequency. The time domain and frequency domain features are combined to construct the electrode-level time-frequency domain feature matrix and the battery-level time-frequency domain feature matrix. The electrode-level time-frequency domain feature matrix and the battery-level time-frequency domain feature matrix are fused with the battery surface temperature distribution time series data to obtain the electrode-level state feature vector and the battery-level state feature vector. The electrode-level state feature vector and the battery-level state feature vector are piecewise linearly interpolated to smooth the data and reduce the impact of noise. The data are normalized to zero mean to eliminate the difference in data magnitude, and the electrode-level charge and discharge data and the battery-level charge and discharge data are obtained.

[0029] Step S2, applying consistent Gaussian noise to the same charge and discharge cycle sequence in the electrode-level charge and discharge data and the battery-level charge and discharge data respectively through a generative adversarial network to obtain an electrode-level expanded data set and a battery-level expanded data set;

[0030] Specifically, the electrode-level charge and discharge data and the battery-level charge and discharge data are segmented into charge and discharge cycles to obtain the electrode-level charge and discharge cycle sequence and the battery-level charge and discharge cycle sequence. Based on the electrode-level and battery-level charge and discharge cycle sequences, the characteristic statistics of each charge and discharge cycle are calculated to obtain the electrode-level characteristic statistics and the battery-level characteristic statistics, which describe the overall characteristics of each charge and discharge cycle, such as average voltage, maximum current, energy efficiency, etc. The characteristic statistics reflect the electrochemical behavior in each charge and discharge cycle, describe the charge and discharge characteristics of the lithium battery from a statistical perspective, and help identify representative patterns and typical behaviors in the data. According to the electrode-level characteristic statistics and the battery-level characteristic statistics, the electrode-level charge and discharge cycle sequence and the battery-level charge and discharge cycle sequence are clustered and analyzed to obtain the electrode-level clustering results and the battery-level clustering results. Through cluster analysis, a large number of charge and discharge cycle sequences are classified according to their characteristics, and charge and discharge cycles with similar characteristics are identified, thereby reducing data redundancy and extracting representative charge and discharge cycles. Based on the results of cluster analysis, a representative cycle sequence is selected from the electrode-level and battery-level charge and discharge cycle sequences as a reference sequence. The benchmark sequences represent typical charge and discharge behaviors in different categories. In order to expand the data set, Gaussian noise is generated based on the benchmark sequences, and the generated noise intensity range is set to 0.1% to 1% of the standard deviation of the original signal. The noise generation method ensures that the applied disturbance intensity is moderate, which can not only expand the diversity of the data, but also does not cause too much deviation from the original signal. By generating Gaussian noise and superimposing it on the benchmark charge and discharge cycle sequences at the electrode level and the battery level, the noisy electrode level and battery level charge and discharge cycle sequences are obtained. Using the generative adversarial network, the noisy electrode level charge and discharge cycle sequence and the noisy battery level charge and discharge cycle sequence are processed respectively to generate new electrode level charge and discharge cycle sequences and new battery level charge and discharge cycle sequences. The generative adversarial network gradually approaches the distribution of the real data through the adversarial process between the training generator and the discriminator, generating data with higher authenticity and diversity. The generated new electrode level charge and discharge cycle sequence and battery level charge and discharge cycle sequence are merged with the original electrode level charge and discharge data and battery level charge and discharge data, respectively, to obtain the electrode level expanded data set and the battery level expanded data set.

[0031] The noisy electrode-level charge-discharge cycle sequence and the battery-level charge-discharge cycle sequence are respectively input into the generator in the generative adversarial network. The generator adopts a U-Net structure, which consists of an encoder and a decoder. The encoder part consists of 5 convolutional layers, each of which uses a 3x3 convolution kernel with a step size of 2, and uses a LeakyReLU activation function and batch normalization. The encoder is used to convert the input noisy sequence into a lower-dimensional feature representation. Through layer-by-layer convolution operations, the encoder captures the multi-level features of the input sequence and reduces the dimension of the data. The encoder output feature sequences of the electrode level and the battery level are generated in the encoder processing. The self-attention mechanism is used to process the encoder output feature sequences of the electrode level and the battery level respectively. The self-attention mechanism is a technology that can capture long-distance dependencies in data. The importance of each feature is determined by calculating the correlation between each position in the sequence and all other positions. In this way, the interdependence of each part of the sequence can be better understood and the quality of the generated data can be improved. By calculating the attention weight matrix, the attention weight matrix of the electrode level and the battery level is obtained to describe the correlation between each position. The electrode-level attention weight matrix is ​​multiplied with the noisy electrode-level charge-discharge cycle sequence to obtain the enhanced electrode-level feature sequence; similarly, the battery-level attention weight matrix is ​​multiplied with the noisy battery-level charge-discharge cycle sequence to obtain the enhanced battery-level feature sequence. By introducing the self-attention mechanism, the important features in the original signal are retained, and the representation ability of the features is significantly enhanced, so that the generator can more accurately restore the original charge-discharge features in the subsequent decoding process. The enhanced electrode-level feature sequence and battery-level feature sequence are respectively input into the decoder part of the generator. The decoder also contains 5 transposed convolution layers, each of which uses a 3x3 convolution kernel with a step size of 2, and uses the ReLU activation function and batch normalization. Through layer-by-layer upsampling and feature concatenation operations, the decoder restores the low-dimensional features to the same high-dimensional space as the input, while retaining and enhancing the feature information captured in the encoding stage. The upsampling and concatenation operations enable the decoder to make full use of the features at different levels in the encoder to generate a more refined charge-discharge cycle sequence, and obtain the initially generated electrode-level and battery-level charge-discharge cycle sequences. The preliminary electrode-level and battery-level charge-discharge cycle sequences generated by the generator are input into the discriminator for discrimination. The discriminator adopts the PatchGAN structure, which is suitable for discriminating local features. The discriminator contains 4 convolutional layers, each of which uses a 4x4 convolution kernel with a step size of 2, and adopts the LeakyReLU activation function and instance normalization. The last layer uses a 1x1 convolution kernel to output the electrode-level and battery-level discrimination results. The role of PatchGAN is to discriminate the generated sequences to evaluate the similarity between these sequences and the real sequences, and to improve the authenticity of the generated data by learning local features.The feedback from the discriminator is used to optimize the generator, prompting it to generate more realistic charge-discharge cycle sequences. Based on the electrode-level and battery-level discrimination results, the generator is back-propagated and gradient updated to continuously improve the quality of the generated sequence until the generated electrode-level and battery-level charge-discharge cycle sequences can be considered by the discriminator to be close to the real data. New electrode-level charge-discharge cycle sequences and new battery-level charge-discharge cycle sequences are output.

[0032] Step S3, respectively extracting the electrode-level time-frequency domain features of the electrode-level extended data set and the battery-level time-frequency domain features of the battery-level extended data set, and performing feature mapping and feature mixing on the electrode-level time-frequency domain features and the battery-level time-frequency domain features by minimizing the maximum average difference loss function to obtain a mapping feature vector;

[0033] Specifically, the electrode-level and battery-level augmented data sets were subjected to short-time Fourier transform, using the Hamming window function, and the window length was set to 256 with an overlap rate of 50%. Through short-time Fourier transform, the time series data was converted into a time-frequency spectrum, which effectively represented the frequency characteristics at different time points during the battery charging and discharging process, and could better capture the dynamic characteristics and local changes of the signal. The electrode-level and battery-level time-frequency spectrum were respectively input into a pre-trained convolutional neural network for feature extraction. The convolutional neural network contains 5 convolutional layers, each of which uses a 3x3 convolution kernel with a step size of 2, and uses the ReLU activation function and batch normalization to accelerate convergence and improve the stability of the model. Through the convolution operation, the local features in the data are effectively extracted, and the deep convolutional layers can gradually construct higher-level and more abstract features to describe the time-frequency domain characteristics of the battery under different charging and discharging conditions. After being processed by the convolutional neural network, the deep time-frequency features at the electrode level and the deep time-frequency features at the battery level are obtained, reflecting the complex changes of the battery in the frequency domain. The extracted electrode-level and battery-level deep time-frequency features are subjected to global average pooling and global maximum pooling operations respectively to reduce the dimension of the features while retaining important statistical information. Global average pooling captures the overall average trend of the features, while global maximum pooling can retain the maximum value information in the features. The results of global average pooling and global maximum pooling are concatenated to obtain the pooled feature vectors at the electrode level and battery level. The electrode-level pooled feature vector and the battery-level pooled feature vector are input into a pre-trained feature mapping network for processing. The feature mapping network contains 3 fully connected layers, each of which uses the LeakyReLU activation function and Dropout regularization. The LeakyReLU activation function can effectively alleviate the problem of "neuron death" in the neural network, so that the network can maintain certain gradient information when facing negative inputs, while the Dropout regularization randomly ignores some neurons during the training process, thereby reducing overfitting and improving the generalization ability of the model. After being processed by the feature mapping network, the electrode-level feature vector after preliminary mapping and the battery-level feature vector after preliminary mapping are obtained. Calculate the maximum mean difference (MMD) loss between the electrode-level feature vector after preliminary mapping and the battery-level feature vector after preliminary mapping. The maximum mean difference loss function is used to measure the difference between the two data distributions. It is calculated by using the weighted sum of the linear kernel function and the RBF kernel function, where the weight of the linear kernel function is 0.3 and the weight of the RBF kernel function is 0.7. The use of the RBF kernel function can better capture nonlinear relationships, while the linear kernel function is used to measure the linear correlation between features. Through weighted combination, the difference between feature vectors is comprehensively measured. The calculated MMD loss value is used to guide the optimization of the feature mapping network. By minimizing the MMD loss, the distribution between the electrode-level features and the battery-level features after feature mapping is ensured to be similar, and the alignment and consistency of the features are achieved.Based on the calculated MMD loss value, the output of the feature mapping network is fine-tuned to obtain an adjusted electrode-level feature vector and an adjusted battery-level feature vector. The adjusted electrode-level feature vector and the battery-level feature vector are concatenated to form a mapping feature vector.

[0034] Step S4, inputting the mapped feature vector into a fusion model of a recurrent neural network and a physical information hybrid neural network to estimate the battery state of charge and state of health, and obtaining an estimated value of the state of charge and an estimated value of the state of health;

[0035] Specifically, the mapping feature vector is segmented and segmented according to a sliding window with a time step of 50 to obtain a time series feature matrix. The feature dimension of each time step is 256. The segmentation can effectively capture the change of features over time and retain the time series information during the charging and discharging process. The time series feature matrix is ​​input into the recurrent neural network. The recurrent neural network adopts a three-layer long short-term memory network (LSTM) structure. Each layer of the LSTM network contains 128 neurons, using a tanh activation function and a dropout rate of 0.5. LSTM achieves selective transmission and forgetting of input information through its gating mechanism, so that LSTM can effectively capture the long-term and short-term dependencies during the charging and discharging process of lithium batteries and retain important time series features. Through the layer-by-layer processing of the three-layer LSTM, the feature sequence output by LSTM is obtained, which has a dimension of 50x128 and contains feature evolution information on the time step. In order to increase the focus on important features, the output feature sequence of LSTM is processed by the attention mechanism. The attention mechanism is a technology that can capture the importance of features at each time step in the sequence. By calculating the attention score, the attention weight vector with a dimension of 50 is obtained. The attention weight vector is weighted and summed with the LSTM output feature sequence to obtain a weighted LSTM feature vector with a dimension of 128. The attention mechanism enables the model to pay more attention to the important time step features for charge state estimation and health state assessment during the charging and discharging process, thereby improving the accuracy of state estimation. The weighted LSTM feature vector is input into the first fully connected layer, whose size is 64, and the ReLU activation function is used to obtain the output feature vector of the recurrent neural network. The ReLU activation function can introduce nonlinearity and enhance the model's ability to fit complex relationships. In the physical information processing part, based on the electrochemical impedance spectroscopy data, the Randles equivalent circuit model is fitted using the complex nonlinear least squares method to extract electrochemical characteristic parameters such as ohmic resistance, charge transfer resistance, and double-layer capacitance. Through these parameters, a physical model parameter vector is constructed. These parameters reflect the electrochemical behavior and electrical performance characteristics inside the battery. The physical model parameter vector is input into the second fully connected layer, whose size is 32, and the Sigmoid activation function is used to obtain the physical information feature vector. The Sigmoid activation function limits the output value between 0 and 1, so that the physical information feature vector is within a specific range, which is convenient for subsequent fusion with other features. The output feature vector of the recurrent neural network is concatenated with the physical information feature vector to form a fused feature vector with a dimension of 96. In order to optimize the fused features, adaptive feature selection is performed on the fused feature vector to obtain the optimized fused feature vector, and the dimension is reduced to 64. The optimized fused feature vector is input into the third fully connected layer with a size of 100, and the Softmax activation function is used to obtain the probability distribution of the state of charge.The Softmax activation function converts the output into a probability form, so that different states of charge can be classified. At the same time, the optimized fused feature vector is input into the fourth fully connected layer, the size of which is 1, and the Sigmoid activation function is used to obtain the health state value. The output of the Sigmoid activation function represents a continuous value of the health state, which is used to describe the health of the battery. Based on the probability distribution of the state of charge, the estimated value of the state of charge is obtained by calculating its expected value. This estimated value is used to describe the current remaining power of the battery. The health state value is directly used as the estimated value of the health state of the battery, reflecting the degree of battery degradation.

[0036] Step S5: construct a battery pack electrothermal model of the lithium battery pack based on the state of charge estimation value and the health state estimation value, calculate the charge and discharge current trajectory and the balance signal, and generate a corresponding charge and discharge control instruction sequence according to the charge and discharge current trajectory and the balance signal.

[0037] Specifically, an equivalent circuit model of a single cell is constructed based on the estimated value of the state of charge (SOC) and the estimated value of the state of health (SOH), and an electrical model of the single cell is obtained. The electrical model includes parameters such as ohmic resistance, charge transfer resistance and double-layer capacitance, which are used to describe the electrochemical characteristics of the battery. The electrical model of the single cell is thermally coupled, and the heat transfer process inside the battery is simulated by introducing thermal capacitance and thermal resistance parameters to establish an electrothermal coupling model of multiple single cells. Multiple single cell electrothermal coupling models are combined in series and parallel to construct an electrothermal model of a lithium battery pack. The electrothermal model of the battery pack comprehensively describes the electrical performance and thermal coupling characteristics of the entire lithium battery pack, so that the electrical interaction between the single cells in the battery pack and the distribution and transfer of heat are considered simultaneously during the charge and discharge management process. Based on the electrothermal model of the battery pack, the charge and discharge process of the entire battery pack is more finely controlled and managed to improve the safety and balance of the lithium battery pack. In order to achieve optimal management of the lithium battery pack, based on the constructed electrothermal model of the battery pack, a model predictive control algorithm is used to design an objective function J to optimize multiple key factors in the charging process. Specifically, the objective function J is defined as J = w1*tf+w2*Σ(SOCi-SOCavg)^2+w3*Σ(Ti-Tavg)^2, where tf represents the charging time, SOCi is the state of charge of the ith battery, SOCavg is the average state of charge of all batteries, Ti is the temperature of the ith battery, Tavg is the average temperature of the battery pack, and w1, w2, and w3 are the corresponding weight coefficients. Through the objective function, a trade-off and optimization are made between the charging time, SOC balance, and temperature balance, which not only ensures the charging speed, but also achieves the uniformity of the state of charge and temperature in the battery pack, and prevents safety problems caused by overcharging or overheating of individual batteries. The objective function is solved, and the sequential quadratic programming algorithm is used to solve it at each time step k to obtain the charging current trajectory and the balanced current distribution plan. Through the sequential quadratic programming algorithm, the objective function is efficiently solved, so that the charging strategy is adjusted in real time so that the battery pack always develops towards the optimal state during the charging process. Through this method, the current is dynamically distributed during the charging process, thereby achieving current balancing between different single cells and preventing imbalance problems caused by differences in the state of charge of individual cells. Based on the charging current trajectory and the balancing current distribution scheme, the charging current and bypass balancing current of each single cell are calculated to obtain the charging and discharging current trajectory and balancing signal of the entire battery pack. The charging current trajectory reflects the actual charging status of each single cell at different time steps, while the balancing signal describes the current distribution scheme used in the bypass circuit to balance the state of charge between different single cells. In order to convert the continuous charging and discharging current trajectory and balancing signal into executable charging and discharging control instructions, they are input into the preset controller mapping function for processing.The mapping function uses fuzzy control rules to convert continuous signals into discrete control instructions to obtain the final charge and discharge control instruction sequence. The control instructions are used to control the charging current of each single cell, the bypass balancing current, and the charge and discharge status of the entire battery pack to ensure the safety, balance, and efficiency of the battery pack during the charging process.

[0038] In the embodiment of the present invention, by combining electrode-level and battery-level data collection and using generative adversarial networks for data enhancement, the present invention significantly improves data utilization efficiency, solves the problem of difficulty in obtaining battery data, uses time-frequency domain feature extraction and maximum average difference loss function for feature mapping, effectively integrates electrode-level and battery-level information, enhances the model's perception of the internal state of the battery, and improves the accuracy of state estimation. Combining the recurrent neural network with the physical information hybrid neural network not only utilizes the powerful fitting ability of the deep learning model, but also incorporates the prior knowledge of the battery physical model, and realizes high-precision estimation of the battery state of charge and health state. Based on the battery pack electrothermal model and model predictive control algorithm, the present invention can simultaneously consider multiple objectives such as charging time, temperature balance and charge state balance, realizes optimized charge and discharge management of the battery pack, improves charging efficiency, and prolongs battery life. By generating a dynamic charge and discharge control instruction sequence, the method can perform adaptive adjustment according to the real-time state of the battery, effectively avoiding safety hazards such as overcharging and overdischarging and temperature runaway, and significantly improving the safety of the battery pack.

[0039] In a specific embodiment, the process of executing step S1 may specifically include the following steps:

[0040] Reference electrodes are installed on the positive and negative electrodes of each single cell in the lithium battery pack, and the positive and negative electrode potentials are collected by a high-precision voltage acquisition module with a sampling frequency of 1kHz to obtain electrode potential time series data;

[0041] The total input and output current of the lithium battery pack is monitored in real time using a Hall current sensor with a sampling frequency of 1kHz. At the same time, a high-precision voltage sensor is used to measure the battery pack terminal voltage with a sampling frequency of 1kHz to obtain the battery pack current and voltage time series data;

[0042] Multiple temperature sensors are evenly arranged on the surface of each single cell in the lithium battery pack, using thermocouples or thermistors with a sampling frequency of 0.1 Hz to obtain the time series data of the battery surface temperature distribution;

[0043] Based on the electrode potential time series data, the electrode-level charge state is calculated using the incremental capacity analysis method, the electrode-level capacity is calculated by the coulomb counting method, and the electrode-level internal resistance obtained by electrochemical impedance spectroscopy is combined to construct the electrode-level state parameter vector.

[0044] Based on the battery pack current and voltage time series data, the extended Kalman filter algorithm is used to estimate the battery-level state of charge. The battery-level capacity is obtained through constant current-constant voltage charge and discharge tests. Combined with the battery-level internal resistance obtained by DC internal resistance measurement, the battery-level state parameter vector is constructed.

[0045] Extract time domain features of the electrode-level state parameter vector and the battery-level state parameter vector, perform fast Fourier transform to obtain frequency domain features, and construct the electrode-level time-frequency domain feature matrix and the battery-level time-frequency domain feature matrix;

[0046] The electrode-level time-frequency domain feature matrix and the battery-level time-frequency domain feature matrix are respectively fused with the battery surface temperature distribution time series data to obtain the electrode-level state feature vector and the battery-level state feature vector;

[0047] The electrode-level state feature vector and the battery-level state feature vector are respectively subjected to piecewise linear interpolation and zero-mean normalization processing to obtain the electrode-level charge-discharge data and the battery-level charge-discharge data.

[0048] Specifically, in the lithium battery pack, reference electrodes are installed on the positive and negative electrodes of each single cell to obtain electrode potential information. The positive and negative electrode potentials are collected through a high-precision voltage acquisition module, and the sampling frequency is set to 1kHz to obtain high-time resolution electrode potential time series data. The use of reference electrodes enables the capture of the reaction potential between the electrode and the electrolyte during the charge and discharge process, which helps to more accurately monitor the dynamic behavior of the electrode. At the same time, the total input and output current of the lithium battery pack is monitored in real time using a Hall current sensor, and the sampling frequency is also 1kHz. The high-frequency acquisition can capture any sudden changes in the current during the charge and discharge process, ensuring that the dynamic response of the battery pack current can be accurately recorded. The terminal voltage of the battery pack is measured using a high-precision voltage sensor, and it is also recorded at a sampling frequency of 1kHz to obtain the time series data of the battery pack current and voltage. In order to evaluate the thermal state of the battery during the charge and discharge process, multiple temperature sensors are evenly arranged on the surface of each single cell of the lithium battery pack, and temperature is measured using thermocouples or thermistors. These sensors collect temperature distribution data on the battery surface at a sampling frequency of 0.1Hz to obtain battery surface temperature distribution time series data. Based on the electrode potential time series data, the incremental capacity analysis method is used to calculate the state of charge (SOC) at the electrode level. The incremental capacity analysis method estimates the SOC of the electrode by analyzing the curve of voltage versus capacity during the charge and discharge process, reflecting the utilization rate of the active material of the electrode. Specifically, the relationship between incremental capacity (ΔQ) and voltage (V) is expressed as:

[0049] ΔQ=C·ΔV;

[0050] Among them, ΔQ represents the incremental capacity, C is the total capacity of the battery, and ΔV is the voltage change. The SOC value of the electrode is obtained by calculating the incremental capacity in different charge and discharge segments. The coulomb counting method is used to calculate the capacity at the electrode level, and the capacity is estimated by accumulating the charge and discharge current, which can more accurately describe the charge storage of the electrode at different time points. Combined with the electrode level internal resistance R obtained by electrochemical impedance spectroscopy e ), construct the electrode-level state parameter vector, which contains parameters such as SOC, capacity and internal resistance to describe the electrochemical state of the electrode. For the processing of battery-level data, the extended Kalman filter algorithm is used to estimate the battery-level state of charge based on the collected battery pack current and voltage time series data. The extended Kalman filter is a state estimation algorithm suitable for nonlinear systems. By linearizing the nonlinear system, the SOC of the battery is estimated more accurately under real-time conditions. Specifically, the SOC estimation is expressed as:

[0051]

[0052] Where SOC(t) represents the state of charge at time t, I(t) is the current at time t, Δt is the sampling time interval, and C nom The battery-level capacity is obtained by constant current-constant voltage charge and discharge test, and the battery-level internal resistance (R dc ), construct a battery-level state parameter vector, which can describe the electrochemical characteristics of the battery at the macro level, including SOC, capacity and internal resistance. For the electrode-level state parameter vector and the battery-level state parameter vector, time domain feature extraction and frequency domain feature extraction are performed respectively. In the time domain, the statistical characteristics of the data, such as mean, variance, peak, etc., are extracted to describe the central tendency and distribution of the characteristics. At the same time, the time domain data is converted to the frequency domain using the fast Fourier transform, and the frequency domain features are extracted to reflect the energy distribution of the signal at different frequencies and capture the periodic and harmonic characteristics of the charging and discharging process. Through these two feature extraction methods, the electrode-level time-frequency domain feature matrix and the battery-level time-frequency domain feature matrix are constructed. The electrode-level and battery-level time-frequency domain feature matrices are fused with the battery surface temperature distribution time series data to obtain the electrode-level state feature vector and the battery-level state feature vector. The electrode-level and battery-level state feature vectors are processed by piecewise linear interpolation and zero-mean normalization respectively. The purpose of piecewise linear interpolation is to fill in missing data or smooth the data to make the feature vector more continuous and smooth, while zero-mean normalization is to adjust the mean of the feature vector to 0 and the standard deviation to 1, thereby eliminating the influence between different feature magnitudes and facilitating subsequent data analysis and modeling. After processing, the electrode-level charge and discharge data and the battery-level charge and discharge data are obtained.

[0053] In a specific embodiment, the process of executing step S2 may specifically include the following steps:

[0054] The electrode-level charge-discharge data and the battery-level charge-discharge data are respectively segmented into charge-discharge cycles to obtain an electrode-level charge-discharge cycle sequence and a battery-level charge-discharge cycle sequence;

[0055] Based on the electrode-level charge-discharge cycle sequence and the battery-level charge-discharge cycle sequence, characteristic statistics of each charge-discharge cycle are calculated to obtain electrode-level characteristic statistics and battery-level characteristic statistics;

[0056] According to the electrode-level characteristic statistics and the battery-level characteristic statistics, the electrode-level charge-discharge cycle sequence and the battery-level charge-discharge cycle sequence are clustered and analyzed to obtain the electrode-level clustering results and the battery-level clustering results;

[0057] Based on the electrode-level clustering results and the battery-level clustering results, representative electrode-level charge-discharge cycle sequences and battery-level charge-discharge cycle sequences are selected as reference sequences, and Gaussian noise is generated on the reference sequences. The noise intensity ranges from 0.1% to 1% of the standard deviation of the original signal to obtain Gaussian noise sequences.

[0058] The Gaussian noise sequence is superimposed on the electrode-level charge-discharge cycle sequence and the battery-level charge-discharge cycle sequence, respectively, to obtain the noisy electrode-level charge-discharge cycle sequence and the noisy battery-level charge-discharge cycle sequence;

[0059] Using a generative adversarial network, the noisy electrode-level charge-discharge cycle sequence and the noisy battery-level charge-discharge cycle sequence are processed respectively to generate a new electrode-level charge-discharge cycle sequence and a new battery-level charge-discharge cycle sequence;

[0060] The new electrode-level charge-discharge cycle sequence and the new battery-level charge-discharge cycle sequence are respectively merged with the electrode-level charge-discharge data and the battery-level charge-discharge data to obtain an electrode-level expanded data set and a battery-level expanded data set.

[0061] Specifically, the electrode-level charge and discharge data and the battery-level charge and discharge data are segmented to obtain the electrode-level charge and discharge cycle sequence and the battery-level charge and discharge cycle sequence. The charge and discharge cycle is segmented according to the periodic changes of parameters such as voltage and current to obtain an independent charge and discharge sequence for each battery, reflecting the electrochemical behavior of a single battery during the charge and discharge process. Characteristic statistics are calculated based on the electrode-level and battery-level charge and discharge cycle sequences to fully characterize the characteristics of each cycle. For example, the statistical characteristics such as the average voltage, maximum current, and capacity of each charge and discharge cycle are calculated. Suppose the data of each charge and discharge cycle is x i (t), where i represents the number of the charge and discharge cycle and t is the time variable, then the average voltage and maximum current Calculated by the following formula:

[0062]

[0063] Among them, μ Vi represents the average voltage of the i-th charge and discharge cycle, T is the duration of the charge and discharge cycle, V i (t) represents the voltage at time t; represents the maximum current in the ith charge-discharge cycle. Based on the electrode-level and battery-level feature statistics, cluster analysis is performed on the charge-discharge cycle sequence to identify different types and patterns of charge-discharge cycles. Use algorithms such as K-means clustering or DBSCAN to cluster the data. Let the set of feature statistics be F = {f1, f2, …, f N}, where f i is the characteristic vector of the i-th charge and discharge cycle. Through cluster analysis, the charge and discharge cycle sequence is divided into several clusters, each cluster represents a charge and discharge cycle with similar characteristics. For example, if the current fluctuation of some charge and discharge cycles is large and the temperature rises rapidly, they belong to the same category, and the clustering results help to identify and understand the performance change pattern of the battery. Based on the results of cluster analysis, representative charge and discharge cycle sequences in each cluster are selected as benchmark sequences. The benchmark sequence represents different types of charge and discharge behaviors and can provide a basis for data expansion. In order to expand the data set, Gaussian noise is generated based on the benchmark sequence, and the intensity of the noise ranges from 0.1% to 1% of the standard deviation of the original signal. By generating Gaussian noise and superimposing it on the baseline sequence, noisy electrode-level and battery-level charge and discharge cycle sequences are obtained. The formula for generating Gaussian noise is expressed as:

[0064]

[0065] Among them, y i (t) represents the sequence after adding noise, x i (t) is the original benchmark sequence, η(t) is a sequence with mean 0 and variance σ 2The Gaussian noise of σ is used, and σ is the standard deviation of the noise, ranging from 0.1% to 1% of the standard deviation of the original signal. This method can increase the diversity of data and improve the generalization ability of the model. The noisy electrode-level and battery-level charge-discharge cycle sequences are processed using a generative adversarial network to generate new electrode-level and battery-level charge-discharge cycle sequences. The generative adversarial network consists of two parts: a generator and a discriminator. The generator generates a new charge-discharge cycle sequence, and the discriminator determines whether the input data comes from real data or generated data. Through adversarial training, the generator can gradually generate a charge-discharge cycle sequence that is similar to the distribution of real data, expanding the original data set. The new electrode-level charge-discharge cycle sequence and the battery-level charge-discharge cycle sequence are merged with the original electrode-level charge-discharge data and the battery-level charge-discharge data, respectively, to obtain the expanded electrode-level and battery-level data sets.

[0066] In a specific embodiment, the execution step uses a generative adversarial network to process the noisy electrode-level charge-discharge cycle sequence and the noisy battery-level charge-discharge cycle sequence respectively, and the process of generating a new electrode-level charge-discharge cycle sequence and a new battery-level charge-discharge cycle sequence can specifically include the following steps:

[0067] The noisy electrode-level charge-discharge cycle sequence and the noisy battery-level charge-discharge cycle sequence are respectively input into the generator of the generative adversarial network to obtain the electrode-level encoder output feature sequence and the battery-level encoder output feature sequence; the generator adopts a U-Net structure, including an encoder and a decoder, wherein the encoder consists of 5 convolutional layers, each of which uses a 3x3 convolution kernel with a step size of 2, a LeakyReLU activation function and batch normalization, and the decoder consists of 5 transposed convolutional layers, each of which uses a 3x3 convolution kernel with a step size of 2, a ReLU activation function and batch normalization;

[0068] The electrode-level encoder output feature sequence and the battery-level encoder output feature sequence are processed by the self-attention mechanism respectively, and the correlation between each position and all other positions in the electrode-level encoder output feature sequence and the battery-level encoder output feature sequence is calculated respectively to obtain the electrode-level attention weight matrix and the battery-level attention weight matrix;

[0069] The electrode-level attention weight matrix is ​​multiplied by the noisy electrode-level charge-discharge cycle sequence to obtain an enhanced electrode-level feature sequence, and the battery-level attention weight matrix is ​​multiplied by the noisy battery-level charge-discharge cycle sequence to obtain an enhanced battery-level feature sequence;

[0070] The enhanced electrode-level feature sequence and the enhanced battery-level feature sequence are respectively input into the decoder of the generator, and a preliminary generated electrode-level charge-discharge cycle sequence and a preliminary generated battery-level charge-discharge cycle sequence are generated through layer-by-layer upsampling and splicing operations;

[0071] The initially generated electrode-level charge-discharge cycle sequence and the initially generated battery-level charge-discharge cycle sequence are input into the discriminator respectively. The discriminator adopts the PatchGAN structure, which contains 4 convolutional layers. Each convolutional layer uses a 4x4 convolution kernel with a step size of 2, a LeakyReLU activation function and instance normalization. The last layer uses a 1x1 convolution kernel to output the electrode-level discrimination results and the battery-level discrimination results.

[0072] According to the electrode level discrimination result and the battery level discrimination result, a new electrode level charge and discharge cycle sequence and a new battery level charge and discharge cycle sequence are output respectively.

[0073] Specifically, the noisy electrode-level charge-discharge cycle sequence and the noisy battery-level charge-discharge cycle sequence are respectively input into the generator of the generative adversarial network to obtain the electrode-level encoder output feature sequence and the battery-level encoder output feature sequence. The generator adopts the U-Net structure, which consists of an encoder and a decoder. The encoder consists of 5 convolutional layers. Each convolutional layer uses a 3x3 convolution kernel with a step size of 2, and uses LeakyReLU activation function and batch normalization to accelerate training and improve stability. The role of the encoder is to extract and compress the input sequence and map the data to a lower-dimensional feature space. The output of the i-th convolution layer of the encoder is expressed as follows:

[0074] f (i) (x) = BatchNorm(LeakyReLU(W (i) *x+b (i) ));

[0075] Among them, f (i) (x) represents the output of the i-th convolution layer, W (i) represents the convolution kernel, * represents the convolution operation, b (i) is the bias term, LeakyReLU(·) represents the Leaky ReLU activation function, and BatchNorm(·) represents the batch normalization operation. Through 5 convolutional layers, the encoder gradually compresses the input charge and discharge cycle sequence and extracts high-dimensional feature information to obtain the encoder output feature sequence at the electrode level and the battery level. The encoder output feature sequences at the electrode level and the battery level are processed by the self-attention mechanism respectively. The self-attention mechanism is a technology that can capture long-distance dependencies in a sequence. It determines the importance of each feature by calculating the correlation between each position in the feature sequence and all other positions. Given a feature sequence X = [x1, x2,…, x n ], the attention weight is expressed as:

[0076]

[0077] Among them, A ijrepresents the attention weight between position i and position j in the feature sequence, Q i and K j are the query vector and key vector of position i and j respectively, and · represents the vector inner product operation. By calculating the attention weight matrix A, the relative importance of each position to other positions is captured, and the relevant information in the sequence is more effectively enhanced. The electrode-level attention weight matrix is ​​multiplied with the noisy electrode-level charge-discharge cycle sequence to obtain the enhanced electrode-level feature sequence. Similarly, the battery-level attention weight matrix is ​​multiplied with the noisy battery-level charge-discharge cycle sequence to obtain the enhanced battery-level feature sequence. The introduction of the attention mechanism enables the generator to better focus on the important parts of the charge-discharge cycle, ignore less important noise and redundant information, and enhance the representation ability of the features. The enhanced electrode-level and battery-level feature sequences are respectively input into the decoder of the generator for upsampling. The decoder consists of 5 transposed convolutional layers, each of which uses a 3x3 convolution kernel with a stride of 2, and adopts the ReLU activation function and batch normalization. The role of the decoder is to map the compressed features back to the original high-dimensional space, and generate preliminary electrode-level and battery-level charge-discharge cycle sequences through layer-by-layer upsampling and splicing operations. Through the layer-by-layer processing of the decoder, a charge and discharge cycle sequence that is closer to the real one is obtained. The generated preliminary electrode-level and battery-level charge and discharge cycle sequences are respectively input into the discriminator for discrimination. The discriminator adopts the PatchGAN structure to determine the authenticity of the sequence by judging the local area of ​​the input. The PatchGAN structure contains 4 convolutional layers, each of which uses a 4x4 convolution kernel with a step size of 2, and uses the LeakyReLU activation function and instance normalization to stabilize the training process. The last layer uses a 1x1 convolution kernel to output the electrode-level and battery-level discrimination results. The role of PatchGAN is to evaluate whether the sequence generated by the generator is similar to the real data, and to improve the quality and authenticity of the generated data by learning the local area. The output of the discriminator is expressed as:

[0078] D(x)=Sigmoid(W (4) *f (3) (x)+b (4) );

[0079] Among them, D(x) represents the discrimination result of the input sequence x, W (4) and b (4) are the weights and biases of the last layer, f (3)(x) is the feature after the first three layers of convolution processing. Sigmoid (·) is used to map the output to between 0 and 1 to represent the authenticity probability of the generated sequence. Through the feedback of the discriminator, the generator continuously optimizes itself so that the generated data is closer to the real charge and discharge cycle sequence. According to the discriminator's judgment results on the electrode-level and battery-level generated sequences, the generator continuously improves the quality of the generated sequence through back propagation and gradient update until the generated electrode-level and battery-level charge and discharge cycle sequences can pass the judgment of the discriminator and are considered to be close to the real data. Finally, a new electrode-level charge and discharge cycle sequence and a new battery-level charge and discharge cycle sequence are obtained.

[0080] In a specific embodiment, the process of executing step S3 may specifically include the following steps:

[0081] The electrode-level expanded data set and the battery-level expanded data set were respectively subjected to short-time Fourier transform, using a Hamming window function with a window length of 256 and an overlap rate of 50%, to obtain an electrode-level time-frequency spectrum and a battery-level time-frequency spectrum;

[0082] The electrode-level time-frequency spectrogram and the battery-level time-frequency spectrogram are respectively input into the pre-trained convolutional neural network to extract the electrode-level deep time-frequency features and the battery-level deep time-frequency features; the pre-trained convolutional neural network contains 5 convolutional layers, each of which uses a 3x3 convolution kernel with a step size of 2, and uses the ReLU activation function and batch normalization;

[0083] Perform global average pooling and global maximum pooling operations on the electrode-level deep time-frequency features and the battery-level deep time-frequency features, respectively, and concatenate the pooling results to obtain the electrode-level pooling feature vector and the battery-level pooling feature vector;

[0084] The electrode-level pooled feature vector and the battery-level pooled feature vector are input into the pre-trained feature mapping network to obtain the electrode-level feature vector and the battery-level feature vector after preliminary mapping; the feature mapping network contains 3 fully connected layers, each layer uses the LeakyReLU activation function and Dropout regularization;

[0085] Calculate the maximum average difference loss between the electrode-level feature vector after preliminary mapping and the battery-level feature vector after preliminary mapping, and the loss function includes the weighted sum of the linear kernel function and the RBF kernel function, where the weight of the linear kernel function is 0.3 and the weight of the RBF kernel function is 0.7;

[0086] Based on the calculated maximum average difference loss value, the output of the feature mapping network is fine-tuned to obtain an adjusted electrode-level feature vector and an adjusted battery-level feature vector, and the adjusted electrode-level feature vector and the adjusted battery-level feature vector are concatenated to obtain a mapping feature vector.

[0087] Specifically, the electrode-level expanded data set and the battery-level expanded data set are respectively subjected to short-time Fourier transform, and the Hamming window function is used to reduce spectrum leakage. The window length is set to 256, and the overlap rate is 50%, which effectively captures the characteristic changes of the signal in the time domain and frequency domain. The result of the short-time Fourier transform is a time-frequency spectrum, which can reflect the energy distribution of different frequency components of the signal at different time points, and can effectively capture the dynamic changes in the charging and discharging process. The short-time Fourier transform is expressed as:

[0088]

[0089] Among them, X(t,f) represents the time-frequency spectrum value at time t and frequency f, x(n) is the original signal, w(n) is the Hamming window function, N is the window length, and the time-frequency spectrum of the electrode level and the battery level is obtained by short-time Fourier transform. The time-frequency spectrum of the electrode level and the battery level are respectively input into a pre-trained convolutional neural network for feature extraction. The convolutional neural network contains 5 convolutional layers, each of which uses a 3x3 convolution kernel with a step size of 2. The output of the convolutional layer is processed by the ReLU activation function and batch normalized to accelerate training and improve model stability. The function of the convolution layer is to extract features from the time-frequency spectrum layer by layer. The low-level convolution kernel extracts local edges and basic shape features, while the high-level convolution kernel can capture more abstract and complex features. The output of the i-th convolution layer is expressed by the following formula:

[0090] f (i) (X) = BatchNorm(ReLU(W (i) *X+b (i) ));

[0091] Among them, f (i) (X) represents the output feature of the i-th convolution layer, W (i) is the convolution kernel, * indicates the convolution operation, b (i) is the bias term, ReLU(·) is the ReLU activation function, and BatchNorm(·) represents batch normalization. After 5 layers of convolution, the deep time-frequency features at the electrode level and battery level are obtained, which effectively reflect the complex dynamic behavior of lithium batteries during the charging and discharging process. In order to reduce the dimension and aggregate the deep time-frequency features, global average pooling and global maximum pooling are performed on the deep time-frequency features at the electrode level and battery level, respectively. Global average pooling calculates the average value of the feature on the entire feature map, reflecting the global trend of the feature. Global maximum pooling calculates the maximum value in the feature map, which can capture the most significant part of the feature. The results of global average pooling and global maximum pooling are concatenated to obtain a pooled feature vector containing global trends and significant features. Assuming the deep feature is F, the global average pooling and maximum pooling are expressed as:

[0092]

[0093] Among them, f avg is the result of global average pooling, f max is the result of global maximum pooling, F i is the i-th element in feature F, and N is the number of elements of the feature. The two results are concatenated to obtain the pooled feature vectors at the electrode level and the battery level. The pooled feature vectors at the electrode level and the battery level are input into the pre-trained feature mapping network. The feature mapping network consists of 3 fully connected layers, each of which uses the LeakyReLU activation function and Dropout regularization to prevent overfitting and enhance the nonlinear expression ability of the model. The purpose of the feature mapping network is to map the features at the electrode level and the battery level to a new feature space so that their feature distributions in this space are closer. The output of the feature mapping network is expressed as:

[0094]

[0095] in, represents the output of the i-th layer mapping network, W (i) and b (i) are the weight and bias of the i-th layer, LeakyReLU(·) is the LeakyReLU activation function, and f input is the input feature vector. After three fully connected layers, the electrode-level feature vector and battery-level feature vector are obtained after preliminary mapping. In order to ensure the distribution consistency of electrode-level and battery-level features in the feature space, the maximum mean difference (MMD) loss between them is calculated. MMD is a measure of the similarity between two distributions. The loss is calculated using the weighted sum of the linear kernel function and the radial basis function (RBF) kernel function. The specific loss function is expressed as:

[0096] MMD(X,Y)=0.3·k linear (X,Y)+0.7·k RBF (X,Y);

[0097] Where MMD(X,Y) represents the maximum mean difference loss between the electrode-level feature vector X and the cell-level feature vector Y, k linear (X,Y) is the result of linear kernel function calculation, k RBF(X,Y) is the result of RBF kernel function calculation, and 0.3 and 0.7 are the weights of each kernel function. By calculating the MMD loss, the similarity between the electrode-level and battery-level features in the feature space is quantified, thereby guiding the fine-tuning of the feature mapping network. Based on the calculated MMD loss value, the parameters of the feature mapping network are fine-tuned through back propagation and gradient descent to minimize the MMD loss, so that the electrode-level and battery-level features are closer in the mapped space, thereby achieving feature alignment. After fine-tuning, the adjusted electrode-level feature vector and battery-level feature vector are obtained. The adjusted electrode-level feature vector and battery-level feature vector are concatenated to obtain the final mapping feature vector, which contains the time-frequency information of the electrode level and the battery level, and the consistency of the two is ensured through feature mapping.

[0098] In a specific embodiment, the process of executing step S4 may specifically include the following steps:

[0099] The mapped feature vector is split into sliding windows with a time step length of 50 to obtain a time series feature matrix with a feature dimension of 256 at each time step;

[0100] The time series feature matrix is ​​input into the recurrent neural network. The recurrent neural network contains three layers of LSTM networks. Each layer of LSTM network contains 128 neurons. The tanh activation function and the dropout rate of 0.5 are used. The information is selectively transmitted and forgotten through the gating mechanism to obtain the LSTM output feature sequence with a dimension of 50x128.

[0101] Perform attention mechanism processing on the LSTM output feature sequence, calculate the attention score, obtain the attention weight vector with a dimension of 50, and perform weighted summation of the attention weight vector and the LSTM output feature sequence to obtain the weighted LSTM feature vector with a dimension of 128;

[0102] The weighted LSTM feature vector is input into the first fully connected layer with a layer size of 64 and a ReLU activation function to obtain the output feature vector of the recurrent neural network.

[0103] Based on the electrochemical impedance spectroscopy data, the Randles equivalent circuit model is fitted using the complex nonlinear least squares method to extract the ohmic resistance, charge transfer resistance and double-layer capacitance, and the physical model parameter vector is constructed. The physical model parameter vector is input into the second fully connected layer with a layer size of 32 and a Sigmoid activation function is used to obtain the physical information feature vector.

[0104] The output feature vector of the recurrent neural network is concatenated with the physical information feature vector to obtain a fused feature vector with a dimension of 96;

[0105] Adaptive feature selection is performed on the fused feature vector to obtain an optimized fused feature vector with a dimension reduced to 64. The optimized fused feature vector is input into the third fully connected layer with a layer size of 100 and a Softmax activation function is used to obtain the state of charge probability distribution. The optimized fused feature vector is input into the fourth fully connected layer with a layer size of 1 and a Sigmoid activation function is used to obtain the health state value.

[0106] The expected value is calculated based on the state of charge probability distribution as the state of charge estimation value, and the health state value is directly used as the health state estimation value.

[0107] Specifically, the mapped feature vector is segmented according to a sliding window with a time step length of 50 to obtain a time series feature matrix. The feature dimension of each time step is 256. This segmentation method can effectively capture the dynamic changes of lithium batteries during the charging and discharging process, retain important information of the data in the time dimension, and form a time series feature matrix containing 50 time steps. The time series feature matrix is ​​input into a recurrent neural network, which consists of a 3-layer long short-term memory network (LSTM). Each LSTM layer contains 128 neurons, using a tanh activation function and a dropout rate of 0.5. The core of the LSTM network lies in its gating mechanism, which realizes selective memory and forgetting of input information through the input gate, forget gate, and output gate, thereby solving the problems of gradient vanishing and gradient explosion in traditional recurrent neural networks in long time series. Specifically, the forget gate f of the LSTM t and output gate o t The calculation is expressed as:

[0108] f t =σ(W f ·[h t-1 ,x t ]+b f ),o t =σ(W o ·[h t-1 ,x t ]+b o );

[0109] Among them, f t represents the output of the forget gate at time step t, o t represents the output of the output gate, σ(·) is the Sigmoid activation function, and W f and W o is the weight matrix, b f and b o is the bias term, h t-1 is the hidden state of the previous time step, x tis the input of the current time step. Through layer-by-layer processing of the 3-layer LSTM, an output feature sequence with a dimension of 50x128 is obtained, which represents the feature representation learned by 128 neurons at each time step. In order to improve the focus on important features, the output feature sequence of LSTM is processed by the attention mechanism. The attention mechanism is a method that can dynamically adjust the model's attention to features at different time steps. By calculating the attention score of each time step, an attention weight vector with a dimension of 50 is obtained. Suppose the feature sequence output by LSTM is H = [h1,h2,…,h 50 ], attention weight a t The calculation is expressed as:

[0110]

[0111] Among them, a t represents the attention weight at the t-th time step, h t is the feature representation of time step t, w is the weight vector to be learned, and · is the vector inner product operation. t and attention weight a t A weighted sum is performed to obtain a weighted LSTM feature vector with a dimension of 128. The weighted feature vector can better capture the dynamic characteristics of the lithium battery state by focusing on important time steps and ignoring unimportant information. The weighted LSTM feature vector is input into the first fully connected layer, the size of which is 64, and the ReLU activation function is used to obtain the output feature vector of the recurrent neural network. The ReLU activation function can introduce nonlinearity and enhance the expressiveness of the model, so that the neural network can better capture complex feature relationships. On this basis, based on the electrochemical impedance spectroscopy (EIS) data, the complex nonlinear least squares method is used to fit the Randles equivalent circuit model to extract the ohmic resistance (R ohm ), charge transfer resistance (R ct ) and double layer capacitance (C dl) and other electrochemical parameters. These parameters can reflect the electrochemical behavior inside the battery, construct the physical model parameter vector, input it into the second fully connected layer, the layer size is 32, and the Sigmoid activation function is used to obtain the physical information feature vector. The Sigmoid activation function limits the output value to between 0 and 1, which is suitable for representing the relative size of physical features. The output feature vector of the recurrent neural network is concatenated with the physical information feature vector to form a fused feature vector with a dimension of 96. In order to optimize the fused features, adaptive feature selection is performed on the fused feature vector to reduce the redundancy of the features and retain the most important features to obtain the optimized fused feature vector with a dimension reduced to 64. The optimized fused feature vector is input into the third fully connected layer with a layer size of 100 and the Softmax activation function is used to obtain the probability distribution of the state of charge. The Softmax activation function can convert the output into a probabilistic form to describe the possibility of the lithium battery being in different states of charge. At the same time, the optimized fused feature vector is input into the fourth fully connected layer with a layer size of 1 and the Sigmoid activation function is used to obtain the state of health value (SOH). In this way, the model can directly predict the health state of the battery and give a specific numerical representation. Based on the obtained state of charge probability distribution, the estimated state of charge is obtained by calculating its expected value:

[0112]

[0113] Among them, SOC est is the estimated value of the state of charge, p i is the probability of the i-th state of charge, SOC i is the corresponding state of charge value. By calculating the expected value of the probability distribution, a more robust SOC estimate is obtained. The health state value is directly used as the SOH estimate to complete a comprehensive assessment of the lithium battery state.

[0114] In a specific embodiment, the process of executing step S5 may specifically include the following steps:

[0115] Based on the estimated value of the state of charge and the estimated value of the state of health, an equivalent circuit model of the single cell is constructed to obtain an electrical model of the single cell. The electrical model of the single cell is thermally coupled, and thermal capacitance and thermal resistance parameters are introduced to establish multiple electrical and thermal coupling models of single cells.

[0116] Combine multiple single-cell electrothermal coupling models in series and parallel to construct an electrothermal model of a lithium battery pack;

[0117] Based on the battery pack electrothermal model, the model predictive control algorithm is used to design the objective function J = w1tf + w2Σ (SOCi-SOCavg) 2 +w3*Σ(Ti-Tavg) 2, where J is the objective function, tf is the charging time, SOCi is the SOC of the ith battery, SOCavg is the average SOC, Ti is the temperature of the ith battery, Tavg is the average temperature, and w1, w2, and w3 are weight coefficients;

[0118] Solve the objective function and use the sequential quadratic programming algorithm to solve it at each time step k to obtain the charging current trajectory and the balanced current distribution plan;

[0119] According to the charging current trajectory and the balancing current distribution scheme, the charging current and bypass balancing current of each single cell are calculated to obtain the charging and discharging current trajectory and the balancing signal. The charging and discharging current trajectory and the balancing signal are input into the preset controller mapping function, and the fuzzy control rules are used to convert the continuous signal into discrete control instructions to obtain the charging and discharging control instruction sequence.

[0120] Specifically, an equivalent circuit model of a single battery is constructed based on the estimated value of the state of charge (SOC) and the estimated value of the state of health (SOH). The equivalent circuit model adopts the Randles equivalent circuit model, including the ohmic resistance (R ohm ), charge transfer resistance (R ct ) and double layer capacitance (C dl ), these parameters can describe the electrochemical process inside the battery and the polarization characteristics of the battery. The electrical model of the single cell is obtained by fitting the electrochemical impedance spectroscopy (EIS) data. According to different SOC and SOH conditions, the various parameters of the equivalent circuit are adjusted to better describe the electrical behavior of the single cell under different conditions. In order to more accurately simulate the performance and safety of the battery, the electrical model is coupled with the thermal characteristics and the heat capacity (C th ) and thermal resistance (R th) parameters are used to describe the thermal dynamic behavior of the battery. Heat capacity is used to describe the ability of the battery to store heat, and thermal resistance is used to describe the transfer of heat between the inside and outside of the battery. Based on the coupling of electrical and thermal characteristics, an electrothermal coupling model of a single cell is established, which can simultaneously reflect the voltage, current, temperature and other characteristics of the battery during the charging and discharging process. The electrothermal coupling models of multiple single cells are combined in series and parallel to construct the battery pack electrothermal model of the entire lithium battery pack. In a battery pack, there are multiple batteries in series, parallel or mixed structures, and these different connection methods will affect the overall performance and thermal dynamic characteristics of the battery pack. When constructing the battery pack electrothermal model, the electrical interaction between the single cells, as well as the heat transfer and temperature changes caused by thermal coupling are considered. The final battery pack electrothermal model can describe the electrical and thermodynamic behavior of the entire battery pack during the charging and discharging process, so that the charging process can be accurately controlled and the safety and consistency of the battery can be improved. Based on the constructed battery pack electrothermal model, in order to achieve the optimal charging and discharging management of the lithium battery pack, a model predictive control (MPC) algorithm is used to design an objective function J to optimize multiple factors in the charging process. In order to achieve the optimal solution of the objective function, the sequential quadratic programming (SQP) algorithm is used to solve at each time step k. Sequential quadratic programming is an iterative algorithm for nonlinear optimization problems. By performing quadratic approximation on the objective function at each time step and solving the local optimal solution, the global optimal solution is gradually approached. The solution is performed at each time step to obtain the charging current trajectory and balancing current distribution scheme of the battery pack. These schemes are used to control the charging current of each single cell and the current balance between each single cell, so that the charging process of the entire battery pack remains in the optimal state. According to the charging current trajectory and balancing current distribution scheme obtained by the solution, the charging current of each single cell and the bypass balancing current are calculated to obtain the charging and discharging current trajectory and balancing signal of the entire battery pack. The charging current trajectory is used to describe the current change of each single cell during the charging and discharging process, and the balancing signal is used to control the bypass balancing circuit. By distributing the balancing current, the SOC and temperature differences between each single cell in the battery pack are reduced to ensure the consistency of the battery pack. In order to convert the continuous charging and discharging current trajectory and balancing signal into specific control instructions, they are input into the preset controller mapping function for processing. The mapping function uses fuzzy control rules to convert continuous current and balance signals into discrete charge and discharge control instructions. Fuzzy control rules are based on empirical knowledge and classify and quantify input signals through fuzzification to generate discrete instructions suitable for control system execution. For example, assuming that the input signal is the charging current I charge and balancing current I balance The fuzzy controller classifies the input signal into "low" and "medium" levels according to its size and change trend, and generates corresponding discrete control instructions u through fuzzy reasoning. charge and ubalance These control instructions are used to actually control the charging current and balancing current of the battery pack. In this way, the resulting charge and discharge control instruction sequence is used to control the charging and balancing operations of each single cell to ensure the safety, consistency and efficiency of the battery pack during the charging and discharging process.

[0121] The above describes the lithium battery charge and discharge detection and management method in the embodiment of the present invention. The following describes the lithium battery charge and discharge detection and management device in the embodiment of the present invention. Figure 2 , an embodiment of a lithium battery charge and discharge detection and management device in an embodiment of the present invention includes:

[0122] The data acquisition module is used to collect data on the charging and discharging process of the lithium battery pack to obtain electrode-level charging and discharging data and battery-level charging and discharging data;

[0123] A data expansion module is used to apply consistent Gaussian noise to the same charge and discharge cycle sequence in the electrode-level charge and discharge data and the battery-level charge and discharge data respectively through a generative adversarial network to obtain an electrode-level expanded data set and a battery-level expanded data set;

[0124] A feature mapping module is used to extract the electrode-level time-frequency domain features of the electrode-level extended data set and the battery-level time-frequency domain features of the battery-level extended data set, respectively, and perform feature mapping and feature mixing on the electrode-level time-frequency domain features and the battery-level time-frequency domain features by minimizing the maximum average difference loss function to obtain a mapping feature vector;

[0125] A state estimation module is used to input the mapping feature vector into a fusion model of a recurrent neural network and a physical information hybrid neural network to estimate the battery state of charge and health state, and obtain an estimated value of the state of charge and an estimated value of the health state;

[0126] The generation module is used to construct a battery pack electrothermal model of the lithium battery pack based on the charge state estimation value and the health state estimation value, and calculate the charge and discharge current trajectory and the balance signal, and generate the corresponding charge and discharge control instruction sequence according to the charge and discharge current trajectory and the balance signal.

[0127] Through the synergy of the above-mentioned components, by combining electrode-level and battery-level data collection, and using generative adversarial networks for data enhancement, the present invention significantly improves data utilization efficiency, solves the problem of difficulty in obtaining battery data, and uses time-frequency domain feature extraction and maximum average difference loss function for feature mapping, effectively integrating electrode-level and battery-level information, enhancing the model's perception of the internal state of the battery, and improving the accuracy of state estimation. Combining the recurrent neural network with the physical information hybrid neural network not only utilizes the powerful fitting ability of the deep learning model, but also incorporates the prior knowledge of the battery physical model, achieving high-precision estimation of the battery state of charge and health state. Based on the battery pack electrothermal model and model predictive control algorithm, the present invention can simultaneously consider multiple objectives such as charging time, temperature balance and charge state balance, realize the optimized charge and discharge management of the battery pack, improve charging efficiency, and extend battery life. By generating a dynamic sequence of charge and discharge control instructions, the method can be adaptively adjusted according to the real-time state of the battery, effectively avoiding safety hazards such as overcharging, overdischarging and temperature runaway, and significantly improving the safety of the battery pack.

[0128] 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.

[0129] 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.

[0130] 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.

[0131] 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 can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many 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.

[0132] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0133] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0134] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A lithium battery charge and discharge detection and management method, characterized in that: The method comprises: Collect data on the charge and discharge process of the lithium battery pack to obtain electrode-level charge and discharge data and battery-level charge and discharge data; Applying consistent Gaussian noise to the same charge-discharge cycle sequence in the electrode-level charge-discharge data and the battery-level charge-discharge data by generating an adversarial network, respectively, to obtain an electrode-level expanded data set and a battery-level expanded data set; Respectively extracting the electrode-level time-frequency domain features of the electrode-level extended data set and the battery-level time-frequency domain features of the battery-level extended data set, and performing feature mapping and feature mixing on the electrode-level time-frequency domain features and the battery-level time-frequency domain features by minimizing the maximum mean difference loss function to obtain a mapping feature vector; Inputting the mapped feature vector into a fusion model of a recurrent neural network and a physical information hybrid neural network to estimate the battery state of charge and state of health, thereby obtaining an estimated value of the state of charge and an estimated value of the state of health; A battery pack electrothermal model of the lithium battery pack is constructed based on the state of charge estimation value and the health state estimation value, and a charge and discharge current trajectory and a balance signal are calculated, and a corresponding charge and discharge control instruction sequence is generated according to the charge and discharge current trajectory and the balance signal.

2. The lithium battery charge and discharge detection and management method according to claim 1, characterized in that: The data collection of the charging and discharging process of the lithium battery pack to obtain electrode-level charging and discharging data and battery-level charging and discharging data includes: Reference electrodes are installed on the positive and negative electrodes of each single cell in the lithium battery pack, and the positive and negative electrode potentials are collected by a high-precision voltage acquisition module with a sampling frequency of 1kHz to obtain electrode potential time series data; The total input and output current of the lithium battery pack is monitored in real time using a Hall current sensor with a sampling frequency of 1kHz. At the same time, a high-precision voltage sensor is used to measure the battery pack terminal voltage with a sampling frequency of 1kHz to obtain the battery pack current and voltage time series data; Multiple temperature sensors are evenly arranged on the surface of each single cell in the lithium battery pack, using thermocouples or thermistors with a sampling frequency of 0.1 Hz to obtain the time series data of the battery surface temperature distribution; Based on the electrode potential time series data, the electrode-level state of charge is calculated using the incremental capacity analysis method, the electrode-level capacity is calculated using the coulomb counting method, and the electrode-level internal resistance obtained by electrochemical impedance spectroscopy is combined to construct an electrode-level state parameter vector; According to the battery pack current and voltage time series data, an extended Kalman filter algorithm is used to estimate the battery-level state of charge, and the battery-level capacity is obtained through constant current-constant voltage charge and discharge test. Combined with the battery-level internal resistance obtained by DC internal resistance measurement, a battery-level state parameter vector is constructed; Extracting time-domain features of the electrode-level state parameter vector and the battery-level state parameter vector, and simultaneously performing fast Fourier transform to obtain frequency-domain features, thereby constructing an electrode-level time-frequency domain feature matrix and a battery-level time-frequency domain feature matrix; The electrode-level time-frequency domain feature matrix and the battery-level time-frequency domain feature matrix are respectively fused with the battery surface temperature distribution time series data to obtain an electrode-level state feature vector and a battery-level state feature vector; The electrode-level state feature vector and the battery-level state feature vector are respectively subjected to piecewise linear interpolation and zero-mean normalization processing to obtain electrode-level charge-discharge data and battery-level charge-discharge data.

3. The lithium battery charge and discharge detection and management method according to claim 2, characterized in that: The method of applying consistent Gaussian noise to the same charge and discharge cycle sequence in the electrode-level charge and discharge data and the battery-level charge and discharge data by generating an adversarial network to obtain an electrode-level expanded data set and a battery-level expanded data set includes: Respectively performing charge and discharge cycle segmentation on the electrode-level charge and discharge data and the battery-level charge and discharge data to obtain an electrode-level charge and discharge cycle sequence and a battery-level charge and discharge cycle sequence; Based on the electrode-level charge-discharge cycle sequence and the battery-level charge-discharge cycle sequence, characteristic statistics of each charge-discharge cycle are calculated to obtain electrode-level characteristic statistics and battery-level characteristic statistics; Performing cluster analysis on the electrode-level charge-discharge cycle sequence and the battery-level charge-discharge cycle sequence according to the electrode-level feature statistics and the battery-level feature statistics to obtain electrode-level clustering results and battery-level clustering results; Based on the electrode-level clustering results and the battery-level clustering results, a representative electrode-level charge-discharge cycle sequence and a battery-level charge-discharge cycle sequence are selected as reference sequences, and Gaussian noise is generated for the reference sequences, with the noise intensity ranging from 0.1% to 1% of the standard deviation of the original signal, to obtain a Gaussian noise sequence; Superimposing the Gaussian noise sequence onto the electrode-level charge-discharge cycle sequence and the battery-level charge-discharge cycle sequence respectively, to obtain a noisy electrode-level charge-discharge cycle sequence and a noisy battery-level charge-discharge cycle sequence; Using a generative adversarial network, the noisy electrode-level charge-discharge cycle sequence and the noisy battery-level charge-discharge cycle sequence are processed respectively to generate a new electrode-level charge-discharge cycle sequence and a new battery-level charge-discharge cycle sequence; The new electrode-level charge-discharge cycle sequence and the new battery-level charge-discharge cycle sequence are respectively merged with the electrode-level charge-discharge data and the battery-level charge-discharge data to obtain an electrode-level expanded data set and a battery-level expanded data set.

4. The lithium battery charge and discharge detection and management method according to claim 3, characterized in that: The method of using a generative adversarial network to process the noisy electrode-level charge-discharge cycle sequence and the noisy battery-level charge-discharge cycle sequence respectively to generate a new electrode-level charge-discharge cycle sequence and a new battery-level charge-discharge cycle sequence includes: The noisy electrode-level charge-discharge cycle sequence and the noisy battery-level charge-discharge cycle sequence are respectively input into the generator of the generative adversarial network to obtain the electrode-level encoder output feature sequence and the battery-level encoder output feature sequence; the generator adopts a U-Net structure, including an encoder and a decoder, wherein the encoder is composed of 5 convolutional layers, each convolutional layer uses a 3x3 convolution kernel with a step size of 2, a LeakyReLU activation function and batch normalization, and the decoder is composed of 5 transposed convolutional layers, each transposed convolutional layer uses a 3x3 convolution kernel with a step size of 2, a ReLU activation function and batch normalization; Performing self-attention mechanism processing on the electrode-level encoder output feature sequence and the battery-level encoder output feature sequence respectively, calculating the correlation between each position and all other positions in the electrode-level encoder output feature sequence and the battery-level encoder output feature sequence respectively, and obtaining an electrode-level attention weight matrix and a battery-level attention weight matrix; Multiplying the electrode-level attention weight matrix with the noisy electrode-level charge-discharge cycle sequence to obtain an enhanced electrode-level feature sequence, and multiplying the battery-level attention weight matrix with the noisy battery-level charge-discharge cycle sequence to obtain an enhanced battery-level feature sequence; The enhanced electrode-level feature sequence and the enhanced battery-level feature sequence are respectively input into the decoder of the generator, and a preliminary generated electrode-level charge-discharge cycle sequence and a preliminary generated battery-level charge-discharge cycle sequence are generated through layer-by-layer upsampling and splicing operations; The initially generated electrode-level charge-discharge cycle sequence and the initially generated battery-level charge-discharge cycle sequence are respectively input into a discriminator, wherein the discriminator adopts a PatchGAN structure, comprises 4 convolutional layers, each convolutional layer uses a 4x4 convolution kernel with a step size of 2, a LeakyReLU activation function and instance normalization, and the last layer uses a 1x1 convolution kernel to output an electrode-level discrimination result and a battery-level discrimination result; According to the electrode-level discrimination result and the battery-level discrimination result, a new electrode-level charge-discharge cycle sequence and a new battery-level charge-discharge cycle sequence are output respectively.

5. The lithium battery charge and discharge detection and management method according to claim 4, characterized in that: The method of extracting the electrode-level time-frequency domain features of the electrode-level extended data set and the battery-level time-frequency domain features of the battery-level extended data set respectively, and performing feature mapping and feature mixing on the electrode-level time-frequency domain features and the battery-level time-frequency domain features by minimizing the maximum average difference loss function to obtain a mapping feature vector includes: Performing short-time Fourier transform on the electrode-level expanded data set and the battery-level expanded data set respectively, using a Hamming window function with a window length of 256 and an overlap rate of 50%, to obtain an electrode-level time-frequency spectrum diagram and a battery-level time-frequency spectrum diagram; The electrode-level time-frequency spectrum and the battery-level time-frequency spectrum are respectively input into a pre-trained convolutional neural network to extract electrode-level deep time-frequency features and battery-level deep time-frequency features; the pre-trained convolutional neural network comprises 5 convolutional layers, each convolutional layer uses a 3x3 convolution kernel, a step size of 2, and uses a ReLU activation function and batch normalization; Performing global average pooling and global maximum pooling operations on the electrode-level deep time-frequency features and the battery-level deep time-frequency features, respectively, and concatenating the pooling results to obtain an electrode-level pooling feature vector and a battery-level pooling feature vector; Inputting the electrode-level pooled feature vector and the battery-level pooled feature vector into a pre-trained feature mapping network to obtain a preliminarily mapped electrode-level feature vector and a preliminarily mapped battery-level feature vector; the feature mapping network comprises three fully connected layers, each layer using a LeakyReLU activation function and Dropout regularization; Calculating the maximum average difference loss between the electrode-level feature vector after the preliminary mapping and the battery-level feature vector after the preliminary mapping, the loss function includes a weighted sum of a linear kernel function and an RBF kernel function, wherein the weight of the linear kernel function is 0.3 and the weight of the RBF kernel function is 0.7; Based on the calculated maximum average difference loss value, the output of the feature mapping network is fine-tuned to obtain an adjusted electrode-level feature vector and an adjusted battery-level feature vector, and the adjusted electrode-level feature vector and the adjusted battery-level feature vector are concatenated to obtain a mapping feature vector.

6. The lithium battery charge and discharge detection and management method according to claim 5, characterized in that: The step of inputting the mapping feature vector into a fusion model of a recurrent neural network and a physical information hybrid neural network to estimate the battery state of charge and state of health to obtain a state of charge estimation value and a state of health estimation value includes: The mapped feature vector is segmented according to a sliding window with a time step length of 50 to obtain a time series feature matrix, and the feature dimension of each time step is 256; The time series feature matrix is ​​input into a recurrent neural network, wherein the recurrent neural network comprises a 3-layer LSTM network, each layer of the LSTM network comprises 128 neurons, a tanh activation function and a dropout rate of 0.5 are used, and selective transmission and forgetting of information are performed through a gating mechanism, so as to obtain an LSTM output feature sequence with a dimension of 50x128; Performing attention mechanism processing on the LSTM output feature sequence, calculating the attention score, obtaining an attention weight vector with a dimension of 50, and performing weighted summation on the attention weight vector and the LSTM output feature sequence to obtain a weighted LSTM feature vector with a dimension of 128; Input the weighted LSTM feature vector into the first fully connected layer with a layer size of 64 and use a ReLU activation function to obtain an output feature vector of the recurrent neural network; Based on the electrochemical impedance spectroscopy data, the Randles equivalent circuit model is fitted using the complex nonlinear least squares method to extract the ohmic resistance, charge transfer resistance and double-layer capacitance, and a physical model parameter vector is constructed. The physical model parameter vector is input into the second fully connected layer with a layer size of 32 and a Sigmoid activation function is used to obtain a physical information feature vector. Concatenate the output feature vector of the recurrent neural network with the physical information feature vector to obtain a fused feature vector with a dimension of 96; Adaptively select the fused feature vector to obtain an optimized fused feature vector, reduce the dimension to 64, input the optimized fused feature vector into the third fully connected layer with a layer size of 100, use the Softmax activation function to obtain the state of charge probability distribution, and input the optimized fused feature vector into the fourth fully connected layer with a layer size of 1, use the Sigmoid activation function to obtain the health state value; An expected value is calculated based on the state of charge probability distribution as a state of charge estimation value, and the health state value is directly used as a health state estimation value.

7. The lithium battery charge and discharge detection and management method according to claim 6, characterized in that: The method of constructing a battery pack electrothermal model of the lithium battery pack based on the state of charge estimation value and the health state estimation value, calculating a charge and discharge current trajectory and a balance signal, and generating a corresponding charge and discharge control instruction sequence according to the charge and discharge current trajectory and the balance signal includes: Based on the estimated state of charge value and the estimated state of health value, an equivalent circuit model of a single cell is constructed to obtain an electrical model of the single cell, and the electrical model of the single cell is thermally coupled to introduce thermal capacitance and thermal resistance parameters to establish multiple electrical-thermal coupling models of single cells; Combine multiple single-cell electrothermal coupling models in series and parallel to construct an electrothermal model of a lithium battery pack; Based on the battery pack electrothermal model, the model predictive control algorithm is used to design the objective function J = w1tf + w2Σ (SOCi-SOCavg) 2 +w3*Σ(Ti-Tavg) 2 , where J is the objective function, tf is the charging time, SOCi is the SOC of the ith battery, SOCavg is the average SOC, Ti is the temperature of the ith battery, Tavg is the average temperature, and w1, w2, and w3 are weight coefficients; Solving the objective function, using a sequential quadratic programming algorithm, solving at each time step k, to obtain a charging current trajectory and a balanced current distribution plan; According to the charging current trajectory and the balancing current distribution scheme, the charging current and bypass balancing current of each single battery are calculated to obtain the charging and discharging current trajectory and the balancing signal, and the charging and discharging current trajectory and the balancing signal are input into a preset controller mapping function, and the continuous signal is converted into a discrete control instruction using a fuzzy control rule to obtain a charging and discharging control instruction sequence.

8. A lithium battery charge and discharge detection and management device, characterized in that: Used to execute the lithium battery charge and discharge detection and management method according to any one of claims 1 to 7, the lithium battery charge and discharge detection and management device comprises: The data acquisition module is used to collect data on the charging and discharging process of the lithium battery pack to obtain electrode-level charging and discharging data and battery-level charging and discharging data; A data expansion module, used to apply consistent Gaussian noise to the same charge and discharge cycle sequence in the electrode-level charge and discharge data and the battery-level charge and discharge data respectively through a generative adversarial network to obtain an electrode-level expanded data set and a battery-level expanded data set; A feature mapping module, used to extract the electrode-level time-frequency domain features of the electrode-level extended data set and the battery-level time-frequency domain features of the battery-level extended data set, respectively, and perform feature mapping and feature mixing on the electrode-level time-frequency domain features and the battery-level time-frequency domain features by minimizing the maximum average difference loss function to obtain a mapping feature vector; A state estimation module, used for inputting the mapping feature vector into a fusion model of a recurrent neural network and a physical information hybrid neural network to estimate the battery state of charge and state of health, and obtaining a state of charge estimation value and a state of health estimation value; A generation module is used to construct a battery pack electrothermal model of the lithium battery pack based on the charge state estimation value and the health state estimation value, and calculate the charge and discharge current trajectory and the balance signal, and generate a corresponding charge and discharge control instruction sequence according to the charge and discharge current trajectory and the balance signal.

9. A computer device, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and is characterized in that when the processor executes the computer program, the lithium battery charge and discharge detection and management method described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor is enabled to execute the lithium battery charge and discharge detection and management method according to any one of claims 1 to 7.

Images