Battery health status comprehensive evaluation method, system, device and storage medium

Abstract

The present invention relates to the technical field of battery health state analysis, and discloses a battery health state comprehensive evaluation method, system, device and storage medium, wherein the method comprises: preprocessing data collected by a battery management system to obtain a standardized data sequence; calculating a first battery characteristic parameter according to the standardized data sequence; performing covariance matrix adaptive Kalman filtering processing on the first battery characteristic parameter to obtain a second battery characteristic parameter; constructing a capacity-internal resistance change rate index and a temperature-efficiency correlation index based on the second battery characteristic parameter to obtain a dual-threshold evaluation matrix; inputting the dual-threshold evaluation matrix into a deep reinforcement learning model, performing forward propagation calculation and iterative training through an Actor network and a Critic network to obtain a feature fusion weight coefficient; performing weighted calculation on the dual-threshold evaluation matrix to generate a battery health state value and a remaining life prediction value, and the method realizes a multi-dimensional evaluation of the battery health state.

Description

Technical Field

[0001] The present invention relates to the technical field of battery health status analysis, and in particular to a battery health status comprehensive evaluation method, system, device and storage medium. Background Art

[0002] As a core component, the health status monitoring and life prediction of power batteries have become key technologies to ensure the safe operation of vehicles. Traditional battery health assessment methods mainly rely on a single capacity or internal resistance feature, which cannot fully reflect the performance degradation of the battery, resulting in insufficient accuracy and reliability of the assessment results.

[0003] At present, the data collected by the battery management system has problems such as noise interference and inconsistent sampling frequency, which directly affects the quality of feature extraction. At the same time, the existing evaluation models often adopt a fixed weight fusion strategy, which is difficult to adapt to the dynamic characteristics of the battery under different working conditions, causing the evaluation results to deviate from the actual state. In addition, battery health status assessment also faces challenges such as quantification of uncertainty in evaluation results and confidence assessment. Especially under complex working conditions, a single evaluation indicator is difficult to accurately reflect the true health status of the battery. It is necessary to establish a comprehensive evaluation method for multi-source data fusion and introduce intelligent algorithms such as deep learning to improve the accuracy and robustness of the evaluation. Summary of the invention

[0004] The present invention provides a method, system, device and storage medium for comprehensive evaluation of battery health status, and the present invention realizes multi-dimensional evaluation of battery health status.

[0005] In a first aspect, the present invention provides a method for comprehensively evaluating the health status of a battery, the method comprising:

[0006] The voltage data, current data and temperature data collected by the battery management system are subjected to wavelet denoising and unified frequency processing to obtain a standardized data sequence;

[0007] According to the standardized data sequence, the internal resistance value is calculated by the pulse response method, the capacity value is calculated by the current integration method, and the efficiency value is calculated by the charge and discharge energy ratio to obtain the first battery characteristic parameter;

[0008] Performing a covariance matrix adaptive Kalman filter process on the first battery characteristic parameter to obtain a second battery characteristic parameter;

[0009] Based on the second battery characteristic parameter, a capacity-internal resistance change rate index and a temperature-efficiency correlation index are constructed to obtain a dual-threshold evaluation matrix;

[0010] The dual threshold evaluation matrix is ​​input into the deep reinforcement learning model, and forward propagation calculation and iterative training are performed through the Actor network and the Critic network to obtain the feature fusion weight coefficient;

[0011] The dual threshold evaluation matrix is ​​weightedly calculated according to the feature fusion weight coefficient to generate a battery health state value and a remaining life prediction value.

[0012] In a second aspect, the present invention provides a battery health status comprehensive evaluation system, the battery health status comprehensive evaluation system comprising:

[0013] The acquisition module is used to perform wavelet noise reduction and unified frequency processing on the voltage data, current data, and temperature data collected by the battery management system to obtain a standardized data sequence;

[0014] A calculation module, used to calculate the internal resistance value by a pulse response method, the capacity value by a current integration method, and the efficiency value by a charge-discharge energy ratio according to the standardized data sequence, so as to obtain a first battery characteristic parameter;

[0015] A processing module, used for performing a covariance matrix adaptive Kalman filter process on the first battery characteristic parameter to obtain a second battery characteristic parameter;

[0016] A construction module, used to construct a capacity-internal resistance change rate index and a temperature-efficiency correlation index based on the second battery characteristic parameter to obtain a dual-threshold evaluation matrix;

[0017] A training module is used to input the dual threshold evaluation matrix into a deep reinforcement learning model, perform forward propagation calculation and iterative training through an Actor network and a Critic network, and obtain a feature fusion weight coefficient;

[0018] A generation module is used to perform weighted calculation on the dual threshold evaluation matrix according to the feature fusion weight coefficient to generate a battery health state value and a remaining life prediction value.

[0019] The third aspect of the present invention provides a battery health status comprehensive assessment 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 battery health status comprehensive assessment device executes the above-mentioned battery health status comprehensive assessment method.

[0020] 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 battery health status comprehensive evaluation method.

[0021] In the technical solution provided by the present invention, the noise interference in the original data is effectively eliminated through wavelet denoising and frequency uniform preprocessing, and the problem of inconsistent sampling frequency of multi-source data is solved. The Kalman filter algorithm with covariance matrix adaptation is adopted to dynamically adjust the noise covariance matrix, improve the filtering accuracy of the characteristic parameters, and enhance the algorithm's resistance to measurement noise. A dual-threshold evaluation matrix is ​​constructed, which comprehensively considers the characteristic indicators of the two dimensions of capacity-internal resistance change rate and temperature-efficiency correlation, and realizes the multi-dimensional evaluation of the battery health state. The TD3 deep reinforcement learning algorithm is introduced to optimize the feature fusion weight coefficient, and the dynamic adaptive adjustment of the weight coefficient is realized through iterative training of timing error, which improves the accuracy of the evaluation results. The confidence analysis of the evaluation results is carried out, and the confidence intervals of the health status value and the remaining life prediction value are given, which provides a quantitative basis for the reliability of the evaluation results. A hierarchical evaluation strategy is adopted to divide the health status into five levels: excellent, good, normal, warning and dangerous, which is convenient for intuitively judging the battery health status and conducive to timely discovering potential safety hazards.

[0022] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0023] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A schematic diagram of an embodiment of a method for comprehensive evaluation of battery health status in an embodiment of the present invention;

[0025] Figure 2 A schematic diagram of an embodiment of a battery health status comprehensive evaluation system in an embodiment of the present invention;

[0026] Figure 3 Schematic diagram of an embodiment of a battery health status comprehensive evaluation device in an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0029] To facilitate understanding of this embodiment, a comprehensive evaluation method for battery health status disclosed in an embodiment of the present invention is first described in detail. Figure 1 As shown, the method comprises the following steps:

[0030] 101. Perform wavelet noise reduction and unified frequency processing on the voltage data, current data, and temperature data collected by the battery management system to obtain a standardized data sequence;

[0031] It is understandable that the execution subject of the present invention may be a battery health status comprehensive evaluation system, 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.

[0032] Specifically, the operating conditions of the target battery are monitored and data is acquired in real time through the battery management system, including the continuous collection of voltage and current data. At the same time, a high-precision temperature sensor is arranged on the battery surface to record the change trend of the battery surface temperature, and obtain the original data sequence consisting of the voltage data sequence, the current data sequence and the temperature data sequence. The original data sequence is subjected to median filtering. Median filtering is a signal smoothing processing technology. By comparing and adjusting the midpoint value of the data window, the abnormal values ​​that are obviously deviated from the normal range are eliminated to obtain a more stable and reliable filtered data sequence. The filtered data sequence is subjected to wavelet decomposition operation to decompose the data sequence into corresponding low-frequency coefficient sequences and high-frequency coefficient sequences, effectively identifying the trend information and high-frequency noise characteristics in the data. The low-frequency coefficient sequence contains the overall change trend of the data, while the high-frequency coefficient sequence contains transient, random and noise information. In order to remove noise and retain key features, the high-frequency coefficients are judged and screened according to the set threshold. For high-frequency coefficients higher than the set threshold, they are appropriately retained in the subsequent reconstruction to ensure that the signal details are not completely lost, while the high-frequency components that are too low below the threshold and are mainly noise are moderately weakened. By performing wavelet reconstruction operations on the low-frequency coefficient sequence and the high-frequency coefficient sequence above the set threshold, the reconstructed data sequence after denoising is obtained. The temperature data in the reconstructed data sequence after denoising is interpolated and the frequency is unified. The reconstructed temperature data sequence after denoising is interpolated to make it evenly distributed on the specified time axis to achieve frequency unification of the temperature data; the reconstructed voltage data and current data after denoising are also interpolated and frequency resampled in the same way to make the time resolution of the voltage data sequence and the current data sequence consistent with the temperature data sequence. The temperature data sequence with unified frequency, the voltage data sequence with unified frequency, and the current data sequence with unified frequency are combined and aligned to construct a set of standardized data sequences.

[0033] 102. Calculate the internal resistance value by a pulse response method, calculate the capacity value by a current integration method, and calculate the efficiency value by a charge-discharge energy ratio according to the standardized data sequence to obtain a first battery characteristic parameter;

[0034] Specifically, based on the standardized data sequence, the voltage data sequence and the current data sequence are subjected to feature extraction under specific experimental conditions. When the battery is in a stable operating state of constant current charging and discharging, a current pulse stimulus with an amplitude of 0.5C is applied to the current input end, so that the battery produces an obvious and measurable transient response under a small disturbance, thereby obtaining a corresponding voltage-current response sequence on the time axis. The response sequence contains subtle changes in the voltage signal before and after the pulse. Using this change law, the internal resistance characteristic value of the battery is calculated according to the measured voltage change and the corresponding current change. This process uses the pulse response principle to process the voltage and current transient data, so as to accurately determine the size of the battery internal resistance under the current working condition. The actual capacity of the battery is analyzed. The actual available capacity value of the battery is obtained by performing time integration operations on the current data sequence in the constant current and constant voltage charging stage and the constant current discharge stage. This capacity value not only represents the actual energy storage capacity of the battery under the current working condition, but also reflects whether the battery has decreased in capacity due to aging, chemical decay or cycle loss. The voltage data sequence and current data sequence in the standardized data sequence are segmented, and the voltage and current changes in the charging and discharging process are respectively intercepted into complete charging and discharging cycle sequences, so as to facilitate the subsequent charging and discharging efficiency calculation. In the charging and discharging efficiency calculation, the voltage data sequence and the current data sequence in the complete charging and discharging cycle sequence are multiplied and integrated to calculate the total input energy value of the charging process and the total output energy value of the discharging process. The discharge output energy value and the charging input energy value are divided to obtain the characteristic value of the charging and discharging efficiency. This efficiency value can directly reflect the energy utilization rate of the battery. The temperature changes during the operation of the battery are statistically analyzed. By extracting the data sequences of the four temperature measurement points in the standardized data sequence, these sequences are statistically operated to calculate the temperature mean, temperature standard deviation and maximum temperature difference. These temperature statistical characteristics can reveal the thermal distribution characteristics of the battery under different working environments and states. For example, the temperature mean reflects the overall heat level, the temperature standard deviation indicates the amplitude of temperature fluctuation, and the maximum temperature difference is used to evaluate whether the temperature distribution is uniform. The battery internal resistance characteristic value, the actual available capacity value, the charge and discharge efficiency characteristic value are combined with the temperature mean, the temperature standard deviation, and the maximum temperature difference to form a first battery characteristic parameter.

[0035] 103. Perform covariance matrix adaptive Kalman filtering on the first battery characteristic parameter to obtain a second battery characteristic parameter;

[0036] Specifically, the first battery characteristic parameter is substituted into the state space equation as a state vector and an observation vector to establish a state space model of the battery system. The state space model is a mathematical framework that describes the dynamic characteristics of the system. Through this model, the change process of the battery health characteristic parameter is expressed in the form of state transfer. After the state space model is constructed, the state transfer matrix is ​​initialized and calculated according to the known system structure to generate the initial state transfer parameters. The state prediction calculation is performed based on the state transfer parameters. State prediction is an important part of Kalman filtering, and the current state information is used to predict the possible state of the system at the next moment. In the process of state prediction, the state transfer covariance matrix is ​​dynamically updated based on the first forgetting factor. The role of the first forgetting factor is to control the influence of historical information on the current prediction. By updating the state transfer covariance matrix, the first adaptive covariance matrix is ​​obtained, so that the model can flexibly adapt to the changes in the battery characteristic parameters. After the state prediction is completed, the first adaptive covariance matrix and the observation matrix are state observed, and the actual measured observation data are compared with the predicted state to identify the true state of the system. In this link, the second adaptive covariance matrix is ​​obtained by updating the observation noise covariance matrix based on the second forgetting factor. The second forgetting factor is used to dynamically adjust the influence of observation noise so that the model can still maintain a high estimation accuracy when facing uncertain noise. The Kalman gain coefficient is calculated according to the first adaptive covariance matrix and the second adaptive covariance matrix to obtain the Kalman correction matrix. The Kalman correction matrix is ​​a dynamically adjusted weight matrix used to balance the error between state prediction and actual observation. When the correction matrix is ​​calculated, it is weighted and fused with the state prediction error to generate a state update sequence. The state update sequence is a correction result of the predicted state, which can more accurately reflect the true dynamics of the battery health characteristic parameters. The state update sequence is calculated by covariance correction to optimize the uncertainty of the state and improve the accuracy of the estimation result, and generate an optimized health characteristic vector. The internal resistance eigenvalue, capacity eigenvalue, efficiency eigenvalue and temperature eigenvalue in the optimized health characteristic vector are combined and reconstructed to generate the second battery characteristic parameter.

[0037] 104. Construct a capacity-internal resistance change rate index and a temperature-efficiency correlation index based on the second battery characteristic parameter to obtain a dual threshold evaluation matrix;

[0038] Specifically, data related to internal resistance and capacity are extracted from the second battery characteristic parameter, and their relative changes are calculated. For internal resistance, the current internal resistance characteristic value is subtracted from the initial internal resistance value of the battery to calculate the absolute change in internal resistance, which is then divided by the initial internal resistance value to obtain the relative change in internal resistance, reflecting the growth rate of the battery internal resistance and revealing the degree of battery aging. At the same time, for the change in capacity, the current capacity characteristic value is subtracted from the initial capacity value to calculate the absolute change in capacity, and then the ratio of the current capacity characteristic value to the initial capacity value is calculated to obtain the relative change in capacity, reflecting the degree of capacity attenuation of the battery and providing key information on changes in battery energy storage performance. By multiplying the relative change in internal resistance by the first weight coefficient, the weighted change in internal resistance is calculated, and at the same time, the relative change in capacity is multiplied by the second weight coefficient to calculate the weighted change in capacity. The setting of the weight coefficient can be adjusted according to the priority and importance in actual applications, so as to more accurately reflect the contribution of internal resistance and capacity to battery health in the calculation. The weighted change of internal resistance and the weighted change of capacity are weighted combined, and the combined operation is realized by linear weighting or nonlinear function to obtain the first threshold index. The second threshold index is generated according to the characteristic value of charge and discharge efficiency and the maximum temperature difference in the second battery characteristic parameter. The characteristic value of charge and discharge efficiency is multiplied by the third weight coefficient to reflect the importance of efficiency characteristics to the overall battery health assessment; the maximum temperature difference in the second battery characteristic parameter is compared with the maximum allowable temperature difference, the difference is calculated, and the difference is multiplied by the fourth weight coefficient to obtain the weighted characteristic value of the temperature difference. The analysis of the temperature difference can reflect the thermal management status and uniformity characteristics of the battery, and is highly related to the battery performance and safety. By combining these weighted characteristic values, a weighted coefficient sequence is generated, and the second threshold index is obtained after the weighted coefficient sequence is summed. The first threshold index is combined with the second threshold index to form a dual threshold evaluation matrix. This matrix not only includes the rate of change of capacity and internal resistance, but also combines the correlation between temperature and efficiency, which can more comprehensively reflect the health status of the battery during use.

[0039] 105. Input the dual threshold evaluation matrix into the deep reinforcement learning model, perform forward propagation calculation and iterative training through the Actor network and the Critic network, and obtain the feature fusion weight coefficient;

[0040] Specifically, the optimized health feature vector, the first threshold index and the second threshold index in the dual threshold evaluation matrix are integrated into a unified state space. The construction of the state space is based on the information contained in the evaluation matrix, which fully reflects the dynamic change characteristics of multiple dimensions such as battery internal resistance, capacity, efficiency and temperature. At the same time, the dynamic weight coefficients β1 and β2 are defined as the action space. The introduction of these two weight coefficients enables the reinforcement learning model to optimize the evaluation results by adjusting the feature weights, forming a deep reinforcement learning input sequence with state-action as the core. By calculating the difference between the input sequence and the reference health state value, the deviation between the health state predicted by the current model and the actual reference value is quantified. In order to smooth this deviation and consider the impact of time series changes, the difference is multiplied by the time series difference of the health state, and a smoothing factor is introduced for addition operation to construct the reward function value of reinforcement learning. The reward function value determines the strategic direction of the model for dynamic weight adjustment during training. After the reward function is constructed, the deep reinforcement learning input sequence is input into the deep reinforcement learning model. The model consists of two key fully connected neural networks, namely the Actor network and the Critic network, which are used to generate action strategies and evaluate the value of actions, respectively. The Actor network consists of 6 layers of fully connected neurons, with the number of neurons in each layer being 256, 128, 64, 32, and 16, respectively. The ReLU activation function is used inside the network to enhance the nonlinear mapping capability. Through forward propagation calculation, the Actor network generates an output sequence for predicting the optimal dynamic weight adjustment strategy. Similarly, the Critic network has the same structure and number of layers as the Actor network. It uses the ReLU activation function to evaluate the value of the input state-action pair and outputs the sequence of the Critic network through forward propagation calculation. The two networks model and evaluate the dynamic weight adjustment from different angles, respectively, providing data support for subsequent policy optimization. After the output sequences of the Actor network and the Critic network are calculated, the TD3 policy gradient is constructed based on these two outputs to form an efficient optimization framework. The TD3 algorithm, namely the Twin Delayed Deep Deterministic Policy Gradient, can effectively suppress the over-estimation problem and improve the stability of policy optimization by introducing two independent Critic networks and a strategy for delaying the update of the target network. According to the TD3 policy gradient, the optimization goal of reinforcement learning is constructed, and the parameters of the network are gradually updated through iterative operations to minimize the temporal difference error. In this process, the adjustment of network parameters can continuously reduce the error between the predicted output and the actual reference value, thereby improving the model's ability to optimize dynamic weights. After completing multiple iterative updates of network parameters, the updated network parameter sequence is input into the experience replay buffer for delayed update.The introduction of experience replay can effectively break the correlation between data, while using past training experience to improve learning efficiency and stability. In the delayed update of the buffer, the dynamic weight coefficients β1 and β2 will gradually converge with the optimization of the network and be extracted in the final stage. The dynamic weight coefficients β1 and β2 are normalized to ensure that their values ​​are on a uniform scale, which is convenient for the subsequent calculation and evaluation of feature fusion, and the feature fusion weight coefficient is obtained to reflect the weight of each feature on the battery health status assessment.

[0041] 106. Perform weighted calculation on the dual threshold evaluation matrix according to the feature fusion weight coefficient to generate the battery health status value and the remaining life prediction value.

[0042] Specifically, the dynamic weight coefficient is used to weight each indicator in the dual threshold evaluation matrix. The first threshold indicator is multiplied by the dynamic weight coefficient β1 to obtain a first weighted evaluation value; at the same time, the second threshold indicator is multiplied by the dynamic weight coefficient β2 to generate a second weighted evaluation value. The two weighted evaluation values ​​respectively reflect the relative contribution ratios of different feature dimensions in the battery health assessment. The first weighted evaluation value and the second weighted evaluation value are summed to obtain a comprehensive health evaluation index. The battery health level is determined for the comprehensive health evaluation index. The level classification standard is set based on the actual application scenario. For example, when the comprehensive health evaluation index is greater than or equal to 0.90, the battery state is judged to be an excellent level; if the index is between 0.80 and 0.90, it is a good level; when the index is between 0.70 and 0.80, the battery is at a normal level; if the index is between 0.60 and 0.70, it enters a warning level; and when the comprehensive health evaluation index is lower than 0.60, it indicates that the battery state is at a dangerous level. Through the judgment process, the user is provided with clear information on the battery health status level. An exponential fitting extrapolation operation is performed based on the time series of the comprehensive health assessment indicators. By fitting the time series of historical data, a health trend curve is generated to reflect the dynamic change trend of the battery health status. A threshold analysis is performed on the health trend curve to determine when the battery will reach a preset health critical value, thereby calculating the remaining service life of the battery. In order to evaluate the confidence level of the health status value and the remaining life prediction, random perturbation values ​​are added to the optimized health feature vector, and multiple Monte Carlo sampling operations are performed to generate multiple groups of health status distribution sequences. The Monte Carlo method can effectively capture the uncertainty in the model prediction results by simulating multiple random samplings. The mean-variance statistical operation is performed on these distribution sequences to obtain the confidence interval range of the battery health status, reflecting the reliability and accuracy of the prediction results. The comprehensive health assessment indicators, health level values, remaining service life and confidence interval range are integrated to generate the battery health status value and remaining life prediction value.

[0043] In the embodiment of the present invention, the noise interference in the original data is effectively eliminated by wavelet denoising and frequency uniform preprocessing, and the problem of inconsistent sampling frequency of multi-source data is solved. The Kalman filter algorithm with covariance matrix adaptation is adopted to dynamically adjust the noise covariance matrix, improve the filtering accuracy of the characteristic parameters, and enhance the algorithm's resistance to measurement noise. A dual-threshold evaluation matrix is ​​constructed, which comprehensively considers the characteristic indicators of the two dimensions of capacity-internal resistance change rate and temperature-efficiency correlation, and realizes the multi-dimensional evaluation of the battery health state. The TD3 deep reinforcement learning algorithm is introduced to optimize the feature fusion weight coefficient, and the dynamic adaptive adjustment of the weight coefficient is realized through iterative training of timing error, which improves the accuracy of the evaluation results. The confidence analysis of the evaluation results is carried out, and the confidence intervals of the health status value and the remaining life prediction value are given, which provides a quantitative basis for the reliability of the evaluation results. A hierarchical evaluation strategy is adopted to divide the health status into five levels: excellent, good, normal, warning and dangerous, which is convenient for intuitively judging the battery health status and conducive to timely discovering potential safety hazards.

[0044] In a specific embodiment, the process of executing step 101 may specifically include the following steps:

[0045] The voltage data and current data are collected through the battery management system, and the temperature data are collected by the temperature sensor arranged on the battery surface to obtain the original data sequence;

[0046] Performing a median filter operation on the original data sequence to obtain a filtered data sequence with outliers removed, and performing a wavelet decomposition operation on the filtered data sequence with outliers removed to obtain a low-frequency coefficient sequence and a high-frequency coefficient sequence;

[0047] Perform wavelet reconstruction operation on the low-frequency coefficient sequence and the high-frequency coefficient sequence above the set threshold to obtain the reconstructed data sequence after noise reduction;

[0048] Performing interpolation and frequency unification on the temperature data in the reconstructed data sequence after denoising to obtain a temperature data sequence with a unified frequency; performing interpolation and frequency unification on the voltage data in the reconstructed data sequence after denoising to obtain a voltage data sequence with a unified frequency; performing interpolation and frequency unification on the current data in the reconstructed data sequence after denoising to obtain a current data sequence with a unified frequency;

[0049] The temperature data sequence with a unified frequency, the voltage data sequence with a unified frequency, and the current data sequence with a unified frequency are combined to obtain a standardized data sequence.

[0050] Specifically, the battery management system monitors the battery operating status in real time and continuously collects voltage data , current data And the temperature data collected by the temperature sensor .in, represents a time series point, Indicates the battery terminal voltage represents the instantaneous value of the current flowing into or out of the battery, while represents the battery surface temperature recorded by the temperature sensor. These raw data constitute a time series . Perform median filtering on the original data sequence. Median filtering is a denoising algorithm used to eliminate sudden outliers. In median filtering, a sliding window length is set , for any time , the data in the window The median replaces the data value at the current position The formula is:

[0051] ;

[0052] in, represents the filtered data, is half the width of the window. For example, when , When , the filtering result is the median value of these data. After this step, the filtered data sequence with outliers removed is obtained. The filtered data sequence is subjected to wavelet decomposition to further remove noise. Wavelet decomposition is a multi-resolution signal analysis method that decomposes the signal into a low-frequency part (representing the main characteristics of the signal) and a high-frequency part (representing details and noise). Suppose the wavelet basis function is , the low-frequency component of the signal and high frequency components It is expressed as:

[0053] ;

[0054] in, is the low frequency scaling function, is the wavelet basis function. Decompose and obtain the low-frequency coefficient sequence and high frequency coefficient sequence . Reconstruct low-frequency coefficients and high-frequency coefficients, but only retain those above the set threshold The high-frequency components of are reconstructed to remove high-frequency noise. For example, the rule of high-frequency reconstruction is expressed as:

[0055] ;

[0056] Reconstructed signal Calculated by the following formula:

[0057] ;

[0058] After the reconstruction operation, a denoised data sequence is generated Perform interpolation on the denoised sequence. Assuming that the time interval of the current sequence is uneven, the interpolation method uses linear interpolation or spline interpolation, with a uniform time interval as The linear interpolation formula is:

[0059] ;

[0060] in is the interpolation time point, and is the most recent known time point, and is the corresponding signal value. After interpolation and frequency resampling, a sequence with uniform frequency is obtained. The temperature data series, voltage data series and current data series with unified frequencies are combined into a standardized data series. The influence of different data magnitudes is eliminated by converting each type of data value into a relative proportional value. Assume that the maximum value of the temperature series is and the minimum value is , the standardized formula is:

[0061] ;

[0062] The voltage and current data are also standardized in a similar way and finally combined into a standardized data sequence .

[0063] In a specific embodiment, the process of executing step 102 may specifically include the following steps:

[0064] Applying a current pulse with an amplitude of 0.5C to the voltage data sequence and the current data sequence in the standardized data sequence during the constant current charge and discharge process to obtain a voltage-current response sequence;

[0065] The ratio of the voltage change to the current change is calculated according to the voltage-current response sequence to obtain the characteristic value of the battery internal resistance;

[0066] Performing time integration operation on the current data sequence in the standardized data sequence in the constant current and constant voltage charging stage and the constant current and constant voltage discharging stage to obtain the actual available capacity value, and performing data segmentation on the voltage data sequence and the current data sequence in the standardized data sequence to obtain a complete charge and discharge cycle sequence;

[0067] Perform product integration operations on the voltage data sequence and the current data sequence in the complete charge and discharge cycle sequence to obtain the charging input energy value and the discharging output energy value;

[0068] The discharge output energy value is divided by the charge input energy value to obtain the charge and discharge efficiency characteristic value, and the temperature data sequence of the four measuring points in the standardized data sequence is statistically calculated to obtain the temperature mean, temperature standard deviation and maximum temperature difference;

[0069] The battery internal resistance characteristic value, the actual available capacity value, the charge and discharge efficiency characteristic value, the temperature mean value, the temperature standard deviation and the maximum temperature difference are combined to obtain the first battery characteristic parameter.

[0070] Specifically, in the battery constant current charge and discharge experiment, the voltage data sequence is extracted from the standardized data sequence and current data series , and apply an amplitude of The current pulse is used as a disturbance. is the nominal capacity of the battery in ampere-hours (Ah). Indicates the current value is For example, for a battery with a nominal capacity of 10Ah, the applied current amplitude is 5A. When a current pulse acts on the battery, the terminal voltage of the battery will respond with a transient change, generating a voltage-current response sequence By recording the voltage change and current change , calculate the internal resistance characteristic value of the battery , the formula is:

[0071] ;

[0072] in , represents the transient change of voltage after the pulse action, is the magnitude of the change in current. For example, if the current pulse causes the voltage to rise from 3.2V to 3.3V instantaneously, and the current changes from 0A to 5A instantaneously, the internal resistance of the battery is This internal resistance characteristic value reflects the conductivity and health status of the battery. Extracting the current data sequence from the standardized data sequence , and perform time integration calculations in the constant current and constant voltage charging stage and the constant current and constant current discharge stage to obtain the actual available capacity value of the battery The capacity calculation formula is:

[0073] ;

[0074] in and is the start and end time of the charge and discharge phase. For example, if the current recorded during constant current discharge is 2A and the discharge duration is 1800s, the actual capacity is:

[0075] ;

[0076] At the same time, the voltage data in the standardized data sequence and current data The data is segmented to extract the complete charge and discharge cycle sequence. The charging input energy values ​​are obtained by multiplying and integrating the segmented data. And discharge output energy value The energy calculation formula is:

[0077] ;

[0078] in and is the time range of charge and discharge. For example, if the charging voltage is 4V, the current is 2A, and the charging duration is 1800s, then the charging energy is:

[0079] ;

[0080] Calculated by the same method as discharge For example, if the discharge energy is 3.6Wh, then the characteristic value of the charge and discharge efficiency is Calculated by the following formula:

[0081] ;

[0082] In this example, the charge and discharge efficiency is Extract the data sequence of 4 temperature measurement points from the standardized data sequence , and perform statistical operations to calculate the temperature mean, temperature standard deviation and maximum temperature difference. The battery internal resistance characteristic value, actual available capacity value, charge and discharge efficiency characteristic value, temperature mean, temperature standard deviation and maximum temperature difference are combined to obtain the first battery characteristic parameter.

[0083] In a specific embodiment, the process of executing step 103 may specifically include the following steps:

[0084] Substituting the first battery characteristic parameter as a state vector and an observation vector into a state space equation to obtain a state space model, and performing initialization calculation on a state transfer matrix according to the state space model to obtain a state transfer parameter;

[0085] Performing state prediction calculation on the state transfer parameters, and updating the state transfer covariance matrix based on the first forgetting factor to obtain a first adaptive covariance matrix;

[0086] Performing state observation calculation on the first adaptive covariance matrix and the observation matrix, and performing an update operation on the observation noise covariance matrix based on the second forgetting factor to obtain a second adaptive covariance matrix;

[0087] The gain coefficient is calculated according to the first adaptive covariance matrix and the second adaptive covariance matrix to obtain a Kalman correction matrix, and the Kalman correction matrix is ​​weightedly fused with the state prediction error to obtain a state update sequence;

[0088] The state update sequence is subjected to covariance correction calculation to obtain an optimized health feature vector, and the internal resistance feature value, capacity feature value, efficiency feature value and temperature feature value in the optimized health feature vector are combined and reconstructed to obtain a second battery characteristic parameter.

[0089] Specifically, the first battery characteristic parameter is constructed as a state vector and an observation vector, where represents the characteristic value of internal resistance, represents the capacity characteristic value, represents the characteristic value of charge and discharge efficiency, Represents the temperature characteristic value. Based on this state vector, the state space equation is constructed, including the state transfer equation and the observation equation, which are expressed as:

[0090] ;

[0091] ;

[0092] in, is the system state vector, is the observation vector, is the state transition matrix, is the observation matrix, and are the state noise and observation noise vectors respectively. Assuming they satisfy zero-mean Gaussian distribution, the covariance matrices are and . State transfer matrix The initialization of is based on assumptions about the physical characteristics of the battery. For example, assuming that the health parameters of the battery degrade at a certain rate, the state transfer matrix is ​​set to:

[0093] ;

[0094] It means that each parameter remains basically stable in a short period of time, but can be dynamically adjusted through subsequent learning. After initialization, the state transition parameters are calculated in combination with the state space model , and combined with noise Complete state prediction. Calculate the covariance of the state prediction value and use the first forgetting factor Update the state transition covariance matrix The update formula of the covariance matrix is:

[0095] ;

[0096] in, represents the prediction covariance matrix, represents the covariance matrix at the previous moment, represents the state noise covariance matrix. The first forgetting factor Control the impact of historical states on current predictions. For example, if , then they are more inclined to trust the change of the current state. After completing the prediction, the predicted value is combined with the observation vector and the observation matrix Perform state observation. Observation covariance matrix The update of , the update formula is:

[0097] ;

[0098] This formula takes real-time observation noise into account and dynamically adjusts the covariance matrix. and the observation covariance matrix , calculate the Kalman gain , the formula is:

[0099] ;

[0100] The Kalman gain is a dynamic weight used to balance the confidence of the state prediction value and the actual observation value. , correct the state prediction value and get the state update sequence:

[0101] ;

[0102] At the same time, the covariance matrix is ​​corrected, and the correction formula is:

[0103] ;

[0104] The above correction process obtains the optimized health feature vector , which contains the updated internal resistance eigenvalue , Capacity characteristic value , efficiency characteristic value and temperature characteristic values The optimized eigenvalues ​​are combined and reconstructed to generate the second battery characteristic parameters.

[0105] In a specific embodiment, the process of executing step 104 may specifically include the following steps:

[0106] Divide the difference between the internal resistance characteristic value in the second battery characteristic parameter and the initial internal resistance value by the initial internal resistance value to obtain a relative change in internal resistance;

[0107] Divide the difference between the capacity characteristic value in the second battery characteristic parameter and its initial capacity value by the initial capacity value to obtain a relative capacity change;

[0108] The relative change in internal resistance is multiplied by the first weight coefficient to obtain a weighted change in internal resistance, and the relative change in capacity is multiplied by the second weight coefficient to obtain a weighted change in capacity;

[0109] Performing a weighted combination operation on the weighted change in internal resistance and the weighted change in capacity to obtain a first threshold indicator;

[0110] The charge and discharge efficiency characteristic value in the second battery characteristic parameter is multiplied by the third weight coefficient, and the difference between the maximum temperature difference in the second battery characteristic parameter and the maximum allowable temperature difference is multiplied by the fourth weight coefficient to obtain a weight coefficient sequence;

[0111] The weighted coefficient sequence is summed to obtain a second threshold index, and the first threshold index is combined with the second threshold index to construct a dual-threshold evaluation matrix.

[0112] Specifically, the internal resistance characteristic value is extracted from the second battery characteristic parameter , Capacity characteristic value , charge and discharge efficiency characteristic value And the maximum temperature difference These characteristic values ​​can reflect the changes in the key performance of the battery and need to be compared with the initial state. , by comparing with the initial internal resistance The difference is normalized to calculate the relative change of internal resistance , the formula is:

[0113] ;

[0114] in, Indicates the internal resistance of the battery at the factory or in its initial state. Similarly, the capacity characteristic value Calculate the relative change and the initial capacity value In contrast, the formula is:

[0115] ;

[0116] in, is the nominal capacity of the battery in its initial state. and the relative change in capacity Perform weighted processing. Introduce the first weight coefficient and the second weight coefficient , calculate the weighted change of internal resistance and capacity-weighted change , the formulas are:

[0117] ;

[0118] Weighted change of internal resistance and capacity-weighted change Perform weighted combination operation to obtain the first threshold index , the formula is:

[0119] ;

[0120] First threshold indicator The change information of internal resistance and capacity is integrated to reflect the main degradation characteristics of battery health status. Extract the charge and discharge efficiency characteristic value from the second battery characteristic parameter and maximum temperature difference .Charge and discharge efficiency characteristic value Multiply by the third weight coefficient , and obtain the weighted efficiency change , the formula is:

[0121] ;

[0122] At the same time, the maximum temperature difference Divide by the maximum allowable temperature difference The difference is multiplied by the fourth weight coefficient , calculate the weighted temperature difference change , the formula is:

[0123] ;

[0124] Combine the above calculation results into a weighted coefficient sequence , and perform a sum operation on it to obtain the second threshold index , the formula is:

[0125] ;

[0126] The first threshold indicator and the second threshold indicator Combine and construct a dual threshold evaluation matrix:

[0127] .

[0128] In a specific embodiment, the process of executing step 105 may specifically include the following steps:

[0129] The optimized health feature vector, the first threshold index and the second threshold index in the dual threshold evaluation matrix are constructed into a state space, and the dynamic weight coefficients β1 and β2 are constructed into an action space to obtain a deep reinforcement learning input sequence;

[0130] The difference between the deep reinforcement learning input sequence and the reference health state value is calculated, and the difference between the health state time series multiplied by the smoothing factor is added to obtain the reward function value;

[0131] The deep reinforcement learning input sequence is input into the 6-layer fully connected Actor network with [256, 128, 64, 32, 16] hidden layer neurons in the deep reinforcement learning model, and the ReLU activation function is used for forward propagation calculation to obtain the Actor network output sequence;

[0132] The deep reinforcement learning input sequence is input into the 6-layer fully connected critic network with [256, 128, 64, 32, 16] hidden layer neurons in the deep reinforcement learning model, and the forward propagation calculation is performed using the ReLU activation function to obtain the critic network output sequence;

[0133] According to the output sequence of the Actor network and the output sequence of the Critic network, the TD3 policy gradient is constructed to obtain the optimization target of the TD3 algorithm, and the temporal difference error minimization iterative operation is performed on the optimization target of the TD3 algorithm to obtain the network parameter update sequence;

[0134] The network parameter update sequence is input into the experience replay buffer for delayed update to obtain the dynamic weight coefficients β1 and β2, and the dynamic weight coefficients β1 and β2 are normalized to obtain the feature fusion weight coefficients.

[0135] Specifically, construct the state space vector , which is composed of the optimized health feature vector and the first and second threshold indicators in the dual threshold evaluation matrix:

[0136] ;

[0137] in, is the characteristic value of internal resistance, is the capacity characteristic value, is the characteristic value of charge and discharge efficiency, is the maximum temperature difference, and are the first and second threshold indicators respectively. The state space fully describes the operating characteristics and evaluation results of the battery. At the same time, the action space is defined Represents the dynamic weight coefficient, which is used to adjust the importance weight of each feature:

[0138] ;

[0139] in, and They are respectively for and The dynamic weight coefficient of the state space and action space Combination, forming the input sequence of deep reinforcement learning :

[0140] ;

[0141] The input sequence describes the overall state of the current system and the weight adjustment strategy. Calculate the reward function , which consists of the input sequence With reference health status value The deviation of , and the smoothness of the health status time series difference:

[0142] ;

[0143] in, is the Euclidean norm, is a smoothing factor that adjusts the effect of temporal continuity on rewards. The first term of the reward function penalizes the extent to which the current strategy deviates from the reference value, and the second term rewards the smoothness of state changes. The input is sent to the Actor network, which consists of 6 fully connected layers. The number of neurons in each layer decreases in sequence, and the network uses the ReLU activation function. The Actor network outputs the optimized action value, i.e., the dynamic weight coefficient, through forward propagation calculation:

[0144] ;

[0145] in, is the strategy function of the Actor network, is the network parameter. The output of the Actor network is the current optimal dynamic weight adjustment strategy. Input to the Critic network, which has a similar structure to the Actor network and is used to evaluate the value of the state-action pair. The Critic network outputs the value function:

[0146] ;

[0147] in, is the parameter of the Critic network. The value function evaluates the contribution of the current action to the optimization goal. Based on the output of the Actor network And the value function of the Critic network , using the TD3 algorithm to optimize network parameters. TD3 outputs two independent Critic networks and To suppress the overestimation of the value function, the calculation formula of the policy gradient is:

[0148] ;

[0149] The critic network updates parameters through temporal difference (TD) error, and the update formula is:

[0150] ;

[0151] in, is the reward value, is the discount factor, and is the state and action at the next moment. TD3 introduces an experience replay buffer for delayed updates. The Critic network recalculates the value function every time it is updated, while the Actor network is updated only after the Critic network is stable. The Actor network is updated less frequently to reduce the instability of the strategy. The dynamic weight coefficient output from the Actor network and The update results are extracted and normalized to ensure that the weight sum is 1. The normalization formula is:

[0152] ;

[0153] The normalized dynamic weight coefficient is used as the feature fusion weight to provide a basis for subsequent health status assessment and life prediction.

[0154] Among them, the TD3 policy gradient is constructed for the Actor network output sequence and the Critic network output sequence to obtain the TD3 algorithm optimization target, and the TD3 algorithm optimization target is iteratively operated by minimizing the temporal difference error to obtain the network parameter update sequence, including: the policy network parameters of the Actor network output sequence are smoothly copied according to the soft update rate τ=0.005 to obtain the target policy network parameters μ'; the value network parameters of the Critic network output sequence are double cloned to obtain two target value network parameters Q1' and Q2'; Gaussian noise perturbation is added to the action output of the target policy network parameter μ' and truncated to obtain the target action sequence at'; the target action sequence at' is input into the two target value networks Q 1' and Q2' are forward calculated, and the minimum value selection operation is performed on the output value to obtain the target Q value Qtarget; the target Q value Qtarget and the immediate reward rt are weighted and summed with the discount coefficient γ=0.99 to obtain the TD target value ytd; the TD target value ytd is differenced with the output value of the current value network Q1 and Q2 to obtain two temporal difference errors δ1 and δ2; the mean square error of the temporal difference errors δ1 and δ2 is calculated, and weighted summation is performed with the same weight to obtain the loss function value L(θQ) of the Critic network; the loss function value L(θQ) is used to perform policy gradient calculation and delayed update on the Actor network parameters, and the parameters are updated every d=2 time steps to obtain the network parameter update sequence.

[0155] In a specific embodiment, the process of executing step 106 may specifically include the following steps:

[0156] Performing a product operation on the first threshold indicator and the dynamic weight coefficient β1 to obtain a first weighted evaluation value, and performing a product operation on the second threshold indicator and the dynamic weight coefficient β2 to obtain a second weighted evaluation value;

[0157] The first weighted evaluation value and the second weighted evaluation value are summed to obtain a comprehensive health evaluation index;

[0158] Perform battery health level determination operation on the comprehensive health assessment index, wherein when the comprehensive health assessment index is greater than or equal to 0.90, it is an excellent level, when the comprehensive health assessment index is greater than or equal to 0.80 and less than 0.90, it is a good level, when the comprehensive health assessment index is greater than or equal to 0.70 and less than 0.80, it is a normal level, when the comprehensive health assessment index is greater than or equal to 0.60 and less than 0.70, it is a warning level, and when the comprehensive health assessment index is less than 0.60, it is a dangerous level, and obtain a health level value;

[0159] Based on the time series of comprehensive health assessment indicators, exponential fitting and extrapolation operations are performed to obtain a health trend curve, and threshold analysis operations are performed on the health trend curve to obtain the remaining service life;

[0160] Add random disturbance values ​​to the optimized health feature vector and perform multiple Monte Carlo sampling operations to obtain multiple groups of health state distribution sequences, and perform mean-variance statistical operations on the multiple groups of health state distribution sequences to obtain confidence interval ranges;

[0161] The comprehensive health assessment indicators, health level values, remaining service life and confidence interval range are integrated to generate battery health status values ​​and remaining service life prediction values.

[0162] Specifically, for the first threshold indicator and dynamic weight coefficient Perform weighted calculation to obtain the first weighted evaluation value The formula is:

[0163] ;

[0164] in, It reflects the comprehensive characteristics of internal resistance and capacity changes. It is for The dynamic weight coefficient of is used to adjust its importance in the comprehensive evaluation. Similarly, for the second threshold indicator and dynamic weight coefficient Perform weighted calculation to obtain the second weighted evaluation value , the formula is:

[0165] ;

[0166] in, Taking into account the effects of efficiency and temperature characteristics, is the corresponding weight coefficient. The first weighted evaluation value and the second weighted evaluation value are summed to obtain the comprehensive health evaluation index. :

[0167] ;

[0168] This indicator comprehensively reflects the multi-dimensional characteristics of the battery health status through weighted integration of dual thresholds. The health level is determined by the preset value range. When the battery status is judged to be excellent; when Good grade when Normal level when It is a warning level when It is a dangerous level. Define the health level value The function expression is:

[0169] ;

[0170] This determination process converts the comprehensive health assessment results of the battery into an easy-to-understand health level. In order to predict the remaining service life of the battery, based on the comprehensive health assessment indicators Time Series Extrapolate the trend. Establish a health trend curve through exponential fitting method :

[0171] ;

[0172] in, is the initial health index value, is the degradation rate parameter. By fitting the time series data, we can determine and The health trend curve is used to predict when the set health threshold will be reached. , the corresponding time The remaining battery life is:

[0173] ;

[0174] In order to quantify the uncertainty of the prediction results, random perturbation values ​​are added to the optimized health feature vector, and multiple Monte Carlo sampling simulations are performed. Assume that the optimized health feature vector is , whose random perturbation is :

[0175] ;

[0176] in, It satisfies the normal distribution A set of health status distribution sequences is generated by multiple sampling , and calculate the mean and variance of the comprehensive health assessment indicators based on these sequences:

[0177] ;

[0178] in, is the number of samples. The mean and variance are used to construct the confidence interval range:

[0179] ;

[0180] in, is the critical value of the standard normal distribution determined according to the confidence level. , Health level value , Remaining useful life and confidence interval range Perform data integration to generate battery health status value and remaining life prediction value.

[0181] The above describes the comprehensive evaluation method of the battery health status in the embodiment of the present invention. The following describes the comprehensive evaluation system of the battery health status in the embodiment of the present invention. Figure 2 , an embodiment of a battery health status comprehensive evaluation system in an embodiment of the present invention includes:

[0182] The acquisition module 201 is used to perform wavelet noise reduction and unified frequency processing on the voltage data, current data, and temperature data collected by the battery management system to obtain a standardized data sequence;

[0183] A calculation module 202 is used to calculate the internal resistance value by a pulse response method, the capacity value by a current integration method, and the efficiency value by a charge-discharge energy ratio according to the standardized data sequence, so as to obtain a first battery characteristic parameter;

[0184] A processing module 203 is used to perform a covariance matrix adaptive Kalman filter process on the first battery characteristic parameter to obtain a second battery characteristic parameter;

[0185] A construction module 204 is used to construct a capacity-internal resistance change rate index and a temperature-efficiency correlation index based on the second battery characteristic parameter to obtain a dual-threshold evaluation matrix;

[0186] The training module 205 is used to input the dual threshold evaluation matrix into the deep reinforcement learning model, perform forward propagation calculation and iterative training through the Actor network and the Critic network, and obtain the feature fusion weight coefficient;

[0187] The generating module 206 is used to perform weighted calculation on the dual threshold evaluation matrix according to the feature fusion weight coefficient to generate a battery health state value and a remaining life prediction value.

[0188] Through the synergy of the above components, wavelet denoising and frequency uniform preprocessing are used to effectively eliminate noise interference in the original data and solve the problem of inconsistent sampling frequency of multi-source data. The Kalman filter algorithm with adaptive covariance matrix is ​​adopted to dynamically adjust the noise covariance matrix, improve the filtering accuracy of characteristic parameters, and enhance the algorithm's resistance to measurement noise. A dual-threshold evaluation matrix is ​​constructed, which comprehensively considers the characteristic indicators of the two dimensions of capacity-internal resistance change rate and temperature-efficiency correlation, and realizes the multi-dimensional evaluation of battery health status. The TD3 deep reinforcement learning algorithm is introduced to optimize the feature fusion weight coefficient. Through iterative training of timing error, the dynamic adaptive adjustment of the weight coefficient is realized, which improves the accuracy of the evaluation results. The confidence analysis of the evaluation results is carried out, and the confidence intervals of the health status value and the remaining life prediction value are given, which provides a quantitative basis for the reliability of the evaluation results. A hierarchical evaluation strategy is adopted to divide the health status into five levels: excellent, good, normal, warning and dangerous, which is convenient for intuitive judgment of the battery health status and conducive to timely discovery of potential safety hazards.

[0189] above Figure 2 The battery health status comprehensive evaluation system in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The battery health status comprehensive evaluation device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0190] Figure 3 3 is a schematic diagram of the structure of a battery health status comprehensive evaluation device provided by an embodiment of the present invention. The battery health status comprehensive evaluation device 300 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 can be short-term storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the battery health status comprehensive evaluation device 300. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the battery health status comprehensive evaluation device 300 to implement the steps of the above-mentioned battery health status comprehensive evaluation method.

[0191] The battery health status comprehensive evaluation device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be understood by those skilled in the art that Figure 3 The structure of the battery health status comprehensive evaluation device shown does not constitute a limitation on the battery health status comprehensive evaluation device provided by the present invention, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0192] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the battery health status comprehensive evaluation method.

[0193] 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.

[0194] 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 is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or 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 and other media that can store program code.

[0195] 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 method for comprehensive evaluation of battery health status, characterized in that: The method comprises: The voltage data, current data and temperature data collected by the battery management system are subjected to wavelet denoising and unified frequency processing to obtain a standardized data sequence; According to the standardized data sequence, the internal resistance value is calculated by the pulse response method, the capacity value is calculated by the current integration method, and the efficiency value is calculated by the charge and discharge energy ratio to obtain the first battery characteristic parameter; Performing a covariance matrix adaptive Kalman filter process on the first battery characteristic parameter to obtain a second battery characteristic parameter; Based on the second battery characteristic parameter, a capacity-internal resistance change rate index and a temperature-efficiency correlation index are constructed to obtain a dual-threshold evaluation matrix; The dual threshold evaluation matrix is ​​input into the deep reinforcement learning model, and forward propagation calculation and iterative training are performed through the Actor network and the Critic network to obtain the feature fusion weight coefficient; The dual threshold evaluation matrix is ​​weightedly calculated according to the feature fusion weight coefficient to generate a battery health state value and a remaining life prediction value.

2. The battery health status comprehensive evaluation method according to claim 1, characterized in that: The voltage data, current data, and temperature data collected by the battery management system are subjected to wavelet noise reduction and unified frequency processing to obtain a standardized data sequence, including: The voltage data and current data are collected through the battery management system, and the temperature data are collected by the temperature sensor arranged on the battery surface to obtain the original data sequence; Performing a median filter operation on the original data sequence to obtain a filtered data sequence with outliers removed, and performing a wavelet decomposition operation on the filtered data sequence with outliers removed to obtain a low-frequency coefficient sequence and a high-frequency coefficient sequence; Performing wavelet reconstruction operation on the low-frequency coefficient sequence and the high-frequency coefficient sequence above a set threshold to obtain a reconstructed data sequence after noise reduction; Performing interpolation operation and frequency unification on the temperature data in the reconstructed data sequence after denoising to obtain a temperature data sequence with a unified frequency; performing interpolation operation and frequency unification on the voltage data in the reconstructed data sequence after denoising to obtain a voltage data sequence with a unified frequency; performing interpolation operation and frequency unification on the current data in the reconstructed data sequence after denoising to obtain a current data sequence with a unified frequency; The temperature data sequence with unified frequency, the voltage data sequence with unified frequency and the current data sequence with unified frequency are combined to obtain a standardized data sequence.

3. The method for comprehensive evaluation of battery health status according to claim 2, characterized in that: The method of calculating the internal resistance value by the pulse response method, the capacity value by the current integration method, and the efficiency value by the charge-discharge energy ratio according to the standardized data sequence to obtain the first battery characteristic parameter includes: Applying a current pulse with an amplitude of 0.5C to the voltage data sequence and the current data sequence in the standardized data sequence during constant current charging and discharging to obtain a voltage-current response sequence; Calculate the ratio of the voltage change to the current change according to the voltage-current response sequence to obtain a characteristic value of the battery internal resistance; Performing time integration operation on the current data sequence in the standardized data sequence in the constant current and constant voltage charging stage and the constant current and constant voltage discharging stage to obtain the actual available capacity value, and performing data segmentation on the voltage data sequence and the current data sequence in the standardized data sequence to obtain a complete charge and discharge cycle sequence; Performing a product-integral operation on the voltage data sequence and the current data sequence in the complete charge-discharge cycle sequence to obtain a charge input energy value and a discharge output energy value; Performing a division operation on the discharge output energy value and the charge input energy value to obtain a charge-discharge efficiency characteristic value, and performing a statistical operation on the temperature data sequence of the four measuring points in the standardized data sequence to obtain a temperature mean, a temperature standard deviation, and a maximum temperature difference; The battery internal resistance characteristic value, the actual available capacity value, the charge and discharge efficiency characteristic value, the temperature mean value, the temperature standard deviation and the maximum temperature difference are combined to obtain a first battery characteristic parameter.

4. The method for comprehensive evaluation of battery health status according to claim 3, characterized in that: The performing covariance matrix adaptive Kalman filtering processing on the first battery characteristic parameter to obtain the second battery characteristic parameter includes: Substituting the first battery characteristic parameter as a state vector and an observation vector into a state space equation to obtain a state space model, and performing initialization calculation on a state transfer matrix according to the state space model to obtain a state transfer parameter; Performing state prediction calculation on the state transfer parameter, and performing update operation on the state transfer covariance matrix based on a first forgetting factor to obtain a first adaptive covariance matrix; Performing state observation calculation on the first adaptive covariance matrix and the observation matrix, and updating the observation noise covariance matrix based on a second forgetting factor to obtain a second adaptive covariance matrix; Calculate the gain coefficient according to the first adaptive covariance matrix and the second adaptive covariance matrix to obtain a Kalman correction matrix, and perform a weighted fusion operation on the Kalman correction matrix and the state prediction error to obtain a state update sequence; A covariance correction calculation is performed on the state update sequence to obtain an optimized health feature vector, and the internal resistance feature value, capacity feature value, efficiency feature value and temperature feature value in the optimized health feature vector are combined and reconstructed to obtain a second battery characteristic parameter.

5. The method for comprehensive evaluation of battery health status according to claim 4, characterized in that: The method of constructing a capacity-internal resistance change rate index and a temperature-efficiency correlation index based on the second battery characteristic parameter to obtain a dual threshold evaluation matrix includes: Dividing the difference between the internal resistance characteristic value and the initial internal resistance value in the second battery characteristic parameter by the initial internal resistance value to obtain a relative change in internal resistance; Dividing the difference between the capacity characteristic value in the second battery characteristic parameter and its initial capacity value by the initial capacity value to obtain a relative capacity change; Multiplying the relative change in internal resistance by a first weight coefficient to obtain a weighted change in internal resistance, and multiplying the relative change in capacity by a second weight coefficient to obtain a weighted change in capacity; Performing a weighted combination operation on the weighted change in internal resistance and the weighted change in capacity to obtain a first threshold indicator; The charge and discharge efficiency characteristic value in the second battery characteristic parameter is multiplied by a third weight coefficient, and a quotient obtained by dividing the maximum temperature difference in the second battery characteristic parameter by the maximum allowable temperature difference, and a difference obtained by subtracting 1 from the quotient is multiplied by a fourth weight coefficient to obtain a weighted coefficient sequence; A sum operation is performed on the weighted coefficient sequence to obtain a second threshold indicator, and the first threshold indicator is combined with the second threshold indicator to construct a dual-threshold evaluation matrix.

6. The method for comprehensive evaluation of battery health status according to claim 5, characterized in that: The dual threshold evaluation matrix is ​​input into the deep reinforcement learning model, and forward propagation calculation and iterative training are performed through the Actor network and the Critic network to obtain the feature fusion weight coefficient, including: The optimized health feature vector, the first threshold index and the second threshold index in the dual threshold evaluation matrix are constructed into a state space, and the dynamic weight coefficients β1 and β2 are constructed into an action space to obtain a deep reinforcement learning input sequence; Calculate the difference between the deep reinforcement learning input sequence and the reference health state value, and add the difference between the health state time series multiplied by the smoothing factor to obtain a reward function value; Input the deep reinforcement learning input sequence into a 6-layer fully connected Actor network with [256, 128, 64, 32, 16] hidden layer neurons in the deep reinforcement learning model, and perform forward propagation calculation using the ReLU activation function to obtain an Actor network output sequence; Input the deep reinforcement learning input sequence into a 6-layer fully connected critic network with [256, 128, 64, 32, 16] hidden layer neurons in the deep reinforcement learning model, and perform forward propagation calculation using the ReLU activation function to obtain a critic network output sequence; Constructing a TD3 policy gradient according to the Actor network output sequence and the Critic network output sequence to obtain a TD3 algorithm optimization target, and performing a temporal difference error minimization iterative operation on the TD3 algorithm optimization target to obtain a network parameter update sequence; The network parameter update sequence is input into the experience replay buffer for delayed update to obtain dynamic weight coefficients β1 and β2, and the dynamic weight coefficients β1 and β2 are normalized to obtain feature fusion weight coefficients.

7. The method for comprehensive evaluation of battery health status according to claim 6, characterized in that: The step of performing weighted calculation on the dual threshold evaluation matrix according to the feature fusion weight coefficient to generate a battery health state value and a remaining life prediction value includes: Performing a product operation on the first threshold indicator and the dynamic weight coefficient β1 to obtain a first weighted evaluation value, and performing a product operation on the second threshold indicator and the dynamic weight coefficient β2 to obtain a second weighted evaluation value; Performing a sum operation on the first weighted evaluation value and the second weighted evaluation value to obtain a comprehensive health evaluation index; A battery health level determination operation is performed on the comprehensive health assessment index, wherein when the comprehensive health assessment index is greater than or equal to 0.90, it is an excellent level, when the comprehensive health assessment index is greater than or equal to 0.80 and less than 0.90, it is a good level, when the comprehensive health assessment index is greater than or equal to 0.70 and less than 0.80, it is a normal level, when the comprehensive health assessment index is greater than or equal to 0.60 and less than 0.70, it is a warning level, and when the comprehensive health assessment index is less than 0.60, it is a danger level, and a health level value is obtained; Performing an exponential fitting extrapolation operation based on the time series of the comprehensive health assessment index to obtain a health trend curve, and performing a threshold analysis operation on the health trend curve to obtain a remaining service life; Adding random disturbance values ​​to the optimized health feature vector to perform multiple Monte Carlo sampling operations to obtain multiple groups of health state distribution sequences, and performing mean-variance statistical operations on the multiple groups of health state distribution sequences to obtain confidence interval ranges; The comprehensive health assessment index, health grade value, remaining service life and confidence interval range are integrated to generate a battery health status value and a remaining service life prediction value.

8. A battery health status comprehensive evaluation system, characterized in that: The system for executing the comprehensive evaluation method for the battery health status according to any one of claims 1 to 7 comprises: The acquisition module is used to perform wavelet noise reduction and unified frequency processing on the voltage data, current data, and temperature data collected by the battery management system to obtain a standardized data sequence; A calculation module, used to calculate the internal resistance value by a pulse response method, the capacity value by a current integration method, and the efficiency value by a charge-discharge energy ratio according to the standardized data sequence, so as to obtain a first battery characteristic parameter; A processing module, used for performing a covariance matrix adaptive Kalman filter process on the first battery characteristic parameter to obtain a second battery characteristic parameter; A construction module, used to construct a capacity-internal resistance change rate index and a temperature-efficiency correlation index based on the second battery characteristic parameter to obtain a dual-threshold evaluation matrix; A training module is used to input the dual threshold evaluation matrix into a deep reinforcement learning model, perform forward propagation calculation and iterative training through an Actor network and a Critic network, and obtain a feature fusion weight coefficient; A generation module is used to perform weighted calculation on the dual threshold evaluation matrix according to the feature fusion weight coefficient to generate a battery health state value and a remaining life prediction value.

9. A battery health status comprehensive assessment device, characterized in that: The battery health status comprehensive assessment device comprises: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory to enable the battery health state comprehensive evaluation device to execute the battery health state comprehensive evaluation method according to any one of claims 1 to 7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the battery health status comprehensive evaluation method as described in any one of claims 1-7 is implemented.

Images