A battery state of health evaluation method, system, computer device and medium

Abstract

The application provides a battery health state evaluation method, system, computer device and medium, and belongs to the technical field of battery health management and life prediction. The method comprises the following steps: performing full-life cycle cyclic charging and discharging test on a reference battery, collecting data samples at fixed sampling intervals, and extracting a fusion feature vector of the reference battery sample; training a reference prediction model by using the reference data sample; collecting a fusion feature vector of a target battery to be tested, calculating the distribution distance of the features of the target battery and the reference battery by using an MMD algorithm, combining the prediction loss of the reference battery to establish a double-loss function, taking the minimization of the total loss as the target, iteratively adjusting the reference model parameters by using back propagation, and obtaining an optimized model suitable for the target battery; and the SOH value of the target battery can be accurately predicted based on the optimized model. The method greatly reduces the data collection cost and model deployment period in the migration stage, and realizes the migration of the prediction model in different battery devices at a low cost and high precision.

Description

Technical Field

[0001] This invention belongs to the field of battery health management and life prediction technology, specifically relating to a battery health status assessment method, system, computer equipment, and medium. Background Technology

[0002] With the rapid development of new energy vehicles, renewable energy storage systems, and intelligent electronic devices, lithium-ion batteries are increasingly widely used as key energy storage components. Accurate assessment of battery state of health (SOH) and prediction of remaining life are crucial core functions in battery management systems, significantly contributing to improved battery efficiency, reduced maintenance costs, and extended system lifespan.

[0003] Currently, lithium-ion battery state of health (SOH) prediction methods are mainly divided into three categories: direct measurement methods, model-driven methods, and data-driven methods. Direct measurement methods, by monitoring physical quantities such as battery internal resistance and conductance, can achieve high-precision estimation under certain ideal operating conditions. However, in complex application scenarios, they are easily affected by sensor accuracy and environmental factors, limiting their practicality. Model-driven methods typically rely on equivalent circuit models or electrochemical mechanism models, possessing a certain theoretical foundation. However, the modeling process requires high levels of professional knowledge and parameter identification accuracy, making it difficult to adapt to different battery types. In contrast, data-driven methods, with their advantages of not relying on physical modeling and strong generalization ability, have become the mainstream direction in current research and engineering practice. Especially with the support of deep learning and big data technologies, they have been widely used for battery state estimation and degradation trend modeling.

[0004] Existing technologies have certain limitations in predicting the health status of lithium-ion batteries. These include the fact that training models often rely on data under specific types or operating conditions, making it difficult to transfer to other batteries or scenarios. Some methods are computationally complex and difficult to run efficiently on resource-constrained embedded platforms, which is not conducive to practical engineering deployment. Summary of the Invention

[0005] To address the problems of difficulty in transferring, low accuracy, and poor adaptability of existing battery health status assessment models when adapting to different battery types, this invention provides a battery health status assessment method, system, computer equipment, and medium.

[0006] To achieve the above objectives, the present invention provides a battery health status assessment method, comprising:

[0007] A cyclic charge-discharge test was performed on the reference battery. At a fixed sampling interval, the charging current and charging voltage with timestamps during the charging phase and the discharging current and discharging voltage with timestamps during the discharging phase were collected synchronously during each charge-discharge cycle. The battery health status (SOH) true label corresponding to each cycle was also collected synchronously.

[0008] Based on the discharge current and discharge voltage, the voltage-discharge capacity curve of the discharge stage is plotted; the differential capacity curve is calculated using the voltage-discharge capacity curve, and the maximum peak value of the differential capacity curve is determined; the maximum characteristic peak region is determined based on the maximum peak value, and the partial discharge capacity of the maximum characteristic peak region is calculated; the maximum voltage time point of the charging voltage is determined; the time point when the current transitions from charging to discharging is determined based on the charging current and the discharging current; the maximum peak value, partial discharge capacity, maximum voltage time point, and the time point when the current transitions from charging to discharging in each cycle are fused according to weights to obtain multiple sets of fused feature vectors.

[0009] A reference prediction model is trained using the multiple sets of fused feature vectors, and the SOH prediction value of the reference battery is output. The prediction loss between the SOH prediction value of the reference battery and the real label is calculated. The fused feature vector of the current charge and discharge data of the target battery is obtained, and the feature distribution distance between the fused feature vectors of the target battery and the reference battery is calculated. The parameters of the reference prediction model are adjusted using the prediction loss and the feature distribution distance to obtain the optimized prediction model.

[0010] Based on the optimized prediction model, the State of Health (SOH) value of the target battery under test is evaluated.

[0011] Preferably, a CNN network is used to fuse the maximum peak value, partial discharge capacity, maximum voltage time point, and the time point when the current transitions from charging to discharging according to weights to obtain a fused feature vector.

[0012] Preferably, the reference prediction model is a gated recurrent unit (GRU) network; furthermore, the prediction results of the GRU network are evaluated using mean square error, root mean square error, mean absolute error, and coefficient of determination.

[0013] Preferably, the feature distribution distance between the fused feature vectors of the target battery and the reference battery is calculated using the MMD algorithm; the parameters of the reference prediction model are adjusted using the prediction loss and the feature distribution distance to obtain an optimized prediction model, specifically including: constructing a double loss function based on the prediction loss and the feature distribution distance; iteratively adjusting the weights and biases of the reference prediction model using the backpropagation algorithm with the goal of minimizing the double loss function; and obtaining the optimized prediction model when the value of the double loss function is less than a set threshold.

[0014] Preferably, the step of plotting the voltage-discharge capacity curve of the discharge stage based on the discharge current and discharge voltage includes: integrating the discharge current of the discharge stage over time to obtain the discharge capacity, and plotting the voltage-discharge capacity curve of the discharge stage using the voltage and discharge capacity of the discharge stage.

[0015] Preferably, the step of determining the maximum characteristic peak region based on the maximum peak value and calculating the partial discharge capacity of the maximum characteristic peak region specifically includes: determining the minimum points adjacent to the left and right of the maximum peak value, and calculating the area of ​​the closed region enclosed by the maximum peak value and the adjacent minimum points; the area of ​​the closed region is the partial discharge capacity of the maximum characteristic peak region; the partial discharge capacity represents the total capacity change within the voltage range corresponding to the maximum characteristic peak.

[0016] Preferably, before acquiring the fused feature vector of the current charge and discharge data of the target battery under test and calculating the feature distribution distance between the fused feature vectors of the target battery under test and the reference battery, the method further includes constructing historical cyclic charge and discharge data of the target battery under test; the historical cyclic charge and discharge data includes the charging current and charging voltage with timestamps for the charging phase of N rounds of charge and discharge of the target battery under test, and the discharging current and discharging voltage with timestamps for the discharging phase; and optimizing the reference prediction model using the feature set of the target battery under test.

[0017] The present invention also includes a battery health status assessment system, comprising:

[0018] The data acquisition module is used to perform cyclic charge-discharge tests on the reference battery. It synchronously acquires the timestamped charging current and charging voltage during the charging phase and the timestamped discharging current and discharging voltage during the discharging phase at fixed sampling intervals. It also synchronously acquires the battery health status (SOH) true label for each cycle.

[0019] The feature fusion module is used to plot the voltage-discharge capacity curve of the discharge stage based on the discharge current and discharge voltage; calculate the differential capacity curve using the voltage-discharge capacity curve and determine the maximum peak value of the differential capacity curve; determine the maximum characteristic peak region based on the maximum peak value and calculate the partial discharge capacity of the maximum characteristic peak region; determine the maximum voltage time point of the charging voltage; determine the time point when the current transitions from charging to discharging based on the charging current and discharging current; and fuse the maximum peak value, partial discharge capacity, maximum voltage time point, and the time point when the current transitions from charging to discharging in each cycle according to weights to obtain multiple sets of fused feature vectors.

[0020] The model training module is used to train a reference prediction model using the multiple sets of fused feature vectors and output the SOH prediction value of the reference battery; calculate the prediction loss between the SOH prediction value of the reference battery and the real label; obtain the fused feature vector of the current charge and discharge data of the target battery under test, and calculate the feature distribution distance between the fused feature vectors of the target battery under test and the reference battery; adjust the parameters of the reference prediction model using the prediction loss and the feature distribution distance to obtain an optimized prediction model; and evaluate the SOH value of the target battery under test based on the optimized prediction model.

[0021] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement any of the steps in the battery health status assessment method.

[0022] The present invention also provides a computer-readable storage medium storing a computer program that, when loaded by a processor, can execute any of the steps in the battery health status assessment method.

[0023] The battery health status assessment method provided by this invention has the following beneficial effects:

[0024] This invention uses transfer learning logic to collect the maximum peak value, partial discharge capacity, maximum voltage time point, and current transition from charging to discharging time point in each cycle, and fuses them according to weights to obtain multiple sets of fused feature vectors. The reference prediction model is then trained using the multiple sets of fused feature vectors. Finally, the reference prediction model is optimized using the target battery under test to obtain a prediction model that can accurately predict the target battery under test. This universal fusion feature vector, trained using reference battery's full lifecycle data, only requires initial cycling data from the target battery to complete model optimization, eliminating the need to wait for the target battery to accumulate a large amount of cycling data, thus significantly shortening the model deployment cycle. It calculates the maximum peak value, partial discharge capacity, maximum voltage time point, and the time point when the current transitions from charging to discharging for each cycle. These data directly correspond to fundamental electrochemical changes such as the loss of active materials in the battery electrodes and ion conduction efficiency, effectively filtering out interference from non-aging factors and providing high-quality input to the model, ensuring prediction accuracy from the source. By utilizing prediction loss and feature distribution distance, it avoids losing the general aging knowledge of the reference battery when adapting the model to the target battery, and allows for targeted adjustment of model parameters to suit the specific characteristics of the target battery. This achieves the goal of low-cost transfer and high-precision evaluation of a single health assessment model across different target batteries. Attached Figure Description

[0025] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart of a battery health status assessment method according to an embodiment of the present invention. Detailed Implementation

[0027] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0028] This invention provides a method for assessing battery health status, specifically as follows: Figure 1 As shown, it includes:

[0029] Considering the significant differences in cross-domain feature distribution among different batteries, a transfer learning mechanism is used to narrow the distribution gap between the source domain (reference battery) and the target domain (target battery under test), thereby completing the lithium-ion battery health status estimation and enabling embedded deployment. In this embodiment of the invention, the source domain uses publicly available lithium-ion battery data from Toyota Research Institute, and the target domain is sampled from publicly available lithium-ion battery data from the University of Maryland.

[0030] The overall process of this invention embodiment is executed in the following six steps: First, the voltage, current, temperature, and other data collected by the sensors are preprocessed, including the handling of missing values, outliers, and noise; Second, features of the battery are extracted from the dataset; Third, the extracted features are adaptively fused using a CNN model to obtain the optimal weights and biases for estimating the battery's health status; Fourth, the dataset is divided into source and target domains, and a GRU model is used for transfer learning to estimate the health status of the lithium-ion battery; Fifth, the prediction results are evaluated and error analyzed using different metrics to verify the accuracy of the prediction model; Sixth, the trained network structure and test data are exported for efficient embedded deployment.

[0031] A cyclic charge-discharge test was performed on the reference battery. At a fixed sampling interval, the charging current and charging voltage with timestamps during the charging phase and the discharging current and discharging voltage with timestamps during the discharging phase were collected synchronously during each charge-discharge cycle. The battery health status (SOH) true label corresponding to each cycle was also collected synchronously.

[0032] A cyclic charge-discharge test was performed on a reference battery. Voltage V, current I, and corresponding timestamp t were synchronously collected during each charge-discharge cycle according to a preset sampling interval. The discharge capacity Q for each discharge cycle was calculated by integrating the current I during the discharge phase over time. Based on the collected voltage V and the calculated discharge capacity Q, a voltage-capacity VQ curve was plotted for each cycle. The voltage difference between adjacent sampling points was used to... Difference with discharge capacity The differential capacity curve (IC curve) is obtained through numerical differentiation calculation. The battery health status label is the ratio of the actual total discharge capacity of the current cycle to the initial total discharge capacity (nominal capacity) of the battery.

[0033] Feature extraction of lithium-ion batteries includes the following steps:

[0034] (1) Obtain the complete voltage, current and discharge capacity of the lithium-ion battery during charging and discharging. Calculate the differential curve based on the discharge voltage and discharge capacity. The calculation formula is as follows:

[0035]

[0036] In the formula, Q represents the battery discharge capacity, and V represents the discharge voltage.

[0037] (2) Four features were extracted: the peak value of the IC curve F1, the area under the peak of the IC curve F2, the time when the voltage reaches its maximum value during charging F3, and the time when the current begins to discharge during charging and discharging F4, for subsequent use. Specifically, based on the discharge current and discharge voltage, the voltage-discharge capacity curve of the discharge stage was plotted; the differential capacity curve was calculated using the voltage-discharge capacity curve, and the maximum peak value of the differential capacity curve F1 was determined; the maximum characteristic peak region was determined based on the maximum peak value, and the partial discharge capacity of the maximum characteristic peak region F2 was calculated; the maximum voltage time point of the charging voltage F3 was determined; and the time point when the current switched from charging to discharging was determined based on the charging current and the discharging current F4.

[0038] (3) The four features obtained are adaptively fused using a CNN neural network model to obtain the optimal weights and biases for subsequent health status estimation. Specifically, the maximum peak value, partial discharge capacity, maximum voltage time point, and current transition from charging to discharging time point of each cycle are fused according to weights to obtain multiple sets of fused feature vectors.

[0039] A reference prediction model is trained using the multiple sets of fused feature vectors, and the predicted state of health (SOH) value of the reference battery is output. The prediction loss between the predicted SOH value of the reference battery and the real label is calculated. The fused feature vector of the current charge and discharge data of the target battery is obtained, and the feature distribution distance between the fused feature vectors of the target battery and the reference battery is calculated. The parameters of the reference prediction model are adjusted using the prediction loss and the feature distribution distance to obtain an optimized prediction model. Based on the optimized prediction model, the SOH value of the target battery is evaluated.

[0040] The fourth and fifth steps of prediction and evaluation specifically include the following steps:

[0041] (1) The dataset with a capacity degradation of up to 60% is divided into the source domain, and the dataset with a capacity degradation of up to 80% is divided into the target domain.

[0042] (2) Use the MMD method to reduce the distribution gap between the source domain and the target domain, and use the GRU neural network model to perform transfer learning from the source domain to the target domain to complete the estimation of the health status of lithium-ion batteries.

[0043] (3) Use different evaluation indicators, such as mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R2), to evaluate the prediction results.

[0044] In the transfer learning process for battery SOH assessment, the following steps are taken: First, the full lifecycle data of a reference battery is used. This full lifecycle data includes fused feature vectors from multiple charge-discharge cycles and corresponding SOH labels. A reference prediction model is trained to understand the correlation between features and SOH, enabling it to identify general aging patterns. Next, for the target battery with scarce initial data and no SOH labels, a small number of fused feature vectors from a few cycles are extracted. The distribution distance between these feature vectors and the reference battery's feature vectors is calculated using the MMD algorithm to quantify the differences in their characteristics. Then, a dual loss function is constructed, fusing the SOH prediction error of the reference model for the reference battery (ensuring the preservation of general aging knowledge) and the aforementioned feature distribution distance (driving adaptation to target battery characteristics) with preset weights to form the total loss. Finally, the backpropagation algorithm iteratively adjusts the parameters of the reference model with the goal of minimizing the total loss. This allows the model to gradually adapt to the specific aging characteristics of the target battery without losing the core patterns learned from the reference battery, ultimately resulting in an optimized prediction model. This achieves the goal of accurately predicting the SOH of a target battery with only a small amount of initial data, solving the problems of insufficient initial data for the target battery and low accuracy of directly trained models. The Feature Distribution Distance (MMD) loss value is calculated using the RBF kernel function for both discrete and continuous sample scenarios. The formula for calculating the loss value for discrete samples is as follows:

[0045]

[0046] in, and j For sample index, For reference dataset number A fused feature vector, M To reference the total number of fused feature vectors in the dataset, For the initial feature set of the target battery, the first j A fused feature vector, N The total number of feature vectors fused from the initial feature set of the target. It is a high-dimensional mapping function of the RBF kernel function. The reference battery feature vector is the mean vector in the high-dimensional space. It is the mean vector of the target battery feature vector in the high-dimensional space.

[0047] The formula for calculating continuous samples is:

[0048] MMD ( P , Q )=

[0049] Among them, MMD ( P , Q ) represents the distribution P With distribution Q The maximum mean difference between them P To reference the probability distribution of the fused feature vector of the battery, Q Let be the probability distribution of the fused feature vector of the target battery. Indicates that the pair follows a distribution P random variables x Find the expected value. Indicates that the pair follows a distribution Q random variables y Find the expected value. x To conform to the characteristic distribution of the reference cell P Let y be a random variable that follows the characteristic distribution of the target battery. Q random variables, It is a high-dimensional mapping function of the RBF kernel function.

[0050] The specific steps for deployment on the embedded platform in step six are as follows:

[0051] (1) The trained GRU model is simplified and the parameters are quantized. The floating-point weight parameters in the model are mapped in a fixed-point format to reduce memory usage and computational complexity.

[0052] (2) Subsequently, the forward inference logic of GRU (including gating mechanism, state update formula and activation function) is gradually converted into C language implementation and encapsulated into callable inference function to achieve efficient calculation of SOH estimation.

[0053] (3) Use the timer module to measure the model inference delay and verify its running efficiency and real-time performance on the embedded platform.

[0054] Based on the above technical solutions and the practical problems they solve, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0055] The transfer learning method proposed in this invention can overcome the limitations of existing technologies to a certain extent, improve the accuracy of lithium-ion battery health status estimation, and achieve efficient deployment on embedded DSP platforms.

[0056] To address the challenges of high data complexity, weak cross-battery generalization ability, and difficult embedded deployment in lithium-ion battery health status estimation, this invention proposes a transfer learning method that combines convolutional neural networks (CNNs) and gated recurrent neural networks (GRUs), along with a complete technical system from data preprocessing to model deployment. This approach innovatively improves two key aspects: feature fusion and cross-domain modeling, significantly enhancing the accuracy of health status prediction, the adaptability of model transfer, and the practicality of system deployment.

[0057] This invention first cleans and repairs the raw sensor data such as voltage, current, and temperature to ensure data quality. Then, key representation indicators are extracted through feature engineering, and CNNs are used to achieve adaptive weighted fusion of features, improving the multi-dimensional data representation capability. Based on this, a GRU transfer learning model is constructed to effectively transfer knowledge between the source and target domains, solving the problem of decreased predictive ability of traditional methods under different battery types. This invention also introduces an error evaluation mechanism to evaluate model performance from multiple perspectives, ensuring the stability and reliability of the results.

[0058] In terms of embedded implementation, this invention exports the trained network structure and parameters to achieve real-time SOH estimation under low power consumption conditions, providing highly reliable health monitoring capabilities for scenarios such as vehicle-mounted and power storage.

[0059] This invention proposes a lithium-ion battery health state estimation method based on transfer learning. By combining CNN and GRU, it fully leverages the expressive power of CNN in feature fusion and the advantages of GRU in processing time-series data and remembering long-term dependencies, effectively modeling the complex dynamic changes during battery operation. Compared with traditional methods that rely on fixed features or empirical parameters for prediction, this invention significantly outperforms methods based on simple statistical regression or non-transfer neural network models in terms of prediction accuracy. Furthermore, its embedded deployment scheme facilitates real-time operation at the edge in resource-constrained environments such as vehicle BMS, power storage systems, and portable devices, demonstrating strong engineering feasibility.

[0060] Therefore, the lithium battery health state estimation method based on transfer learning proposed in this invention not only has innovative algorithm architecture and practical engineering implementation, but also has cross-scenario universality and wide application promotion value, demonstrating strong comprehensive technical advantages. This method is adaptable to multiple types of batteries and complex scenario solutions and is not limited to specific types of batteries (such as ternary lithium batteries, lithium iron phosphate batteries, lithium cobalt oxide batteries, etc.). It only requires changing the reference battery data of the corresponding type.

[0061] This invention enables battery health system management by assessing the health status of lithium-ion batteries. It helps users understand battery status in real time, take timely measures to extend battery life, reduce production problems caused by battery aging, lower production costs, and improve the reliability and stability of production equipment. In fields ranging from consumer electronics to energy storage, this invention manages and ensures the high efficiency and optimal performance of these systems, thereby guaranteeing the reliability and safety of these lithium-ion battery systems.

[0062] The technical solution of this invention adopts a more systematic feature extraction and fusion strategy, fully considering the multi-dimensional signal features generated by the battery throughout the charging and discharging process, including key parameters such as terminal voltage, current, and temperature. In the feature processing stage, a convolutional neural network (CNN) is used to efficiently fuse the original features and extract potential correlation patterns. In the modeling stage, a gated recurrent unit (GRU) is introduced and combined with a transfer learning mechanism to effectively improve the model's adaptability to cross-battery data and its prediction accuracy. This comprehensive method not only enhances the model's ability to perceive battery degradation features but also significantly improves the accuracy and generalization performance of health status estimation. Furthermore, this invention has excellent embedded implementation capabilities and can be integrated into various battery management systems to achieve intelligent management and efficient prediction of lithium-ion battery health status in multiple scenarios, providing important technical support for realizing full lifecycle battery management.

[0063] Based on the same inventive concept, the present invention also provides a battery health status assessment system, comprising:

[0064] The data acquisition module is used to perform cyclic charge-discharge tests on the reference battery. It synchronously acquires the timestamped charging current and charging voltage during the charging phase and the timestamped discharging current and discharging voltage during the discharging phase at fixed sampling intervals. It also synchronously acquires the battery health status (SOH) true label for each cycle.

[0065] The feature fusion module is used to plot the voltage-discharge capacity curve of the discharge stage based on the discharge current and discharge voltage; calculate the differential capacity curve using the voltage-discharge capacity curve and determine the maximum peak value of the differential capacity curve; determine the maximum characteristic peak region based on the maximum peak value and calculate the partial discharge capacity of the maximum characteristic peak region; determine the maximum voltage time point of the charging voltage; determine the time point when the current transitions from charging to discharging based on the charging current and discharging current; and fuse the maximum peak value, partial discharge capacity, maximum voltage time point, and the time point when the current transitions from charging to discharging in each cycle according to weights to obtain multiple sets of fused feature vectors.

[0066] The model training module is used to train a reference prediction model using the multiple sets of fused feature vectors and output the SOH prediction value of the reference battery; calculate the prediction loss between the SOH prediction value of the reference battery and the real label; obtain the fused feature vector of the current charge and discharge data of the target battery under test, and calculate the feature distribution distance between the fused feature vectors of the target battery under test and the reference battery; adjust the parameters of the reference prediction model using the prediction loss and the feature distribution distance to obtain an optimized prediction model; and evaluate the SOH value of the target battery under test based on the optimized prediction model.

[0067] This invention also provides a computer device. At the hardware level, the computer device includes a processor, an internal bus, a network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the battery health status assessment method provided above.

[0068] The present invention also provides a computer-readable storage medium storing a computer program that can be used to execute the battery health status assessment method provided above.

[0069] Specific limitations regarding the battery health status assessment method calculation system can be found in the limitations of the battery health status assessment method described above, and will not be repeated here. Each module in the aforementioned battery health status assessment system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0070] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. Furthermore, the above embodiments only illustrate several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for assessing battery health status, characterized in that, include: A cyclic charge-discharge test was performed on the reference battery. At a fixed sampling interval, the charging current and charging voltage with timestamps during the charging phase and the discharging current and discharging voltage with timestamps during the discharging phase were collected synchronously. Synchronously collect the battery health status (SOH) real label for each cycle; Based on the discharge current and discharge voltage, plot the voltage-discharge capacity curve during the discharge stage; The differential capacity curve is calculated using the voltage-discharge capacity curve, and the maximum peak value of the differential capacity curve is determined. The region of maximum characteristic peak is determined based on the maximum peak value, and the partial discharge capacity of the region of maximum characteristic peak is calculated. Determine the time point when the charging voltage reaches its maximum value; determine the time point when the current transitions from charging to discharging based on the charging and discharging currents; and fuse the maximum peak value, partial discharge capacity, maximum voltage time point, and current transition time point of each cycle according to weights to obtain multiple sets of fused feature vectors. The reference prediction model is trained using the multiple sets of fused feature vectors, and the predicted SOH value of the reference battery is output. Calculate the predicted SOH value of the reference battery and the prediction loss of the real label; obtain the fusion feature vector of the current charge and discharge data of the target battery, and calculate the feature distribution distance between the fusion feature vectors of the target battery and the reference battery; construct a double loss function based on the prediction loss and the feature distribution distance, and iteratively adjust the weights and biases of the reference prediction model by minimizing the double loss function. When the value of the double loss function is less than a set threshold, the optimized prediction model is obtained. Based on the optimized prediction model, the State of Health (SOH) value of the target battery under test is evaluated.

2. The battery health status assessment method according to claim 1, characterized in that, The maximum peak value, partial discharge capacity, maximum voltage time point, and the time point when the current transitions from charging to discharging are fused according to weights using a CNN network to obtain a fused feature vector.

3. The battery health status assessment method according to claim 1, characterized in that, The reference prediction model is a gated recurrent unit (GRU) network; This includes evaluating the prediction results of GRU networks using mean square error, root mean square error, mean absolute error, and coefficient of determination.

4. The battery health status assessment method according to claim 1, characterized in that, The feature distribution distance between the target battery under test and the reference battery is calculated using the MMD algorithm.

5. The battery health status assessment method according to claim 1, characterized in that, The step of plotting the voltage-discharge capacity curve for the discharge stage based on the discharge current and discharge voltage includes: integrating the discharge current of the discharge stage over time to obtain the discharge capacity, and plotting the voltage-discharge capacity curve for the discharge stage using the voltage and discharge capacity of the discharge stage.

6. The battery health status assessment method according to claim 1, characterized in that, The step of determining the maximum characteristic peak region based on the maximum peak value and calculating the partial discharge capacity of the maximum characteristic peak region specifically includes: determining the minimum points adjacent to the left and right of the maximum peak value, and calculating the area of ​​the closed region enclosed by connecting the maximum peak value and the adjacent minimum points; the area of ​​the closed region is the partial discharge capacity of the maximum characteristic peak region; the partial discharge capacity represents the total capacity change within the voltage range corresponding to the maximum characteristic peak.

7. The battery health status assessment method according to claim 1, characterized in that, Before acquiring the fusion feature vector of the current charge and discharge data of the target battery under test and calculating the feature distribution distance between the fusion feature vectors of the target battery under test and the reference battery, the method further includes constructing historical cyclic charge and discharge data of the target battery under test; the historical cyclic charge and discharge data includes the charging current and charging voltage with timestamps for the charging stage of N rounds of charge and discharge of the target battery under test, and the discharging current and discharging voltage with timestamps for the discharging stage; the reference prediction model is optimized using the feature set of the target battery under test.

8. A battery health status assessment system, characterized in that, include: The data acquisition module is used to perform cyclic charge-discharge tests on the reference battery. It synchronously acquires the timestamped charging current and charging voltage during the charging phase and the timestamped discharging current and discharging voltage during the discharging phase of each charge-discharge cycle at a fixed sampling interval. Synchronously collect the battery health status (SOH) real label for each cycle; The feature fusion module is used to plot the voltage-discharge capacity curve during the discharge phase based on the discharge current and discharge voltage. The differential capacity curve is calculated using the voltage-discharge capacity curve, and the maximum peak value of the differential capacity curve is determined. The region of maximum characteristic peak is determined based on the maximum peak value, and the partial discharge capacity of the region of maximum characteristic peak is calculated. Determine the time point when the charging voltage reaches its maximum value; determine the time point when the current transitions from charging to discharging based on the charging and discharging currents; and fuse the maximum peak value, partial discharge capacity, maximum voltage time point, and current transition time point of each cycle according to weights to obtain multiple sets of fused feature vectors. The model training module is used to train a reference prediction model using the multiple sets of fused feature vectors and output the SOH prediction value of the reference battery. Calculate the predicted SOH value of the reference battery and the prediction loss of the actual label; obtain the fused feature vector of the current charge and discharge data of the target battery, and calculate the feature distribution distance between the fused feature vectors of the target battery and the reference battery; construct a double loss function based on the prediction loss and the feature distribution distance, and iteratively adjust the weights and biases of the reference prediction model by minimizing the double loss function. When the value of the double loss function is less than a set threshold, the optimized prediction model is obtained; evaluate the SOH value of the target battery based on the optimized prediction model.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is loaded by the processor, it is able to perform the steps of the method according to any one of claims 1 to 7.

Images