Method for estimating state of health of power battery of battery swap station based on small data sample

Abstract

The invention discloses a battery swap station power battery health state estimation method based on a small data sample, and the method comprises the steps: obtaining the use data of a battery swap station power battery, carrying out the analysis to obtain a temperature feature value, a capacity increment and a charging time feature value, carrying out the analysis to obtain a DTV and IC curve, processing the features extracted from the original time sequence data through a CNN through BiLSTM, and carrying out the calculation of the health state of the battery swap station power battery. After an LSTM layer, applying an attention mechanism to dynamically select a related hidden state in a whole time sequence, using transfer learning to solve the problem of limited data in a target domain, and using Bayesian optimization to optimize model hyper-parameters in a pre-training stage with sufficient data; acquiring data of the power battery, analyzing to obtain a discharge characteristic value, calculating to obtain a temperature characteristic value, a capacity increment and a charging time characteristic value of the power battery, analyzing to obtain an abnormal data proportion of the power battery, analyzing a charging and discharging condition of the power battery to obtain a charging and discharging state value, and comprehensively analyzing the data to obtain a health characteristic value.

Description

Technical Field

[0001] The present invention relates to the technical field of battery health assessment, and in particular to a method for estimating the health status of power batteries in a battery swap station based on small data samples. Background Art

[0002] With the continuous advancement of related technologies, the electrification of automobiles has formed a booming trend. The electrification of automobiles has also become one of the key ways to help achieve carbon neutrality. However, the problem of slow charging still seriously restricts the practicality of electric vehicles.

[0003] Battery swapping technology is an effective solution to the problem of slow charging. However, for automotive power batteries that use battery swapping technology, a battery often circulates in different battery swapping stations and vehicles, and its usage data is often very scattered, making it difficult to obtain relatively large amounts of complete data for battery state of health SOH estimation. The generalization of battery state of health estimation methods is also one of the issues that need to be improved. The prediction accuracy of existing battery state of health prediction methods is heavily dependent on the underlying battery degradation model, and it is difficult to take into account both the model prediction accuracy and the generalization under different working conditions. Data-driven methods are non-mechanical and non-parametric, and due to their powerful ability to handle nonlinear mappings, they have higher modeling capabilities than model-based methods. Flexibility; Long Short-Term Memory Model Deep learning models with powerful data processing capabilities are also used for battery health status estimation. Existing methods usually require more than 25% of the data of the entire battery life for parameter estimation and model training to achieve accurate capacity trajectory prediction. Machine learning usually assumes that the data of the training set and the test set have the same distribution, but the on-board power battery may experience a variety of complex and changeable operating conditions, including different temperatures, charge and discharge rates, ambient humidity, and frequent charge and discharge cycles. These conditions will significantly affect the health status and performance of the battery, resulting in a decrease in prediction accuracy. Therefore, there is a lack of a prediction model that can adapt to multiple operating conditions with a small amount of data support and has strong generalization capabilities.

[0004] To this end, the present invention provides a method for estimating the health status of power batteries in a battery swap station based on small data samples. Summary of the invention

[0005] The object of the present invention is to provide a method for estimating the health status of power batteries in a battery swap station based on small data samples, so as to solve at least one of the above-mentioned problems of the prior art.

[0006] In a first aspect, the present invention provides a method for estimating the health status of a power battery in a battery swap station based on a small data sample, comprising the following steps: By acquiring the usage data of the power battery in the battery swap station, the temperature characteristic value, capacity increment and charging time characteristic value are analyzed and obtained, and the DTV and IC curves are analyzed; By using BiLSTM to process the features extracted by CNN from the original time series data, an attention mechanism is applied after the LSTM layer to dynamically select relevant hidden states in the entire time series; We chose to use transfer learning to solve the problem of limited data in the target domain. In the pre-training stage with sufficient data, we used Bayesian optimization to optimize the model hyperparameters and develop neural network models. The discharge characteristic value is obtained by analyzing the DTV and IC curves, and the temperature characteristic value, capacity increment and charging time characteristic value of the power battery are calculated and analyzed to obtain the abnormal data ratio of the power battery; Analyze the charging and discharging conditions of the power battery to obtain the charging and discharging status value, comprehensively analyze the data to obtain the health characteristic value, and perform classification analysis.

[0007] Beneficial effects of the present invention: 1. A multi-feature analysis method combining DTV, IC, and CVT is used to more fully explore the characteristics that characterize battery aging in order to obtain more accurate SOH prediction results. A network is constructed by combining CNN-BiLSTM, attention mechanism, Bayesian optimization and other methods to improve the efficiency and effectiveness of the SOH prediction framework for battery data learning. The transfer learning method is used, and only 10% of the data is needed after sufficient pre-training to achieve accurate prediction, ensuring the generalization of the SOH prediction model and realizing prediction using small sample data. It is suitable for situations where the on-board power battery data of battery swap stations is scattered and missing, enhancing the flexibility and adaptability of the model.

[0008] 2. Simultaneously monitor the temperature, voltage and current of the power battery to ensure the integrity and accuracy of the data, integrate the discharge characteristic value FT and the charging time deviation ratio PC, and obtain the overall charge and discharge performance score. Based on the abnormal data ratio YC and the charge and discharge state value CF, obtain a comprehensive indicator to measure the battery health. According to whether the health characteristic value JK exceeds the preset threshold JK 0 , automatically distinguishes between two different aging modes, making it easier to distinguish and understand, and supports machine learning algorithms to build a more accurate power battery SOH prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 is a flowchart of the steps of a method for estimating the health status of a power battery in a battery swap station based on a small data sample provided in Embodiment 1 of the present invention; Figure 2 is a flowchart of the steps of a method for estimating the health status of a power battery in a battery swap station based on a small data sample provided in Embodiment 2 of the present invention; Figure 3 It is a structural schematic diagram of a computer device of a method for estimating the health status of a power battery in a battery swap station based on a small data sample provided in Embodiment 3 of the present invention; Figure Notes: 3 - Computer device, 301 - Processor, 302 - Memory, 303 - Computer program; Figure 4 It is a power battery data change diagram of the power battery health status estimation method of the battery swap station based on a small data sample provided in Example 1 of the present invention; Figure 5 It is a schematic diagram of transfer learning of a method for estimating the health status of a power battery in a battery swap station based on a small data sample provided in Embodiment 1 of the present invention; Figure 6 It is a statistical graph of correlation coefficients of the method for estimating the health status of power batteries in a battery swap station based on a small data sample provided in Example 1 of the present invention; Figure 7 is a prediction structure diagram of a method for estimating the health status of a power battery in a battery swap station based on a small data sample provided in Example 1 of the present invention; Figure 8 It is a statistical graph of lithium iron phosphate battery data of the method for estimating the health status of power batteries in a battery swap station based on a small data sample provided in Example 1 of the present invention; Fig. 9 It is a prediction performance diagram of the method for estimating the health status of power batteries in a battery swap station based on a small data sample provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0011] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present invention.

[0012] Example 1

[0013] Figure 1A flow chart of a method for estimating the health status of power batteries in a battery swap station based on small data samples is provided in Example 1 of the present invention. The embodiment of the present invention can be applied to estimating the health status of power batteries in a battery swap station based on small data samples. The method for estimating the health status of power batteries in a battery swap station based on small data samples can be executed by a system for estimating the health status of power batteries in a battery swap station based on small data samples. The system for estimating the health status of power batteries in a battery swap station based on small data samples can be implemented by software and / or hardware, and the embodiment of the present invention does not limit this.

[0014] like Figure 1 and Figure 4 As shown, the method for estimating the health status of a power battery in a battery swap station based on a small data sample provided by an embodiment of the present invention specifically includes the following steps: Step 1: Obtain the usage data of the power battery in the battery swap station, use differential thermovoltammetry to analyze the temperature characteristic value, capacity increment and charging time characteristic value, and analyze to obtain the DTV and IC curves; It should be noted that the usage data includes but is not limited to battery service time, battery charging current, battery charging voltage, battery charging capacity and battery charging temperature; In some embodiments, lithium batteries exhibit various performance characteristics under complex and changeable working conditions, so a single feature often cannot fully reflect its true state and health. Therefore, the present invention uses a multi-feature analysis method to mine the internal information of the battery from multiple angles, including but not limited to temperature, voltage, and current characteristics, to improve the generalization ability of the model and enable the model to better adapt to different usage scenarios; Differential thermal voltammetry (DTV) is used to mine temperature features. The calculation formula is as follows:

[0015] Where T is the battery temperature at time t, V is the voltage, and t is the sampling time; As an important characteristic analysis method, the capacity increment IC of the power battery is calculated as another characteristic analysis method, through the formula

[0016] Get the capacity increment IC, where Q represents the discharge capacity, I represents the current, V represents the voltage, and t represents the time; In the process of calculating the DTV and IC curves, the measured values ​​of temperature, current and voltage are accompanied by noise, which will produce undesirable changes in the DTV and IC curves, thereby affecting the extraction of features; therefore, the present invention uses the Savitzky-Golay filtering method to filter the IC curve; Get the 3.8V-3.9V constant current charging time through the formula

[0017] The charging time characteristic value is calculated, where CTV 1 It indicates the charging time for the power battery charging voltage to reach 3.8V, CTV 2 It indicates the charging time for the power battery charging voltage to reach 3.9V. CTV is defined as the charging time in the current voltage range. It should be noted that, considering the long duration of constant current charging and the time cost of data collection, the 3.8V-3.9V charging time window corresponds to the middle of charging, when the voltage change is relatively stable and can effectively reflect the time difference between cycles. Therefore, the present invention selects CTV 3.9V -CTV 3.8V As an effective feature of battery degradation; Step 2: Improve the accuracy of lithium-ion battery capacity trajectory prediction by processing the features extracted by CNN from the original time series data using BiLSTM. Apply the attention mechanism after the LSTM layer to dynamically select relevant hidden states in the entire time series. Choose to use transfer learning to solve the problem of limited data in the target domain. In the pre-training stage with sufficient data, use Bayesian optimization to optimize the model hyperparameters and develop the neural network model. Use CNN to extract features from raw time series data. BiLSTM consists of two layers of LSTM, where the predicted system state refers to the output sequence from the bidirectional incoming LSTM layer, and the prediction results are merged and assigned to the next LSTM layer; After the second LSTM layer, the final prediction is determined by both forward and backward propagation. BiLSTM is used to process the features extracted by CNN from the original time series data to improve the accuracy of lithium-ion battery capacity trajectory prediction; Focusing on relevant states across time series by applying an attention mechanism after the LSTM layer to dynamically select relevant hidden states throughout the time series; For the hidden layer output H = [h1, h2,…,hT] of LSTM, the formula is:

[0018] Calculate the attention weight , where T represents the number of hidden layer outputs, represents the state vector at time step t, and They respectively represent and The exponential function of , and , represents the weight matrix to be learned; After obtaining the weights, the weighted sum is calculated by the formula Get c, and then use the formula through the fully connected layer Get the final result, where c represents the relevant hidden state, represents the weight matrix, and b represents the bias vector; Solve the problem of limited data in the target domain by using transfer learning, which includes two main steps: a pre-training phase using source domain data and a fine-tuning phase using target domain data; The goal of the fine-tuning phase is to adjust the parameters of the model to better adapt to the specific degradation mode of the target domain; the transfer learning fine-tuning strategy selected by the present invention is as follows: Figure 5 As shown, in the fine-tuning stage, only the parameters of BiLSTM are opened for training; It should be noted that BiLSTM, as the core of the framework's time series prediction, needs to be retrained under different battery operating conditions and different battery chemical compositions. In the case of small samples, it can adapt to the battery degradation mode of the target domain. At the same time, only the parameter training of the BiLSTM block is open, which reduces the complexity of model training in the fine-tuning stage. In the pre-training stage with sufficient data, Bayesian optimization is used to optimize model hyperparameters; By using Gaussian process to model the target, the parameter range of Bayesian optimization is shown in the following table (1): Table (1) shows the parameter range of Bayesian optimization corresponding to different hyperparameters;

[0019] Use NASA’s public battery dataset for pre-training, and then fine-tune the network using only the first 10% of the target battery data to predict the target battery SOH; For example, the Pearson correlation coefficient is widely used as a linear correlation indicator, and its definition for variables X and Y is:

[0020] in, Expressed as the Pearson correlation coefficient, and denote the average values ​​of the eigenvalue and battery capacity value, respectively; n denotes the sample size; the correlation coefficient ranges from -1 to 1, where values ​​close to 1 or -1 indicate a strong relationship, and values ​​close to 0 indicate a weak relationship; Figure 6 The correlation coefficient between the features extracted by the multi-feature extraction method and the battery capacity is shown; 7 features with a correlation coefficient above 0.8 are finally selected as the feature input for predicting the battery capacity; the small sample prediction performance, in the verification stage of the present invention, the model is first pre-trained, and then the model is fine-tuned using the first 10% of the target monomer data in the fine-tuning stage. The prediction structure is as follows Figure 7As shown in Figure 2, the predicted results of SOH of ternary lithium battery under different temperature conditions are shown in Figure 2. Figure 7 In the figure, Figure (a) corresponds to the temperature condition of -20°C, Figure (b) corresponds to the temperature condition of -10°C, Figure (c) corresponds to the temperature condition of 0°C, and Figure (d) corresponds to the temperature condition of 25°C. It can be seen that under different small sample data predictions, the present invention has achieved excellent prediction results; Figure 8 The results are obtained on the lithium iron phosphate battery dataset; when the battery electrode chemical composition and operating conditions are different, the model still achieves an RMSE error of 0.025Ah, which further proves the generalization of the proposed method; the prediction performance of different features, in order to fully verify the effectiveness of the proposed multi-feature fusion method and different feature extraction strategies, a series of comparative experiments are designed to analyze the performance of these methods on the experimental dataset, such as Fig. 9 As shown in Figure 2, Table (2) further summarizes the key values ​​of the experimental results. The experimental results show that although the single feature extraction method can capture some aging patterns, its performance is limited when dealing with complex working conditions and diverse data features. The proposed multi-feature fusion method significantly improves the overall prediction ability of the model and shows the best prediction effect under all working conditions. This shows that the multi-feature fusion method can effectively combine the advantages of different features, more comprehensively characterize the battery aging characteristics, and provide a more reliable solution for complex prediction tasks.

[0021] Table (2) shows the experimental results corresponding to different working temperatures;

[0022] The technical solution of the present invention is: the embodiment of the present invention obtains the usage data of the power battery of the battery swap station, uses differential thermovoltammetry to analyze the temperature characteristic value, capacity increment and charging time characteristic value, analyzes to obtain the DTV and IC curves, and uses BiLSTM to process the features extracted from the original time series data by CNN to improve the accuracy of the lithium-ion battery capacity trajectory prediction, and applies the attention mechanism after the LSTM layer to dynamically select relevant hidden states in the entire time series, and chooses to use transfer learning to solve the problem of limited data in the target domain. In the pre-training stage with sufficient data, Bayesian optimization is used to optimize the model hyperparameters; the prediction of the battery health status under small sample data is realized, and the generalization of the method is ensured, which is suitable for the management of on-board power batteries in battery swap stations. In the case of incomplete battery usage data in battery swap stations, accurate prediction is achieved to ensure the safe and reliable operation of battery swap stations, and to help further promote the electrification of automobiles and the realization of carbon neutrality. The accuracy of lithium-ion battery capacity trajectory prediction is improved through multi-feature analysis and complex machine learning models, and the flexibility and adaptability of the model are enhanced.

[0023] Example 2 like Figure 2 As shown, based on the acquisition of charging maintenance data of the power battery in the battery swap station in Example 1, the method for estimating the health status of the power battery in the battery swap station based on a small data sample provided by the embodiment of the present invention specifically includes the following steps: Step 3: Obtain power battery data, analyze and process to generate a discharge time curve, analyze to obtain discharge characteristic values, calculate to obtain temperature characteristic values, capacity increment and charging time characteristic values ​​of the power battery, analyze the abnormal data ratio of the power battery, analyze the charging and discharging conditions of the power battery to obtain the charging and discharging status values, comprehensively analyze the data to obtain health characteristic values, and perform classification analysis; It should be noted that the preset voltage is the power battery voltage set by technicians in the field of the present invention to obtain more characteristic data to analyze the charging and discharging status of the power battery. The power battery data includes but is not limited to the duration of continuous discharge to reach the preset voltage, the number of times the power battery is discharged, the temperature, voltage and current of the power battery, etc.; In some embodiments, based on the obtained power battery aging voltage change curve, the duration of continuous discharge of the power battery to reach a preset voltage and the number of times the power battery is discharged are obtained, and an XY two-dimensional coordinate system is established with the number of times the power battery is discharged as the abscissa and the duration of continuous discharge of the power battery to reach the preset voltage as the ordinate to generate a discharge duration and number curve; Set the discharge time curve to , through the formula

[0024] Calculate the curvature , where the obtained curvatures are all non-negative numbers, and the curvatures of all discharge times corresponding points are obtained through the formula

[0025] The sum of the curvatures corresponding to all discharge times is calculated, where n is the total number of kinetic battery discharges; The obtained sum of curvatures is processed by ratio with the corresponding total number of discharges to obtain the average curvature QL of the curve, and the variation range of the discharge time between the discharge times of the power battery is analyzed; Obtain the total number of times the power battery discharges when the power battery continuously discharges to reach a preset voltage and reaches a preset minimum duration, and perform ratio processing on the preset minimum duration and the total number of times the power battery discharges to obtain a discharge loss value SZ; It should be noted that the preset minimum duration is set by professionals in the field of the present invention based on historical experience, and is a standard discharge duration for judging the degree of aging of the power battery when the power battery is continuously discharged to reach a preset voltage; The average curvature QL and discharge loss value SZ obtained by comprehensive analysis are calculated by the formula

[0026] The discharge characteristic value FT is calculated, where a, b and c are preset correlation coefficients, a is 0.137, b is 0.22, and c is 1.724; It should be noted that the discharge characteristic value FT is calculated by the average curvature QL of the curve and the discharge loss value SZ. The smaller the average curvature QL of the curve is, the more stable the change range of the discharge time between the discharge times of the power battery is, the more stable the aging process of the power battery is, and the smaller the discharge loss value SZ is, the more the total number of power battery discharges that reach the preset shortest time when the power battery continuously discharges to the preset voltage is, and the more times the power battery is used before aging. The characteristic performance of the battery discharge process can be obtained through comprehensive analysis; Calculate and obtain the temperature characteristic value and capacity increment of the power battery, compare the obtained temperature characteristic value with a preset temperature characteristic threshold, and obtain the number of temperature characteristic values ​​greater than the temperature characteristic threshold; The number of temperature characteristic values ​​greater than the temperature characteristic threshold is ratioed with the total number of obtained temperature characteristic values ​​to obtain the temperature anomaly ratio; The obtained capacity increment is compared with a preset capacity increment threshold, and the number of capacity increments less than the capacity increment threshold is obtained; The number of capacity increments that are less than the capacity increment threshold is ratioed to the total number of capacity increments obtained to obtain the capacity abnormality ratio; The obtained temperature abnormality ratio and capacity abnormality ratio are weighted summed to obtain the abnormal data ratio YC; Obtain a charging time characteristic value CTV of the power battery, perform a difference processing between the obtained charging time characteristic value CTV and the charging time standard value and take the absolute value to obtain a charging time deviation value, perform a ratio processing between the charging time deviation value and the charging time standard value to obtain a charging time deviation ratio PC; Comprehensive analysis of the discharge characteristic value FT and the charging time deviation ratio PC, through the formula

[0027] The charge and discharge status value CF is calculated, where d and e are preset correlation coefficients, the value of d is 1.127, and the value of e is 1.879; It should be noted that the charge and discharge state value CF is calculated by the discharge characteristic value FT and the charge time deviation ratio PC. The larger the discharge characteristic value FT is, the better the discharge state of the power battery is. The smaller the charge time deviation ratio PC is, the smaller the deviation of the charge time characteristic value CTV relative to the charge time standard value is, which indicates that the charge state of the power battery is better, which is helpful for analyzing the basis of the health state of the power battery. The abnormal data proportion YC and the charge and discharge status value CF are comprehensively analyzed, and the formula

[0028] The health characteristic value JK is calculated, where and is the preset scale factor, The value of is 0.417, The value of is 0.539; The obtained health characteristic value JK and the preset health characteristic threshold JK 0 Compare and analyze to identify reasonable and useful power batteries as data basis; Specifically, if the health characteristic value JK is greater than or equal to the preset health characteristic threshold JK 0 , it indicates that the power battery data is suitable for the model training data of normal aging of power batteries, and the corresponding power battery data is marked as normal aging training data; If the health characteristic value JK is less than the preset health characteristic threshold JK 0 , it indicates that the power battery data is suitable for the training data of the power battery abnormal aging model, and the corresponding power battery data is marked as abnormal aging data; The technical solution of the embodiment of the present invention is: obtaining the duration of continuous discharge of the power battery to reach a preset voltage and the number of times the power battery is discharged, analyzing and processing to generate a curve, analyzing to obtain a discharge characteristic value, obtaining the temperature, voltage and current of the power battery synchronously, calculating to obtain the temperature characteristic value, capacity increment and charging time characteristic value of the power battery, analyzing the abnormal data proportion of the power battery, analyzing the charging and discharging conditions of the power battery to obtain the charging and discharging state value, comprehensively analyzing the data to obtain a health characteristic value, and performing classification analysis; at the same time, monitoring the temperature, voltage and current of the power battery to ensure the integrity and accuracy of the data, integrating the discharge characteristic value FT and the charging time deviation ratio PC to obtain an overall charging and discharging performance score, and obtaining a comprehensive indicator for measuring battery health based on the abnormal data proportion YC and the charging and discharging state value CF, and determining whether the health characteristic value JK exceeds a preset threshold value JK 0 , automatically distinguishes between two different aging modes, making it easier to distinguish and understand, and supports machine learning algorithms to build a more accurate power battery SOH prediction model.

[0029] Example 3

[0030] Reference Figure 3 The embodiment of the present invention further provides a computer device 3, comprising: a memory 302 and a processor 301 and a computer program 303 stored in the memory 302. When the computer program 303 is executed on the processor 301, a method for estimating the health status of a power battery in a battery swap station based on a small data sample as described in any one of the above methods is implemented.

[0031] The computer device 3 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server; the computer device 3 may include, but is not limited to, a processor 301 and a memory 302, which can be understood by those skilled in the art; Figure 3 It is only an example of computer device 3 and does not constitute a limitation on computer device 3. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components, for example, it may also include input and output devices, network access devices, etc.

[0032] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

[0033] In some embodiments, the memory 302 may be an internal storage unit of the computer device 3, such as a hard disk or memory of the computer device 3. In other embodiments, the memory 302 may also be an external storage device of the computer device 3, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device 3; further, the memory 302 may also include both an internal storage unit and an external storage device of the computer device 3. The memory 302 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program; the memory 302 may also be used to temporarily store data that has been output or is to be output.

[0034] Example 4

[0035] An embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for estimating the health status of a power battery in a battery swap station based on a small data sample as described in any one of the above methods is implemented.

[0036] In this embodiment, 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 present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the camera / terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. For example, USB flash drive, mobile hard disk, disk or optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0037] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0038] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0039] In the embodiments disclosed in the present application, it should be understood that the disclosed apparatus / terminal equipment and method can be implemented in other ways. For example, the apparatus / terminal equipment embodiments described above are merely illustrative. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. One point, the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interface, device or unit, which may be electrical, mechanical or other forms.

[0040] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0041] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0042] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A method for estimating the health status of power batteries in a battery swap station based on small data samples, characterized in that: The following steps are involved: By acquiring the usage data of the power battery in the battery swap station, the temperature characteristic value, capacity increment and charging time characteristic value are analyzed and obtained, and the DTV and IC curves are analyzed; By using BiLSTM to process the features extracted by CNN from the original time series data, an attention mechanism is applied after the LSTM layer to dynamically select relevant hidden states in the entire time series; We chose to use transfer learning to solve the problem of limited data in the target domain. In the pre-training stage with sufficient data, we used Bayesian optimization to optimize the model hyperparameters and develop neural network models. The discharge characteristic value is obtained by analyzing the DTV and IC curves, and the temperature characteristic value, capacity increment and charging time characteristic value of the power battery are calculated and analyzed to obtain the abnormal data ratio of the power battery; Analyze the charging and discharging conditions of the power battery to obtain the charging and discharging status value, comprehensively analyze the data to obtain the health characteristic value, and perform classification analysis.

2. The method for estimating the health status of power batteries in a battery swap station based on a small data sample according to claim 1, characterized in that: The specific process of obtaining the temperature characteristic value, capacity increment and charging time characteristic value is as follows: Differential thermovoltammetry (DTV) is used to mine temperature features. The calculation formula is as follows: ; Where T is the battery temperature at time t, V is the voltage, and t is the sampling time; Calculate the capacity increment IC of the power battery as another characteristic analysis method, through the formula: ; Where Q represents the discharge capacity, I represents the current, V represents the voltage, and t represents the time; In the process of calculating the DTV and IC curves, the measured values ​​of temperature, current and voltage are accompanied by noise, which will produce undesirable changes in the DTV and IC curves, thus affecting the extraction of features; the Savitzky-Golay filtering method is used to filter the IC curve; Get the 3.8V-3.9V constant current charging time through the formula The charging time characteristic value is calculated, where CTV1 represents the charging time when the power battery charging voltage reaches 3.8V, CTV2 represents the charging time when the power battery charging voltage reaches 3.9V, and CTV is defined as the charging time in the current voltage range.

3. The method for estimating the health status of power batteries in a battery swap station based on a small data sample according to claim 1, characterized in that: The specific process of processing the features is as follows: Use CNN to extract features from raw time series data. BiLSTM consists of two layers of LSTM, where the predicted system state refers to the output sequence from the bidirectional incoming LSTM layer, and the prediction results are merged and assigned to the next LSTM layer; After the second LSTM layer, the final prediction is determined by both forward and backward propagation. BiLSTM is chosen to process the features extracted by CNN from the original time series data to improve the accuracy of lithium-ion battery capacity trajectory prediction; By applying an attention mechanism after the LSTM layer to dynamically select relevant hidden states in the entire time series, focusing on relevant states across the time series; For the hidden layer output H = [h1, h2,…,hT] of LSTM, the formula is: Calculate the attention weight , where T represents the number of hidden layer outputs, represents the state vector at time step t, and They respectively represent and The exponential function of , and , represents the weight matrix to be learned; After obtaining the weights, the weighted sum is calculated by the formula Get c, and then use the formula through the fully connected layer Get the final result, where c represents the relevant hidden state, represents the weight matrix and b represents the bias vector.

4. The method for estimating the health status of power batteries in a battery swap station based on a small data sample according to claim 1, characterized in that: The specific steps of the transfer learning are: Solve the problem of limited data in the target domain by using transfer learning, which includes two main steps: a pre-training phase using source domain data and a fine-tuning phase using target domain data; The goal of the fine-tuning phase is to adjust the parameters of the model to better adapt to the specific degradation pattern of the target domain; select the transfer learning fine-tuning strategy, and only open the parameters of BiLSTM for training during the fine-tuning phase; In the pre-training stage with sufficient data, Bayesian optimization is used to optimize the model hyperparameters, and the target is modeled by using Gaussian process; The public battery dataset NASA is used for pre-training, and then only the first 10% of the target battery data is used to fine-tune the network to predict the target battery SOH.

5. The method for estimating the health status of power batteries in a battery swap station based on a small data sample according to claim 1, characterized in that: The specific process of obtaining the discharge characteristic value is as follows: The average curvature QL and discharge loss value SZ obtained by comprehensive analysis are calculated by the formula The discharge characteristic value FT is calculated, where a, b and c are all preset correlation coefficients.

6. The method for estimating the health status of power batteries in a battery swap station based on a small data sample according to claim 5, characterized in that: The specific process of obtaining the average curvature of the curve is as follows: Based on the obtained power battery aging voltage change curve, the duration of continuous discharge of the power battery to reach the preset voltage and the number of times the power battery is discharged are obtained, and an XY two-dimensional coordinate system is established with the number of times the power battery is discharged as the horizontal coordinate and the duration of continuous discharge of the power battery to reach the preset voltage as the vertical coordinate to generate a discharge duration and number curve; Set the discharge time curve to , through the formula Calculate the curvature , where the obtained curvatures are all non-negative numbers, and the curvatures of all discharge times corresponding points are obtained through the formula ; The sum of the curvatures corresponding to all discharge times is calculated, where n is the total number of kinetic battery discharges; The obtained sum of curvatures is ratioed with the corresponding total number of discharges to obtain the average curvature QL of the curve, and the variation range of the discharge time between the discharge times of the power battery is analyzed.

7. The method for estimating the health status of power batteries in a battery swap station based on a small data sample according to claim 5, characterized in that: The specific process of obtaining the discharge loss value is as follows: The total number of times the power battery discharges when the power battery continuously discharges to reach a preset voltage and reaches a preset minimum duration is obtained, and the preset shortest duration is ratioed to the total number of times the power battery discharges to obtain a discharge loss value SZ.

8. The method for estimating the health status of power batteries in a battery swap station based on a small data sample according to claim 1, characterized in that: The specific process of obtaining the health characteristic value is as follows: The abnormal data proportion YC and the charge and discharge status value CF are comprehensively analyzed, and the formula The health characteristic value JK is calculated, where and is the preset proportional coefficient; Compare the obtained health characteristic value JK with the preset health characteristic threshold JK0, and analyze and divide reasonable and useful power batteries as data basis; Specifically, if the health characteristic value JK is greater than or equal to the preset health characteristic threshold value JK0, it indicates that the power battery data is suitable for the model training data of normal aging of the power battery, and the corresponding power battery data is marked as normal aging training data; If the health characteristic value JK is less than the preset health characteristic threshold JK0, it indicates that the power battery data is suitable for the power battery abnormal aging model training data, and the corresponding power battery data is marked as abnormal aging data.

9. The method for estimating the health status of power batteries in a battery swap station based on a small data sample according to claim 8, characterized in that: The specific process of obtaining the abnormal data ratio is as follows: Calculate and obtain the temperature characteristic value and capacity increment of the power battery, compare the obtained temperature characteristic value with a preset temperature characteristic threshold, and obtain the number of temperature characteristic values ​​greater than the temperature characteristic threshold; The number of temperature characteristic values ​​greater than the temperature characteristic threshold is ratioed with the total number of obtained temperature characteristic values ​​to obtain the temperature anomaly ratio; The obtained capacity increment is compared with a preset capacity increment threshold, and the number of capacity increments less than the capacity increment threshold is obtained; The number of capacity increments that are less than the capacity increment threshold is ratioed to the total number of capacity increments obtained to obtain the capacity abnormality ratio; The obtained temperature anomaly ratio and capacity anomaly ratio are weightedly summed to obtain the abnormal data ratio YC.

10. The method for estimating the health status of power batteries in a battery swap station based on a small data sample according to claim 8, characterized in that: The specific acquisition process of the charging and discharging status value is as follows: Obtain the charging time characteristic value CTV of the power battery, perform a difference processing between the obtained charging time characteristic value CTV and the charging time standard value and take the absolute value to obtain a charging time deviation value, perform a ratio processing between the charging time deviation value and the charging time standard value to obtain a charging time deviation ratio PC; Comprehensive analysis of the discharge characteristic value FT and the charging time deviation ratio PC, through the formula The charge and discharge state value CF is calculated, where d and e are preset correlation coefficients.

Images