A battery life prediction method, device, storage medium and equipment

Abstract

This application discloses a battery life prediction method, apparatus, storage medium, and device. The method includes: acquiring initial battery characteristic data of a target battery under operating conditions; preprocessing the initial battery characteristic data to obtain target battery characteristic data; performing wavelet transform processing on the target battery characteristic data to obtain effective feature data reflecting the lifespan degradation trend of the target battery; extracting time-series correlation features from the effective feature data to obtain time-series correlation feature data; and predicting the target battery life based on the time-series correlation feature data. This application can improve the accuracy and efficiency of battery life prediction.

Description

Technical Field

[0001] This application relates to the field of battery technology, and more particularly to a battery life prediction method, apparatus, storage medium, and device. Background Technology

[0002] With the development of power tools (such as electric vehicles, electric trains, and electric bicycles), the health status and life prediction of battery systems have become particularly important. The problem of predicting the remaining life of batteries has become one of the research hotspots and challenges in the field of electronic system failure prediction and health management. Achieving accurate estimation of remaining life helps to improve system reliability and has important research and practical value.

[0003] Currently, due to the ever-changing usage conditions of batteries and the influence of numerous factors such as ambient temperature, user habits, and equipment status, the amount of feature data collected for predicting battery life is quite large. Directly using a large amount of feature data to predict battery life results in low prediction efficiency and is easily affected by invalid feature data, leading to low accuracy of the prediction results. Summary of the Invention

[0004] The technical problem to be solved by the embodiments of this application is to provide a battery life prediction method, apparatus, storage medium and device, which can improve the accuracy and efficiency of battery life prediction.

[0005] One embodiment of this application provides a battery life prediction method, including:

[0006] The initial battery characteristic data of the target battery under the operating conditions is obtained, and the initial battery characteristic data is preprocessed to obtain the target battery characteristic data.

[0007] Wavelet transform is performed on the target battery feature data to obtain effective feature data that reflects the life decay trend of the target battery.

[0008] Temporal correlation feature extraction is performed on the effective feature data to obtain temporal correlation feature data;

[0009] Based on time-series correlation feature data, the target battery life is predicted.

[0010] The process involves preprocessing the initial battery feature data to obtain the target battery feature data, including:

[0011] The initial battery feature data is converted to a different data type to obtain battery feature data with the target data type. Battery feature data that is associated with battery life degradation is then selected from the battery feature data with the target data type and used as associated feature data.

[0012] Abnormal feature data in the associated feature data is deleted to obtain processed associated feature data. The sampling features of the processed associated feature data are obtained. The processed associated feature data is then segmented according to the sampling features to obtain associated sub-feature data corresponding to K time intervals; K is a positive integer greater than 1.

[0013] Normalize the associated sub-feature data corresponding to the K time intervals to obtain the standard feature data corresponding to the K time intervals.

[0014] The standard feature data corresponding to the K time intervals are determined as the target battery feature data.

[0015] Specifically, wavelet transform processing is performed on the target battery feature data to obtain effective feature data reflecting the lifespan degradation trend of the target battery, including:

[0016] By using the wavelet transform layer in the target life prediction model, wavelet decomposition is performed on the target battery feature data to obtain the wavelet decomposition coefficients corresponding to the target battery feature data.

[0017] The wavelet decomposition coefficients are denoised by adjusting the threshold according to the target, and the denoised wavelet decomposition coefficients are obtained.

[0018] Wavelet reconstruction is performed on the denoised wavelet decomposition coefficients to obtain effective feature data that reflects the life degradation trend of the target battery.

[0019] Among them, the wavelet decomposition coefficients corresponding to the target battery feature data include frequencies in the first frequency set and frequencies in the second frequency set, with the frequencies in the first frequency set being greater than the frequencies in the second frequency set;

[0020] The wavelet decomposition coefficients are denoised based on the target threshold adjustment to obtain the denoised wavelet decomposition coefficients, including:

[0021] Based on the target, the frequency in the first frequency set is threshold quantized to obtain the quantized first frequency set.

[0022] The frequencies in the first and second frequency sets after quantization are determined as the wavelet decomposition coefficients after denoising.

[0023] Specifically, temporal correlation feature extraction is performed on the effective feature data to obtain temporal correlation feature data, including:

[0024] By using the temporal correlation feature extraction layer in the target lifetime prediction model, the data correlation degree between the effective feature data at time t and the effective feature data at time t-1 is obtained; both the feature data at time t and the feature data at time t-1 are considered effective feature data.

[0025] Based on the data correlation degree, the effective feature data at time t-1 is weighted to obtain the first weighted feature data. The input control parameters in the time-series correlation feature extraction layer are obtained. Based on the input control parameters, the effective feature data at time t is weighted to obtain the second weighted feature data.

[0026] The first weighted feature data and the second weighted feature data are fused to obtain the important feature data in the effective feature data at time t;

[0027] Based on the weight matrix and bias parameters in the temporal correlation feature extraction layer, features are extracted from important feature data to obtain the correlation feature data corresponding to time t.

[0028] Until the associated feature data corresponding to the valid feature data is obtained, the associated feature data corresponding to the valid feature data is determined as the time-series associated feature data.

[0029] The methods also include:

[0030] Acquire the initial life prediction model, sample battery feature data under operating conditions, and the corresponding battery life label for the sample battery; associate the sample battery with the target battery;

[0031] Based on the characteristic data of the sample battery, the initial life prediction model is used to predict the life of the sample battery, and the predicted battery life of the sample battery is obtained.

[0032] Based on the predicted battery life and battery life label, the model parameters in the initial life prediction model are adjusted. When the initial life prediction model with adjusted parameters meets the convergence condition, the initial life prediction model with adjusted parameters is determined as the target life prediction model.

[0033] Specifically, based on the characteristic data of the sample batteries, an initial lifespan prediction model is used to predict the lifespan of the sample batteries, resulting in the predicted battery lifespan. This includes:

[0034] The sample battery feature data is preprocessed using an initial lifetime prediction model to obtain preprocessed sample battery feature data. The preprocessed sample battery feature data is then divided into M time intervals to obtain sample sub-feature data corresponding to each time interval; M is a positive integer greater than 1.

[0035] By using the wavelet transform layer in the initial lifetime prediction model, wavelet transform processing is performed on the sample sub-feature data corresponding to M time intervals respectively to obtain the effective sample feature data corresponding to M time intervals respectively.

[0036] By using the temporal correlation feature extraction layer in the initial life prediction model, temporal correlation features are extracted from the effective feature data of the samples corresponding to M time intervals, and the temporal correlation feature data of the samples corresponding to M time intervals is obtained. Based on the temporal correlation feature data of the samples corresponding to M time intervals, the predicted battery life of the sample battery is predicted.

[0037] The preprocessed sample battery feature data is divided into M time intervals, each corresponding to a sub-feature data point, including:

[0038] Obtain the sampling features of the preprocessed sample battery feature data;

[0039] Based on the sampling characteristics of the preprocessed sample battery feature data, determine the adaptive sliding window corresponding to the preprocessed sample feature data;

[0040] An adaptive sliding window corresponding to the preprocessed sample feature data is used to divide the preprocessed sample battery feature data, resulting in sample sub-feature data corresponding to M time intervals.

[0041] One embodiment of this application provides a battery life prediction device, including:

[0042] The preprocessing module is used to acquire the initial battery characteristic data of the target battery under the operating conditions, and to perform data preprocessing on the initial battery characteristic data to obtain the target battery characteristic data.

[0043] The wavelet transform module is used to perform wavelet transform processing on the target battery feature data to obtain effective feature data that reflects the life decay trend of the target battery.

[0044] The feature extraction module is used to extract temporal correlation features from effective feature data to obtain temporal correlation feature data;

[0045] The first prediction module is used to predict the target battery life of the target battery based on time-series correlated feature data.

[0046] The preprocessing module includes:

[0047] The type conversion unit is used to convert the initial battery feature data into a battery feature data with the target data type, and to filter out the battery feature data that is related to battery life degradation from the battery feature data with the target data type as the associated feature data.

[0048] The deletion unit is used to delete abnormal feature data in the associated feature data to obtain processed associated feature data. It then obtains the sampling features of the processed associated feature data and segments the processed associated feature data according to the sampling features to obtain associated sub-feature data corresponding to K time intervals; K is a positive integer greater than 1.

[0049] The normalization unit is used to normalize the associated sub-feature data corresponding to K time intervals to obtain the standard feature data corresponding to K time intervals.

[0050] The first determining unit is used to determine the standard feature data corresponding to the K time intervals as the target battery feature data.

[0051] The wavelet transform module includes:

[0052] The wavelet decomposition unit is used to perform wavelet decomposition on the target battery feature data through the wavelet transform layer in the target lifetime prediction model to obtain the wavelet decomposition coefficients corresponding to the target battery feature data.

[0053] The noise reduction processing unit is used to adjust the threshold according to the target to perform noise reduction processing on the wavelet decomposition coefficients, and obtain the noise-reduced wavelet decomposition coefficients.

[0054] The wavelet reconstruction unit is used to reconstruct the wavelet decomposition coefficients after noise reduction, so as to obtain effective feature data that reflects the life degradation trend of the target battery.

[0055] Among them, the wavelet decomposition coefficients corresponding to the target battery feature data include frequencies in the first frequency set and frequencies in the second frequency set, with the frequencies in the first frequency set being greater than the frequencies in the second frequency set;

[0056] The noise reduction processing unit is specifically used for:

[0057] Based on the target, the frequency in the first frequency set is threshold quantized to obtain the quantized first frequency set.

[0058] The frequencies in the first and second frequency sets after quantization are determined as the wavelet decomposition coefficients after denoising.

[0059] The feature extraction module includes:

[0060] The acquisition unit is used to obtain the data correlation degree between the effective feature data at time t and the effective feature data at time t-1 through the time-series correlation feature extraction layer in the target lifetime prediction model; both the feature data at time t and the feature data at time t-1 are effective feature data.

[0061] The weighted processing unit is used to perform weighted processing on the effective feature data at time t-1 according to the data correlation degree to obtain the first weighted feature data, obtain the input control parameters in the time-series correlation feature extraction layer, and perform weighted processing on the effective feature data at time t according to the input control parameters to obtain the second weighted feature data.

[0062] The fusion unit is used to fuse the first weighted feature data and the second weighted feature data to obtain the important feature data in the effective feature data at time t;

[0063] The first extraction unit is used to extract features from important feature data based on the weight matrix and bias parameters in the temporal correlation feature extraction layer, so as to obtain the correlation feature data corresponding to time t.

[0064] The second determining unit is used to determine the associated feature data corresponding to the valid feature data as time-series associated feature data until the associated feature data corresponding to the valid feature data is obtained. The battery life prediction device further includes:

[0065] The acquisition module is used to acquire the initial life prediction model, the sample battery feature data under the operating conditions, and the battery life label corresponding to the sample battery; the sample battery is associated with the target battery.

[0066] The second prediction module is used to predict the lifespan of the sample battery based on the sample battery feature data using the initial lifespan prediction model, and obtain the predicted battery lifespan of the sample battery.

[0067] The determination module is used to adjust the model parameters in the initial life prediction model based on the predicted battery life and the battery life label. When the initial life prediction model after parameter adjustment meets the convergence condition, the initial life prediction model after parameter adjustment is determined as the target life prediction model.

[0068] The second prediction model includes:

[0069] The preprocessing unit is used to preprocess the sample battery feature data through the initial lifetime prediction model to obtain preprocessed sample battery feature data. The preprocessed sample battery feature data is then divided to obtain sample sub-feature data corresponding to M time intervals; M is a positive integer greater than 1.

[0070] The wavelet transform unit is used to perform wavelet transform processing on the sample sub-feature data corresponding to M time intervals through the wavelet transform layer in the initial lifetime prediction model, so as to obtain the effective sample feature data corresponding to M time intervals respectively.

[0071] The second extraction unit is used to extract time-series correlation features from the effective feature data of samples corresponding to M time intervals through the time-series correlation feature extraction layer in the initial life prediction model, so as to obtain the time-series correlation feature data of samples corresponding to M time intervals, and predict the predicted battery life of the sample battery based on the time-series correlation feature data of samples corresponding to M time intervals.

[0072] Specifically, the preprocessing unit is used for:

[0073] Obtain the sampling features of the preprocessed sample battery feature data;

[0074] Based on the sampling characteristics of the preprocessed sample battery feature data, determine the adaptive sliding window corresponding to the preprocessed sample feature data;

[0075] An adaptive sliding window corresponding to the preprocessed sample feature data is used to divide the preprocessed sample battery feature data, resulting in sample sub-feature data corresponding to M time intervals.

[0076] One embodiment of this application provides a computer device, including: a processor and a memory;

[0077] The processor is connected to a memory, which stores a computer program. When the computer program is executed by the processor, it causes the computer device to perform the method provided in the embodiments of this application.

[0078] One aspect of this application provides a computer-readable storage medium storing a computer program adapted to be loaded and executed by a processor, so that a computer device having the processor performs the method provided in this application.

[0079] In this embodiment, data preprocessing is performed on the initial battery feature data of the target battery to reduce abnormal data and remove irrelevant feature data, facilitating subsequent battery life prediction processing and reducing the computational load of subsequent predictions, thus improving the efficiency of life prediction. Furthermore, wavelet transform processing is applied to the preprocessed target battery feature data to extract effective feature data reflecting the life decay trend of the target battery. It can be seen that through preprocessing and wavelet transform, effective feature data affecting the life decay of the target battery can be extracted from the initial battery feature data. Simultaneously, wavelet transform processing can fit the target battery feature data, making the obtained effective feature data smoother and effectively reducing the noise impact caused by discrete data in the target battery feature data. Furthermore, temporal correlation feature extraction is performed on the effective feature data to obtain temporal correlation feature data. By considering the temporal correlation between effective feature data, the accuracy of feature extraction can be improved. Predicting the target battery life based on the temporal correlation feature data can improve the accuracy of battery life prediction. Attached Figure Description

[0080] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, 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.

[0081] Figure 1 This is a schematic diagram of the architecture of a battery life prediction system provided in an embodiment of this application;

[0082] Figure 2 This is a schematic flowchart of a battery life prediction method provided in an embodiment of this application;

[0083] Figure 3 This is a schematic diagram illustrating a comparison between target battery life and actual battery life provided in an embodiment of this application;

[0084] Figure 4 This is a schematic flowchart of a battery life prediction method provided in an embodiment of this application;

[0085] Figure 5 This is a schematic diagram of a dataset partitioning provided in an embodiment of this application;

[0086] Figure 6 This is a schematic diagram illustrating the training of an initial lifetime prediction model provided in an embodiment of this application;

[0087] Figure 7This is a schematic diagram of the structure of a battery life prediction device provided in an embodiment of this application;

[0088] Figure 8 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0089] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0090] Please see Figure 1 , Figure 1 This is a schematic diagram of the battery life prediction system in the embodiments of this application. Figure 1 The battery life prediction system shown includes a cluster of terminal devices, which may include one or at least two terminal devices; the number of terminal devices is not limited here. Figure 1 As shown, the terminal device cluster may include terminal device 100a, terminal device 100b, terminal device 100c, ..., terminal device 100n.

[0091] Optionally, terminal devices 100a, 100b, 100c, ..., 100n can communicate with each other by connecting to a server. For example... Figure 1 As shown, Figure 1 The battery life prediction system shown also includes a server 10, with each terminal device establishing a communication connection with the server 10. In other words, the server 10 can serve as an intermediary device for communication between the various terminal devices. Specifically, the server 10 can refer to a backend server that processes data from the terminal devices. This server 10 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. Each terminal device in the terminal device cluster can include: in-vehicle terminal devices, smartphones, tablets, laptops, desktop computers, wearable devices, smart home devices, head-mounted devices, and other intelligent terminal devices with battery life prediction functions.

[0092] Understandably, any terminal device in the terminal device cluster can collect initial battery feature data of the target battery and send it to server 10. For ease of understanding, taking terminal device 100a in the terminal device cluster as an example, terminal device 100a can send the initial battery feature data of the target battery to server 10. Server 10 can receive the initial battery feature data of the target battery sent by the terminal device, perform wavelet transform processing on the target battery feature data, and obtain effective feature data reflecting the lifespan degradation trend of the target battery. Further, server 10 can perform time-series correlation feature extraction on the effective feature data to obtain time-series correlation feature data, and predict the target battery lifespan of the target battery based on the time-series correlation feature data. After the server predicts the target battery lifespan, it can send the target battery lifespan of the target battery to terminal device 100a, or it can send the target battery lifespan of the target battery to other terminal devices in the terminal device cluster.

[0093] Thus, by preprocessing the initial battery feature data, outliers can be reduced, facilitating subsequent battery life prediction and improving its accuracy. Simultaneously, wavelet transform processing of the target battery feature data reduces noise, effectively addressing the impact of discrete noise on prediction accuracy. Furthermore, extracting temporal correlation features from the effective feature data and predicting target battery life based on these temporally correlated features further enhances prediction accuracy and efficiency.

[0094] Please see Figure 2 , Figure 2 This is a schematic flowchart of a battery life prediction method provided in an embodiment of this application. The battery life prediction method can be... Figure 1 It can be executed by any terminal device in the system, or by... Figure 1 Server 10 in the middle can be used to execute it, and it can also be executed by Figure 1 Any terminal device in and Figure 1 The server 10 in the application executes the method together, and the device used to execute the method in this application can be collectively referred to as a computer device. For example... Figure 2 As shown, the battery life prediction method may include, but is not limited to, the following steps:

[0095] S101, Obtain the initial battery characteristic data of the target battery under the operating conditions, perform data preprocessing on the initial battery characteristic data, and obtain the target battery characteristic data.

[0096] Specifically, the computer equipment can acquire initial battery characteristic data of the target battery under operating conditions. These operating conditions can include actual charging, discharging, and standby conditions. The initial battery characteristic data can include multiple sets of voltage values, multiple sets of current values, charging temperature, number of charging cycles, and charging frequency under charging conditions; discharge current, number of discharge cycles, discharge frequency, and discharge temperature under discharging conditions; and standby time and standby temperature under standby conditions. Of course, the initial battery characteristic data can also include battery internal resistance, battery type, and number of charge-discharge cycles. The target battery can be a lithium battery. The computer equipment can preprocess the initial battery characteristic data to obtain the target battery characteristic data. This preprocessing reduces outliers in the initial battery characteristic data, facilitating subsequent battery life prediction and improving the accuracy of life prediction. Optionally, the computer equipment can preprocess the initial battery characteristic data using a target life prediction model to obtain the target battery characteristic data.

[0097] Optionally, the computer device may preprocess the initial battery feature data to obtain the target battery feature data in the following ways: converting the initial battery feature data to a data type with the target data type; selecting battery feature data related to battery life degradation from the target data type as associated feature data; deleting abnormal feature data from the associated feature data to obtain processed associated feature data; acquiring the sampling features of the processed associated feature data; segmenting the processed associated feature data according to the sampling features to obtain K time intervals corresponding to associated sub-feature data, where K is a positive integer greater than 1; normalizing the K time intervals corresponding to the associated sub-feature data to obtain K time intervals corresponding to standard feature data; and determining the K time intervals corresponding to the standard feature data as the target battery feature data.

[0098] Specifically, computer equipment can convert the initial battery characteristic data into data types, obtaining battery characteristic data with a specific data type. Understandably, the target data type can be determined based on subsequent data processing needs; specifically, the target data type can refer to a numerical type that facilitates model prediction. The computer equipment can convert the initial battery characteristic data into battery characteristic data with a numerical type; in other words, it can convert non-numerical battery characteristic data from the initial battery characteristic data into battery characteristic data with a numerical type. Furthermore, the computer equipment can filter out battery characteristic data related to battery life degradation from the battery characteristic data with the target data type, using this as correlated characteristic data. Specifically, the computer equipment can obtain the correlation between the target battery's battery life degradation and the battery characteristic data, deleting battery characteristic data with a correlation less than the target correlation, such as deleting battery characteristic data unrelated to battery life degradation. For example, the model of the charging machine is unrelated to battery life degradation; different charging guns or different charging piles do not affect battery capacity, so the model of the charging machine can be deleted as an irrelevant factor.

[0099] It should be understood that computer equipment can delete anomalous data from the associated feature data to obtain processed associated feature data. Specifically, the computer equipment can construct a data distribution map corresponding to the associated feature data, such as using a normal distribution and box plots. Furthermore, the computer equipment can delete anomalously distributed data from the data distribution map to obtain processed associated feature data. For example, data outside the 25% and 75% ranges can be defined as anomalous feature data and deleted. This can improve the accuracy of subsequent lifetime prediction. Simultaneously, the computer equipment can acquire the sampling characteristics of the processed associated feature data, which may include sampling time, sampling content, and sampling frequency.

[0100] Furthermore, the computer device can use the sampling characteristics of the processed associated feature data to segment the processed associated feature data, obtaining K associated sub-feature data corresponding to each time interval, where K is a positive integer greater than 1. Specifically, the influencing factors of battery life degradation (i.e., battery capacity degradation) differ at different stages. Since the battery feature data of the target battery is collected and recorded according to system time, the processed associated feature data can be segmented according to the sampling time of the processed associated feature data to obtain K associated sub-feature data corresponding to each time interval. For example, the computer device can segment the battery feature data of the target battery during a single charge into associated sub-feature data corresponding to one time interval. Alternatively, when the target battery has a cycle, the computer device can segment according to the cycle of the target battery, and combine the cycle steps of the target battery to associate the various related time periods. For example, if the cycle is 2 hours, all data needs to be segmented into 2-hour cycles, i.e., one time interval is 2 hours.

[0101] Specifically, the computer equipment can normalize the associated sub-feature data corresponding to each of the K time intervals to obtain standard feature data corresponding to each of the K time intervals, and then determine the standard feature data corresponding to each of the K time intervals as the target battery feature data. Since the distribution of the associated sub-feature data corresponding to the K time intervals is different, normalizing (or standardizing) the associated sub-feature data corresponding to each of the K time intervals can effectively unify the associated sub-feature data into a certain data distribution, which can improve the accuracy of subsequent battery life prediction. It should be noted that other preprocessing methods can also be used to preprocess the initial battery feature data, and this embodiment does not impose any limitations on this.

[0102] S102, perform wavelet transform processing on the target battery feature data to obtain effective feature data that reflects the life decay trend of the target battery.

[0103] Specifically, computer equipment can perform wavelet transform processing on the target battery feature data to obtain effective feature data reflecting the battery's lifespan degradation trend. Wavelet transform processing can effectively highlight the effective feature data reflecting the target battery's lifespan degradation trend, enabling localized analysis of time (space) frequency. Through scaling and translation operations, it progressively refines the feature data at multiple scales, ultimately achieving time subdivision at high frequencies and frequency subdivision at low frequencies. This automatically adapts to the requirements of time-frequency feature data analysis, thus focusing on the effective feature data reflecting the target battery's lifespan degradation trend, thereby improving the accuracy of subsequent battery lifespan prediction. Therefore, by preprocessing the initial battery feature data of the target battery and then performing wavelet transform processing on the preprocessed target battery feature data, effective feature data reflecting the target battery's lifespan degradation trend can be found, thereby improving the accuracy of target battery lifespan prediction.

[0104] Optionally, the specific method by which the computer device performs wavelet transform processing on the target battery feature data may include: performing wavelet decomposition on the target battery feature data through a wavelet transform layer in the target lifetime prediction model to obtain the wavelet decomposition coefficients corresponding to the target battery feature data; performing noise reduction processing on the wavelet decomposition coefficients according to the target adjustment threshold to obtain the noise-reduced wavelet decomposition coefficients; and performing wavelet reconstruction on the noise-reduced wavelet decomposition coefficients to obtain effective feature data reflecting the lifetime degradation trend of the target battery.

[0105] Specifically, the computer equipment can use the wavelet transform layer in the target life prediction model to perform wavelet decomposition of the target battery feature data at a specified number of layers using the target wavelet function, obtaining the wavelet decomposition coefficients corresponding to the target battery feature data. These wavelet decomposition coefficients contain important feature data from the target battery feature data. The target wavelet function can be a Haar wavelet function (i.e., an orthogonal wavelet function), a Meyer wavelet function, a sym8 wavelet function, etc., and the specified number of layers can be 2 or 3, etc. Furthermore, the computer equipment can adjust the threshold according to the target to perform noise reduction processing on the wavelet decomposition coefficients, obtaining the noise-reduced wavelet decomposition coefficients.

[0106] Furthermore, the computer equipment can employ a wavelet reconstruction algorithm to reconstruct the denoised wavelet decomposition coefficients, obtaining effective feature data that reflects the lifespan degradation trend of the target battery. In this way, wavelet transform can locally optimize the target battery feature data, effectively fitting and smoothing the time-series discrete values, effectively mitigating the noise impact of inefficient time-series discrete values, and effectively solving the problem of discrete noise data affecting the accuracy of battery lifespan prediction. This makes the effective feature data more stable and better reflects the lifespan degradation trend of the target battery.

[0107] Optionally, the computer device can determine an adaptive sliding window corresponding to the target battery feature data based on the sampling features corresponding to the target battery feature data, and perform wavelet transform processing on the target battery feature data based on the adaptive sliding window. Specifically, the target battery feature data includes standard feature data corresponding to K time intervals, and the computer device can acquire time interval K from the K time intervals. i The corresponding standard feature data corresponds to an adaptive sliding window for the time interval K. i The corresponding standard feature data is divided to obtain the time interval K. i The N time series in the dataset correspond to standard sub-feature data, where i is a positive integer less than or equal to K. Furthermore, the computer device can perform wavelet transform on the standard sub-feature data corresponding to each of the N time series. For example, if the target battery feature data is sampled in minutes, the target adaptive sliding window can be 10 minutes, i.e., a sliding window of 10 minutes. If the target battery feature data is sampled in hours, the target adaptive sliding window can be 12 hours, i.e., a sliding window of 12 hours. The specific sliding window size can be customized or adjusted according to the actual prediction situation. It is evident that wavelet transform can provide a "time-frequency" adaptive sliding window that changes with frequency, enabling separate processing of feature data collected at different stages, localized analysis of time (space) frequency, achieving time subdivision at high frequencies and frequency subdivision at low frequencies, automatically adapting to the analysis requirements of time-frequency feature data, and obtaining effective feature data reflecting the lifespan degradation trend of the target battery.

[0108] Optionally, the wavelet decomposition coefficients corresponding to the target battery feature data include frequencies in a first frequency set and frequencies in a second frequency set, where the frequencies in the first frequency set are greater than those in the second frequency set. The frequencies in the first frequency set can refer to decomposition coefficients with frequencies greater than a target frequency threshold (i.e., high-frequency decomposition coefficients), and frequencies in the first frequency set can refer to decomposition coefficients with frequencies less than or equal to the target frequency threshold (i.e., low-frequency decomposition coefficients). The specific method by which the computer equipment performs noise reduction processing on the wavelet decomposition coefficients based on the target adjustment threshold can include: threshold quantizing the frequencies in the first frequency set according to the target adjustment threshold to obtain a quantized first frequency set. The frequencies in the quantized first frequency set and the frequencies in the second frequency set are then determined as the noise-reduced wavelet decomposition coefficients.

[0109] Specifically, since low-frequency decomposition coefficients can effectively reflect the overall trend characteristics of the data, computer equipment can determine the target adjustment threshold corresponding to the target battery feature data using threshold estimation criteria and the content characteristics of the target battery feature data. Based on the target adjustment threshold, the frequencies in the first frequency set are then threshold-quantized to obtain the quantized first frequency set. For example, the decomposition coefficients S in wavelet decomposition coefficients... j If the deviation from the normal decomposition coefficients is too large, it can be adjusted by using a target adjustment threshold to make the decomposition coefficients S... j Approximately normal decomposition coefficients. The frequencies in the first and second frequency sets after quantization are determined as the denoised wavelet decomposition coefficients.

[0110] S103, extract time-series correlation features from the effective feature data to obtain time-series correlation feature data.

[0111] Specifically, computer equipment can perform time-series correlation feature extraction on valid feature data to obtain time-series correlation feature data. Understandably, computer equipment can perform time-series correlation feature extraction on valid feature data based on the sampling time of the valid feature data. For example, when analyzing the feature information of valid feature data at the current moment, the computer equipment can combine it with the valid feature data from the previous moment, thus improving the accuracy of data analysis.

[0112] Optionally, the specific method for the computer device to perform temporal correlation feature extraction on valid feature data may include: obtaining the data correlation degree between valid feature data at time t and valid feature data at time t-1 through the temporal correlation feature extraction layer in the target lifetime prediction model; both feature data at time t and feature data at time t-1 are considered valid feature data. Based on the data correlation degree, the valid feature data at time t-1 is weighted to obtain first weighted feature data; the input control parameters in the temporal correlation feature extraction layer are obtained; and the valid feature data at time t is weighted based on the input control parameters to obtain second weighted feature data. The first weighted feature data and the second weighted feature data are fused to obtain important feature data among the valid feature data at time t. Based on the weight matrix and bias parameters in the temporal correlation feature extraction layer, features are extracted from the important feature data to obtain the correlated feature data corresponding to time t, until the correlated feature data corresponding to the valid feature data is obtained, and the correlated feature data corresponding to the valid feature data is determined as the temporal correlation feature data.

[0113] Specifically, the computer device can obtain the data correlation degree between the effective feature data at time t and the effective feature data at time t-1 through the temporal correlation feature extraction layer in the target life prediction model; both the feature data at time t and the feature data at time t-1 are considered effective feature data. Specifically, the temporal correlation feature extraction layer can refer to LSTM (Long Short-Term Memory, a type of recurrent neural network). The computer device can obtain the data correlation degree between the effective feature data at time t and the effective feature data at time t-1 through the forget gate in the temporal correlation feature extraction layer. Specifically, the data correlation degree can be determined according to the following formula (1).

[0114] f t =σ(W f [h t-1 ,x t ]+b f (1)

[0115] Wherein, h in formula (1) t-1 x represents the battery life (e.g., battery capacity information) corresponding to the target battery stored at time t-1. t W represents the valid feature data input to the target battery at time t. f b f f represents the model parameters. t σ represents the correlation between data, and σ represents the sigmoid function (i.e., classification function).

[0116] Furthermore, the computer device can use the input gate, classification layer, and hidden layer in the temporal correlation feature extraction layer to weight the effective feature data at time t-1 according to the data correlation degree to obtain the first weighted feature data. Simultaneously, the computer device can obtain the input control parameters in the temporal correlation feature extraction layer, and use these parameters to weight the effective feature data at time t to obtain the second weighted feature data. The first and second weighted feature data are then fused to obtain the important feature data among the effective feature data at time t. Specifically, the important feature data can be determined according to the following formula (2).

[0117] C t =f t C t-1 +σtanh(W c [h t-1 ,x t ]+b c (2)

[0118] Wherein, W in formula (2) c and b cC is the input control parameter in the temporal correlation feature extraction layer. t-1 C represents the key characteristic data of the target battery retained in the memory cell state at time t-1. t f represents important feature data among the effective feature data of the target battery retained in the memory cell state at time t. t Indicates the correlation between data, σ represents the sigmoid function (i.e., classification function), and h t-1 x represents the battery life (e.g., battery capacity information) corresponding to the target battery stored at time t-1. t This represents the valid feature data input to the target battery at time t.

[0119] Furthermore, the computer device can extract features from important feature data based on the weight matrix and bias parameters in the temporal correlation feature extraction layer, obtaining the correlated feature data corresponding to time t. Thus, through the temporal correlation feature extraction layer, effective feature data can be extracted from the temporal correlation dimension, improving the accuracy of feature extraction and consequently enhancing the accuracy of subsequent battery life prediction.

[0120] S104, based on time-series correlation feature data, predicts the target battery life of the target battery.

[0121] Specifically, computer equipment can predict the target battery life of a target battery based on time-series correlated feature data. Optionally, the computer equipment can predict the target battery life of the target battery based on time-series correlated feature data through the input layer of the target battery life prediction model. In this way, the wavelet transform layer in the target battery life prediction model can perform noise reduction processing on the target battery feature data, making the obtained effective feature data more suitable and smoother. This effectively solves the influence of low-time-series discrete values, thus facilitating better feature extraction by the time-series correlated feature extraction layer in the target battery life prediction model, improving the accuracy of feature extraction. Therefore, based on the time-series correlated feature data extracted by the time-series correlated feature extraction layer, the target battery life of the target battery can be accurately predicted.

[0122] The specific formula is shown in formula (3) below.

[0123] h t =σ(W o [h t-1 ,x t ]+b o )tanh(c t (3)

[0124] In formula (3), σ represents the sigmoid classification function, W0 represents the weight matrix, and b o C represents the bias parameter.t h represents important feature data among the effective feature data of the target battery retained in the memory cell state at time t. t-1 x represents the battery life (e.g., battery capacity information) corresponding to the target battery stored at time t-1. t This represents the valid feature data input to the target battery at time t.

[0125] like Figure 3 As shown, Figure 3 This is a schematic diagram illustrating a comparison between target battery life and actual battery life provided in an embodiment of this application, as shown below. Figure 3 As shown, the longer the target battery is used, the smaller its capacity (i.e., battery life). The target battery life predicted by this application is quite close to the actual battery life, indicating that this application can improve the accuracy of target battery life prediction through a target life prediction model.

[0126] In this embodiment, data preprocessing is performed on the initial battery feature data of the target battery to reduce abnormal data and remove irrelevant feature data, facilitating subsequent battery life prediction processing and reducing the computational load of subsequent predictions, thus improving the efficiency of life prediction. Furthermore, wavelet transform processing is applied to the preprocessed target battery feature data to extract effective feature data reflecting the life decay trend of the target battery. It is evident that preprocessing and wavelet transform can extract effective feature data affecting the life decay of the target battery from the initial battery feature data. Simultaneously, wavelet transform processing can fit the target battery feature data, making the obtained effective feature data smoother and effectively reducing the noise impact of discrete data in the target battery feature data. Further, temporal correlation feature extraction is performed on the effective feature data to obtain temporal correlation feature data. By considering the temporal correlation between effective feature data, the accuracy of feature extraction can be improved. Predicting the target battery life based on the temporal correlation feature data can improve the accuracy of battery life prediction. Simultaneously, this embodiment predicts the target battery life of the target battery using a target life prediction model, which can improve the accuracy and efficiency of life prediction.

[0127] Please see Figure 4 , Figure 4 This is a schematic flowchart of a battery life prediction method provided in an embodiment of this application. The battery life prediction method can be... Figure 1 It can be executed by any terminal device in the system, or by... Figure 1 Server 10 in the middle can be used to execute it, and it can also be executed by Figure 1 Any terminal device in and Figure 1The server 10 in the application executes the method together, and the device used to execute the method in this application can be collectively referred to as a computer device. For example... Figure 4 As shown, the battery life prediction method may include, but is not limited to, the following steps:

[0128] S201, obtain the initial life prediction model, the sample battery feature data under the operating conditions, and the battery life label corresponding to the sample battery.

[0129] Specifically, the computer equipment can acquire characteristic data of the sample battery under operating conditions, including charging, discharging, and standby conditions during testing. Similarly, this characteristic data can include multiple sets of charging voltage values, multiple sets of charging current values, charging temperature, number of charging cycles, charging frequency, discharging current, number of discharging cycles, discharging frequency, discharging temperature, standby time, and standby temperature. Of course, the initial battery characteristic data can also include battery capacity (i.e., discharge capacity Ah), discharge energy Wh, battery internal resistance, battery type, and number of charge-discharge cycles. The sample battery is associated with the target battery; for example, the sample battery type is the same as the target battery type, such as both being lithium batteries.

[0130] Optionally, the computer equipment can set charging time, resting time, discharging time, and pause time according to a standardized constant current constant voltage charge-discharge protocol to form a complete charge-discharge cycle. Furthermore, the computer equipment can perform a complete charge-discharge cycle test based on nominal parameters such as the rated capacity, charge-discharge cutoff voltage, and maximum continuous charge-discharge current of the test battery. The charging process is as follows: using a constant current constant voltage (CC-CV) charging protocol, the battery is first charged at a constant current CC mode at a charging rate of 0.2–0.5C until the upper limit cutoff voltage is reached, then stopped. Next, a constant voltage CV stage charging model is performed until the charging current decreases to 0.02C, at which point the test stops; then the battery is paused for 30–60 minutes. The discharging process is as follows: the battery is discharged at the standard cell discharge current until the voltage drops to the lower limit cutoff voltage, at which point the test stops; then the battery is paused for 30–60 minutes. This cycle is repeated multiple times. If the remaining capacity of the sample battery is 80% of its initial capacity, the sample battery is considered to have failed and is deemed unusable. The computer equipment can record characteristic data such as the number of cycles, discharge capacity (i.e., battery capacity) (Ah), discharge energy (Wh), internal resistance (mΩ), temperature (°C), voltage (V), and current during each cycle test, which are used as the sample battery characteristic data of the sample battery.

[0131] It should be understood that the computer device can define the model structure of the initial life prediction model. Specifically, the computer device can define the basic model structure of the initial life prediction module based on the magnitude of the sample battery feature data. The computer device can determine the number of wavelet transform layers and temporal correlation feature extraction layers in the target initial life prediction model based on the amount of sample battery feature data. Specifically, the most basic model of the temporal correlation feature extraction layer is set as a single structure, including one input gate, one output gate, and one forget gate. The number of different single structures required is determined based on different magnitudes; for example, 10 single structures can be used when the feature data is on the order of 50,000-100,000. The computer device can determine the battery capacity obtained during the testing process as the battery life label corresponding to the sample battery.

[0132] S202, based on the characteristic data of the sample battery, the initial life prediction model is used to predict the life of the sample battery and obtain the predicted battery life of the sample battery.

[0133] Specifically, computer equipment can use an initial life prediction model to predict the life of a sample battery based on its characteristic data, thereby obtaining the predicted battery life of the sample battery.

[0134] Optionally, the specific method for the computer device to predict the lifespan of the sample battery may include: preprocessing the sample battery feature data using an initial lifespan prediction model to obtain preprocessed sample battery feature data; dividing the preprocessed sample battery feature data into M time intervals to obtain sample sub-feature data corresponding to each time interval; where M is a positive integer greater than 1. Using a wavelet transform layer in the initial lifespan prediction model, performing wavelet transform processing on the sample sub-feature data corresponding to each of the M time intervals to obtain effective sample feature data corresponding to each of the M time intervals. Using a time-series correlation feature extraction layer in the initial lifespan prediction model, extracting time-series correlation features from the effective sample feature data corresponding to each of the M time intervals to obtain sample time-series correlation feature data corresponding to each of the M time intervals; and predicting the predicted battery lifespan of the sample battery based on the sample time-series correlation feature data corresponding to each of the M time intervals.

[0135] Specifically, the computer equipment can invoke the initial lifetime prediction model to preprocess the sample battery feature data, obtaining preprocessed sample battery feature data. Similarly, refer to the above. Figure 2Step S101, the specific method for preprocessing the sample battery feature data, may include: converting the sample battery feature data to obtain sample battery feature data with a target data type; simultaneously, selecting sample battery feature data related to battery life degradation from the sample battery feature data with the target data type as sample associated feature data; deleting abnormal feature data from the sample associated feature data to obtain processed sample battery feature data. Further, the computer device can acquire the sampling features of the processed sample battery feature data and perform associated segmentation based on the sampling features of the processed associated battery feature data. For example, the feature data collected within the same cycle can be segmented into a dataset. For instance, if the cycle is 2 hours, all data needs to be segmented into 2-hour cycles, i.e., a time interval of 2 hours. The computer device may also perform normalization and other processing on the sample battery feature data.

[0136] Furthermore, the computer device can divide the preprocessed sample battery feature data to obtain sample sub-feature data corresponding to M time intervals, where M is a positive integer greater than 1. The computer device can perform wavelet transform processing on the sample sub-feature data corresponding to the M time intervals respectively through the wavelet transform layer in the initial lifetime prediction model to obtain effective sample feature data corresponding to the M time intervals. Refer to step S102 above; this embodiment will not be repeated here. Furthermore, the computer device can perform time-series correlation feature extraction on the effective sample feature data corresponding to the M time intervals respectively through the time-series correlation feature extraction layer in the initial lifetime prediction model to obtain sample time-series correlation feature data corresponding to the M time intervals. Refer to step S103 above; this embodiment will not be repeated here. Furthermore, the computer device can predict the predicted battery lifetime corresponding to the sample battery based on the sample time-series correlation feature data corresponding to the M time intervals.

[0137] Optionally, the specific method by which the computer device divides the preprocessed sample battery feature data may include acquiring the sampling features of the preprocessed sample battery feature data. Based on the sampling features of the preprocessed sample battery feature data, an adaptive sliding window corresponding to the preprocessed sample feature data is determined. Using the adaptive sliding window corresponding to the preprocessed sample feature data, the preprocessed sample battery feature data is divided to obtain sample sub-feature data corresponding to M time intervals. It should be understood that the computer device can acquire the sampling features of the preprocessed sample battery feature data, which may include sampling time, sampling content, and sampling frequency, etc.

[0138] Furthermore, the computer equipment can determine the adaptive sliding window corresponding to the preprocessed sample battery feature data based on the sampling characteristics of the preprocessed sample battery feature data. For example, since the data collected at different time stages differ significantly, the computer equipment can determine the adaptive sliding window based on the collection time. The target adaptive sliding window can be 10 minutes (i.e., 10 minutes per sliding window) or 12 hours (i.e., 12 hours per sliding window), which can be set according to specific needs or based on the model prediction loss of the initial lifetime prediction model. Furthermore, the computer equipment can use the adaptive sliding window corresponding to the preprocessed sample battery feature data to divide the preprocessed sample battery feature data into M time intervals, each corresponding to a sub-feature data segment. The computer equipment can also determine the adaptive sliding window corresponding to the preprocessed sample battery feature data based on the sampling content of the preprocessed sample battery feature data, such as the charge / discharge cycle period. This better ensures that the time-series correlation feature extraction layer adapts to multiple rounds of repetitive charge / discharge training data. It is evident that by adding a wavelet transform layer, the overall learning performance of the initial life prediction model can be improved, and by combining it with an adaptive sliding window to optimize the overall model's data learning for charge and discharge cycles, the accuracy of battery life prediction can be enhanced.

[0139] Optionally, before partitioning the preprocessed sample battery feature data, the computer device can partition the preprocessed sample battery feature data into a training set, a validation set, and a test set. The feature data in each of these sets is then further partitioned to obtain sample sub-feature data corresponding to M time intervals. These M time interval sub-feature data include the sample sub-feature data corresponding to the time intervals in the training set, the validation set, and the test set.

[0140] Specifically, the training set typically comprises more than 60% of the preprocessed sample battery feature data, but the exact percentage depends on the size of the preprocessed sample battery feature data. For example, 70% of the preprocessed sample battery feature data can be used as the training set. After each training round, to better ensure the generalization and robustness of the initial life prediction model, a validation set is used to test the model. The validation set typically comprises about 10-20% of the preprocessed sample battery feature data, but the exact percentage depends on the size of the preprocessed sample battery feature data. For example, 10% of the preprocessed sample battery feature data can be used as the training set. After the entire model training is complete, the performance of the saved model needs to be verified, which requires using a test set to validate the model. The validation set typically comprises about 20-30% of the total data, but the exact percentage depends on the size of the preprocessed sample battery feature data. For example, 20% of the preprocessed sample battery feature data can be used as the training set. For example, the training set, validation set, and test set are divided in a 7:1:2 ratio. The training set is used to train the model, the validation set is used to verify the model's performance during training, and the test set is used to test the model's performance after training.

[0141] Optionally, for the training set: the computer device can use an adaptive sliding window corresponding to the training set to transform the discrete data distribution into a smooth and continuous training set by sliding the window. For the validation set: the computer device can use an adaptive sliding window corresponding to the validation set to transform the discrete data distribution into a smooth and continuous validation set by sliding the window. For the test set: the computer device can use an adaptive sliding window corresponding to the test set to transform the discrete data distribution into a smooth and continuous test set by sliding the window.

[0142] Optionally, the specific method by which the computer device performs wavelet transform processing on the sample sub-feature data corresponding to the M time intervals through the wavelet transform layer in the initial life prediction model can include: the computer device can use the sym8 wavelet function to perform wavelet decomposition of the sample sub-feature data corresponding to the M time intervals with a layer of 2, then use rigrsure (i.e., unbiased risk adaptive threshold estimation criterion) to determine the target adjustment threshold of the wavelet coefficients of each layer, and then use a soft thresholding function to process the wavelet decomposition coefficients corresponding to each sample sub-feature data according to the target adjustment threshold of the wavelet coefficients of each layer to obtain estimated wavelet coefficients, and obtain the denoised effective feature data of the samples based on the estimated wavelet coefficients. Further, the computer device can output the effective feature data of the samples corresponding to the M time intervals into a hybrid neural network prediction layer based on LSTM and wavelet transform to predict the predicted battery life of the sample battery. Optionally, the computer device can train the initial life prediction model using the sample sub-feature data corresponding to the M time intervals in rounds according to the set epoch.

[0143] S203. Based on the predicted battery life and battery life label, adjust the model parameters in the initial life prediction model. When the initial life prediction model after parameter adjustment meets the convergence condition, determine the initial life prediction model after parameter adjustment as the target life prediction model.

[0144] Specifically, the computer equipment can determine the model prediction loss of the initial lifespan prediction model based on the predicted battery life and battery life label of the sample battery. Based on this model prediction loss, the model parameters in the initial lifespan prediction model are adjusted. When the parameter-adjusted initial lifespan prediction model meets the convergence condition, it is determined as the target lifespan prediction model. The convergence condition can be that the model prediction loss of the initial lifespan prediction model is less than the target loss threshold, or that the number of parameter adjustments to the initial lifespan prediction model reaches the target number.

[0145] Specifically, computer equipment can train the initial lifetime prediction model using sample sub-feature data corresponding to time intervals in the training set and the sample sub-feature data corresponding to time intervals in the validation set, and test the initial lifetime prediction model using sample sub-feature data corresponding to time intervals in the test set.

[0146] Optionally, the computer equipment can acquire the prediction evaluation parameters of the initial lifetime prediction model and determine the model prediction loss of the initial lifetime prediction model based on these parameters. The prediction evaluation parameters may include the following: Parameter 1: Mean-square error (Mse): Also known as the mean square error, it is a measure of the difference between the estimator and the estimated quantity. Mse can evaluate the degree of data variation; the smaller the Mse value, the better the predictive ability of the prediction model. Parameter 2: Mean absolute error (Mae): Also known as the mean absolute error, it is a measure of the difference between the predicted value and the true value; it is the expected value of the square of the difference between the predicted value and the true value. Parameter 3: Mean Absolute Percentage Error (Mape): Also known as the average of the absolute values ​​of the relative percentage errors, it can be used to measure the goodness of a model's prediction results. R²: Coefficient of determination. Parameter 4: The larger the R², the smaller the sum of squared residuals, indicating higher model accuracy and a better fit to the data. Its value range is usually between (0,1), and the closer to 1, the better the model fits the data. Parameter 4: Loss: Also known as the loss value, it is used to judge the difference between the model's predicted value and the actual data, thereby helping the model to make adjustments.

[0147] Optionally, when a computer device tests an initial lifetime prediction model using a test set, it can detect the performance of the initial lifetime prediction model based on the corresponding result parameters, including prediction labels and accuracy.

[0148] Optionally, the network structure and parameter settings of the temporal correlation feature extraction layer (i.e., LSTM) are shown in Table 1 below:

[0149] lithium battery sample Data volume Training set length 850 Validation set length 123 test set length 233 Input step size 5 Number of iterations 100 Hidden layer 48 Number of fully connected layers 2 Droupout layer deactivation rate 0.2 Number of samples per batch (BatchSize) 64 Learning Rate 0.001

[0150] Table 1

[0151] The computer equipment uses the parameters in Table 1 to train the initial life prediction model, and the prediction estimation parameters are shown in Table 2.

[0152] Mse Mae Mape <![CDATA[R 2 ]]> 0.0736 0.0604 0.0008 0.9695

[0153] Table 2

[0154] As shown in Table 2, the values ​​of Mse, Mae, and Mape are all much less than 0, indicating that the prediction accuracy of the initial lifetime prediction model is high. At the same time, the R2 is 0.9695. The larger the R2, the higher the prediction accuracy of the model and the better the data fit.

[0155] like Figure 5 As shown, Figure 5This is a schematic diagram of a dataset partitioning provided in an embodiment of this application, such as... Figure 5 As shown, the longer the cycle period, the smaller the capacity (i.e., battery life) of the sample battery; that is, the battery life of the sample battery decreases with increasing cycle period. Figure 5 As shown, the training set comprises a large proportion, typically over 60% of the total data, but the exact percentage depends on the total dataset size. The validation set comprises a smaller proportion, generally around 10-20% of the total data, also depending on the total dataset size. The validation set is used to better ensure the generalization and robustness of the initial lifetime prediction model. The test set is larger than the validation set but smaller than the training set. After the initial lifetime prediction model is trained, it's necessary to verify the performance of the saved target lifetime prediction model. The test set is used to validate the model, typically comprising around 20-30% of the total data, but the exact percentage depends on the total dataset size.

[0156] like Figure 6 As shown, Figure 6 This is a schematic diagram illustrating the training of an initial lifetime prediction model provided in an embodiment of this application, as shown below. Figure 6 As shown, the computer equipment can acquire multiple sets of sample batteries and execute step 1 to obtain the characteristic data of the sample batteries. Specifically, the computer equipment can set the charging time, resting time, discharging time, and shelving time according to the standardized constant current constant voltage charge-discharge protocol to form a complete charge-discharge cycle. Further, the computer equipment can perform a complete charge-discharge cycle test based on the nominal parameters such as the rated capacity, charge-discharge cutoff voltage, and maximum continuous charge-discharge current of the test battery. The charging process is as follows: using the constant current constant voltage (CC-CV) charging protocol, the sample battery is first charged at a constant current CC mode at a charging rate of 0.2–0.5C until the upper limit cutoff voltage is reached, then stopped. Next, a constant voltage CV stage charging model is performed until the charging current decreases to 0.02C, at which point the test stops; then it is shelved for 30–60 minutes. The discharging process is as follows: the battery is discharged at the standard cell discharge current, and the test stops when the voltage drops to the lower limit cutoff voltage; then it is shelved for 30–60 minutes. This cycle test is performed multiple times. If the remaining capacity of the sample battery is 80% of the initial capacity, the sample battery is determined to be defective and scrapped. The computer equipment can record battery test data such as the number of cycles, discharge capacity (i.e., battery capacity) (Ah), discharge energy (Wh), internal resistance (mΩ), temperature (°C), voltage (V), and current during each cycle test, which can be used as sample battery characteristic data for the sample battery.

[0157] Further, the computer device can perform step 2, data cleaning (i.e., data preprocessing) on ​​the sample battery feature data. Data cleaning may include: 201 type conversion (i.e., conversion to the target data type), 202 deletion of irrelevant / abnormal data, 203 null value imputation, 204 time segmentation and correlation, and 205 normalization / standardization. For details, please refer to step S202 above; this embodiment will not repeat them here. Further, the computer device can perform step 3, defining the initial lifetime prediction model, i.e., determining the model structure of the initial lifetime prediction model. The computer device can perform step 4, dividing the sample battery feature data obtained from data cleaning into a dataset, obtaining a training set, a validation set, and a test set. Further, the computer device can perform step 5, using an adaptive sliding window 501 to construct time-series data for the training set, validation set, and test set, obtaining sample sub-feature data corresponding to M time intervals.

[0158] Further, the computer device can execute step 6 to build an initial lifetime prediction model, which can be composed of wavelet transform and LSTM (Long Short-Term Memory network), and execute step 7 to train the initial lifetime prediction model using sample sub-feature data corresponding to M time intervals. Further, the computer device can execute step 8 to verify and test the initial lifetime prediction model, and detect its performance. Further, the computer device can execute step 9 to evaluate the initial lifetime prediction model, and if the initial lifetime prediction model meets the convergence condition, it is determined as the target lifetime prediction model.

[0159] S204. Obtain the initial battery characteristic data of the target battery under operating conditions. Preprocess the initial battery characteristic data using the target life prediction model to obtain the target battery characteristic data.

[0160] S205, perform wavelet transform processing on the target battery feature data to obtain effective feature data that reflects the life decay trend of the target battery.

[0161] S206, extract time-series correlation features from the effective feature data to obtain time-series correlation feature data.

[0162] S207, based on time-series correlation feature data, predicts the target battery life of the target battery.

[0163] Specifically, the contents of steps S204-S207 in this embodiment can be found above. Figure 2 The contents of steps S101-S104 are not described again in this embodiment of the application.

[0164] In this embodiment, by training the initial life prediction model, a high-performance target life prediction model is obtained, which can improve the accuracy of battery life prediction. Data preprocessing is performed on the initial battery feature data of the target battery to reduce outlier data and remove irrelevant feature data, facilitating subsequent battery life prediction processing and reducing the computational load of subsequent predictions, thus improving the efficiency of life prediction. Furthermore, wavelet transform processing is performed on the preprocessed target battery feature data to extract effective feature data reflecting the life decay trend of the target battery. It can be seen that through preprocessing and wavelet transform, effective feature data affecting the life decay of the target battery can be extracted from the initial battery feature data. Simultaneously, wavelet transform processing can fit the target battery feature data, making the obtained effective feature data smoother and effectively reducing the noise impact of discrete data in the target battery feature data. Furthermore, temporal correlation feature extraction is performed on the effective feature data to obtain temporal correlation feature data. By considering the temporal correlation between effective feature data, the accuracy of feature extraction can be improved. Predicting the target battery life based on the temporal correlation feature data can improve the accuracy of battery life prediction. Meanwhile, this application implements a target battery life prediction model to predict the target battery life, which can improve the accuracy and efficiency of life prediction.

[0165] Please see Figure 7 , Figure 7 This is a schematic diagram of a battery life prediction device provided in an embodiment of this application. The aforementioned battery life prediction device can be a computer program (including program code) running on a computer device; for example, the battery life prediction device is application software. The battery life prediction device can be used to execute corresponding steps in the battery life prediction method provided in the embodiments of this application. Figure 7 As shown, the battery life prediction device may include: a preprocessing module 11, a wavelet transform module 12, a feature extraction module 13, a first prediction module 14, an acquisition module 15, a second prediction module 16, and a determination module 17.

[0166] Preprocessing module 11 is used to acquire the initial battery characteristic data of the target battery under the operating conditions, perform data preprocessing on the initial battery characteristic data, and obtain the target battery characteristic data.

[0167] Wavelet transform module 12 is used to perform wavelet transform processing on the target battery feature data to obtain effective feature data that reflects the life decay trend of the target battery.

[0168] Feature extraction module 13 is used to extract temporal correlation features from effective feature data to obtain temporal correlation feature data;

[0169] The first prediction module 14 is used to predict the target battery life of the target battery based on time-series correlated feature data.

[0170] The preprocessing module 11 includes:

[0171] The type conversion unit 1101 is used to convert the initial battery feature data to a target data type to obtain battery feature data with a target data type, and to filter out battery feature data related to battery life degradation from the battery feature data with the target data type as associated feature data.

[0172] The deletion unit 1102 is used to delete abnormal feature data in the associated feature data to obtain the processed associated feature data, obtain the sampling features of the processed associated feature data, and divide the processed associated feature data according to the sampling features of the processed associated feature data to obtain the associated sub-feature data corresponding to K time intervals respectively; K is a positive integer greater than 1.

[0173] Normalization unit 1103 is used to normalize the associated sub-feature data corresponding to K time intervals respectively, so as to obtain the standard feature data corresponding to K time intervals respectively.

[0174] The first determining unit 1104 is used to determine the standard feature data corresponding to K time intervals as target battery feature data.

[0175] The wavelet transform module 12 includes:

[0176] Wavelet decomposition unit 1201 is used to perform wavelet decomposition on target battery feature data through wavelet transform layer in target lifetime prediction model to obtain wavelet decomposition coefficients corresponding to target battery feature data;

[0177] The noise reduction processing unit 1202 is used to perform noise reduction processing on the wavelet decomposition coefficients according to the target adjustment threshold to obtain the noise-reduced wavelet decomposition coefficients.

[0178] Wavelet reconstruction unit 1203 is used to reconstruct the wavelet decomposition coefficients after noise reduction to obtain effective feature data that reflects the life decay trend of the target battery.

[0179] Among them, the wavelet decomposition coefficients corresponding to the target battery feature data include frequencies in the first frequency set and frequencies in the second frequency set, with the frequencies in the first frequency set being greater than the frequencies in the second frequency set;

[0180] The noise reduction processing unit 1202 is specifically used for:

[0181] Based on the target, the frequency in the first frequency set is threshold quantized to obtain the quantized first frequency set.

[0182] The frequencies in the first and second frequency sets after quantization are determined as the wavelet decomposition coefficients after denoising.

[0183] The feature extraction module 13 includes:

[0184] The acquisition unit 1301 is used to acquire the data correlation degree between the effective feature data at time t and the effective feature data at time t-1 through the time-series correlation feature extraction layer in the target lifetime prediction model; both the feature data at time t and the feature data at time t-1 are effective feature data.

[0185] The weighted processing unit 1302 is used to perform weighted processing on the effective feature data at time t-1 according to the data correlation degree to obtain the first weighted feature data, obtain the input control parameters in the time-series correlation feature extraction layer, and perform weighted processing on the effective feature data at time t according to the input control parameters to obtain the second weighted feature data.

[0186] The fusion unit 1303 is used to fuse the first weighted feature data and the second weighted feature data to obtain the important feature data in the effective feature data at time t;

[0187] The first extraction unit 1304 is used to extract features from important feature data based on the weight matrix and bias parameters in the temporal correlation feature extraction layer, so as to obtain the correlation feature data corresponding to time t.

[0188] The second determining unit 1305 is used to determine the associated feature data corresponding to the valid feature data as time-series associated feature data until the associated feature data corresponding to the valid feature data is obtained.

[0189] The battery life prediction device also includes:

[0190] The acquisition module 15 is used to acquire the initial life prediction model, the sample battery feature data under the operating conditions, and the battery life label corresponding to the sample battery; the sample battery is associated with the target battery;

[0191] The second prediction module 16 is used to predict the life of the sample battery based on the sample battery feature data through the initial life prediction model, and obtain the predicted battery life of the sample battery.

[0192] The determination module 17 is used to adjust the model parameters in the initial life prediction model based on the predicted battery life and the battery life label. When the initial life prediction model after parameter adjustment meets the convergence condition, the initial life prediction model after parameter adjustment is determined as the target life prediction model.

[0193] The second prediction module 16 includes:

[0194] The preprocessing unit 1601 is used to preprocess the sample battery feature data through the initial lifetime prediction model to obtain the preprocessed sample battery feature data, and to divide the preprocessed sample battery feature data into M time intervals to obtain sample sub-feature data corresponding to each time interval; M is a positive integer greater than 1.

[0195] Wavelet transform unit 1602 is used to perform wavelet transform processing on the sample sub-feature data corresponding to M time intervals respectively through the wavelet transform layer in the initial lifetime prediction model to obtain the sample effective feature data corresponding to M time intervals respectively.

[0196] The second extraction unit 1603 is used to extract time-series correlation features from the effective feature data of samples corresponding to M time intervals through the time-series correlation feature extraction layer in the initial life prediction model, to obtain the time-series correlation feature data of samples corresponding to M time intervals, and to predict the predicted battery life of the sample battery based on the time-series correlation feature data of samples corresponding to M time intervals.

[0197] Specifically, the preprocessing unit 1601 is used for:

[0198] Obtain the sampling features of the preprocessed sample battery feature data;

[0199] Based on the sampling characteristics of the preprocessed sample battery feature data, determine the adaptive sliding window corresponding to the preprocessed sample feature data;

[0200] An adaptive sliding window corresponding to the preprocessed sample feature data is used to divide the preprocessed sample battery feature data, resulting in sample sub-feature data corresponding to M time intervals.

[0201] According to one embodiment of this application, Figure 7 The various modules in the battery life prediction device shown can be individually or entirely combined into one or more units, or one or more of these units can be further divided into multiple functionally smaller sub-units to achieve the same operation without affecting the technical effects of the embodiments of this application. The above modules are based on logical functional division; in practical applications, the function of one module can be implemented by multiple units, or the function of multiple modules can be implemented by one unit. In other embodiments of this application, the battery life prediction device may also include other units; in practical applications, these functions can also be implemented with the assistance of other units, and can be implemented collaboratively by multiple units.

[0202] In this embodiment, by training the initial life prediction model, a high-performance target life prediction model is obtained, which can improve the accuracy of battery life prediction. Data preprocessing is performed on the initial battery feature data of the target battery to reduce outlier data and remove irrelevant feature data, facilitating subsequent battery life prediction processing and reducing the computational load of subsequent predictions, thus improving the efficiency of life prediction. Furthermore, wavelet transform processing is performed on the preprocessed target battery feature data to extract effective feature data reflecting the life decay trend of the target battery. It can be seen that through preprocessing and wavelet transform, effective feature data affecting the life decay of the target battery can be extracted from the initial battery feature data. Simultaneously, wavelet transform processing can fit the target battery feature data, making the obtained effective feature data smoother and effectively reducing the noise impact of discrete data in the target battery feature data. Furthermore, temporal correlation feature extraction is performed on the effective feature data to obtain temporal correlation feature data. By considering the temporal correlation between effective feature data, the accuracy of feature extraction can be improved. Predicting the target battery life based on the temporal correlation feature data can improve the accuracy of battery life prediction. Meanwhile, this application implements a target battery life prediction model to predict the target battery life, which can improve the accuracy and efficiency of life prediction.

[0203] Please see Figure 8 , Figure 8 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Figure 8 As shown, the computer device 1000 may include a processor 1001, a network interface 1004, and a memory 1005. Furthermore, the computer device 1000 may also include a user interface 1003 and at least one communication bus 1002. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen and a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be high-speed RAM or non-volatile memory, such as at least one disk storage device. Optionally, the memory 1005 may also be at least one storage device located remotely from the processor 1001. Figure 8 As shown, the memory 1005, which is a computer-readable storage medium, may include an operating system, a network communication module, a user interface module, and a device control application.

[0204] exist Figure 8In the computer device 1000 shown, the network interface 1004 provides network communication functionality; the user interface 1003 is mainly used to provide an input interface for the user; and the processor 1001 can be used to call the device control application stored in the memory 1005 to achieve:

[0205] The initial battery characteristic data of the target battery under the operating conditions is obtained, and the initial battery characteristic data is preprocessed to obtain the target battery characteristic data.

[0206] Wavelet transform is performed on the target battery feature data to obtain effective feature data that reflects the life decay trend of the target battery.

[0207] Temporal correlation feature extraction is performed on the effective feature data to obtain temporal correlation feature data;

[0208] Based on time-series correlation feature data, the target battery life is predicted.

[0209] It should be understood that the computer device 1000 described in the embodiments of this application can perform the foregoing... Figure 2 or Figure 4 The description of the battery life prediction method in the corresponding embodiments can also be performed as described above. Figure 7 The description of the battery life prediction device in the corresponding embodiments will not be repeated here.

[0210] Furthermore, it should be noted that this application embodiment also provides a computer-readable storage medium, which stores a computer program executed by the aforementioned battery life prediction device. The computer program includes program instructions, and when the processor executes the program instructions, it can execute the aforementioned... Figure 2 and Figure 4 The description of the battery life prediction method in the corresponding embodiments will not be repeated here.

[0211] Furthermore, the beneficial effects of using the same method will not be repeated here. For technical details not disclosed in the computer-readable storage medium embodiments involved in this application, please refer to the description of the method embodiments of this application. As an example, program instructions may be deployed and executed on a computer device, or on multiple computer devices located in one location, or on multiple computer devices distributed in multiple locations and interconnected through a communication network. Multiple computer devices distributed in multiple locations and interconnected through a communication network can constitute a blockchain system.

[0212] It should be noted that if the above embodiments of this application need to use the operation information and behavior logs of the logged-in object, permission or consent from the logged-in object is required, and relevant laws and regulations of the relevant countries and regions must be complied with.

[0213] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0214] The steps in the method of this application embodiment can be adjusted, combined, or deleted according to actual needs.

[0215] The modules in the device of this application embodiment can be merged, divided, and deleted according to actual needs.

[0216] Those skilled in the art will understand that implementing all or part of the processes in the above embodiments is possible.

[0217] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0218] The above description discloses only preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A battery life prediction method, characterized by, The method comprises the following steps: obtaining initial battery characteristic data of a target battery under a use condition, performing data preprocessing on the initial battery characteristic data to obtain target battery characteristic data; the target battery characteristic data comprises standard characteristic data corresponding to K time intervals respectively; By using the wavelet transform layer in the target lifetime prediction model, time interval K is obtained from K time intervals. i The corresponding standard feature data corresponds to an adaptive sliding window for the time interval K. i The corresponding standard feature data is divided to obtain the time interval K. i The standard sub-feature data corresponding to the time series in the data, where i is a positive integer less than or equal to K, are subjected to wavelet transform on the standard sub-feature data corresponding to each time series to obtain the wavelet decomposition coefficients corresponding to the target battery feature data; the wavelet decomposition coefficients corresponding to the target battery feature data include the frequencies in the first frequency set and the frequencies in the second frequency set, where the frequencies in the first frequency set are greater than the frequencies in the second frequency set; performing threshold quantization on the frequencies in the first frequency set according to a target adjustment threshold to obtain a quantized first frequency set; determining the frequencies in the quantized first frequency set and the frequencies in the second frequency set as wavelet decomposition coefficients after noise reduction; performing wavelet reconstruction on the wavelet decomposition coefficients after noise reduction to obtain effective characteristic data reflecting a life attenuation trend of the target battery; performing time sequence correlation characteristic extraction on the effective characteristic data to obtain time sequence correlation characteristic data; based on the time sequence correlation characteristic data, predicting target battery life of the target battery.

2. The method of claim 1, wherein, The data preprocessing on the initial battery characteristic data to obtain target battery characteristic data comprises the following steps: performing data type conversion on the initial battery characteristic data to obtain battery characteristic data with a target data type, screening battery characteristic data associated with battery life attenuation from the battery characteristic data with the target data type as correlation characteristic data; performing deletion processing on abnormal characteristic data in the correlation characteristic data to obtain processed correlation characteristic data, obtaining sampling characteristics of the processed correlation characteristic data, and performing segmentation on the processed correlation characteristic data according to the sampling characteristics of the processed correlation characteristic data to obtain correlation sub-characteristic data corresponding to K time intervals respectively; K is a positive integer greater than 1; performing normalization processing on the correlation sub-characteristic data corresponding to the K time intervals respectively to obtain standard characteristic data corresponding to the K time intervals respectively; determining the standard characteristic data corresponding to the K time intervals respectively as the target battery characteristic data.

3. The method of claim 1, wherein, The time sequence correlation characteristic extraction on the effective characteristic data to obtain time sequence correlation characteristic data comprises the following steps: obtaining a data correlation degree between effective characteristic data at time t and effective characteristic data at time t-1 through a time sequence correlation characteristic extraction layer in a target life prediction model; the characteristic data at time t and the characteristic data at time t-1 both belong to the effective characteristic data; performing weighted processing on the effective characteristic data at time t-1 according to the data correlation degree to obtain first weighted characteristic data, obtaining an input control parameter in the time sequence correlation characteristic extraction layer, performing weighted processing on the effective characteristic data at time t according to the input control parameter to obtain second weighted characteristic data; performing fusion on the first weighted characteristic data and the second weighted characteristic data to obtain important characteristic data in the effective characteristic data at time t; performing characteristic extraction on the important characteristic data according to a weight matrix and a bias parameter in the time sequence correlation characteristic extraction layer to obtain correlation characteristic data corresponding to time t. Until the associated feature data corresponding to the effective feature data is acquired, the associated feature data corresponding to the effective feature data is determined as the time sequence associated feature data.

4. The method of claim 3, wherein, The method further comprises: acquiring an initial life prediction model, sample battery feature data of a sample battery under a use condition, and a battery life label corresponding to the sample battery; the sample battery is associated with the target battery; performing life prediction on the sample battery based on the sample battery feature data by using the initial life prediction model, to obtain a predicted battery life of the sample battery; adjusting model parameters in the initial life prediction model according to the predicted battery life and the battery life label, and determining the initial life prediction model after parameter adjustment as the target life prediction model when the initial life prediction model after parameter adjustment meets a convergence condition.

5. The method of claim 4, wherein, The method further comprises: performing data preprocessing on the sample battery feature data by using the initial life prediction model, to obtain preprocessed sample battery feature data, and dividing the preprocessed sample battery feature data to obtain sample sub-feature data corresponding to M time intervals respectively; M is a positive integer greater than 1; performing wavelet transform processing on the sample sub-feature data corresponding to the M time intervals respectively by using a wavelet transform layer in the initial life prediction model, to obtain sample effective feature data corresponding to the M time intervals respectively; performing time sequence associated feature extraction on the sample effective feature data corresponding to the M time intervals respectively by using a time sequence associated feature extraction layer in the initial life prediction model, to obtain sample time sequence associated feature data corresponding to the M time intervals respectively, and predicting the predicted battery life corresponding to the sample battery based on the sample time sequence associated feature data corresponding to the M time intervals respectively.

6. The method of claim 5, wherein, The method further comprises: acquiring sampling features of the preprocessed sample battery feature data; determining an adaptive sliding window corresponding to the preprocessed sample feature data according to the sampling features of the preprocessed sample battery feature data; dividing the preprocessed sample battery feature data by using the adaptive sliding window corresponding to the preprocessed sample feature data, to obtain the sample sub-feature data corresponding to the M time intervals respectively.

7. A battery life prediction apparatus characterized by comprising: The method further comprises: a preprocessing module, configured to acquire initial battery feature data of a target battery under a use condition, and perform data preprocessing on the initial battery feature data to obtain target battery feature data; the target battery feature data comprises standard feature data corresponding to K time intervals respectively; a wavelet transform module comprising a wavelet decomposition unit, a noise reduction processing unit, and a wavelet reconstruction unit; The wavelet decomposition unit is configured to obtain, through a wavelet transform layer in the target life prediction model, time interval K i corresponding to the adaptive sliding window, time interval K i corresponding to the standard feature data, obtain standard sub-feature data corresponding to each time sequence in the time interval K i , i is a positive integer less than or equal to K, and wavelet transform is performed on the standard sub-feature data corresponding to each time sequence to obtain wavelet decomposition coefficients corresponding to the target battery feature data; the wavelet decomposition coefficients corresponding to the target battery feature data include frequencies in a first frequency set and frequencies in a second frequency set, the frequencies in the first frequency set are greater than the frequencies in the second frequency set. the noise reduction processing unit, configured to perform threshold quantization on frequencies in the first frequency set according to a target adjustment threshold, to obtain a quantized first frequency set; The wavelet reconstruction unit is configured to determine the frequencies in the quantized first frequency set and the frequencies in the second frequency set as wavelet decomposition coefficients after noise reduction, and perform wavelet reconstruction on the wavelet decomposition coefficients after noise reduction to obtain effective feature data reflecting the life attenuation trend of the target battery. The feature extraction module is configured to perform time sequence correlation feature extraction on the effective feature data to obtain time sequence correlation feature data. The first prediction module is configured to predict a target battery life of the target battery based on the time sequence correlation feature data.

8. A computer device, comprising: The computer program is stored in the computer readable storage medium and is adapted to be loaded and executed by the processor, so that the computer device having the processor performs the method in any one of claims 1-6. The computer program is stored in the computer readable storage medium and is adapted to be loaded and executed by the processor, so that the computer device having the processor performs the method in any one of claims 1-6. ​ 9. A computer-readable storage medium, characterized in that, ​

Images