Battery SOH estimation method, energy storage device and computer device

Abstract

The invention discloses a battery SOH estimation method, energy storage equipment and computer equipment, relates to the technical field of energy storage, and aims to solve the problem of how to accurately estimate the SOH of a battery. The battery SOH estimation method comprises the following steps: acquiring battery SOH related data in a target time period; and data segmentation is performed on the battery SOH related data in the target time period to obtain a target data sequence, the target data sequence comprises data of a plurality of time slices, and the duration of each time slice is equal. And inputting the target data sequence into a pre-trained target N-BEATS model to obtain a battery SOH predicted value. And determining the battery SOH as a battery SOH predicted value. The target data sequence is target domain data, and the pre-trained target N-BEATS model is applied to the target domain data through transfer learning, so that the target N-BEATS model not only can meet the requirement of SOH prediction precision under the historical working condition corresponding to the source domain data, but also can meet the requirement of SOH prediction precision under the target working condition corresponding to the target domain data; therefore, the accuracy and reliability of battery SOH estimation are improved.

Description

Technical Field

[0001] The present application relates to the field of energy storage technology, and in particular to a battery SOH estimation method, energy storage equipment and computer equipment. Background Art

[0002] The state of health (SOH) of a battery reflects the battery's ability to charge or discharge, efficiency, and voltage characteristics, and is a key indicator for evaluating the battery's current performance relative to its initial performance. The battery's SOH is affected by many factors, such as the voltage, current, and temperature of the charge or discharge. As the battery's usage time increases, the battery's SOH gradually decreases, which manifests as battery aging and performance degradation. Accurately estimating the battery's SOH can help users take measures in advance, which helps avoid battery failures and thus improves the safety and reliability of the power system. However, current battery models generally only meet the requirements for SOH prediction accuracy under historical conditions, and the prediction accuracy of SOH under new conditions is poor. Summary of the invention

[0003] In view of this, the embodiments of the present application provide a battery SOH estimation method, an energy storage device, and a computer device, aiming to solve the problem of how to accurately estimate the battery SOH.

[0004] The first aspect of the embodiment of the present application provides a battery SOH estimation method, the method comprising: obtaining battery SOH related data of a target time period. Performing data segmentation on the battery SOH related data of the target time period to obtain a target data sequence, the target data sequence including data of multiple time segments, each time segment having the same duration. Inputting the target data sequence into a pre-trained target N-BEATS model to obtain a battery SOH prediction value. Determining the battery SOH as the battery SOH prediction value.

[0005] In this embodiment, the target data sequence is target domain data, and the target domain data is independent of the source domain data used for N-BEATS model training. The pre-trained target N-BEATS model is applied to the target domain data through transfer learning, so that the target N-BEATS model can not only meet the requirements of SOH prediction accuracy under the historical operating conditions corresponding to the source domain data, but also meet the requirements of SOH prediction accuracy under the target operating conditions corresponding to the target domain data, thereby improving the accuracy and reliability of battery SOH estimation.

[0006] In one embodiment, determining the battery SOH as the battery SOH prediction value includes: calculating the deviation between the battery SOH prediction value and the battery SOH target value. When the deviation satisfies a deviation condition, determining the battery SOH as the battery SOH prediction value.

[0007] In another embodiment, the method further includes: when the deviation does not meet the deviation condition, optimizing the target N-BEATS model by adjusting the model parameters to adjust the battery SOH prediction value until the deviation between the battery SOH prediction value and the battery SOH target value meets the deviation condition.

[0008] In another embodiment, the method further includes: obtaining battery SOH related data for a training time period. Performing data segmentation on the battery SOH related data for the training time period to obtain a training data sequence. Dividing the training data sequence into a training data set and a verification data set. Training the N-BEATS model using the training data set. Verifying the accuracy of the N-BEATS model using the verification data set. When the accuracy of the N-BEATS model meets the accuracy condition, storing the N-BEATS model. Determining a target N-BEATS model based on the stored N-BEATS model.

[0009] In this embodiment, the training data sequence is source domain data. By training the N-BEATS model on the source domain data, the common features and trends of the source domain data are obtained, so that the N-BEATS model can deeply understand the deep-level features of the source domain data, thereby improving the prediction ability of the N-BEATS model on similar data. Moreover, since the N-BEATS model is composed of a fully connected layer and has a lightweight characteristic, it can reduce the occupation of computing resources in battery SOH estimation, thereby improving the efficiency of battery SOH estimation.

[0010] In another embodiment, determining a target N-BEATS model according to a stored N-BEATS model includes: when the number of iterations reaches a maximum number of iterations and the number of stored N-BEATS models is greater than 0, determining the target N-BEATS model to be the N-BEATS model with the largest number of iterations among the stored N-BEATS models. When the number of iterations reaches a maximum number of iterations and the number of stored N-BEATS models is 0, determining the target N-BEATS model to be the N-BEATS model with the maximum number of iterations.

[0011] In another embodiment, the accuracy of the N-BEATS model satisfies the accuracy condition including: the current prediction error is less than the minimum prediction error of the previous iteration, or the current prediction error is less than the error threshold, or the current prediction error is within the target error range. The current prediction error is the prediction error of the current iteration number, and the prediction error includes the mean square error or the mean absolute error.

[0012] In another embodiment, the deviation satisfies the deviation condition including: the deviation is less than a deviation threshold, or the deviation is within a target deviation range.

[0013] In another embodiment, after obtaining the battery SOH related data of the target time period, the method further includes: preprocessing the battery SOH related data of the target time period, the preprocessing including data cleaning and standardization. Data cleaning includes filtering data through a target value range, the target value range includes a target lower limit value and a target upper limit value, the target lower limit value is obtained by subtracting the average value of the battery SOH related data from 3 times the standard deviation, and the target upper limit value is obtained by adding the average value of the battery SOH related data to 3 times the standard deviation.

[0014] The second aspect of an embodiment of the present application provides an energy storage device, which includes a processor and a battery pack connected to the processor, the battery pack includes at least one battery, and the processor is used to: obtain battery SOH related data of at least one battery in a target time period. Perform data segmentation on the battery SOH related data of the target time period to obtain a target data sequence, the target data sequence includes multiple time segments, and the duration of each time segment is equal. Input the target data sequence into a pre-trained target N-BEATS model to obtain a battery SOH prediction value. Determine the battery SOH as the battery SOH prediction value.

[0015] A third aspect of an embodiment of the present application provides a computer device, comprising a processor, the processor being electrically connected to a battery pack, the battery pack comprising at least one battery, the processor being used to: obtain battery SOH related data of at least one battery in a target time period. Perform data segmentation on the battery SOH related data of the target time period to obtain a target data sequence, the target data sequence comprising a plurality of time segments, each of which has an equal duration. Input the target data sequence into a pre-trained target N-BEATS model to obtain a battery SOH prediction value. Determine the battery SOH as the battery SOH prediction value.

[0016] It can be understood that the beneficial effects of the energy storage device provided in the second aspect and the computer device provided in the third aspect of the embodiment of the present application are substantially the same as the beneficial effects of the battery SOH estimation method provided in the first aspect, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the structure of the N-BEATS model provided as an example.

[0018] Figure 2 An example is provided Figure 1 Schematic diagram of the structure of stacked blocks in the N-BEATS model shown.

[0019] Figure 3 An example is provided Figure 1 Schematic diagram of the structure of blocks in stacking blocks shown.

[0020] Figure 4is a flowchart of the training process of the N-BEATS model provided by an example.

[0021] Figure 5 is a flowchart of the transfer learning process of the N-BEATS model provided as an example. DETAILED DESCRIPTION

[0022] It should be noted that in the embodiments of the present application, "multiple" refers to two or more than two. The terms "first", "second", etc. in the specification, claims and drawings of the present application are used to distinguish similar objects, rather than to describe a specific order or sequence. The method disclosed in the embodiments of the present application or the method shown in the flowchart includes one or more steps for implementing the method. Without departing from the scope of the claims, the execution order of multiple steps can be interchangeable with each other, and some of the steps can also be deleted.

[0023] The state of health (SOH) of a battery can reflect the battery's ability to charge or discharge, efficiency, and voltage characteristics, and is a key indicator for evaluating the battery's current performance relative to its initial performance. However, the battery's SOH cannot be measured directly, and the ratio of the battery's current capacity to its initial capacity is usually used to represent the battery's SOH. The battery's SOH is affected by many factors, such as the voltage, current, and temperature of the charge or discharge. As the battery's usage time increases, the battery's SOH gradually decreases, which is manifested as battery aging and performance degradation. Accurately estimating the battery's SOH can help users take measures in advance, which is conducive to avoiding battery failures, thereby improving the safety and reliability of the power system. For example, in the field of electric vehicles, accurately estimating the battery's SOH can extend the battery's service life and optimize the charging or discharging strategy, thereby improving driving safety and reducing maintenance costs. In the field of portable electronic devices, accurately estimating the battery's SOH can prevent accidental power outages, thereby improving the user experience. In the field of renewable energy systems, accurately estimating the battery's SOH helps stabilize power supply, thereby improving the reliability of renewable energy systems.

[0024] The estimation of battery SOH can be done by electrochemical model methods, data-driven model methods, etc. The electrochemical model method can be modeled according to the internal mechanism of battery aging, and the battery SOH can be estimated by simulating the physical and chemical changes inside the battery. However, the electrochemical model method requires detailed battery parameters, such as electrode materials, electrolyte composition, etc., to calibrate and verify the model, and data acquisition is difficult. Moreover, the electrochemical model method usually uses complex partial differential equations, resulting in high requirements for memory and computing power. The data-driven model method can use tools such as machine learning and deep learning to process large-scale data and complex working conditions, and estimate the battery SOH through data analysis. The data-driven model method requires high-quality historical data for model training. The model estimates the battery SOH under historical working conditions more accurately, but has poor compatibility with different working conditions.

[0025] The embodiment of the present application uses a pre-trained Neural Basis Expansion Analysis for Interpretable Time Series Forecasting (N-BEATS) model for interpretable time series estimation to estimate the battery SOH. By training the N-BEATS model on the source domain data, the common features and trends of the source domain data are obtained, so that the N-BEATS model can deeply understand the deep-level features of the source domain data, thereby improving the prediction ability of the N-BEATS model on similar data. By applying the pre-trained N-BEATS model to the target domain data through transfer learning, the target N-BEATS model can not only meet the requirements of SOH prediction accuracy under the historical working conditions corresponding to the source domain data, but also meet the requirements of SOH prediction accuracy under the target working conditions corresponding to the target domain data, thereby improving the accuracy and reliability of the battery SOH estimation.

[0026] The structural diagram of the N-BEATS model is described in detail below.

[0027] like Figure 1 As shown, the N-BEATS model includes N stack blocks, such as 1 , Stack 2 ,…,Stack N , N is a positive integer. Stack 1 The input of the model is the data sequence, which is a time series sorted by time, such as the data sequence , the data sequence X includes T time segments of data, each of which is obtained by slicing the data set according to the time window. The time window can be set as needed, and T is a positive integer. The output of each stacking block includes the stacking block prediction value (Stackforecast) and the stacking block residual (Stack residual). Stacking block Stack 2 -Stack N The input of each stacking block is the stacking block prediction value output by the previous stacking block, for example, 2 The input is the stack block Stack 1 Output stacked block prediction value. All stacked blocks Stack 1 -Stack N The output stacked block prediction values ​​are added together to obtain the model prediction value (Modelforecast).

[0028] Stack 2 For example. Figure 2 As shown, the stack block Stack 2 It includes M blocks, for example, 1 , Block 2 ,…,Block M , M is a positive integer. Block 1 The input is the stacking block prediction value output by the previous stacking block, and the previous stacking block is the stacking block Stack 1 The output of each block includes the block forecast and the block backcast. 2 -Block M The input of each block is the result of subtracting the reconstruction value of the previous block from its input. For example, 2 The input is a block 1 The block reconstruction value and block Block 1 The result of the calculation is obtained by subtracting the input of Block. M The result of subtracting the block reconstruction value from its input is the stacking block Stack 2 The stacked block residuals of all blocks 1 -Block M The block prediction values ​​are added to get the stacked block Stack 2 The predicted value of the stacked blocks. All stacked blocks Stack 1 -Stack N The stacked block residuals are used to evaluate the model accuracy.

[0029] Stack2 Block 2 For example. Figure 3 As shown, Block 2 Includes multiple fully connected layers (FC), such as the fully connected layer FC 1 , FC 2 ,…,FC K , K is a positive integer. Fully connected layer FC 1 The input is a block 2 Input. Fully connected layer FC 2 -FC K The input of each fully connected layer is the output of the previous fully connected layer, for example, the fully connected layer FC 2 The input is the fully connected layer FC 1 Output of the fully connected layer FC K The output of is linearly fitted to obtain the fitted value, calling function g f Process the fitted values ​​to get the block Block 2 The block prediction value of the function g is called b Process the fitted values ​​to get the block Block 2 The block reconstruction value of . Function g f and function g b It is used to make the block learn specific properties of the time series to enhance the interpretability of the N-BEATS model. For example, the function g f and function g b Trigonometric functions, such as cosine (cos) or sine (sin), can be included to allow the block to learn trend and seasonality of a time series. Trend is the long-term directional change in a time series over time. Seasonality is a pattern in a time series that repeats periodically over time.

[0030] For example, in block i Including four fully connected layers FC 1 -FC 4 For example, i is a positive integer, and i≤K. Block i The input is , fully connected layer FC 1 The output is , fully connected layer FC 2 The output is , fully connected layer FC 3 The output is , fully connected layer FC 4 The output is , Block i The block prediction value is , Block iThe block reconstruction value is , the function Linear is a linear fitting function.

[0031] The above is a detailed description of the structure of the N-BEATS model. The following is a detailed description of the training process of the N-BEATS model.

[0032] like Figure 4 As shown in Figure 1, the training process of the N-BEATS model includes the following steps: S101. Obtain battery SOH related data during a training period.

[0033] The battery SOH related data includes the voltage, current and temperature of the battery during charging or discharging. The battery SOH can be expressed as SOH = f(V, I, T), where V is the voltage of the battery during charging or discharging, I is the current of the battery during charging or discharging, and T is the temperature of the battery during charging or discharging. The duration of the training period can be set as needed.

[0034] S102: pre-process the battery SOH related data in the training time period to obtain training pre-processed data.

[0035] In this embodiment, preprocessing includes data cleaning and standardization. Data cleaning may include removing noise and outliers, and data cleaning may improve the quality and consistency of data. Standardization may be replaced by normalization, and the scale of data may be unified through standardization or normalization.

[0036] Exemplarily, data cleaning may adopt the 3σ method as shown in formula (1). The 3σ method is to filter data by a target value range, that is, to remove data outside the target value range. The target value range includes a target lower limit value and a target upper limit value.

[0037] (1) Among them, lower_bound is the target lower limit value, upper_bound is the target upper limit value, mean is the average value of battery SOH related data, and σ is the standard deviation of battery SOH related data.

[0038] In another embodiment, the preprocessing further includes feature extraction. Feature extraction is used to extract key data, and the quality of data can be improved through feature extraction.

[0039] In other embodiments, step S102 may not be performed.

[0040] S103: Segment the training preprocessed data to obtain a training data sequence.

[0041] In this embodiment, data segmentation of the training preprocessed data includes slicing the training preprocessed data according to the time window, dividing the training preprocessed data into data of multiple time segments to form a training data sequence. For example, the training data sequence , the training data sequence X includes data of T time segments. The training data sequence has time series characteristics and meets the requirements of the N-BEATS model for input data.

[0042] It can be understood that the duration of each time segment is equal, and the duration of the time segment is equal to the duration of the time window. The duration of the time window can be set as needed.

[0043] S104: Divide the training data sequence into a training data set and a validation data set.

[0044] In this embodiment, the training data sequence can be divided in proportion. For example, the training data sequence is divided in proportion a:b, and a / (a+b) of the data in the training data sequence is divided into a training data set, and the remaining b / (a+b) of the data is divided into a verification data set.

[0045] In other embodiments, the training data sequence can also be divided by time segments. For example, the training data sequence can be divided by the jth time segment, and the data in the time segments before the jth time segment and the data in the jth time segment in the training data sequence can be divided into the training data set, and the data in the time segments after the jth time segment can be divided into the verification data set.

[0046] S105. Train the N-BEATS model using the training data set.

[0047] In this embodiment, the goal of model training is to optimize model parameters through continuous iteration to reduce the loss value until the number of iterations reaches the maximum number of iterations. Model parameters may include the number of stacked blocks, the number of blocks, the number of neurons in the fully connected layer, the duration of the training data sequence, the duration of the time segment, the learning rate, and the number of iterations.

[0048] The loss value is calculated by a loss function. For example, a loss function as shown in formula (2) may be used.

[0049] (2) Among them, Loss is the loss value, MSE is the mean square error, n is the number of observations, is the true value of the ith observation, is the predicted value of the ith observation.

[0050] In another example, a loss function as shown in formula (3) may be used.

[0051] (3) Among them, Loss is the loss value, MAE is the mean absolute error, n is the number of observations, is the true value of the ith observation, is the predicted value of the ith observation.

[0052] S106. Use the validation data set to verify the accuracy of the N-BEATS model.

[0053] In this embodiment, the goal of model validation is to optimize the model parameters through continuous iteration to reduce the prediction error until the number of iterations reaches the maximum number of iterations. The prediction error can be expressed as mean square error or mean absolute error.

[0054] S107. Determine whether the accuracy of the N-BEATS model meets the accuracy requirement.

[0055] If yes, execute steps S108-S109; if no, execute step S109.

[0056] In this embodiment, the model accuracy is measured by the prediction error. The model accuracy meets the accuracy conditions including: the current prediction error is less than the minimum prediction error of the previous iteration, or the current prediction error is less than the error threshold, or the current prediction error is within the target error range. Among them, the current prediction error is the prediction error of the current iteration. The error threshold and the target error range can be set as needed.

[0057] Exemplarily, in the rth iteration, when determining the prediction error E of the rth iteration r Less than the minimum prediction error E before the rth iteration min , indicating that the accuracy of the N-BEATS model at the rth iteration meets the accuracy condition. Or, when determining the prediction error E of the rth iteration r is less than the error threshold, indicating that the accuracy of the N-BEATS model at the rth iteration meets the accuracy condition. r Within the target error range, it means that the accuracy of the N-BEATS model at the rth iteration meets the accuracy condition. Where r is a positive integer, and 2≤r≤max(r), max(r) is the maximum number of iterations.

[0058] S108. Store the N-BEATS model.

[0059] In this embodiment, when the accuracy of the N-BEATS model meets the accuracy condition, it means that the N-BEATS model of the current iteration number is better than the previous N-BEATS model.

[0060] S109: Determine whether the iteration is finished.

[0061] If yes, execute step S110; if no, return to execute step S105.

[0062] In this embodiment, the end of iteration means that the number of iterations reaches the maximum number of iterations.

[0063] S110. Determine the target N-BEATS model.

[0064] In this embodiment, when the number of iterations reaches the maximum number of iterations and the number of stored N-BEATS models is greater than 0, the target N-BEATS model is the N-BEATS model with the largest number of iterations among the stored N-BEATS models. When the number of iterations reaches the maximum number of iterations and the number of stored N-BEATS models is 0, the target N-BEATS model is the N-BEATS model with the maximum number of iterations.

[0065] The above is a detailed description of the training process of the N-BEATS model. In this embodiment, the training data sequence is source domain data. By training the N-BEATS model on the source domain data, the common features and trends of the source domain data are obtained, so that the N-BEATS model can deeply understand the deep-level features of the source domain data, thereby improving the prediction ability of the N-BEATS model on similar data. Moreover, since the N-BEATS model is composed of a fully connected layer and has a lightweight characteristic, it can reduce the usage of computing resources in battery SOH estimation, thereby improving the efficiency of battery SOH estimation.

[0066] Among them, the deep-level features of source domain data include operating conditions and key features. Operating conditions may include temperature conditions, charge and discharge conditions, load conditions and special conditions. Temperature conditions include high temperature conditions and low temperature conditions. Charge and discharge conditions include deep discharge conditions, shallow charge and discharge conditions, fast charge conditions and slow charge conditions. Load conditions include high load conditions and low load conditions. Special conditions include extreme conditions and abnormal conditions. Key features may include time correlation, trend, seasonality, randomness, stability, autocorrelation, mutation points and outliers.

[0067] With the diversification of battery types and application scenarios, battery models trained based on a single scenario are difficult to meet the performance requirements under different working conditions. Therefore, by introducing transfer learning technology, the target N-BEATS model can efficiently adapt to different battery types and working conditions, reducing the time and data requirements for retraining, thereby improving the battery SOH prediction accuracy and model generalization capabilities. In addition, by performing transfer learning on the target N-BEATS model, the model deployment efficiency can be improved, and the state estimation in different environments can be optimized, such as improving the prediction accuracy of key parameters such as SOH and state of charge (SOC). Therefore, integrating transfer learning technology on the basis of the target N-BEATS model is crucial to improving the adaptability of battery systems and reducing engineering development costs.

[0068] The following describes the process of transfer learning in detail based on the pre-trained target N-BEATS model.

[0069] like Figure 5 As shown in Figure 1, the process of transfer learning includes the following steps: S201. Obtain battery SOH related data for a target time period.

[0070] The target time period is any time period after the training time period, and the length of the target time period can be set as needed.

[0071] S202 : Preprocess the battery SOH related data in the target time period to obtain target preprocessed data.

[0072] S203: Segment the target preprocessed data to obtain a target data sequence.

[0073] The target data sequence includes data of multiple time segments.

[0074] The implementation method of steps S201-S203 is Figure 4 The steps S101-S103 shown are the same and will not be repeated here.

[0075] S204: Input the target data sequence into the pre-trained target N-BEATS model to obtain the battery SOH prediction value.

[0076] Among them, the battery SOH prediction value is the model prediction value output by the target N-BEATS model.

[0077] S205 , calculating the deviation between the battery SOH prediction value and the battery SOH target value.

[0078] The battery SOH target value can be set as needed.

[0079] S206: Determine whether the deviation satisfies the deviation condition.

[0080] If yes, execute step S207; if no, execute step S208.

[0081] In this embodiment, the deviation satisfies the deviation condition including: the deviation is less than the deviation threshold, or the deviation is within the target deviation range. The deviation threshold and the target deviation range can be set as required.

[0082] S207 . Determine the battery SOH as a battery SOH prediction value.

[0083] In this embodiment, when the deviation satisfies the deviation condition, it indicates that the target N-BEATS model has a higher accuracy under the current working condition.

[0084] S208. Optimize the target N-BEATS model by adjusting model parameters.

[0085] In this embodiment, when the deviation does not meet the deviation condition, it means that the accuracy of the target N-BEATS model under the current working condition is low. The battery SOH prediction value is adjusted by optimizing the target N-BEATS model to reduce the deviation between the battery SOH prediction value and the battery SOH target value.

[0086] Methods for adjusting model parameters may include Bayesian optimization, learning curve analysis, cross-validation, and adjusting the learning rate.

[0087] After executing step S208, return to execute steps S204-S206 until the deviation meets the deviation condition.

[0088] The above specifically describes the process of transfer learning. In this embodiment, the target data sequence is the target domain data. The pre-trained target N-BEATS model is applied to the target domain data through transfer learning, so that the target N-BEATS model can not only meet the requirements of SOH prediction accuracy under the historical working conditions corresponding to the source domain data, but also meet the requirements of SOH prediction accuracy under the target working conditions corresponding to the target domain data, thereby improving the accuracy and reliability of battery SOH estimation. Furthermore, according to the prediction results of the target N-BEATS model, it is determined whether to fine-tune the target N-BEATS model to improve the accuracy of the target N-BEATS model, so that the target N-BEATS model has higher accuracy under different working conditions, thereby improving the accuracy and compatibility of the target N-BEATS model.

[0089] The various steps in the above embodiments may be applied to energy storage devices, computer devices, chips, computer-readable storage media, computer program products, and the like.

[0090] The energy storage device includes a processor and a battery pack. The processor is electrically connected to the battery pack, and the battery pack includes at least one battery. The processor can read computer instructions stored in the memory, and when the processor executes the computer instructions, each step in the above embodiment is implemented.

[0091] The computer device includes a processor. The processor is electrically connected to a battery pack, and the battery pack includes at least one battery. The processor can read computer instructions stored in a memory, and when the processor executes the computer instructions, each step in the above embodiment is implemented.

[0092] The chip includes a processor and an interface circuit, and the processor is electrically connected to the interface circuit. The interface circuit can read the computer instructions stored in the memory and send the computer instructions to the processor. When the processor executes the computer instructions, each step in the above embodiment is implemented.

[0093] The processor may include a central processing unit (CPU), a neural processing unit (NPU), an application processor (AP) and a digital signal processor (DSP), etc.

[0094] The computer-readable storage medium stores computer instructions, and when the processor executes the computer instructions, each step in the above embodiment is implemented.

[0095] Computer-readable storage media include volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules or other data. Computer-readable storage media include, but are not limited to, Random Access Memory (RAM), Read-Only Memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), flash memory or other memory, Compact Disc Read-Only Memory (CD-ROM), Digital Versatile Disc (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer.

[0096] The computer program product includes computer instructions. When a processor executes the computer instructions, each step in the above embodiments is implemented.

[0097] The embodiments of the present application are described in detail above in conjunction with the accompanying drawings, but the present application is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in the relevant technical field without departing from the purpose of the present application.

Claims

1. A battery SOH estimation method, characterized in that: The method comprises: Obtain battery SOH related data for the target time period; Segmenting the battery SOH-related data of the target time period to obtain a target data sequence, wherein the target data sequence includes data of multiple time segments, and each time segment has the same duration; Input the target data sequence into the pre-trained target N-BEATS model to obtain a battery SOH prediction value; The battery SOH is determined to be the battery SOH prediction value.

2. The battery SOH estimation method according to claim 1, characterized in that: The determining the battery SOH as the battery SOH prediction value comprises: Calculating the deviation between the battery SOH prediction value and the battery SOH target value; When the deviation satisfies the deviation condition, the battery SOH is determined to be the battery SOH prediction value.

3. The battery SOH estimation method according to claim 2, characterized in that: The method further comprises: When the deviation does not satisfy the deviation condition, the target N-BEATS model is optimized by adjusting the model parameters to adjust the battery SOH prediction value until the deviation between the battery SOH prediction value and the battery SOH target value satisfies the deviation condition.

4. The battery SOH estimation method according to any one of claims 1 to 3, characterized in that: The method further comprises: Obtain battery SOH related data during the training period; Segmenting the battery SOH related data in the training time period to obtain a training data sequence; Dividing the training data sequence into a training data set and a validation data set; Using the training data set to train the N-BEATS model; Using the validation data set to verify the accuracy of the N-BEATS model; When the accuracy of the N-BEATS model meets the accuracy condition, storing the N-BEATS model; The target N-BEATS model is determined according to the stored N-BEATS model.

5. The battery SOH estimation method according to claim 4, characterized in that: The step of determining the target N-BEATS model according to the stored N-BEATS model includes: When the number of iterations reaches the maximum number of iterations, and the number of stored N-BEATS models is greater than 0, determining that the target N-BEATS model is the N-BEATS model with the largest number of iterations among the stored N-BEATS models; When the number of iterations reaches the maximum number of iterations and the number of stored N-BEATS models is 0, the target N-BEATS model is determined to be the N-BEATS model with the maximum number of iterations.

6. The battery SOH estimation method according to claim 4 or 5, characterized in that: The accuracy of the N-BEATS model meets the accuracy conditions including: The current prediction error is less than the minimum prediction error of the previous iteration, or the current prediction error is less than the error threshold, or the current prediction error is within the target error range; wherein the current prediction error is the prediction error of the current iteration number, and the prediction error includes the mean square error or the mean absolute error.

7. The battery SOH estimation method according to claim 2 or 3, characterized in that: The deviation satisfies the deviation condition including: the deviation is less than a deviation threshold, or the deviation is within a target deviation range.

8. The battery SOH estimation method according to any one of claims 1 to 7, characterized in that: After obtaining the battery SOH related data for the target time period, the method further includes: Preprocessing the battery SOH related data in the target time period, wherein the preprocessing includes data cleaning and standardization; The data cleaning includes filtering data through a target value range, wherein the target value range includes a target lower limit value and a target upper limit value, wherein the target lower limit value is obtained by subtracting the average value of the battery SOH related data from three times the standard deviation, and the target upper limit value is obtained by adding the average value of the battery SOH related data to three times the standard deviation.

9. An energy storage device, characterized in that: The device comprises a processor and a battery pack connected to the processor, wherein the battery pack comprises at least one battery, and the processor is used for: Acquiring battery SOH related data of the at least one battery in a target time period; Segmenting the battery SOH-related data of the target time period to obtain a target data sequence, wherein the target data sequence includes data of multiple time segments, and each time segment has the same duration; Input the target data sequence into the pre-trained target N-BEATS model to obtain a battery SOH prediction value; The battery SOH is determined to be the battery SOH prediction value.

10. A computer device, characterized in that: The device comprises a processor, the processor is electrically connected to a battery pack, the battery pack comprises at least one battery, and the processor is used to: Acquiring battery SOH related data of the at least one battery in a target time period; Segmenting the battery SOH-related data of the target time period to obtain a target data sequence, wherein the target data sequence includes data of multiple time segments, and each time segment has the same duration; Input the target data sequence into the pre-trained target N-BEATS model to obtain a battery SOH prediction value; The battery SOH is determined to be the battery SOH prediction value.

Images