Battery capacity prediction method and system based on multi-scale wavelet decomposition
Through the combination of multi-scale wavelet decomposition and ARIMA-LSTM-Informer model, the problem of capturing medium- and long-term trends and fluctuations in battery capacity prediction is solved, and high-precision and robust battery capacity prediction are achieved.
Patent Information
- Application Number
- CN202510547095.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-28
- Publication Date
- 2025-08-12
AI Technical Summary
When the existing battery capacity prediction methods process complex nonlinear characteristics in battery capacity data, it is difficult to capture long-term trends and short-term fluctuations at the same time, resulting in insufficient prediction accuracy and adaptability, especially in the case of limited data volume or high noise.
The multi-scale wavelet decomposition method is used to decompose the battery capacity time series into low-frequency approximate subsequences and high-frequency detail subsequences. The low-frequency approximate subsequence is predicted by the ARIMA-LSTM combination model, and the Informer model predicts the high-frequency detail subsequence, and the final battery capacity prediction value is obtained through data reconstruction.
It significantly improves the robustness and accuracy of battery capacity prediction, can effectively process complex nonlinear characteristics in battery capacity data, and improves the adaptability and prediction accuracy of the model.
Smart Images
Figure CN120468665A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of battery management, and in particular to a battery capacity prediction method and system based on multi-scale wavelet decomposition. Background Art
[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.
[0003] In electric vehicles, energy storage systems, and portable devices, battery state of health (SOH) is a crucial indicator for evaluating battery performance, particularly for accurately predicting battery capacity. Battery capacity decay is an irreversible process that occurs over time. Predicting battery capacity is not only crucial for battery life management but also effectively extends the battery's lifespan and improves the user experience.
[0004] Existing battery capacity prediction methods can be roughly divided into the following categories: 1) Complex nonlinear electrochemical models: These models simulate the physical and chemical properties of batteries and generally provide high prediction accuracy. However, these models are computationally complex and have stringent requirements for real-time online applications.
[0005] 2) Empirical models: These models usually use simplified mathematical models to describe the battery capacity decay process. Although they have a simple structure and high computational efficiency, they have poor prediction accuracy, are easily affected by noise, and have poor adaptability under changing working conditions.
[0006] 3) Data-Driven Models: Traditional battery capacity prediction methods typically use raw data directly for modeling, but this approach has significant limitations. Battery capacity exhibits complex nonlinear characteristics over the lifespan, making it difficult for traditional methods to effectively capture this nonlinearity, especially in battery degradation patterns. Furthermore, these models are susceptible to high-frequency noise when processing battery data, leading to overfitting. This can lead to poor generalization, especially when data is limited or noise is high. Battery capacity data contains both long-term, low-frequency trends and short-term, high-frequency fluctuations. A single model struggles to capture both, thus impacting prediction accuracy.
[0007] Based on the defects of the above methods, there is an urgent need for a battery capacity prediction method that can effectively capture the battery capacity degradation characteristics, has high accuracy and strong adaptability, so as to predict the future health status of the battery under different cycle conditions. Summary of the Invention
[0008] To address the shortcomings of existing technologies, the present invention provides a battery capacity prediction method and system based on multi-scale wavelet decomposition. This method can effectively predict the battery's future capacity under a specific cycle condition, even when only a partial initial capacity dataset is collected. This method effectively handles the complex nonlinear characteristics of battery capacity data, improving the model's robustness and prediction accuracy.
[0009] In order to achieve the above object, the present invention is implemented through the following technical solutions: A first aspect of the present invention provides a battery capacity prediction method based on multi-scale wavelet decomposition, comprising the following steps: Obtaining operating data of the battery to be tested and calculating the preliminary battery capacity of the battery to be tested; The preliminary battery capacity time series is decomposed using wavelet transform to obtain low-frequency approximate subsequences and high-frequency detail subsequences; Using the first prediction model to predict the low-frequency approximate subsequence, to obtain a low-frequency approximate subsequence prediction value; The second prediction model is used to capture the feature sequence of the high-frequency detail subsequence to obtain the prediction value of the high-frequency detail subsequence; Data is reconstructed based on the low-frequency approximate subsequence prediction value and the high-frequency detail subsequence prediction value to obtain the final battery capacity prediction value.
[0010] Furthermore, the specific steps of obtaining the operating data of the battery to be tested and calculating the preliminary battery capacity of the battery to be tested are as follows: Select the power battery to be tested, formulate the working conditions according to the actual application requirements, conduct battery aging experiments on the power battery to be tested, obtain the operating data of the battery under various working conditions, and calculate the number of cycles and the corresponding battery capacity based on the operating data.
[0011] Furthermore, the specific steps of using wavelet transform to decompose the preliminary battery capacity time series are as follows: Select db3 as the wavelet basis function and perform wavelet transform on the original battery capacity sequence to obtain the first layer of low-frequency approximate subsequence and the first high-frequency detail subsequence; Perform wavelet transform on the first layer low-frequency approximate subsequence to obtain the second layer low-frequency approximate subsequence and the second high-frequency detail subsequence; The second-layer low-frequency approximate subsequence is subjected to wavelet transform to obtain a third-layer low-frequency approximate subsequence and a third high-frequency detail subsequence.
[0012] Furthermore, the first prediction model includes an ARIMA model and an LSTM model. The ARIMA model is used to predict the residual sequence of the low-frequency approximate subsequence, and the LSTM model is used to predict the nonlinear part and short-term fluctuations in the residual sequence; the second prediction model includes an Informer model. The Informer model is used to capture the nonlinear part and short-term fluctuations of the high-frequency detail subsequence.
[0013] Furthermore, the specific steps of using the first prediction model to predict the low-frequency approximate subsequence are as follows: Use the ARIMA model to calculate the residual sequence of the low-frequency approximate subsequence and obtain the residual prediction sequence; Build an LSTM model and use the residual sequence to train the LSTM model; Use the trained LSTM model to predict the low-frequency approximate subsequence and obtain the predicted data; The predicted data is added to the residual prediction sequence to obtain the final low-frequency approximate subsequence prediction value.
[0014] Furthermore, the specific steps of using the second prediction model to capture the feature sequence of the high-frequency detail subsequence are as follows: Build an Informer model with a self-attention mechanism and train the Informer model using a known sample set; The trained Informer model is used to extract the features of the high-frequency detail subsequence and obtain the predicted value of the high-frequency detail subsequence.
[0015] Furthermore, the inverse wavelet transform is used to restore the low-frequency approximate subsequence layer by layer, thereby completing data reconstruction.
[0016] A second aspect of the present invention provides a battery capacity prediction system based on multi-scale wavelet decomposition, comprising: a data acquisition module configured to acquire operating data of the battery to be tested and calculate a preliminary battery capacity of the battery to be tested; a wavelet decomposition module configured to decompose the preliminary battery capacity time series using wavelet transform to obtain a low-frequency approximate subsequence and a high-frequency detail subsequence; a low-frequency prediction module configured to predict the low-frequency approximate subsequence using the first prediction model to obtain a low-frequency approximate subsequence prediction value; a high-frequency prediction module configured to capture a feature sequence of the high-frequency detail subsequence using a second prediction model to obtain a prediction value of the high-frequency detail subsequence; The data reconstruction module is configured to reconstruct data based on the low-frequency approximate subsequence prediction value and the high-frequency detail subsequence prediction value to obtain a final battery capacity prediction value.
[0017] A third aspect of the present invention provides a medium having a program stored thereon, which, when executed by a processor, implements the steps of the battery capacity prediction method based on multi-scale wavelet decomposition as described in the first aspect of the present invention.
[0018] The fourth aspect of the present invention provides a device comprising a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps of the battery capacity prediction method based on multi-scale wavelet decomposition as described in the first aspect of the present invention are implemented.
[0019] One or more of the above technical solutions have the following beneficial effects: The present invention discloses a battery capacity prediction method and system based on multi-scale wavelet decomposition. By performing three-layer wavelet decomposition on the battery capacity time series, the original sequence is decomposed into a low-frequency approximate subsequence and a high-frequency detail subsequence. The low-frequency approximate subsequence reflects the long-term trend and law of the battery capacity. The ARIMA-LSTM combination model can effectively capture the patterns of these long-term changes. The high-frequency detail subsequence usually contains complex and nonlinear short-term fluctuations. The Informer model can accurately capture and model the nonlinear changes of these high-frequency parts through the self-attention mechanism, thereby significantly improving the prediction accuracy of short-term fluctuations. Finally, the predicted value of the low-frequency approximate subsequence is added to the predicted value of the high-frequency detail subsequence to reconstruct the final battery capacity prediction value. This method can effectively handle the complex nonlinear characteristics in battery capacity data, and improve the robustness and prediction accuracy of the model.
[0020] The specific advantages are: (1) This paper uses a multi-scale wavelet decomposition method (using the db3-based wavelet) to decompose the battery capacity sequence into a low-frequency approximate subsequence and a high-frequency detail subsequence. This method can effectively reduce the complexity of the battery capacity sequence and extract fluctuations at different time scales separately, allowing the prediction task of each part to be specialized, thereby improving the overall prediction accuracy.
[0021] (2) The present invention adopts the ARIMA-LSTM combined model to predict low-frequency approximate sequences, where ARIMA first effectively captures the long-term trend and linear components in the low-frequency sequence, while LSTM is used to model the nonlinear part and short-term fluctuations in the ARIMA model prediction residual, which can improve the overall prediction accuracy and stability of the low-frequency approximate sequence.
[0022] (3) The present invention uses Informer to predict high-frequency detail sequences and models them through the self-attention mechanism of the Informer model. It can effectively identify and predict complex and nonlinear fluctuations in the short term, thereby improving the model's adaptability and accuracy to changes in battery capacity.
[0023] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.
[0025] Figure 1 This is a flow chart of a battery capacity prediction method based on multi-scale wavelet decomposition in Example 1 of the present invention; Figure 2 Schematic diagram of multi-scale wavelet decomposition in Example 1 of the present invention. DETAILED DESCRIPTION
[0026] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.
[0027] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or their combinations; Example 1: Embodiment 1 of the present invention provides a battery capacity prediction method based on multi-scale wavelet decomposition, comprising the following steps: S1: Obtain operating data of the battery to be tested and calculate the preliminary battery capacity of the battery to be tested.
[0028] S2: Use wavelet transform to decompose the preliminary battery capacity time series to obtain low-frequency approximate subsequences and high-frequency detail subsequences.
[0029] S3: Use the first prediction model to predict the low-frequency approximate subsequence to obtain a low-frequency approximate subsequence prediction value.
[0030] S4: Use the second prediction model to capture the feature sequence of the high-frequency detail subsequence to obtain a prediction value of the high-frequency detail subsequence.
[0031] S5: Reconstruct data based on the low-frequency approximate subsequence prediction value and the high-frequency detail subsequence prediction value to obtain the final battery capacity prediction value.
[0032] In S1, the power battery to be tested is selected, and the working conditions are formulated according to the actual application requirements. Battery aging experiments are carried out on the power batteries to be tested, and the operating data of the battery under various working conditions is obtained. The number of cycles is calculated based on the operating data. Corresponding battery capacity .
[0033] In this embodiment, the cycle conditions of the working condition include the current rate C-rate, discharge depth DOD, temperature and cut-off voltage of the battery charge and discharge; the discharge capacity under each cycle condition is calculated by the ampere-hour integration method and other methods, and the cycle number is obtained through this process. Corresponding battery capacity . In S2, according to the obtained battery capacity , select db3 wavelet as the base wavelet, wavelet transform decomposes the battery capacity time series into low-frequency approximate subsequences and high-frequency detail subsequences , where the low-frequency approximate subsequence refers to the low-frequency components extracted from the battery capacity time series by wavelet transform, which usually represents the long-term change trend of the battery capacity; while the high-frequency detail subsequence refers to the high-frequency components extracted by wavelet transform, which represents the short-term fluctuation or noise of the battery capacity. Figure 2 As shown, the specific steps are: S21: Select db3 (Daubechies 3) as the wavelet basis function, and calculate the original battery capacity sequence Perform wavelet transform to obtain the first layer low-frequency approximate subsequence A1[n] and the first high-frequency detail subsequence D1[n]. The calculation formula is: .
[0034] Among them, h0[n] and g0[n] are low-pass filter and high-pass filter respectively, which are determined by the selected wavelet basis function db3. Represents the original battery capacity time series data Data at a point in time, The index of the filter shift is used to define the convolution operation between the signal and the filter.
[0035] S22: Perform wavelet transform on the first-layer low-frequency approximate subsequence A1[n] to obtain the second-layer low-frequency approximate subsequence A2[n] and the second high-frequency detail subsequence D2[n]. The calculation formula is: .
[0036] S23: Perform wavelet transform on the second-level low-frequency approximate subsequence A2[n] to obtain the third-level low-frequency approximate subsequence A3[n] and the third high-frequency detail subsequence D3[n]. The calculation formula is: .
[0037] In S3, the first prediction model includes the ARIMA model and the LSTM model. The ARIMA model is used to predict the residual sequence of the low-frequency approximate subsequence, and the LSTM model is used to predict the nonlinear part and short-term fluctuations in the residual sequence. , determine the ARIMA model parameters (p, d, q) for prediction, obtain the residual sequence to train the LSTM model, and establish the ARIMA-LSTM model.
[0038] The specific steps of using the first prediction model to predict the low-frequency approximate subsequence are as follows: S31: Use the ARIMA model to calculate the residual sequence of the low-frequency approximate subsequence to obtain the residual prediction sequence.
[0039] S311: For the third layer low-frequency approximate subsequence Establish ARIMA (p, d, q) to fit its linear part. The calculation formula is: .
[0040] in, p represents the order of the autoregressive term (i.e. how many past observations are used to make predictions), d represents the number of differences (used to make the time series stationary), q represents the order of the moving average (MA) term (i.e. how many past forecast errors are used to adjust the current value), represents the value of the time series, is the hysteresis operator (i.e., is the parameter of the AR autoregressive model, which represents the relationship between the current value and the past value. is the parameter of the MA moving average model, which represents the relationship between the current value and the past error term. is white noise or error term, is a constant term.
[0041] S312: The stationarity of the sequence is determined by performing an ADF test (Augmented Dickey-Fuller Test) on the third-level low-frequency approximate subsequence A3. If the p-value of the ADF test is less than the set significance level (such as 0.05), the null hypothesis is rejected, indicating that the sequence is stationary; if the p-value is greater than the significance level, the null hypothesis cannot be rejected, indicating that the sequence is non-stationary. In this case, the data is differentiated until the ADF test shows that the data is stationary, and the parameters are determined. .
[0042] S313: Based on the minimum information criterion, the Bayesian Information Criterion (BIC) is used. The calculation formula is: .
[0043] in, n is the sample size; k is the number of free parameters in the model; It is the maximum likelihood estimate of the model. The smaller the BIC, the more accurate the model fit. The grid search method is used to find the optimal model parameters that minimize BIC. .
[0044] It should be noted that the methods used include but are not limited to BIC, and other methods can also be used to optimize parameters.
[0045] S314: The third low-frequency approximate subsequence is approximated according to the determined optimal parameters ARIMA (p, d, q). Make predictions and get low-frequency approximate prediction sequences ; S315: Calculate the residual sequence e of the ARIMA forecast model. The calculation formula is: .
[0046] S32: Build an LSTM model and train it using the residual sequence.
[0047] S321: Construct a three-layer long short-term memory network model consisting of an input layer, a single hidden layer, and an output layer. The input layer is set to a node, including but not limited to the grid search method to determine the step size time_steps, the number of hidden layer neurons, the learning rate, the number of model iterations epochs, and the training batch batch_size to capture time series characteristics. The output layer is set to a node, and the activation functions of the input gate, forget gate, and output gate are set to the Sigmoid function. The unit also includes a candidate memory unit whose activation function is set to the hyperbolic tangent function Tanh, which is used to generate the final prediction result.
[0048] S322: The residual sequence obtained from S315 is divided into a training set and a test set, where the training set accounts for 70% of the data set. During the training process, an appropriate optimization algorithm is used, such as Optimization algorithm to minimize the prediction error, where the loss function during model training is Defined as Model Output predicted low-frequency approximate subsequence residual and residuals The mean square error MSE is calculated as follows: .
[0049] S33: Using the trained LSTM model to approximate low-frequency subsequences The residual sequence is predicted to obtain the predicted data .
[0050] LSTM captures nonlinearities and short-term fluctuations in the residual sequence, primarily improving forecast accuracy. This is particularly true for handling nonlinear dynamics and local variations in the residual. In subsequent steps, the LSTM forecast data is combined with the ARIMA residual forecast to produce more accurate low-frequency approximate subsequence forecasts.
[0051] S34: Add the predicted data to the residual prediction sequence to obtain the final low-frequency approximate subsequence prediction value.
[0052] In a specific embodiment, by The third layer of the model's low-frequency approximate subsequence Forecast data and The model approximates the third layer of low-frequency subsequences Forecast data of residual series Add them together to get the final low-frequency approximate subsequence prediction value , the calculation formula is: .
[0053] In S4, the second prediction model includes the Informer model, which is used to capture high-frequency detail subsequences. The nonlinear part and short-term fluctuations.
[0054] The specific steps of using the second prediction model to capture the feature sequence of the high-frequency detail subsequence are as follows: S41: Build an Informer model with a self-attention mechanism and train the Informer model using a known sample set.
[0055] In a specific embodiment, the high-frequency detail subsequence The known sample set is divided into a training set and a test set, where the training set accounts for 70% of the data set. A sliding window strategy is used to divide the data. The training set and the test set are then divided with a specific window size of 10, and each window slides for 1 cycle.
[0056] S42: Use the trained Informer model to extract the features of the high-frequency detail subsequence and obtain the predicted value of the high-frequency detail subsequence.
[0057] S421: By extracting the units, tens, and hundreds digits of the sequence cycle number as the global timestamp, normalizing the sequence data, and concatenating the local time and global time information with the normalized sequence data, a complete input matrix is constructed to help the model capture sequence features.
[0058] S422: The concatenated input matrix is mapped to a high dimension and input into the Informer encoder. Features are extracted through multiple ProbSparse self-attention calculations and distillation to generate a feature matrix and pass it to the decoder.
[0059] In this embodiment, the Informer encoder extracts nonlinear features, time-related features, and multi-scale features of high-frequency detail subsequences. These features enhance the model's ability to capture short-term fluctuations and periodic patterns. Compared with the original sequence , which more comprehensively reflects the dynamic changes of high-frequency details and is more accurate in trend and fluctuation predictions.
[0060] S423: By taking the sub-matrix as a token and splicing all-0 placeholders, the output matrix is generated using masked self-attention calculation and distillation operation, mapped to the label length through the fully connected layer and activation function, and the model parameters are optimized through the MSE loss function.
[0061] S424: After optimizing hyperparameters using the grid search method, a multi-step rolling prediction method is used to gradually predict high-frequency detail subsequences. The failure threshold is used to determine the number of failed cycles and obtain the high-frequency detail subsequence prediction value: The first-layer high-frequency detail subsequence prediction value , the second layer high frequency detail subsequence prediction value and the third layer high frequency detail subsequence prediction value .
[0062] The number of failure cycles is used as a metric to evaluate model performance. By counting the number of failure cycles over different time periods, the model's performance in long-term predictions can be assessed. A lower number of failure cycles indicates a relatively stable model that accurately captures fluctuations in high-frequency details. A higher number of failure cycles may indicate potential model issues and require further optimization. In this embodiment, if the number of failure cycles in a high-frequency detail subsequence exceeds a set threshold, the model may need to be updated or retrained. This prevents gradual deviations caused by error accumulation, thereby maintaining prediction accuracy.
[0063] It should be noted that in addition to the grid search method, other existing methods can also be used to implement the hyperparameter optimization process.
[0064] In S5, the inverse wavelet transform is used to restore the low-frequency approximate subsequence layer by layer, thereby completing data reconstruction.
[0065] S51: By building the ARIMA-LSTM combined model and the Informer model above, the third layer of low-frequency approximate subsequences are respectively and high-frequency detail subsequences Make predictions and get the prediction results of each model.
[0066] S52: Use and Restore the low-frequency part of the second layer by inverse wavelet transform , the calculation formula is: .
[0067] S53: Use and Restore the low-frequency part of the first layer by inverse wavelet transform , the calculation formula is: .
[0068] S53: Use and Restoring battery capacity prediction sequence by inverse wavelet transform , the calculation formula is: .
[0069] Example 2: A second embodiment of the present invention provides a battery capacity prediction system based on multi-scale wavelet decomposition, including: a data acquisition module configured to acquire operating data of the battery to be tested and calculate a preliminary battery capacity of the battery to be tested; a wavelet decomposition module configured to decompose the preliminary battery capacity time series using wavelet transform to obtain a low-frequency approximate subsequence and a high-frequency detail subsequence; a low-frequency prediction module configured to predict the low-frequency approximate subsequence using the first prediction model to obtain a low-frequency approximate subsequence prediction value; a high-frequency prediction module configured to capture a feature sequence of the high-frequency detail subsequence using a second prediction model to obtain a prediction value of the high-frequency detail subsequence; The data reconstruction module is configured to reconstruct data based on the low-frequency approximate subsequence prediction value and the high-frequency detail subsequence prediction value to obtain a final battery capacity prediction value.
[0070] Example 3: A third embodiment of the present invention provides a medium having a program stored thereon. When the program is executed by a processor, the steps of the battery capacity prediction method based on multi-scale wavelet decomposition as described in the first embodiment of the present invention are implemented.
[0071] Example 4: Embodiment 4 of the present invention provides a device, including a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps of the battery capacity prediction method based on multi-scale wavelet decomposition as described in Embodiment 1 of the present invention are implemented.
[0072] The steps involved in the above embodiments 2, 3 and 4 correspond to those in the method embodiment 1. For the specific implementation methods, please refer to the relevant description part of the embodiment 1.
[0073] Those skilled in the art will appreciate that the modules or steps of the present invention described above can be implemented using a general-purpose computer device. Alternatively, they can be implemented using program code executable by a computing device, which can then be stored in a storage device and executed by the computing device. Alternatively, they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.
[0074] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.
Claims
1. A battery capacity prediction method based on multi-scale wavelet decomposition, characterized in that: The following steps are involved: Obtaining operating data of the battery to be tested and calculating the preliminary battery capacity of the battery to be tested; The preliminary battery capacity time series is decomposed using wavelet transform to obtain low-frequency approximate subsequences and high-frequency detail subsequences; Using the first prediction model to predict the low-frequency approximate subsequence, to obtain a low-frequency approximate subsequence prediction value; The second prediction model is used to capture the feature sequence of the high-frequency detail subsequence to obtain the prediction value of the high-frequency detail subsequence; Data is reconstructed based on the low-frequency approximate subsequence prediction value and the high-frequency detail subsequence prediction value to obtain the final battery capacity prediction value.
2. The battery capacity prediction method based on multi-scale wavelet decomposition according to claim 1, characterized in that: The specific steps for obtaining the operating data of the battery to be tested and calculating the preliminary battery capacity of the battery to be tested are as follows: Select the power battery to be tested, formulate the working conditions according to the actual application requirements, conduct battery aging experiments on the power battery to be tested, obtain the operating data of the battery under various working conditions, and calculate the number of cycles and the corresponding battery capacity based on the operating data.
3. The battery capacity prediction method based on multi-scale wavelet decomposition according to claim 1, characterized in that: The specific steps of decomposing the preliminary battery capacity time series using wavelet transform are: Select db3 as the wavelet basis function and perform wavelet transform on the original battery capacity sequence to obtain the first layer of low-frequency approximate subsequence and the first high-frequency detail subsequence; Perform wavelet transform on the first layer low-frequency approximate subsequence to obtain the second layer low-frequency approximate subsequence and the second high-frequency detail subsequence; The second-layer low-frequency approximate subsequence is subjected to wavelet transform to obtain a third-layer low-frequency approximate subsequence and a third high-frequency detail subsequence.
4. The battery capacity prediction method based on multi-scale wavelet decomposition according to claim 1, characterized in that: The first prediction model includes the ARIMA model and the LSTM model. The ARIMA model is used to predict the residual sequence of the low-frequency approximate subsequence, and the LSTM model is used to predict the nonlinear part and short-term fluctuations in the residual sequence; the second prediction model includes the Informer model, which is used to capture the nonlinear part and short-term fluctuations of the high-frequency detail subsequence.
5. The battery capacity prediction method based on multi-scale wavelet decomposition according to claim 4, characterized in that: The specific steps of using the first prediction model to predict the low-frequency approximate subsequence are as follows: Use the ARIMA model to calculate the residual sequence of the low-frequency approximate subsequence and obtain the residual prediction sequence; Build an LSTM model and use the residual sequence to train the LSTM model; Use the trained LSTM model to predict the low-frequency approximate subsequence and obtain the predicted data; The predicted data is added to the residual prediction sequence to obtain the final low-frequency approximate subsequence prediction value.
6. The battery capacity prediction method based on multi-scale wavelet decomposition according to claim 4, characterized in that: The specific steps of using the second prediction model to capture the feature sequence of the high-frequency detail subsequence are as follows: Build an Informer model with a self-attention mechanism and train the Informer model using a known sample set; The trained Informer model is used to extract the features of the high-frequency detail subsequence and obtain the predicted value of the high-frequency detail subsequence.
7. The battery capacity prediction method based on multi-scale wavelet decomposition according to claim 1, characterized in that: The inverse wavelet transform is used to restore the low-frequency approximate subsequence layer by layer to complete data reconstruction.
8. A battery capacity prediction system based on multi-scale wavelet decomposition, characterized in that: include: a data acquisition module configured to acquire operating data of the battery to be tested and calculate a preliminary battery capacity of the battery to be tested; a wavelet decomposition module configured to decompose the preliminary battery capacity time series using wavelet transform to obtain a low-frequency approximate subsequence and a high-frequency detail subsequence; a low-frequency prediction module configured to predict the low-frequency approximate subsequence using the first prediction model to obtain a low-frequency approximate subsequence prediction value; a high-frequency prediction module configured to capture a feature sequence of the high-frequency detail subsequence using a second prediction model to obtain a prediction value of the high-frequency detail subsequence; The data reconstruction module is configured to reconstruct data based on the low-frequency approximate subsequence prediction value and the high-frequency detail subsequence prediction value to obtain a final battery capacity prediction value.
9. A computer-readable storage medium, characterized in that A plurality of instructions are stored therein, and the instructions are suitable for being loaded by a processor of a terminal device and executing the battery capacity prediction method based on multi-scale wavelet decomposition according to any one of claims 1 to 7.
10. A terminal device, characterized in that: The invention comprises a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded by the processor and executing the battery capacity prediction method based on multi-scale wavelet decomposition according to any one of claims 1 to 7.