Battery RUL prediction method, device, equipment and medium based on prediction error correction

Abstract

The present application discloses a battery RUL prediction method, device, equipment and medium based on prediction error correction, which relates to the field of battery technology, including: obtaining the initial battery remaining service life, obtaining the original error sequence according to the initial battery predicted service life, processing the original error sequence by discrete wavelet transform to obtain low-frequency features, optimizing the support vector regression model by weighted particle swarm according to the low-frequency features, obtaining the predicted error sequence, correcting the initial battery predicted service life according to the predicted error sequence, and obtaining the target battery remaining service life. By using discrete wavelet transform to extract the low-frequency features of the original error sequence, combining the support vector regression model optimized by weighted particle swarm to predict and correct the initial battery RUL, the prediction accuracy and reliability of RUL are improved.

Description

Technical Field

[0001] The present invention relates to the field of battery technology, and in particular to a battery RUL prediction method, device, equipment and medium based on prediction error correction. Background Art

[0002] Lithium-ion batteries are widely used in consumer electronics, electric vehicles, and space systems because they have higher energy density, longer cycle life, and lower self-discharge rate. However, as the use time increases, the performance of lithium batteries will gradually decline. This irreversible change affects the safety and reliability of the battery.

[0003] In terms of RUL prediction of lithium batteries, the method based on support vector regression (SVR) is a commonly used one. The traditional SVR method does not give a reasonable confidence space when predicting RUL, making it difficult for users to evaluate the credibility and uncertainty of the prediction results, and over time, the prediction error may gradually accumulate, causing the final prediction result to deviate from the actual situation. In addition, due to the existence of individual differences in batteries, the general model may not be applicable to all types of batteries. In order to solve these problems, an optimization scheme of the fusion method combined with prediction error correction is proposed to improve the accuracy of RUL prediction. Summary of the invention

[0004] The main purpose of this application is to provide a battery RUL prediction method, device, equipment and medium based on prediction error correction, aiming to solve the technical problem of how to improve the accuracy of battery RUL prediction.

[0005] To achieve the above objectives, the present application proposes a battery RUL prediction method based on prediction error correction, the method comprising: Get the initial remaining battery life; Obtaining an original error sequence according to the initial battery predicted service life; Processing the original error sequence by discrete wavelet transform to obtain low-frequency features; According to the low-frequency features, a prediction error sequence is obtained by optimizing a support vector regression model through a weighted particle swarm optimization; The initial predicted battery service life is corrected according to the prediction error sequence to obtain a target battery remaining service life.

[0006] In one embodiment, before the step of obtaining a prediction error sequence by optimizing a support vector regression model through a weighted particle swarm optimization according to the low-frequency features, the step includes: Obtain low-frequency feature samples and establish an initial support vector regression model; The initial support vector regression model is optimized according to the low-frequency feature samples using a particle swarm optimization algorithm to obtain an optimized support vector regression model.

[0007] In one embodiment, the step of obtaining a prediction error sequence by optimizing a support vector regression model through a weighted particle swarm optimization according to the low-frequency features comprises: Get the preset starting point cycle period and the selected cycle period; When the selected cycle period exceeds the cycle period of the preset starting point, the selected cycle period is substituted into the optimized support vector regression model to obtain a prediction error sequence.

[0008] In one embodiment, the step of obtaining an original error sequence according to the initial predicted battery service life includes: Get the actual battery life; Calculate the error value based on the initial predicted battery service life and the actual battery service life; The error values ​​are arranged in chronological order to obtain an original error sequence.

[0009] In one embodiment, the step of processing the original error sequence by discrete wavelet transform to obtain low-frequency features includes: Performing multi-layer discrete wavelet decomposition on the original error sequence to obtain multiple approximate components; Comparing the plurality of approximate components, and extracting the approximate component that meets the preset requirements; The approximate component is taken as the low-frequency feature.

[0010] In one embodiment, the step of correcting the initial predicted battery service life according to the prediction error sequence to obtain the target battery remaining service life includes: Calculate according to the prediction error sequence to obtain the mean and variance; Obtaining a corrected service life based on the mean value and the initial predicted battery service life; Analyzing the corrected useful life according to the variance to determine a confidence space; When the confidence space meets the verification requirement, the corrected service life is used as the target battery remaining service life.

[0011] In one embodiment, the step of obtaining the initial remaining battery life includes: Obtain battery status signal and obtain particle swarm; Select a cycle period, sample the particle group, and obtain multiple particles; The collected particles are used to obtain observation values ​​according to the relationship between the particle states and the observation values; Correcting the observed value to obtain a posterior probability density; Obtaining a battery capacity estimation value according to the cycle period and the posterior probability density; The initial remaining battery life is calculated based on the estimated battery capacity.

[0012] In addition, to achieve the above-mentioned purpose, the present application also proposes a battery RUL prediction device based on prediction error correction, and the battery RUL prediction device based on prediction error correction includes: An acquisition module is used to obtain the initial remaining battery life; A construction module is used to obtain an original error sequence according to the initial battery predicted service life; An extraction module, used for processing the original error sequence by discrete wavelet transform to obtain low-frequency features; A calculation module, used for obtaining a prediction error sequence by optimizing a support vector regression model through a weighted particle swarm according to the low-frequency features; The correction module is used to correct the initial battery predicted service life according to the prediction error sequence to obtain the target battery remaining service life.

[0013] In addition, to achieve the above-mentioned purpose, the present application also proposes a medium, which is a computer-readable medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the battery RUL prediction method based on prediction error correction as described above are implemented.

[0014] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the battery RUL prediction method based on prediction error correction as described above.

[0015] This application obtains the initial battery remaining service life, obtains the original error sequence based on the initial battery predicted service life, processes the original error sequence through discrete wavelet transform to obtain low-frequency features, obtains the predicted error sequence based on the low-frequency features through weighted particle swarm optimization support vector regression model, corrects the initial battery predicted service life based on the predicted error sequence, and obtains the target battery remaining service life. By using discrete wavelet transform to extract the low-frequency features of the original error sequence, combined with the weighted particle swarm optimized support vector regression model to predict and correct the initial battery RUL, the prediction accuracy and reliability of RUL are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0017] Figure 1 This is a flow chart of a first embodiment of a battery RUL prediction method based on prediction error correction of the present application; Figure 2 The DWT decomposed low-frequency signal of the first embodiment of the battery RUL prediction method based on prediction error correction of this application; Figure 3 The DWT decomposed high frequency signal of the first embodiment of the battery RUL prediction method based on prediction error correction of this application; Figure 4 This is a result diagram of the discrete wavelet transform of the A01 battery in the first embodiment of the battery RUL prediction method based on prediction error correction of this application; Figure 5 This is a prediction method performance evaluation result diagram of the first embodiment of the battery RUL prediction method based on prediction error correction of the present application; Figure 6 This is a battery RUL prediction result diagram of the first embodiment of the battery RUL prediction method based on prediction error correction of the present application; Figure 7 This is a flow chart of a second embodiment of a battery RUL prediction method based on prediction error correction of the present application; Figure 8 This is a flow chart of a third embodiment of a battery RUL prediction method based on prediction error correction of the present application; Fig. 9 This is a flow chart of a fourth embodiment of a battery RUL prediction method based on prediction error correction of the present application; Fig.10 This is a schematic diagram of the module structure of a battery RUL prediction device based on prediction error correction in an embodiment of the present application; Fig.11 Schematic diagram of the device structure of the hardware operating environment involved in the battery RUL prediction method based on prediction error correction in an embodiment of the present application.

[0018] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0019] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0020] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0021] The main solution of the embodiment of the present application is: by obtaining the initial battery remaining service life, the original error sequence is obtained according to the initial battery predicted service life, the original error sequence is processed by discrete wavelet transform to obtain low-frequency features, and the support vector regression model is optimized by weighted particle swarm optimization according to the low-frequency features to obtain a predicted error sequence, and the initial battery predicted service life is corrected according to the predicted error sequence to obtain the target battery remaining service life.

[0022] Based on this, the embodiment of the present application provides a battery RUL prediction method based on prediction error correction, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the battery RUL prediction method based on prediction error correction of the present application.

[0023] In this embodiment, the battery RUL prediction method based on prediction error correction includes steps S10 to S50: Step S10, obtaining the initial remaining battery life.

[0024] It should be noted that in this embodiment, the experimental data comes from 4 lithium batteries of the battery data set A01, A02, A03, and A04 provided by Stanford University. Before step S10, it is necessary to determine the battery capacity failure threshold. The standard is that when the battery capacity drops to approximately 80% of its initial nominal capacity, the battery is considered to have reached its end of life (EOL). This standard is based on considerations of battery performance and safety, because below this threshold, the battery no longer meets application requirements and will experience faster performance degradation. However, in some special cases, depending on the specific application scenarios, such as electric vehicles, aerospace and other fields, different failure thresholds will be set to meet higher safety requirements or optimize cost-effectiveness.

[0025] Further, in this embodiment, by applying the improved particle filter (CE-PF) method, the future degradation trend of the battery is estimated based on historical data and current state information to obtain the remaining service life (RUL) of the initial battery. Step S10 also includes: obtaining a battery state signal to obtain a particle swarm; selecting a cycle period, sampling the particle swarm to obtain a plurality of particles; obtaining an observation value based on the relationship between the particle state and the observation value of the collected particles; correcting the observation value to obtain a posterior probability density; obtaining a battery capacity estimation value based on the cycle period and the posterior probability density; and calculating the initial battery remaining service life based on the battery capacity estimation value. Specifically, first, data on the battery charging and discharging process is collected through sensors or experimental equipment, and these data may include parameters such as operating voltage, current, and temperature. Then, based on the prior probability distribution, the particle swarm is sampled to generate a group of particles representing different possible states. Each particle corresponds to a possible battery state and is assigned a certain weight. After selecting the cycle period, the particle swarm is sampled at a selected time interval to obtain a plurality of particles. Then, based on the relationship between the particle state and the observation value, the behavior of the actual system is simulated by the state equation and the observation equation, and the corresponding observation value is calculated. This step involves mapping the state of the particle to a measurable physical quantity, such as battery capacity or internal resistance. In order to improve the accuracy of the estimate, the observations need to be corrected to account for the effects of measurement noise and other uncertainties. The corrected observations are used to update the weights of the particles to obtain a posterior probability density that is closer to the actual situation. This process is usually done using Bayes' theorem, which allows the particle weights to be adjusted in combination with previous knowledge and new observation information. Once the posterior probability density is obtained, it can be used to estimate the battery capacity at the current moment. Specifically, the estimated value of the battery capacity can be obtained by calculating the weighted average of all particles. Over time, the above process is repeated until a preset number of cycles is reached. Finally, based on the estimated battery capacity and a known failure threshold (for example, the capacity drops to 80% of the initial value), the RUL of the initial battery can be calculated.

[0026] Step S20, obtaining an original error sequence according to the initial predicted battery service life.

[0027] It should be noted that the actual battery capacity or health state of the battery is obtained, and then the actual battery capacity or health state is compared with the initial predicted battery service life, and the difference between the two is calculated, thereby forming an original error sequence.

[0028] Specifically, step S20 includes: obtaining the actual service life of the battery; calculating according to the initial predicted service life of the battery and the actual service life of the battery to obtain an error value; arranging the error values ​​in chronological order to obtain an original error sequence. First, the actual service life of the battery needs to be obtained. The above steps usually involve data collected through experiments or field applications, recording the time point from the start of battery use until its capacity drops to a preset threshold (such as 80% of the original capacity). These data can come from a battery management system (BMS) or other monitoring equipment, which can provide detailed charge and discharge cycle information, operating temperature, and capacity measurement results after each cycle. Next, the predicted service life of the battery calculated based on the initial prediction model is compared with the actual service life. The initial prediction here may be an estimate based on historical data and a specific algorithm, such as a particle filter (PF), an improved particle filter (CE-PF), or other mathematical models. A series of error values ​​can be obtained by subtracting the corresponding predicted service life value from the actual service life value at each time point. These error values ​​reflect the accuracy of the prediction model at different time nodes. Then, these error values ​​are arranged in chronological order to form an original error sequence. This sequence not only reflects the accuracy of the prediction model, but also may reveal potential patterns or changes that the model fails to capture. For example, in the life cycle of lithium batteries, the degradation rate is relatively slow in the early stages, but accelerates in the middle and late stages, which may cause some simple prediction models to deviate from the performance at different stages.

[0029] Step S30, processing the original error sequence by discrete wavelet transform to obtain low-frequency features.

[0030] It should be noted that discrete wavelet transform (DWT) plays a vital role in processing the original error sequence to extract low-frequency features. DWT can effectively remove high-frequency noise and retain low-frequency information reflecting long-term trends by decomposing the signal into approximate and detailed components of different scales. This is a very critical step for error correction in battery RUL prediction. When DWT is applied to the original error sequence, a suitable wavelet basis function, such as Haar wavelet, is first selected, and then multi-layer decomposition is performed according to the Mallat algorithm. In each round of decomposition, the original signal is passed to two complementary filters to generate approximate components and detailed components, respectively. As the number of decomposition layers increases, fewer and fewer approximate components are obtained, and the variance of the reconstructed signal decreases accordingly. For battery RUL prediction, a decomposition depth of about 6 layers is usually selected to obtain sufficient low-frequency details. From these low-frequency components, a smooth and denoised long-term trend of error evolution is reconstructed. This is because high-frequency components are often considered unimportant and they have little influence on the development direction of the overall error sequence. By retaining only the low-frequency part and recombining the signal, we are able to obtain clearer trend features, which helps to improve the accuracy of subsequent predictions of weighted particle swarm optimization support vector regression models or other data-driven algorithms.

[0031] Step S40, obtaining a prediction error sequence by optimizing a support vector regression model through weighted particle swarm optimization according to the low-frequency features.

[0032] It should be noted that before using the weighted particle swarm optimization support vector (APSO-SVR) model for prediction, two important parameters need to be determined: the penalty factor and the kernel parameter. These two parameters determine the complexity of the APSO-SVR model and the degree of fit to the training data. First, a single-input single-output SVR model with cycle period as input and residual as output is constructed. The adaptive inertia weighted particle swarm optimization algorithm (APSO) is introduced here, which can dynamically adjust the speed and position of particles to search for the optimal solution space. APSO continuously iterates and updates the position and speed of particles, and evaluates the performance of each particle according to the fitness function to find the best penalty factor and kernel parameter, and then obtains the optimized SVR model based on the best penalty factor and kernel parameter, and uses this model to obtain the prediction error sequence.

[0033] Furthermore, step S40 also includes: obtaining a preset starting point cycle period and a selected cycle period; when the selected cycle period exceeds the cycle period of the preset starting point, substituting the selected cycle period into the optimized support vector regression model to obtain a prediction error sequence. Specifically, the preset starting point cycle period refers to the reference number of cycles used to start RUL prediction, which is usually determined based on when the battery capacity decays to a certain level, such as when the capacity drops to 90% or 85% of the initial value. The choice of this starting point depends on application requirements and safety standards. The selected cycle period refers to the time series data points after the preset starting point, which will be used to train the APSO-SVR model test to ensure that the prediction results can reflect the actual degradation trend of the battery. At the same time, the preset cycle period and the selected cycle period are obtained. When the battery capacity prediction value is greater than the battery capacity failure threshold, the selected cycle period is accumulated by the preset cycle period as a new cycle period to obtain the battery capacity prediction value of the new cycle period, and so on, until the battery capacity prediction value of the new cycle period does not exceed the battery capacity failure threshold. The preset cycle refers to the fixed number of cycles that advances forward each time the prediction is made. When the predicted value of the battery capacity is still higher than the set failure threshold, we will select the cycle plus the preset cycle as the new cycle, and use the APSO-SVR model again to predict the battery capacity under the new cycle. Whenever the prediction result shows that the battery capacity is still within the safe range, that is, greater than the failure threshold, continue to increase the preset cycle and recalculate. In this way, we can gradually approach the actual end of the battery's life until a prediction shows that the predicted value of the battery capacity in the new cycle does not exceed the above failure threshold.

[0034] Step S50, correcting the initial predicted battery service life according to the prediction error sequence to obtain the target battery remaining service life.

[0035] It should be noted that the predicted error sequence is applied to the initial battery predicted service life for correction. Specifically, the predicted error value at each time point is added or subtracted from the corresponding initial predicted RUL to adjust the initial predicted value. This can compensate for the systematic deviations in the prediction model, making the corrected RUL prediction closer to reality. In addition, considering the nonlinear changes and emergencies that may occur during battery degradation, such as sudden capacity drop, additional mechanisms can be introduced to further improve the correction process. For example, a segmented empirical degradation model or other data-driven methods can be used to ensure that high prediction accuracy is maintained throughout the life cycle. At the same time, for the corrected RUL prediction results, the corresponding probability density distribution can also be provided to better understand and quantify the uncertainty of the prediction. Finally, through the above series of steps, we get the remaining service life of the target battery.

[0036] In this embodiment, the experimental data are selected from the battery data sets A01, A02, A04, and A04 of Stanford University. In this embodiment, the A01 lithium battery is subjected to discrete wavelet transform using haar wavelet to extract 6 layers of low-frequency detail components and high-frequency detail components. Figure 2 The DWT decomposition low frequency signal diagram and Figure 3 The DWT decomposed high-frequency signal diagram shown in the figure, since the high-frequency components are considered to be unimportant to the evolution trend of the error sequence, the reconstructed error sequence can be composed of low-frequency components. Therefore, the low-frequency component reconstructed signal of the sixth layer (the original signal minus the high-frequency component signals of the first six layers) is selected as the long-term trend of the error evolution. By deleting some low-correlated high-frequency signals from the residual components, the data quality is improved. Figure 4 The A01 battery discrete wavelet transform result shown in the figure shows that the global fluctuation of the residual component and the trend characteristics of the reconstructed low-frequency signal are contained in a smoother and less noisy error sequence. Therefore, it is used in the data-driven algorithm, laying a solid foundation for the subsequent successful prediction of support vector regression.

[0037] Specifically, for the battery data set of Stanford University, two methods, namely the standard PF-CEEMD-GPR and the PF-DWT-SVR of this embodiment, are used to compare and analyze the battery RUL prediction performance of four batteries A01, A02, A03, and A04 of Stanford University, and the RMSE, MAE, MAPE, ERUL and Five indicators are used to verify the effectiveness. The Gaussian regression equation (GPR) algorithm is a nonlinear regression method derived from the random process of the Bayesian algorithm. Like the RVM algorithm, it can also give the confidence interval of the prediction results. Complementary ensemble empirical modal decomposition (CEEMD) is a particle resampling algorithm, such as Figure 5 The prediction method performance evaluation results shown in the figure show that the prediction performance of the fusion algorithm is better than that of other algorithms, which is quantitatively reflected in the evaluation indicators of the four batteries. These will be strong evidence to verify the ability of the method to predict the accelerated battery capacity decay under these operating conditions.

[0038] In this embodiment, the preset starting point period is set to 219 periods. Figure 6The battery RUL prediction result diagram shown in the figure obtains the RUL prediction results of four batteries. The predicted RULs of A01, A02, A03 and A04 are 515, 532, 430 and 479 respectively. In addition, assuming that the prediction results conform to the Gaussian distribution, their variances are 18.5, 11.4, 10.2 and 10.4 respectively. Through statistical regression analysis of the prediction results, the effectiveness of the prediction error correction method includes root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE) and maximum abs error.

[0039] This embodiment provides a battery RUL prediction method based on prediction error correction, by obtaining the initial battery remaining service life, obtaining the original error sequence according to the initial battery predicted service life, processing the original error sequence through discrete wavelet transform to obtain low-frequency features, optimizing the support vector regression model through weighted particle swarm according to the low-frequency features, obtaining the prediction error sequence, correcting the initial battery predicted service life according to the prediction error sequence, and obtaining the target battery remaining service life. By using discrete wavelet transform to extract the low-frequency features of the original error sequence, combined with the weighted particle swarm optimized support vector regression model to predict and correct the initial battery RUL, the prediction accuracy and reliability of RUL are improved.

[0040] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can refer to the above introduction, and will not be repeated later. Figure 7 The battery RUL prediction method based on prediction error correction further includes steps S401-S402 before step S40: Step S401, obtaining low-frequency feature samples and establishing an initial support vector regression model.

[0041] It should be noted that low-frequency feature samples need to be extracted from the original error sequence. This step is achieved through discrete wavelet transform (DWT), which decomposes the signal into approximate and detail components at different scales, thereby removing high-frequency noise and retaining information reflecting long-term trends. After selecting a suitable wavelet basis function (such as Haar wavelet), multi-layer decomposition is performed according to the Mallat algorithm until enough low-frequency details are obtained. Usually, a decomposition depth of about 6 layers is selected to obtain a smooth and denoised long-term trend of error evolution. Once these low-frequency feature samples are obtained, the next step is to establish an initial support vector regression (SVR) model.

[0042] Step S402 , optimizing the initial support vector regression model according to the low-frequency feature samples using a particle swarm optimization algorithm to obtain an optimized support vector regression model.

[0043] It should be noted that before obtaining the optimized support vector regression model, its parameters, namely the parameter penalty factor and the kernel parameter, need to be determined.

[0044] Furthermore, before step S403, it includes: obtaining the particle filter in the training set feature data; initializing the algorithm parameters, setting the species measurement threshold; dividing the particle population in the particle filter into multiple subgroups; calculating the fitness value of each particle in each subgroup, and initializing the optimal position of each individual subgroup and the optimal position of the group; updating the fitness value of each individual subgroup, the optimal position of each individual subgroup and the optimal position of the group according to the functional relationship; judging whether the species measurement value exceeds the species measurement threshold, if the particle species measurement value is less than the species measurement threshold, re-planning the particle population; judging whether the maximum number of iterations has been reached, if so, outputting the model parameters; after inputting the model parameters into the nonlinear fitting function, obtaining the parameter penalty factor and kernel parameters for constructing the optimized support vector model. Specifically, the algorithm parameters are initialized. Including the total number of particles M, the maximum number of iterations , the maximum value of inertia weight and minimum value , learning factors c1, c2, particle position value , the speed value is , set the population diversity metric threshold. The particle population in the particle filter is divided into multiple subgroups to explore the search space more effectively. The particles in each subgroup interact with each other according to their own position update rules and calculate their respective fitness values. The fitness function here is a standard used to measure the quality of the particle solution to a given problem. For example, in this study, it can be an error indicator between the actual battery capacity. After the initialization phase, each subgroup will record the individual best position PP(t) and the group best position PG(t), which represent the best solutions found so far by a single particle and the entire group, respectively. As the iterative process progresses, the speed and position of the particles in each subgroup are continuously updated according to the pre-defined functional relationship, and the fitness value is recalculated and the best position of the individual and the group is adjusted accordingly. This update mechanism is designed to guide the particles to move in the direction of a better solution and gradually approach the global optimal solution. During this period, it is also necessary to regularly check whether the particle species metric exceeds the preset threshold. If the particle species metric is found to be lower than the threshold, it means that the particle diversity is insufficient, which may lead to premature phenomenon. At this time, it is necessary to re-plan the particle population and introduce new mutations or crossover operations to increase diversity. When the maximum number of iterations is reached, the optimization process ends, and the output is the optimal model parameters obtained after multiple rounds of iterations - penalty factor C and kernel parameter g. These parameters are then input into the nonlinear fitting function to construct the final optimized support vector regression model.

[0045] Furthermore, step S403 also includes: obtaining a particle swarm in the training set data information; initializing the position and velocity of particles in the particle swarm and calculating the fitness of particles; setting the number of iterations, calculating the historical optimal value of particles; obtaining the global optimal value of the particle swarm by comparing the fitness of particles in each iteration; updating the position and velocity of particles according to the historical optimal value and the global optimal value; judging whether the position and velocity of particles meet the preset range, if not, performing boundary condition processing on particles; if the number of iterations is reached or the error is less than the preset error, an optimized support vector regression model is obtained. Specifically, in the initial stage, an initial position and velocity are randomly assigned to each particle. The position represents the coordinates of the particle in the search space, corresponding to a set of SVR parameters; and the velocity determines the direction and distance of the particle in subsequent iterations. At the same time, the fitness value of each particle is calculated, which is usually by using the position of the particle as an APSO-SVR model parameter to evaluate the degree of fit of the model to the training data, such as minimizing the mean square error (MSE) or maximizing the cross-validation score. Next, a reasonable number of iterations is set as one of the termination conditions, and the iteration process is started. In each iteration, the fitness value of the current particle is calculated based on its position and compared with the historical optimal value of the particle. If the new fitness is better, the historical optimal value is updated. At the same time, the historical optimal values ​​of all particles are used to update the global optimal value, that is, the best solution found by the entire group. The global optimal value reflects the best parameter combination explored by all particles so far. As the iteration proceeds, the position and speed of each particle are adjusted according to a specific speed update formula based on the historical optimal value and the global optimal value. This process simulates the information sharing and learning mechanism between particles, allowing particles to move towards a better solution. For example, the new speed of a particle can be a weighted sum of the current speed, the attraction toward the historical optimal value, and the attraction toward the global optimal value. In order to ensure that the particle remains within a reasonable range, it is necessary to determine whether the position and speed of the particle meet the preset range. If not, the particle must be subjected to boundary condition processing, such as rebound, truncation, or re-randomization, to prevent the particle from flying out of the effective search space, thereby ensuring the effectiveness and stability of the algorithm. When the set number of iterations is reached or the error is less than the preset error, it indicates that the optimization process is over and the optimized support vector regression model is output.

[0046] In this embodiment, the key characteristic data of battery capacity are divided into a training set and a test set, and an initial support vector regression model is established by using the training set and optimized by a particle swarm optimization algorithm to improve the prediction accuracy and model robustness.

[0047] Based on the first embodiment of the present application, in the third embodiment of the present application, the same or similar contents as those in the first embodiment can be referred to the above description, and will not be described in detail later. Figure 8The battery RUL prediction method based on prediction error correction step S30 also includes steps S301 to S303: Step S301 , performing multi-layer discrete wavelet decomposition on the original error sequence to obtain multiple approximate components.

[0048] It should be noted that through DWT, the error signal can be decomposed into multiple approximate components and detail components of different frequencies. This process is based on the Mallat algorithm, which uses a pair of complementary filters to analyze the signal layer by layer. In each layer of decomposition, the original signal is divided into low-frequency approximate components and high-frequency detailed components. As the number of decomposition layers increases, the approximate components obtained gradually decrease, and the variance of the reconstructed signal also decreases accordingly.

[0049] Furthermore, choosing a suitable wavelet basis function (such as Haar wavelet) is crucial to obtain effective low-frequency features. Typically, we perform DWT on the error sequence with about six layers to ensure that sufficiently rich long-term trend information can be captured. High-frequency components are considered to be secondary because they mainly contain short-term fluctuations and noise and have little impact on the overall error evolution. Therefore, when reconstructing the error sequence, we can choose to ignore these high-frequency parts and only retain the low-frequency approximate components to construct a smoother and less noisy new sequence. By further processing these approximate components, we can remove random perturbations in the original error sequence and highlight its inherent trend pattern. This not only improves the accuracy of subsequent model predictions such as support vector regression (SVR), but also provides valuable insights into understanding the dynamic characteristics of battery degradation. For example, after removing some high-frequency signals with low correlation with the dominant information of error evolution from the residual components, the data quality is significantly improved, and the trend characteristics of global fluctuations and reconstructed low-frequency signals are better captured in a new, smoother sequence. Finally, the low-frequency approximate components after multi-layer DWT processing can be used as the input of the improved particle filter (CE-PF) algorithm or the adaptive inertia weight particle swarm optimization support vector regression (APSO-SVR) model to predict future error changes.

[0050] Step S302: compare multiple approximate components and extract approximate components that meet preset requirements.

[0051] It should be noted that through discrete wavelet transform (DWT), the original error sequence is decomposed into a series of approximate components and high-frequency detail components at different scales. These approximate components represent the low-frequency part of the signal, which usually contains important information about long-term trends, while high-frequency detail components are mostly short-term fluctuations and noise. In order to select the most suitable approximate components for subsequent analysis, it is necessary to establish certain standards or rules to evaluate the quality of each component. First, by calculating the energy distribution of each approximate component, that is, the proportion of each layer component in the total energy, those components with a large and stable proportion can be selected as candidates. Secondly, consider the correlation between the component and the actual physical meaning. For example, components within a certain frequency range may better reflect the key features of the battery degradation process. In addition, based on statistical characteristics such as variance, mean and other indicators, components with smaller volatility can be screened out to ensure that the selected components can better represent the overall trend of change.

[0052] Step S303: taking the approximate component as a low-frequency feature.

[0053] This embodiment performs multi-layer discrete wavelet decomposition on the original error sequence, extracts approximate components that meet preset requirements as low-frequency features, optimizes the prediction model, and improves the RUL prediction accuracy and stability.

[0054] Based on the first embodiment of the present application, in the third embodiment of the present application, the same or similar contents as those in the first embodiment can be referred to the above description, and will not be described in detail later. Fig. 9 The battery RUL prediction method based on prediction error correction step S50 also includes steps S501 to S504: Step S501, calculating according to the prediction error sequence to obtain the mean and variance.

[0055] It should be noted that all values ​​in the forecast error sequence are added together and divided by the number of data points. This average value represents the systematic deviation between the forecast results and the actual values. If the mean is close to zero, it means that the forecast model has no obvious deviation. However, relying solely on the mean is not enough to fully understand the distribution characteristics of the forecast error, so the variance needs to be calculated. The variance measures the degree of difference between each error value and its mean, which reveals the volatility and uncertainty of the forecast error. A smaller variance means that the forecast results are more concentrated and stable, while a larger variance indicates that there are more changes or noise.

[0056] Step S502, obtaining a corrected service life according to the average value and the initial predicted service life of the battery.

[0057] It should be noted that after obtaining the mean of the prediction error sequence, this average deviation is applied to the initial predicted battery service life for correction. Specifically, if the mean is positive, it means that the predicted RUL is too short and the corresponding time or number of cycles needs to be increased; conversely, if the mean is negative, it means that the predicted RUL is too long and the estimated life should be reduced. The purpose of this is to compensate for the systematic deviations in the prediction model so that the corrected RUL is closer to the actual situation.

[0058] Step S503: Analyze the corrected service life according to the variance to determine the confidence space.

[0059] It should be noted that based on the variance, we can construct a confidence interval to express the uncertainty range of the predicted RUL. Usually, we choose 95% as the confidence level, which means that in multiple repeated experiments, there is a 95% probability that the true value will fall within this interval. In this way, we can not only give a specific RUL estimate, but also clearly indicate its possible range of variation, which is crucial for risk assessment and maintenance planning. In addition, when faced with batteries of different batches or models, since there may be performance differences between them, it is necessary to calculate their respective variances separately and establish corresponding confidence intervals. This helps to identify which types of batteries are more suitable for specific application scenarios, and it is also convenient to formulate personalized management strategies. For example, for batteries with large variances and strong prediction uncertainties, you can consider increasing the monitoring frequency or arranging replacement plans in advance to ensure the safety and stability of the system.

[0060] Step S504: when the confidence space meets the verification requirement, the corrected service life is used as the target battery remaining service life.

[0061] It should be noted that the confidence interval is constructed based on the variance of the prediction error sequence, and a 95% confidence level is usually selected to express the uncertainty range of the prediction results. If the width of the confidence interval is found to be within an acceptable range after statistical analysis and meets the preset verification criteria, then we can confirm that the revised RUL is reliable. At this point, this revised RUL can be regarded as the true remaining service life of the target battery.

[0062] This embodiment calculates the mean and variance of the prediction error sequence, corrects the initial battery prediction service life, and determines the confidence space based on variance analysis. If the confidence space meets the verification requirements, the corrected service life is used as the target RUL to improve the prediction accuracy and reliability.

[0063] This application also provides a battery RUL prediction device based on prediction error correction, please refer to Fig.10 , the device comprises: An acquisition module 10 is used to acquire the initial remaining battery life; A construction module 20 is used to obtain an original error sequence according to the initial battery predicted service life; An extraction module 30 is used to process the original error sequence through discrete wavelet transform to obtain low-frequency features; A calculation module 40 is used to obtain a prediction error sequence by optimizing a support vector regression model through a weighted particle swarm according to low-frequency features; The correction module 50 is used to correct the initial battery predicted service life according to the prediction error sequence to obtain the target battery remaining service life.

[0064] The battery RUL prediction device based on prediction error correction provided by the present application adopts the battery RUL prediction method based on prediction error correction in the above embodiment, which can solve the technical problem of how to improve the accuracy of battery RUL prediction. Compared with the prior art, the beneficial effects of the battery RUL prediction device based on prediction error correction provided by the present application are the same as the beneficial effects of the battery RUL prediction method based on prediction error correction provided by the above embodiment, and the other technical features in the battery RUL prediction device based on prediction error correction are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0065] In one embodiment, the calculation module 40 is further used to obtain low-frequency feature samples and establish an initial support vector regression model; the initial support vector regression model is optimized according to the low-frequency feature samples using a particle swarm optimization algorithm to obtain an optimized support vector regression model.

[0066] In one embodiment, the calculation module 40 is also used to obtain the preset starting point cycle period and the selected cycle period; when the selected cycle period exceeds the cycle period of the preset starting point, the selected cycle period is substituted into the optimized support vector regression model to obtain a prediction error sequence.

[0067] In one embodiment, the construction module 20 is also used to obtain the actual service life of the battery; calculate according to the initial predicted battery service life and the actual battery service life to obtain an error value; and arrange the error values ​​in chronological order to obtain an original error sequence.

[0068] In one embodiment, the extraction module 30 is further used to perform multi-layer discrete wavelet decomposition on the original error sequence to obtain multiple approximate components; compare the multiple approximate components, extract the approximate components that meet preset requirements, and use the approximate components as low-frequency features.

[0069] In one embodiment, the correction module 50 is also used to calculate according to the prediction error sequence to obtain a mean and a variance; obtain a corrected service life according to the mean and the initial battery predicted service life; analyze the corrected service life according to the variance to determine the confidence space; when the confidence space meets the verification requirements, the corrected service life is used as the remaining service life of the target battery.

[0070] In one embodiment, the acquisition module 10 is also used to acquire a battery status signal to obtain a particle swarm; select a cycle period, sample the particle swarm, and obtain a plurality of particles; obtain an observation value based on the relationship between the particle state and the observation value of the collected particles; correct the observation value to obtain a posterior probability density; obtain a battery capacity estimate based on the cycle period and the posterior probability density; and calculate the initial battery remaining service life based on the battery capacity estimate.

[0071] The present application provides a battery RUL prediction device based on prediction error correction, and the battery RUL prediction device based on prediction error correction includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the battery RUL prediction method based on prediction error correction in the above-mentioned embodiment one.

[0072] Reference below Fig.11 , which shows a schematic diagram of the structure of a battery RUL prediction device based on prediction error correction suitable for implementing an embodiment of the present application. The battery RUL prediction device based on prediction error correction in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Fig.11 The battery RUL prediction device based on prediction error correction shown is only an example and should not bring any limitation to the function and scope of use of the embodiments of the present application.

[0073] like Fig.11As shown, the battery RUL prediction device based on prediction error correction may include a processing device 1001 (such as a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM: Read Only Memory) 1002 or the program loaded from the storage device 1003 to the random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the battery RUL prediction device based on prediction error correction are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other via a bus 1005. The input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 can allow the battery RUL prediction device based on prediction error correction to communicate wirelessly or wired with other devices to exchange data. Although the figure shows a battery RUL prediction device based on prediction error correction with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or provided alternatively.

[0074] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0075] The battery RUL prediction device based on prediction error correction provided by the present application adopts the battery RUL prediction method based on prediction error correction in the above embodiment, which can solve the technical problem of how to improve the accuracy of battery RUL prediction. Compared with the prior art, the beneficial effects of the battery RUL prediction device based on prediction error correction provided by the present application are the same as the beneficial effects of the battery RUL prediction method based on prediction error correction provided by the above embodiment, and the other technical features in the battery RUL prediction device based on prediction error correction are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0076] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0077] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0078] The present application provides a computer-readable medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the battery RUL prediction method based on prediction error correction in the above-mentioned embodiment.

[0079] The computer-readable medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM: Random Access Memory), a read-only memory (ROM: Read Only Memory), an erasable programmable read-only memory (EPROM: Erasable Programmable Read Only Memory or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM: CD-Read Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency: Radio Frequency), etc., or any suitable combination of the above.

[0080] The computer-readable medium may be included in the battery RUL prediction device based on prediction error correction; or may exist independently without being assembled into the battery RUL prediction device based on prediction error correction.

[0081] The computer-readable medium carries one or more programs. When the one or more programs are executed by the battery RUL prediction device based on prediction error correction, the battery RUL prediction device based on prediction error correction can be written in one or more programming languages ​​or a combination thereof to perform computer program codes for the operation of the present application. The programming languages ​​include object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0082] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0083] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0084] The readable medium provided in this application is a computer-readable medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned battery RUL prediction method based on prediction error correction, and can solve the technical problem of how to improve the accuracy of battery RUL prediction. Compared with the prior art, the beneficial effects of the computer-readable medium provided in this application are the same as the beneficial effects of the battery RUL prediction method based on prediction error correction provided in the above-mentioned embodiment, and will not be repeated here.

[0085] The present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the battery RUL prediction method based on prediction error correction as described above.

[0086] The computer program product provided in this application can solve the technical problem of how to improve the accuracy of battery RUL prediction. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as the beneficial effects of the battery RUL prediction method based on prediction error correction provided in the above embodiment, which will not be repeated here.

[0087] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A battery RUL prediction method based on prediction error correction, characterized in that: The method comprises: Get the initial remaining battery life; Obtaining an original error sequence according to the initial battery predicted service life; Processing the original error sequence by discrete wavelet transform to obtain low-frequency features; According to the low-frequency features, a prediction error sequence is obtained by optimizing a support vector regression model through a weighted particle swarm optimization; The initial predicted battery service life is corrected according to the prediction error sequence to obtain a target battery remaining service life.

2. The method according to claim 1, characterized in that Before the step of obtaining a prediction error sequence by optimizing a support vector regression model through a weighted particle swarm optimization according to the low-frequency features, the method includes: Obtain low-frequency feature samples and establish an initial support vector regression model; The initial support vector regression model is optimized according to the low-frequency feature samples using a particle swarm optimization algorithm to obtain an optimized support vector regression model.

3. The method according to claim 1, characterized in that The step of obtaining a prediction error sequence by optimizing a support vector regression model through a weighted particle swarm optimization according to the low-frequency features comprises: Get the preset starting point cycle period and the selected cycle period; When the selected cycle period exceeds the cycle period of the preset starting point, the selected cycle period is substituted into the optimized support vector regression model to obtain a prediction error sequence.

4. The method according to claim 1, characterized in that The step of obtaining an original error sequence according to the initial predicted battery service life comprises: Get the actual battery life; Calculate the error value based on the initial predicted battery service life and the actual battery service life; The error values ​​are arranged in chronological order to obtain an original error sequence.

5. The method according to claim 1, characterized in that The step of processing the original error sequence by discrete wavelet transform to obtain low-frequency features includes: Performing multi-layer discrete wavelet decomposition on the original error sequence to obtain multiple approximate components; Comparing the plurality of approximate components, and extracting the approximate component that meets the preset requirements; The approximate component is taken as the low-frequency feature.

6. The method according to claim 1, characterized in that The step of correcting the initial battery predicted service life according to the prediction error sequence to obtain the target battery remaining service life includes: Calculate according to the prediction error sequence to obtain the mean and variance; Obtaining a corrected service life based on the mean value and the initial predicted battery service life; Analyzing the corrected useful life according to the variance to determine a confidence space; When the confidence space meets the verification requirement, the corrected service life is used as the target battery remaining service life.

7. The method according to claim 1, characterized in that The step of obtaining the initial remaining battery life includes: Obtain battery status signal and obtain particle swarm; Select a cycle period, sample the particle group, and obtain multiple particles; The collected particles are used to obtain observation values ​​according to the relationship between the particle states and the observation values; Correcting the observed value to obtain a posterior probability density; Obtaining a battery capacity estimation value according to the cycle period and the posterior probability density; The initial remaining battery life is calculated based on the estimated battery capacity.

8. A battery RUL prediction method and device based on prediction error correction, characterized in that: The device comprises: An acquisition module is used to obtain the initial remaining battery life; A construction module is used to obtain an original error sequence according to the initial battery predicted service life; An extraction module, used for processing the original error sequence by discrete wavelet transform to obtain low-frequency features; A calculation module, used for obtaining a prediction error sequence by optimizing a support vector regression model through a weighted particle swarm according to the low-frequency features; The correction module is used to correct the initial battery predicted service life according to the prediction error sequence to obtain the target battery remaining service life.

9. A battery RUL prediction method and device based on prediction error correction, characterized in that: The device includes: a memory, a processor, and a battery RUL prediction method program based on prediction error correction stored in the memory and running on the processor, wherein the battery RUL prediction method program based on prediction error correction is configured to implement the steps of the battery RUL prediction method based on prediction error correction as described in any one of claims 1-7.

10. A storage medium, characterized in that: The storage medium stores a battery RUL prediction method program based on prediction error correction, and when the battery RUL prediction method program based on prediction error correction is executed by the processor, the steps of the battery RUL prediction method based on prediction error correction as described in any one of claims 1-7 are implemented.

Images