A method, device, equipment and medium for predicting battery capacity attenuation

Abstract

The present invention discloses a method, device, equipment and medium for predicting battery capacity attenuation. The method comprises: obtaining operating parameter information and current operating condition information of a battery to be detected; performing feature extraction based on the operating parameter information to determine a battery characteristic parameter set; determining a health status evaluation value of the battery to be detected based on the current operating condition information, a pre-trained state evaluation model set and a battery characteristic parameter set; and determining a capacity attenuation prediction value based on the health status evaluation value, the battery characteristic parameter set and a pre-trained time convolutional network model. By first determining the pre-trained state evaluation model corresponding to the current operating condition of the battery to be detected, determining the health status evaluation value in combination with the battery characteristic parameter set, and then determining the capacity attenuation prediction value in combination with the pre-trained time convolutional network model. The impact of operating condition fluctuations on the prediction is reduced, and the prediction accuracy, robustness and long-term adaptability are improved, laying the foundation for subsequently extending battery life and reducing operation and maintenance costs.

Description

Technical Field

[0001] The present invention relates to the technical field of battery status detection, and in particular to a battery capacity attenuation prediction method, device, equipment and medium. Background Art

[0002] With the rapid development of the global electric vehicle and energy storage markets, lithium-ion batteries, as core energy storage devices, are widely used in electric bicycles, electric vehicles, consumer electronics, and renewable energy storage systems. However, the performance of lithium-ion batteries gradually degrades with age and changes in operating conditions, manifesting as capacity fade and a decrease in state of health (SoH). This degradation not only shortens the battery's service life but also directly impacts the safety, reliability, and economic benefits of the equipment. Therefore, accurately assessing the battery's SoH and predicting its capacity fade trends have become key issues in battery management systems.

[0003] Traditional battery health management technologies rely on simple experience-based models such as equivalent circuit models and electrochemical analysis models.

[0004] While these methods played a role in early research, their limitations became increasingly apparent as application scenarios became more complex. Equivalent circuit models are insensitive to dynamic changes in the environment and operating conditions, making them difficult to adapt to diverse operating conditions. While electrochemical analysis models offer high accuracy, they require complex experimental equipment and lengthy testing, making them difficult to meet the real-time demands of actual operation. Furthermore, as battery capacity gradually degrades, operating condition fluctuations (such as temperature and load variations) significantly impact prediction accuracy, making traditional methods unable to effectively address data noise issues under multiple operating conditions. Summary of the Invention

[0005] The present invention provides a battery capacity attenuation prediction method, device, equipment and medium to achieve accurate prediction of battery capacity attenuation under different operating conditions.

[0006] According to a first aspect of the present invention, a method for predicting battery capacity attenuation is provided, comprising:

[0007] Obtain operating parameter information and current working condition information of the battery to be tested;

[0008] Perform feature extraction based on the operating parameter information to determine a battery feature parameter set;

[0009] Determine a health status assessment value of the battery to be tested based on the current operating condition information, the pre-trained state assessment model set, and the battery characteristic parameter set;

[0010] A capacity decay prediction value is determined based on the health status assessment value, the battery characteristic parameter set, and a pre-trained time convolutional network model.

[0011] According to a second aspect of the present invention, there is provided a battery capacity attenuation prediction device, comprising:

[0012] An information acquisition module is used to obtain operating parameter information and current operating condition information of the battery to be tested;

[0013] A first determining module, configured to extract features based on the operating parameter information and determine a battery feature parameter set;

[0014] A second determination module is used to determine the health status evaluation value of the battery to be tested based on the current operating condition information, the pre-trained state evaluation model set and the battery characteristic parameter set;

[0015] The third determination module is used to determine the capacity decay prediction value based on the health status assessment value, the battery characteristic parameter set and the pre-trained time convolution network model.

[0016] According to a third aspect of the present invention, there is provided an electronic device, comprising:

[0017] at least one processor; and

[0018] a memory communicatively connected to the at least one processor; wherein,

[0019] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the battery capacity attenuation prediction method described in any embodiment of the present invention.

[0020] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the battery capacity attenuation prediction method described in any embodiment of the present invention when executed.

[0021] According to a fifth aspect of the present invention, an embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the battery capacity attenuation prediction method of any embodiment of the present invention.

[0022] The technical solution of the embodiment of the present invention obtains the operating parameter information and current operating condition information of the battery to be tested; performs feature extraction based on the operating parameter information to determine the battery characteristic parameter set; determines the health status evaluation value of the battery to be tested based on the current operating condition information, the pre-trained state evaluation model set and the battery characteristic parameter set; and determines the capacity attenuation prediction value based on the health status evaluation value, the battery characteristic parameter set and the pre-trained time convolutional network model. By first determining the pre-trained state evaluation model corresponding to the current operating condition of the battery to be tested, determining the health status evaluation value in combination with the battery characteristic parameter set, and then determining the capacity attenuation prediction value in combination with the pre-trained time convolutional network model. This reduces the impact of operating condition fluctuations on the prediction, improves the prediction accuracy, robustness and long-term adaptability, and lays the foundation for subsequently extending battery life and reducing operation and maintenance costs.

[0023] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 This is a flow chart of a battery capacity attenuation prediction method provided according to the first embodiment of the present invention;

[0026] Figure 2 This is a flow chart of a battery capacity attenuation prediction method provided according to the second embodiment of the present invention;

[0027] Figure 3 This is a schematic structural diagram of a battery capacity attenuation prediction device provided according to a third embodiment of the present invention;

[0028] Figure 4 It is a schematic structural diagram of an electronic device implementing an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] Example 1

[0032] Figure 1 A flowchart of a battery capacity attenuation prediction method is provided for the first embodiment of the present invention. This embodiment is applicable to the case of predicting the capacity attenuation value of a battery under different working conditions. The method can be executed by a battery capacity attenuation prediction device, which can be implemented in the form of hardware and / or software. The battery capacity attenuation prediction device can be configured in an electronic device. Figure 1 As shown, the method includes:

[0033] S110: Obtain operating parameter information and current operating condition information of the battery to be tested.

[0034] In this embodiment, the battery to be tested can be understood as a battery for which capacity fade prediction is required, such as a lithium-ion battery. The operating parameter information can be understood as parameters that change during the use of the battery to be tested, such as voltage, current, temperature, and internal resistance. The current operating condition information can be understood as information used to characterize the current operating state of the battery to be tested.

[0035] Specifically, the processor can obtain operating parameter information and current operating conditions through the Battery Management System (BMS) and external sensors. The data sampling frequency can be selected based on the application scenario: high-frequency sampling (e.g., once per second) is suitable for analyzing short-term dynamic characteristics, while low-frequency sampling (e.g., once per minute) is suitable for long-term trend prediction. This can be set based on actual needs. This data can be stored in a time series format, including timestamps and multivariate data.

[0036] Specifically, operating parameter information may include the following key data:

[0037] Voltage: Records the changes in the terminal voltage of the battery under different loads and charge and discharge conditions to reflect the current capacity state of the battery. Current: Collects instantaneous discharge or charge current information to analyze the electrochemical reaction rate during the charge and discharge process. Temperature: Monitors the operating temperature on the surface or inside of the battery to evaluate battery safety and thermal characteristics changes. Internal resistance: Indicates the equivalent internal resistance of the battery, which is an important indicator of aging status. Among them, the internal resistance acquisition adopts a high-frequency pulse test method, combined with real-time dynamic environmental test data, to ensure accurate acquisition under both static and dynamic conditions. The specific steps are as follows:

[0038] Pulse test principle: Apply current pulses by programming the DC load module to collect the instantaneous voltage changes at both ends of the battery. The calculation formula is:

[0039]

[0040] in, is the voltage change, is the current change.

[0041] Use a high-speed data acquisition module (such as NI DAQ) to record voltage fluctuations at a frequency of more than 10 kHz to ensure that transient characteristics are captured.

[0042] Dynamic scene acquisition: Under actual electric vehicle riding conditions, MEMS sensors are used to synchronously record riding acceleration, speed, and external environmental data (such as temperature and humidity) to capture the impact of dynamic load changes on battery performance. Under dynamic load conditions such as acceleration and braking, current and voltage data pairs are collected at multiple points to calculate transient internal resistance in real time. Specifically, the real-time load data is used to match instantaneous current changes, combined with transient voltage fluctuations, and the equivalent internal resistance of the battery is calculated using the above formula to ensure the accuracy and availability of internal resistance data in dynamic environments.

[0043] Equipment calibration and environmental compensation: To ensure the accuracy of the collected data, the collection equipment needs to be calibrated regularly to control the measurement error within 1%. At the same time, in order to reduce the influence of ambient temperature on the internal resistance measurement, a dynamic compensation method is used for correction. The temperature correction formula R corrected =R measured +k(T ambient -T reference ), temperature compensation is performed on the measured internal resistance value, where k is the temperature compensation coefficient, obtained by experimental fitting to improve the reliability of the internal resistance data under different environmental conditions, and the unit is usually Ω / C; R corrected Represents the temperature-corrected internal resistance of the battery, that is, the standardized internal resistance value obtained after temperature compensation; R measured The measured internal resistance of the battery is at ambient temperature T ambient The internal resistance of the battery directly collected under T ambientis the current ambient temperature, that is, the ambient temperature when measuring the battery internal resistance, in °C; T reference The reference temperature is usually set to 25°C or the laboratory standard temperature, which serves as the benchmark for correction calculations.

[0044] For example, the current operating condition information may include data such as riding speed, acceleration, load changes, etc., reflecting the dynamic changes of the battery in actual use; it also includes environmental parameters such as external temperature, humidity and terrain conditions, reflecting the impact of the external environment on battery performance.

[0045] S120 : Extract features based on the operating parameter information to determine a battery feature parameter set.

[0046] In this embodiment, the battery characteristic parameter set can be understood as parameter characteristics obtained after screening.

[0047] Specifically, the processor can preprocess the operating parameter information and then group and label the preprocessed data when training the model. In actual applications, the processor can directly extract and select features from the preprocessed data to determine the set of battery characteristic parameters most relevant to health status assessment.

[0048] S130 : Determine a health status evaluation value of the battery to be tested based on current operating condition information, a pre-trained state evaluation model set, and a battery characteristic parameter set.

[0049] In this embodiment, the pre-trained state assessment model set can be understood as a set of multiple trained models for health state assessment under different operating conditions. The health state assessment value can be understood as a value used to quantify the degree to which the current battery performance is maintained relative to the initial rated performance.

[0050] Specifically, the processor can determine the current working condition of the battery to be tested through the current working condition information, and then select the pre-trained state evaluation model that is most suitable for the working condition from the pre-trained state evaluation model set, and obtain the model output result as the health status evaluation value of the battery to be tested by inputting the battery characteristic parameter set into the pre-trained state evaluation model.

[0051] S140. Determine a capacity attenuation prediction value based on the health status assessment value, the battery characteristic parameter set, and the pre-trained time convolutional network model.

[0052] In this embodiment, the pre-trained temporal convolutional network model can be understood as the trained temporal convolutional network model. The capacity decay prediction value can be understood as a value used to characterize the capacity change under a set number of charge and discharge cycles.

[0053] Specifically, the processor can input the health status assessment value and battery characteristic parameters into a pre-trained time convolutional network model, and determine the output result to obtain a capacity decay prediction value.

[0054] The technical solution of the embodiment of the present invention obtains the operating parameter information and current operating condition information of the battery to be tested; performs feature extraction based on the operating parameter information to determine the battery characteristic parameter set; determines the health status evaluation value of the battery to be tested based on the current operating condition information, the pre-trained state evaluation model set and the battery characteristic parameter set; and determines the capacity attenuation prediction value based on the health status evaluation value, the battery characteristic parameter set and the pre-trained time convolutional network model. By first determining the pre-trained state evaluation model corresponding to the current operating condition of the battery to be tested, determining the health status evaluation value in combination with the battery characteristic parameter set, and then determining the capacity attenuation prediction value in combination with the pre-trained time convolutional network model. This reduces the impact of operating condition fluctuations on the prediction, improves the prediction accuracy, robustness and long-term adaptability, and lays the foundation for subsequently extending battery life and reducing operation and maintenance costs.

[0055] As a first optional embodiment of the first embodiment, after determining the capacity fade prediction value according to the health status assessment value and the battery characteristic parameter set, the method further includes:

[0056] The pre-trained temporal convolutional network model is adaptively optimized to determine an updated pre-trained temporal convolutional network model.

[0057] Specifically, to ensure model accuracy, the processor can adaptively optimize the pre-trained temporal convolutional network model over a period of time or after multiple predictions, dynamically optimizing the model's parameters and weights to determine an updated pre-trained temporal convolutional network model. This allows the processor to address dynamic changes in battery operating data and improve the adaptability and accuracy of the prediction model.

[0058] Furthermore, based on the above embodiment, the steps of adaptively optimizing the pre-trained temporal convolutional network model and determining the updated pre-trained temporal convolutional network model can be refined as follows:

[0059] The predicted values ​​and true values ​​of capacity decay in historical periods are obtained to form a capacity decay prediction sequence and a true value sequence respectively; the global trend and local fluctuation of the capacity decay prediction sequence are extracted through discrete wavelet decomposition; the fused prediction sequence is determined based on the global trend, local fluctuation and fusion weight; the model parameters of the pre-trained temporal convolutional network model are adjusted based on the fused prediction sequence, the true value sequence and the improved whale optimization algorithm to obtain an updated pre-trained temporal convolutional network model; the prediction error is adjusted based on the fused prediction sequence, the true value sequence and the fusion weight.

[0060] In this embodiment, the historical period can be understood as a period for which predictions have been made and actual values ​​are available. The capacity decay prediction sequence can be understood as a chronological sequence of capacity decay prediction values ​​from the historical period. The actual value sequence can be understood as a time series consisting of actual values ​​corresponding to the capacity decay prediction values. The global trend can be understood as the decomposed low-frequency component. The local fluctuation can be understood as the decomposed high-frequency component. Model parameters can include weights and biases. The fusion weight is used to fuse the prediction values ​​determined by the two components.

[0061] Specifically, the capacity decay prediction value and the true value over a period of time can be stored in the storage medium, and the processor can obtain the capacity decay prediction value and the true value over a period of time from the current time, which respectively constitute the capacity decay prediction sequence and the true value sequence. The processor can extract the global trend and local fluctuation of the capacity decay prediction sequence through discrete wavelet decomposition. The processor can use the pre-trained time convolution network model to model based on the global trend to determine the long-term capacity prediction sequence, and use other models to model the local fluctuations to determine the capacity prediction sequence under short-term fluctuations. The processor can fuse the long-term capacity prediction sequence and the capacity prediction sequence under short-term fluctuations through fusion weights to determine the fused prediction sequence. The processor can adjust the model parameters of the pre-trained time convolution network model based on the fused prediction sequence, the true value sequence and the improved whale optimization algorithm to obtain an updated pre-trained time convolution network model; and adjust the prediction error based on the fused prediction sequence, the true value sequence and the fusion weight.

[0062] For example, the capacity decay data is decomposed into a global trend and local fluctuations through wavelet decomposition, and the global trend and local fluctuations are modeled separately to reduce the impact of noise.

[0063] The capacity decay time series data y(t) is decomposed into the global trend c corresponding to the high-frequency component by discrete wavelet decomposition (DWT). h (t) and the local fluctuation c1(t) corresponding to the low-frequency component:

[0064] y(t)=c1(t)+c h (t)

[0065] Where c1(t) represents the local fluctuation, which is used to describe the global attenuation trend; c h (t) represents the global trend, which is used to describe the local fluctuation characteristics. The local fluctuation c1(t) is modeled using a prediction model (such as the pre-trained temporal convolutional network TCN mentioned above) to model the global trend and obtain the long-term capacity prediction value; the global trend c h(t) Use another model (such as the pre-trained extreme learning machine mentioned above) to model local dynamic characteristics and capture short-term fluctuations. By separating global and local characteristics and modeling the two trends separately, the interference of short-term fluctuations on long-term trend prediction is reduced, improving the accuracy of the overall prediction.

[0066] For example, parameter optimization and dynamic adjustment can use the improved whale optimization algorithm (IWOA) to optimize the parameters of the above model (such as weights and biases) to ensure that the model can more accurately fit the current data distribution. The specific steps may include: 1. Initialize the optimization parameters. Initialize the whale population, each individual represents a parameter combination (such as Gaussian kernel parameters , feature weight α). These include: individual position : The current value of the model parameters; 2. The fitness function can use the model prediction error (MSE). The fitness function is calculated for each individual using the MSE algorithm; 3. Position and velocity are updated. The individual position is updated according to the whale optimization formula:

[0067]

[0068] Among them, p represents the current global optimal position; A is used to adjust the distance between the current individual and the optimal individual.

[0069] 4. Dynamic convergence factor. Introducing a nonlinear convergence factor allows the population to focus on the optimal solution in the later stages of iteration:

[0070]

[0071] Where T max is the maximum number of iterations, and t is the current number of iterations.

[0072] 5. Output optimization results. Apply the optimal parameter combination to the prediction model to improve model fitting capabilities. By combining global search with local convergence, the prediction model's ability to fit capacity trends is improved, avoiding being trapped in local optimality.

[0073] Furthermore, dynamic feature weight allocation can adjust the weights of input features (such as incremental capacity features and internal resistance growth rate) in real time, making the model more sensitive to key features. The specific adjustment process may include:

[0074] 1. Feature weight calculation. Assign weights based on the feature variance of the input data:

[0075]

[0076] where w j represents the weight of feature j, represents the variance of feature j.

[0077] 2. Dynamic Adjustment: During operation, feature weights are recalculated based on real-time data to enhance the influence of the incremental capacity feature (ICA) or internal resistance growth feature (RGR) in specific scenarios.

[0078] 3. Model update: Feed the weight adjustment results back to the model input layer to optimize the contribution of feature combinations to the prediction results.

[0079] This step is mainly to enhance the model's dependence on key features and improve the prediction accuracy under special working conditions such as high temperature and low temperature.

[0080] Furthermore, model integration and error feedback can also be included. By integrating the global trend forecast and local fluctuation forecast results, errors can be further reduced and the long-term performance of the model can be dynamically optimized. The specific steps can be:

[0081] 1. Integration formula. The global trend prediction value determined above is and local fluctuation forecasts Fusion:

[0082]

[0083] in, represents the fusion prediction sequence, and β represents the dynamically adjusted fusion weight.

[0084] 2. Error feedback and adjustment. Based on historical errors Adjust the fusion weight:

[0085]

[0086] in, represents the i-th fusion prediction value in the fusion prediction sequence, y i Represents the i-th true value in the true value sequence. In this step, the error compensation value is used to correct the next prediction.

[0087] 3. Iterative optimization: Continuously optimize model parameters and fusion mechanisms to reduce cumulative errors.

[0088] By integrating global and local results, the model's predictive capabilities for short-term fluctuations and long-term trends are balanced.

[0089] In the first optional embodiment of this embodiment one, the global trend is used to capture the long-term trend of changes in battery capacity decay and reduce the interference of short-term fluctuations on the prediction, while the local fluctuation focuses on short-term abnormal changes to enhance the model's sensitivity and correction ability to short-term changes. The extraction of global trends and local fluctuations is used to optimize the temporal convolutional network model to enable it to learn the capacity decay pattern more accurately, rather than being directly used for the final prediction. Although both are derived from the capacity decay prediction sequence, their role is to reduce the impact of noise and enhance the extraction of time series features, thereby improving the long-term prediction stability and short-term correction ability of the model.

[0090] Example 2

[0091] Figure 2 This is a flow chart of a battery capacity attenuation prediction method provided by the second embodiment of the present invention. This embodiment is a further refinement of the above embodiment. Figure 2 As shown, the method includes:

[0092] S201: Obtain operating parameter information and current operating condition information of the battery to be tested.

[0093] S202: Preprocess the operating parameter information to obtain intermediate operating parameter information.

[0094] In this embodiment, the intermediate operating parameter information can be understood as pre-processed parameter information.

[0095] Specifically, the processor may filter noise, fill in missing values, and correct abnormal values ​​on the operating parameter information to obtain intermediate operating parameter information that is more suitable for the input model.

[0096] For example, the processor can first perform noise filtering on the operating parameter information to remove random interference signals generated during the acquisition process. The processor can achieve this goal through two methods: low-pass filtering and Kalman filtering. Among them, low-pass filtering is suitable for smoothing signal fluctuations. The original time series of voltage or current can be input into the filter, and a suitable window size (for example, 10 data points) can be selected for calculation to generate a smoothed signal, thereby reducing the impact of high-frequency noise on the prediction. For more complex time series signals, such as temperature or internal resistance change trends, the processor can use Kalman filtering. First, the state value and covariance matrix in the Kalman filter are initialized, and the estimated value is updated through real-time iteration to dynamically optimize the signal. This filtering method adjusts the estimated value according to the dynamic characteristics of the observed data.

[0097] For example, after noise filtering, the filtered data can be cleaned to handle missing values ​​and outliers. For missing data, the processor can use the Lagrange interpolation method, which constructs an interpolation polynomial based on existing data points to fill in the missing values. In the voltage data, if some timestamp records are missing, interpolation prediction can be performed based on the data of the two previous and next time points to generate a continuous time series. At the same time, the processor can be based on The principle is to detect and remove abnormal values ​​in temperature or internal resistance data. In terms of data format standardization, the processor can store all collected data in a time series format, indexed by timestamp, and record relevant features, including voltage (V), current (I) and filtered internal resistance data (R filtered ), its structure definition can be:

[0098] DataSet={(T1, V, I, R filtered )}

[0099] Among them, T1 is the timestamp, V is the processed voltage, I is the processed current, R filtered The above method further improves the accuracy and robustness of the internal resistance data under dynamic conditions, providing high-quality input data for subsequent analysis.

[0100] Because the numerical ranges of different parameters can vary significantly, directly inputting them into the model may result in certain features dominating the results. To avoid this problem, the standardized data can be normalized using the Min-Max algorithm. For example, for voltage data, normalization can be performed based on the battery's maximum and minimum operating voltage ranges, generating standardized eigenvalues ​​in the range [0, 1]. If the data distribution characteristics need to be preserved, a normalization algorithm can be used instead. Normalization is particularly effective for data such as internal resistance and temperature, ensuring that the feature mean is 0 and the standard deviation is 1, thus avoiding imbalance between features. The final step is dimensionality reduction, especially for multivariate, high-dimensional datasets. Principal component analysis (PCA) can be used to reduce redundant features and retain essential information. The key to this method is calculating the covariance matrix, performing eigenvalue decomposition on the covariance matrix, and selecting the principal eigenvectors as the reduced features. Within the operating parameter information, the processor can reduce the multi-dimensional parameters into a few principal components, significantly reducing data complexity.

[0101] Furthermore, after preprocessing, data grouping and labeling steps can also be included to structure the intermediate operating parameter information after preprocessing so that it can be directly used for subsequent model training or prediction. The purpose of grouping is to divide the continuous time series into appropriate time periods or periodic blocks, and the purpose of labeling is to generate label values ​​required for supervised learning based on the target variable. Among them, the core of data grouping is to divide the original time series data into meaningful segments according to certain rules. The processor can group according to time windows or charge and discharge cycles. For example, when a fixed time window (such as every 10 minutes) is used for grouping, it is necessary to extract continuous segments from the time series and store these segments as independent input data. The processor can traverse the time series data, extract subsets at fixed time intervals, and form a set of data blocks. The format of each data block is as follows:

[0102] [V(t), I(t), T(t), R(t),…]

[0103] Where V(t) is the voltage, I(t) is the current, T(t) is the temperature, R(t) is the internal resistance, and t is the time window.

[0104] Another grouping method is to segment data by charge and discharge cycle. This approach is suitable for generating a set of data blocks for each complete charge and discharge cycle of the battery. In practice, cycle boundaries can be identified by the timestamps of the start and end of charging, and the data within these intervals can be stored as a single data block. Regardless of the grouping method chosen, each data block provides independent input to the model.

[0105] Through labeling, each data block is assigned a target variable, which serves as a supervisory signal for the model. In battery health assessment and capacity decay prediction, the label is typically the battery's state of health (SoH) or remaining capacity. The processor can calculate the health assessment value using the formula and assign it as a label to the data block:

[0106]

[0107] Among them, Q current The actual capacity of the battery can be obtained through discharge test or estimation method; Q rated The rated capacity of the battery, typically provided by the battery specifications, is calculated for each data block and associated with the grouped data. For example, if the actual capacity measured during charge and discharge for a data block is 950 mAh, while the rated capacity is 1000 mAh, the SoH value is 95%.

[0108] In addition to SoH labels, classification and regression labels can also be introduced. For classification tasks, the health status assessment value can be divided into several intervals, such as healthy (90%-100%), slightly degraded (70%-90%), and severely degraded (<70%), and each interval can be assigned a corresponding category. For example, if the health status assessment value of a data block is 85%, its classification label is "slightly degraded." For regression tasks, the continuous health status assessment value (such as 95%) can be directly used as the regression label.

[0109] Through the above grouping and labeling steps, the preprocessed data is converted into a standard training set that can be directly used by the model. A complete sample consists of input features (such as voltage and current series within a time window) and target labels (such as health status assessment values ​​or categories). Taking time window grouping as an example, the final dataset structure may be as follows:

[0110] surface Battery health status and characteristic sequence grouping table

[0111]

[0112] Through this structured grouping and labeling, subsequent models can easily use input features to predict target variables, thereby achieving accurate assessment of battery health status and prediction of capacity decay trends.

[0113] Through the above process, the collected and preprocessed standardized data can be used as input for machine learning models. The core of this process is to ensure data quality (noise-free and missing data), consistency (feature normalization and dimensionality uniformity), and validity (feature extraction and label generation), providing a solid foundation for subsequent feature analysis and predictive modeling.

[0114] S203: Extract features from the intermediate operating parameter information to obtain intermediate parameter features.

[0115] In this embodiment, the intermediate parameter feature can be understood as the result of feature extraction.

[0116] Specifically, since the determination of two numerical values ​​includes the health status assessment value and capacity attenuation prediction value of the battery to be tested, the processor can extract important features that can reflect the battery health status and capacity attenuation from the intermediate operating parameter information, and use the algorithm to screen the optimal feature subset as the intermediate parameter features to improve the performance and robustness of the model.

[0117] For example, the first step of feature extraction is the calculation of basic statistical features, which quantifies the overall profile of the data distribution. By calculating the mean, standard deviation, and range of voltage or current, the stability and fluctuation of the battery's operating state can be understood. Among them, the mean is used to evaluate the average performance of the battery over a period of time. For example, within a 10-minute time window, the voltage data is averaged to obtain the typical voltage level within that period. The standard deviation is used to measure the degree of fluctuation of the voltage data and to determine whether there are unstable fluctuations during the operation of the battery. The range reflects the maximum range of variation of the voltage or temperature data and can quickly identify extreme situations such as overcharging or over-discharging.

[0118] The second step is to extract the incremental capacity characteristics, which is an important indicator of battery aging. The incremental capacity curve describes the charging capacity Voltage The incremental capacity formula is:

[0119]

[0120] To extract this feature, it is necessary to collect voltage and capacity data during the charging process and calculate the change in both. If the incremental capacity decreases with increasing cycle number, it indicates that the battery has aged. The peak height, position, and rate of change of the incremental capacity curve can also be used as important features.

[0121] The third step is to extract the internal resistance characteristics. The internal resistance characteristics are another important indicator of the battery's health status. Especially under the condition of changing ambient temperature, its changing trend can better reveal the aging of the battery. The formula for the internal resistance temperature gradient is:

[0122]

[0123] in, Indicates the change in internal resistance, To represent the change in ambient temperature, when extracting the internal resistance characteristics, you can record the internal resistance value of the battery under different temperature conditions and calculate its gradient with temperature change. If the gradient increases significantly, it may indicate that the battery is seriously aged.

[0124] The fourth step involves extracting time series features. The processor uses a Fourier transform (FFT) to extract frequency-domain characteristics and identify periodic patterns or dominant frequency components. The processor can perform an FFT on voltage or current data to extract the dominant frequency and amplitude. This method is particularly suitable for detecting periodic signals in batteries under specific load conditions.

[0125] The fifth step is to extract the charge and discharge characteristics to evaluate the battery's energy efficiency and its changing trend. The following is the charge and discharge efficiency formula:

[0126]

[0127] in, Indicates the charge and discharge efficiency, the percentage of battery energy utilization; E discharge Indicates the total energy provided by the battery during discharge (output energy); E charge Represents the total energy input to the battery during charging (input energy). To extract this feature, it is necessary to record the energy input and output during charging and discharging. Energy can be calculated using the following formula:

[0128]

[0129] Here, E refers to the energy accumulated during the battery's charge or discharge process; V refers to the battery's instantaneous voltage during the charge or discharge process; I refers to the battery's instantaneous current during the charge or discharge process; and dt refers to the time differential, which represents the integration of time. Substituting the charge and discharge energies into the efficiency formula yields the charge and discharge efficiency. If the efficiency decreases significantly with the number of cycles, it indicates battery performance degradation.

[0130] For a better understanding, here is an application example of electric bicycle battery feature extraction, as shown in the following table.

[0131] surface E-bike battery feature extraction application example

[0132] S204 : Screening the intermediate parameter features based on a preset algorithm set to determine a battery characteristic parameter set, where the battery characteristic parameter set includes basic battery operation data and dynamic characteristic parameters.

[0133] In this embodiment, the preset algorithm set can be understood as a collection of preset algorithms for feature screening. Different preset algorithms can be applied to different situations and scenarios and can be selected according to needs. The battery characteristic parameter set can be understood as the characteristic parameters used to determine the subsequent health status assessment value and the parameters used to determine the capacity decay prediction value. It can include basic battery operation data and dynamic characteristic parameters. Basic battery operation data can be understood as processed battery operation data, and dynamic characteristic parameters can be understood as characteristic parameters within a certain time period.

[0134] Specifically, the goal of feature selection is to select the most representative features from the extracted features for the target (i.e., subsequent health status assessment value or capacity decay prediction value) while removing redundant or irrelevant features, thereby improving model training efficiency and predictive performance. The processor can use a preset algorithm set to screen intermediate parameter features, quantify the importance or relevance of each feature, and thus determine the optimal set of battery feature parameters.

[0135] Exemplary algorithms in the preset algorithm set may include recursive feature elimination, principal component analysis, LASSO regularization, and feature importance analysis. Recursive Feature Elimination (RFE) is a commonly used feature selection method that optimizes a feature subset by recursively removing unimportant features. In operation, the model is first trained using all features, and the importance score of each feature (such as the weight of a linear model or the split score of a tree model) is calculated. The feature with the lowest score is then removed, and this process is repeated until a preset number of features is reached. The specific steps are as follows: initialize the model and train it using all features; sort the features based on model weights or feature importance, and remove the features with the lowest scores; retrain the model and repeat the sorting and removal steps until the required number of features is met, outputting the remaining feature subset as the final selection. In practical applications, RFE is suitable for scenarios with a large number of features. For example, when selecting 100 features, the number of features can be gradually reduced (e.g., by removing 10 at a time) to ultimately find the feature subset with the greatest impact on the target variable. This method is particularly suitable for models such as random forests and support vector machines that have the ability to calculate feature importance. Principal Component Analysis (PCA) is a commonly used dimensionality reduction method. By projecting high-dimensional feature data into a low-dimensional space, it extracts the main direction of variation in the data, retaining key information while reducing redundancy. The core of PCA is to extract principal components through eigenvalue decomposition. The main steps are as follows:

[0136] Calculate the covariance matrix C:

[0137]

[0138] Among them, x i is the data vector and μ is the mean.

[0139] Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors:

[0140]

[0141] in, is the eigenvalue, is the feature vector.

[0142] Arrange in descending order of eigenvalue size and select eigenvectors with cumulative contribution rates exceeding 90%.

[0143] The cumulative contribution rate can be expressed as:

[0144]

[0145] PCA is suitable for scenarios with very high feature dimensionality but redundancy. For example, when the feature dimension exceeds 1000, PCA can reduce the data to 10-20 principal components while retaining over 90% of the information. Tools such as the Python Scikit-learn library can be used to quickly perform PCA dimensionality reduction and input the principal components as new features into the machine learning model.

[0146] LASSO (Least Absolute Shrinkage and Selection Operator) is a feature selection method based on regression models. It adds an L1 regularization term to the model's loss function, reducing the weights of some features to 0, thereby achieving automated feature selection. The optimization objective formula of LASSO is relatively common, so it will not be described here.

[0147] In practical applications, the processor can control the sparsity of the model by adjusting the parameters in the LASSO optimization objective formula. When the parameters are small, most features are retained; as the parameters increase, the weights of more features are compressed to 0, automatically filtering out the most important features. For example, when applying LASSO to select battery features, different parameter values ​​can be set and the results observed, ultimately selecting features with non-zero weights as input.

[0148] Feature importance analysis can be an intuitive and efficient feature selection method based on tree models (such as random forest or XGBoost). These models calculate the contribution of each feature to model performance through the gain of node splitting and use this to rank the importance of features.

[0149] During the process, a pre-trained random forest model is first trained using all features. The scores for each feature are then extracted and sorted from high to low by score, with the top several important features selected as input. Feature importance analysis is suitable for high-dimensional datasets and requires no additional parameter tuning, resulting in fast computation and stable results.

[0150] Feature selection can efficiently identify the most representative feature subsets through recursive feature elimination (RFE), principal component analysis (PCA), LASSO regularization, and feature importance analysis. In specific applications, the appropriate method can be selected based on the data size and target task. For example, PCA is suitable for dimensionality reduction of high-dimensional data, LASSO is suitable for sparse models, and random forest feature importance is more suitable for nonlinear problems. Combining these methods can provide more efficient and accurate input data for model training, thereby improving overall performance.

[0151] In practice, the processor can first calculate each feature dimension in the data block, such as extracting the mean and standard deviation through basic statistical methods. Subsequently, based on the task objectives (such as SoH prediction or capacity decay analysis), the most relevant feature subsets are screened using methods such as recursive feature elimination (RFE) or principal component analysis (PCA). Ultimately, these feature subsets will be used to construct the input of the machine learning model. For example, in battery health prediction, if charge and discharge efficiency and incremental capacity peak height are selected as important features, only these key features need to be provided when training the model, which not only reduces computational complexity but also improves the model's generalization ability. Through this process, high relevance and low redundancy of features can be ensured, providing optimal input for subsequent model training.

[0152] S205: Determine the operating condition type to which the current operating condition information belongs. The operating condition types include dynamic operating conditions, static operating conditions, and resource-constrained scenarios.

[0153] In this embodiment, the operating condition types can be divided into dynamic operating conditions, static operating conditions and resource-constrained scenarios.

[0154] For example, during the operation of an electric bicycle, if the load changes dramatically (such as frequent starting and stopping), it can be judged as a dynamic operating condition; under uniform road conditions, it corresponds to a static operating condition; in situations such as low power, low computing costs are required to maintain power, which corresponds to a resource-constrained scenario.

[0155] Specifically, the processor can determine the type of working condition based on the current working condition information. The working condition types include dynamic working conditions, static working conditions, and resource-constrained scenarios.

[0156] S206. Filter a target state assessment model that matches the working condition type from a set of pre-trained state assessment models; the set of pre-trained state assessment models includes a pre-trained lightweight long short-term memory network model, a pre-trained extreme learning machine, and a pre-trained random forest model.

[0157] In this embodiment, the pre-trained state assessment model set can be understood as including models used to determine health state assessment values, wherein each operating condition type has a corresponding relationship with the corresponding pre-trained state assessment model, and the target state assessment model can be understood as a model adapted to the current operating condition type for determining the health state assessment value. During the performance verification and optimization process of the above models, the model performance can be quantified using mean square error, mean absolute error, or mean absolute percentage error for subsequent optimization.

[0158] Specifically, the processor can filter the target state evaluation model that matches the working condition type according to the corresponding relationship recorded in the pre-trained state evaluation model set; the pre-trained state evaluation model set includes a pre-trained lightweight long short-term memory network model, a pre-trained extreme learning machine and a pre-trained random forest model.

[0159] For example, a pre-trained lightweight long short-term memory network model is used for dynamic working conditions, which can effectively process high-frequency changing data and is suitable for real-time prediction under complex operating conditions. A pre-trained random forest model is used for static working conditions, which is suitable for evaluation under constant load or single environmental conditions and has stability. A pre-trained extreme learning machine is used in resource-constrained scenarios to achieve low computing cost requirements through fast training and prediction. For example, during the operation of an electric bicycle, if the load changes drastically (such as frequent starting and stopping), it corresponds to a dynamic working condition. The pre-trained lightweight long short-term memory network model can use time series information to capture complex dynamics; under uniform road conditions, it corresponds to a static working condition. The pre-trained random forest model can quickly give accurate evaluation results; when the battery is low, it corresponds to a resource-constrained scenario. The use of a pre-trained extreme learning machine can make predictions efficiently with low resource usage.

[0160] Among them, the pre-trained lightweight long short-term memory network model can learn the time series characteristics of battery operation data and is suitable for predicting health status assessment values ​​under dynamic conditions. Its formula is:

[0161]

[0162] in, Represents the hidden state at the current time, representing the time-dependent information captured by the LSTM model; is the input feature at the current time (such as voltage, current, etc.); 、 is a weight matrix obtained through training. When training this model, historical voltage and current data can first be segmented (for example, into 10-minute time windows) to construct a time series input matrix. This input is then passed to the lightweight long short-term memory network model, where hidden states are calculated layer by layer. Finally, a fully connected layer maps the hidden states to health status assessment values. For example, based on voltage and current changes over a 10-minute period, the lightweight long short-term memory network model can output a health status assessment value (e.g., 92%) as the assessment result for that period. By optimizing the lightweight long short-term memory network model structure (for example, reducing the number of network layers or optimizing weight pruning), efficient operation can be achieved on embedded hardware.

[0163] Among them, the pre-trained extreme learning machine is an efficient model that does not require iterative training and can perform fast predictions. During training, the weight matrix contained therein can be optimized by introducing the BES algorithm to obtain a pre-trained extreme learning machine.

[0164] Among them, random forest is an integrated learning model that performs regression prediction through multiple decision trees.

[0165] In this model, characteristic data (such as voltage, current, and internal resistance) are input into multiple decision trees. Each tree makes independent predictions and the average value is taken as the health status assessment value. The pre-trained random forest model is particularly suitable for health status assessment values ​​under static working conditions (such as constant load or uniform temperature). In actual deployment, the number of trees can be adjusted to and maximum depth, optimizing the computational cost and prediction accuracy of the model.

[0166] Furthermore, based on the above embodiment, the training steps of the pre-trained extreme learning machine may include:

[0167] Obtain a feature matrix of samples to be input and an initial extreme learning machine; randomly initialize the weight matrix of the initial extreme learning machine using a vulture search optimization algorithm to obtain an initial weight matrix set; determine the prediction error of each initial weight matrix in the initial weight matrix set based on the initial extreme learning machine and the feature matrix of the samples to be input; optimize each initial weight matrix using the prediction error to obtain a candidate weight matrix set and perform iterative optimization until a termination condition is met to obtain a final weight matrix; determine a pre-trained extreme learning machine based on the final weight matrix and the initial extreme learning machine.

[0168] In this embodiment, the feature matrix of the input samples can be understood as the feature matrix formed after the historical samples are extracted through the above-mentioned feature extraction, which is used for model training. The initial extreme learning machine can be understood as an untrained extreme learning machine. The vulture search optimization algorithm can be understood as an algorithm for solving complex optimization problems. By simulating the searching, hunting, and migration behaviors of vultures, it explores the solution space and finds the global optimal solution. The initial weight matrix set can be understood as the unoptimized weight matrix set. The prediction error can be understood as the error value used to measure the accuracy of the weights. The termination condition can be understood as the condition used to determine whether the current weight matrix set can be used as the final result. For example, it can be the maximum number of iterations or the prediction error is within the error threshold.

[0169] Specifically, the extreme learning machine is an efficient model that does not require iterative training and can quickly predict health status assessment values. Its formula is:

[0170] Hβ1=Y

[0171] Where H is the input feature matrix, which contains all health indicators (such as internal resistance growth and incremental capacity peak); β1 is the weight matrix, which represents the model parameters and is determined through a one-time optimization; Y is the predicted output, that is, the health status assessment value.

[0172] Among them, the Bald Eagle Search (BES) algorithm is introduced here to optimize the weight matrix. The BES algorithm is a new optimization algorithm based on the foraging behavior of vultures, which is used to solve complex optimization problems. It explores the solution space and finds the global optimal solution by simulating the search, hunting and migration behavior of vultures. The BES algorithm is used to optimize the weight matrix β1 in the extreme learning machine model to improve the prediction accuracy and robustness of the model. The traditional extreme learning machine uses the least squares method to solve directly, but this method may lead to insufficient generalization ability of the model under high-dimensional features or noisy data. Therefore, the BES algorithm is introduced to perform global optimization of β1 to improve the robustness and adaptability of the model. The operating principle of the BES algorithm is as follows:

[0173] Search phase: Global exploration of the solution space. In this phase, the vulture simulates observing its prey from high altitude and conducts a random but guided search of the solution space. The formula is as follows:

[0174]

[0175] in, Indicates the value of the current solution; X best Represents the current global optimal solution; r1 represents the random factor, which controls the search step size.

[0176] When optimizing the weight matrix, this stage is mainly used to generate multiple candidate weight matrices in the initial solution space.

[0177] Hunting phase: Local development. Once the vulture locks on its prey, the algorithm will locally develop the solution by adjusting parameters to find a better solution. The formula is as follows:

[0178]

[0179] Among them, X mean represents the average of all solutions; r2 represents another random factor used to introduce moderate perturbations. In this stage, the weight matrix is ​​refined through a refined search to effectively reduce the prediction error.

[0180] Migration phase: To avoid being trapped in a local optimum, the vulture will migrate after a long search to explore new areas. The formula is as follows:

[0181]

[0182] Among them, X max 、Xmin Indicates the upper and lower limits of the current solution; r3 represents the random perturbation factor. During the optimization process, this stage can prevent the optimization from falling into the local optimum.

[0183] The specific steps for using the BES algorithm to optimize the weight matrix in the extreme learning machine are as follows:

[0184] First, the processor can initialize the candidate solutions and randomly generate an initial set of weight matrices (multiple candidate solutions) based on the input feature matrix H and the target output Y. The processor can then perform iterative search, hunting, and migration phases to gradually optimize each candidate matrix, and then calculate the objective function of each candidate solution (such as the prediction error MSE), and then update the current optimal solution X based on the error. best .

[0185] When the error is lower than the preset threshold or reaches the maximum number of iterations, the termination condition is met, the optimization is stopped, and the final weight matrix β is output. optima .

[0186] The BES algorithm combines global search with local development to find the globally optimal weight matrix in complex, high-dimensional solution spaces, avoiding local optima. Compared to traditional optimization methods, the BES algorithm utilizes random factors and migration mechanisms to accelerate the convergence of the weight matrix. The BES-optimized ELM model exhibits enhanced generalization capabilities under noisy data and diverse operating conditions, ensuring the stability of health status assessment values.

[0187] For example, in this model, the processor can first input a set of battery characteristic parameters (for example, characteristics collected for a specific battery under different temperature and load conditions) into a pre-trained extreme learning machine model to directly calculate the current health status assessment value. Because the pre-trained extreme learning machine model does not require backpropagation, training and inference speed are faster, making it suitable for resource-constrained embedded environments.

[0188] Ultimately, the BES-optimized pre-trained extreme learning machine model can efficiently predict the current battery's state of health assessment value, with significantly lower error than traditional methods. Combined with the BES-optimized pre-trained extreme learning machine model, it offers significant advantages in prediction accuracy, convergence speed, and computational efficiency, providing a highly effective solution for battery health assessment.

[0189] S207: Input the battery characteristic parameter set into the target state assessment model to obtain the model output result as the health state assessment value of the battery to be tested.

[0190] In this embodiment, the model output result can be understood as the result output after each model is calculated.

[0191] Specifically, the processor can input the characteristic parameters of the battery characteristic parameter set used for calculating the health status assessment value (for example, the basic battery operation data therein) into the target state assessment model, obtain the model output result and use it as the health status assessment value of the battery to be tested.

[0192] S208 : Determine a time series input matrix according to the health status evaluation value and the dynamic characteristic parameters in the battery characteristic parameter set.

[0193] In this embodiment, the time sequence input matrix can be understood as an input matrix after time sequence arrangement.

[0194] Specifically, the processor can determine the corresponding timing sequence input matrix for each charge and discharge cycle based on the health status assessment value and the dynamic characteristic parameters in the battery characteristic parameter set (for example, dynamic characteristic parameters such as incremental capacity characteristics and internal resistance growth rate) and other characteristics such as the number of charge and discharge cycles, voltage and temperature in the basic battery operation data.

[0195] Among them, the internal resistance growth rate reflects the growth trend of internal resistance with the number of cycles through linear fitting, and the calculation formula is:

[0196] R t =R0+kt

[0197] Where R0 is the initial internal resistance, k is the growth rate, and t is the time or number of cycles. In practice, the change in the battery's internal resistance is measured regularly, fitted into a linear relationship, and the growth rate k is extracted as a dynamic input feature.

[0198] S209: Input the time series input matrix into the pre-trained time convolutional network model to determine the predicted value of capacity attenuation to be corrected under the set number of cycles in the future.

[0199] In this embodiment, the pre-trained temporal convolutional network model is a trained model used to determine the capacity decay prediction value. The temporal convolutional network model captures the long-term dependencies of time series data through extended convolution, while also being parallel and efficient, making it suitable for processing matrices of long sequence data. The future set number of cycles can be understood as a set number of charge and discharge cycles, such as 50 cycles in the future. The capacity decay prediction value to be corrected can be understood as the inaccurate capacity prediction value directly output by the model.

[0200] Specifically, the processor can input the time series input matrix into the pre-trained time convolutional network model to obtain the output result of the model as the predicted value of capacity attenuation to be corrected under the future set number of cycles.

[0201] For example, the processor can input a time series input matrix formed by the most recent 50 cycles, and the model can predict the battery capacity after the next 10 cycles, helping users plan maintenance in advance. For example, for an electric bicycle battery in operation, its current health assessment value is known to be 90%. Combined with the incremental capacity characteristics and internal resistance growth rate of the most recent 50 cycles, the pre-trained temporal convolutional network model can predict the capacity change trend after the next 100 cycles and prompt the user to schedule maintenance before the capacity drops to 80%.

[0202] Among them, the basic formula of the pre-trained time convolutional network model is as follows:

[0203]

[0204] Among them, y t is the predicted value of capacity attenuation to be corrected at time t; Indicates time Input features; w i represents the convolution kernel weight; d is the expansion step size to capture long-term dependencies.

[0205] When using a pretrained temporal convolutional network model, the time series input matrix is ​​organized into fixed-length time series and fed into the model. The dilated convolutional layer extracts capacity variation patterns over long timeframes by gradually increasing the step size. Ultimately, the model outputs capacity predictions for several future cycles.

[0206] S210: Determine a historical error feedback value based on historical prediction results and historical actual results.

[0207] In this embodiment, the historical prediction results can be understood as the results obtained when the historical predictions are made using the pre-trained temporal convolutional network model. The historical actual results can be understood as the actual results corresponding to the historical prediction results. The historical error feedback value can be understood as the battery capacity value used to correct the error.

[0208] In actual operation, due to changes in environmental conditions or the influence of data noise, the model's prediction may deviate. To this end, a dynamic error correction mechanism is introduced. The processor can determine the historical error feedback value based on historical prediction results and historical actual results to adjust the prediction results in real time. The formula for determining the historical error feedback value is:

[0209]

[0210] in, Represents the historical error feedback value averaged over the past m predictions, is the i-th historical prediction result, P iThe corresponding historical actual results. By correcting the current forecast value with historical error feedback, the model can dynamically adjust the current forecast value, significantly reducing the impact of cumulative errors.

[0211] S211 . Determine a capacity decay prediction value based on historical error feedback values ​​and the capacity decay prediction value to be corrected.

[0212] Specifically, the processor may add the historical error feedback value to the capacity decay prediction value to be corrected to obtain a final capacity decay prediction value.

[0213] The ultimate goal of this step is to obtain the predicted value of battery capacity decay to quantify the change in battery health status. Therefore, it is necessary to further process the predicted future capacity and calculate the corresponding predicted value of capacity decay to provide a basis for subsequent error correction and optimization. The basic formula for calculating the predicted value of capacity decay is:

[0214]

[0215] in, is the predicted value of capacity decay, which indicates the capacity loss compared to the initial state; y0 is the initial rated capacity of the battery; y t is the battery capacity at a certain moment in the future predicted by the TCN model.

[0216] The technical solution of the embodiments of the present invention preprocesses and extracts features from operating parameter information to obtain intermediate parameter features. These intermediate parameter features are then filtered based on a preset algorithm set to determine a battery characteristic parameter set, reducing the computational complexity of subsequent models and providing optimal input for subsequent models. By determining the operating condition type to which the current operating condition information belongs and selecting a target state assessment model adapted to that operating condition type from a set of pretrained state assessment models, model adaptability is ensured, thereby reducing the impact of operating condition fluctuations on predictions. By inputting the battery characteristic parameter set into the target state assessment model to obtain a health state assessment value, this completes the entire process from input feature processing to multi-model prediction. By using different pretrained state assessment models for prediction, a good balance is achieved between prediction accuracy, efficiency, and adaptability, providing reliable support for battery health management. The dynamic features in the battery characteristic parameter set enhance the input feature's ability to describe capacity decay. Dilated convolution captures temporal dependencies, improving the efficiency and accuracy of capacity prediction. Historical error feedback values ​​are determined based on historical results to adjust the prediction value in real time, ensuring the model's long-term stability and multi-scenario adaptability. The prediction accuracy, robustness and long-term adaptability are improved, laying the foundation for subsequently extending battery life and reducing operation and maintenance costs.

[0217] Example 3

[0218] Figure 3This is a schematic diagram of the structure of a battery capacity attenuation prediction device provided by the third embodiment of the present invention. Figure 3 As shown, the device includes: an information acquisition module 31, a first determination module 32, a second determination module and a third determination module 34.

[0219] An information acquisition module 31 is used to obtain operating parameter information and current operating condition information of the battery to be tested;

[0220] A first determining module 32 is configured to extract features based on the operating parameter information and determine a battery feature parameter set;

[0221] A second determining module 33 is configured to determine a health status evaluation value of the battery to be tested based on the current operating condition information, the pre-trained state evaluation model set, and the battery characteristic parameter set;

[0222] The third determination module 34 is used to determine a capacity decay prediction value based on the health status assessment value, the battery characteristic parameter set and the pre-trained time convolution network model.

[0223] The technical solution of the embodiment of the present invention obtains the operating parameter information and current operating condition information of the battery to be tested; performs feature extraction based on the operating parameter information to determine the battery characteristic parameter set; determines the health status evaluation value of the battery to be tested based on the current operating condition information, the pre-trained state evaluation model set and the battery characteristic parameter set; and determines the capacity attenuation prediction value based on the health status evaluation value, the battery characteristic parameter set and the pre-trained time convolutional network model. By first determining the pre-trained state evaluation model corresponding to the current operating condition of the battery to be tested, determining the health status evaluation value in combination with the battery characteristic parameter set, and then determining the capacity attenuation prediction value in combination with the pre-trained time convolutional network model. This reduces the impact of operating condition fluctuations on the prediction, improves the prediction accuracy, robustness and long-term adaptability, and lays the foundation for subsequently extending battery life and reducing operation and maintenance costs.

[0224] Furthermore, the first determining module 32 is specifically configured to:

[0225] Preprocessing the operating parameter information to obtain intermediate operating parameter information;

[0226] Performing feature extraction on the intermediate operating parameter information to obtain intermediate parameter features;

[0227] The intermediate parameter characteristics are screened based on a preset algorithm set to determine a battery characteristic parameter set, where the battery characteristic parameter set includes basic battery operation data and dynamic characteristic parameters.

[0228] Furthermore, the second determining module 33 is specifically configured to:

[0229] Determining the operating condition type to which the current operating condition information belongs, wherein the operating condition types include dynamic operating conditions, static operating conditions, and resource-constrained scenarios;

[0230] A target state assessment model that matches the working condition type is selected from a set of pre-trained state assessment models; the set of pre-trained state assessment models includes a pre-trained lightweight long short-term memory network model, a pre-trained extreme learning machine, and a pre-trained random forest model:

[0231] The battery characteristic parameter set is input into the target state assessment model to obtain a model output result as the health state assessment value of the battery to be tested.

[0232] The device also includes a model training unit.

[0233] The model training unit is specifically used for:

[0234] Obtain the feature matrix of the sample to be input and the initial extreme learning machine;

[0235] Randomly initializing the weight matrix of the initial extreme learning machine using a vulture search optimization algorithm to obtain an initial weight matrix set;

[0236] Determining a prediction error of each initial weight matrix in the initial weight matrix set according to the initial extreme learning machine and the feature matrix of the sample to be input;

[0237] Optimizing each of the initial weight matrices using the prediction error to obtain a candidate weight matrix set and performing iterative optimization until a termination condition is met to obtain a final weight matrix;

[0238] A pre-trained extreme learning machine is determined according to the final weight matrix and the initial extreme learning machine.

[0239] Furthermore, the third determining module 34 is specifically configured to:

[0240] Determining a time series input matrix according to the health status assessment value and the dynamic characteristic parameters in the battery characteristic parameter set;

[0241] Inputting the time series input matrix into a pre-trained temporal convolutional network model to determine a predicted value of capacity decay to be corrected under a set number of future cycles;

[0242] Determine the historical error feedback value based on historical prediction results and historical actual results;

[0243] A capacity decay prediction value is determined according to the historical error feedback value and the capacity decay prediction value to be corrected.

[0244] Optionally, the device further includes:

[0245] The model optimization module is used to adaptively optimize the pre-trained temporal convolutional network model and determine an updated pre-trained temporal convolutional network model.

[0246] The model optimization module is specifically used to:

[0247] Obtain the capacity decay prediction value and actual value in the historical period to form the capacity decay prediction sequence and actual value sequence respectively;

[0248] Extracting the global trend and local fluctuation of the capacity decay prediction sequence through discrete wavelet decomposition;

[0249] Determining a fusion prediction sequence according to the global trend, the local fluctuation and the fusion weight;

[0250] According to the fused prediction sequence, the true value sequence and the improved whale optimization algorithm, the model parameters of the pre-trained temporal convolutional network model are adjusted to obtain an updated pre-trained temporal convolutional network model;

[0251] The prediction error is adjusted according to the fused prediction sequence, the true value sequence and the fusion weight.

[0252] The battery capacity attenuation prediction device provided in the embodiment of the present invention can execute the battery capacity attenuation prediction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0253] Example 4

[0254] Figure 4 A schematic diagram of an electronic device 40 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0255] like Figure 4As shown, electronic device 40 includes at least one processor 41 and memory, such as ROM 42 and RAM 43, communicatively connected to at least one processor 41. The memory stores computer programs executable by the at least one processor, and processor 41 can perform various appropriate actions and processes based on the computer programs stored in ROM 42 or loaded from storage unit 48 into RAM 43. RAM 43 can also store various programs and data required for the operation of electronic device 40. Processor 41, ROM 42, and RAM 43 are interconnected via bus 44. An I / O interface 45 is also connected to bus 44.

[0256] Multiple components in the electronic device 40 are connected to the I / O interface 45, including an input unit 46, such as a keyboard, a mouse, etc.; an output unit 47, such as various types of displays, speakers, etc.; a storage unit 48, such as a magnetic disk, an optical disk, etc.; and a communication unit 49, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 49 allows the electronic device 40 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0257] Processor 41 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of processor 41 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any other suitable processor, controller, microcontroller, etc. Processor 41 executes the various methods and processes described above, such as the battery capacity decay prediction method.

[0258] In some embodiments, the battery capacity degradation prediction method can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as storage unit 48. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 40 via ROM 42 and / or communication unit 49. When the computer program is loaded into RAM 43 and executed by processor 41, one or more steps of the battery capacity degradation prediction method described above can be performed. Alternatively, in other embodiments, processor 41 can be configured to execute the battery capacity degradation prediction method in any other appropriate manner (e.g., by means of firmware).

[0259] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system comprising at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0260] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the computer program is executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program may be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0261] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, device, or apparatus. A computer-readable storage medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or apparatus, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media may include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0262] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device that has: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0263] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0264] A computing system may include clients and servers. The clients and servers are typically remote from each other and typically interact via a communication network. This client-server relationship arises through computer programs running on the respective computers, creating a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host. This server is a hosting product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosting and VPS services.

[0265] In one embodiment, the present invention further includes a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the battery capacity attenuation prediction method of any embodiment of the present invention.

[0266] During implementation, the computer program product may be written in one or more programming languages ​​or a combination thereof to perform the operations of the present invention. The programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0267] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This is not limited herein.

[0268] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A method for predicting battery capacity attenuation, characterized in that: include: Obtain operating parameter information and current working condition information of the battery to be tested; Perform feature extraction based on the operating parameter information to determine a battery feature parameter set; Determine a health status assessment value of the battery to be tested based on the current operating condition information, the pre-trained state assessment model set, and the battery characteristic parameter set; Determining a capacity decay prediction value based on the health status assessment value, the battery characteristic parameter set, and a pre-trained temporal convolutional network model; The pre-trained state evaluation model set includes a pre-trained lightweight long short-term memory network model, a pre-trained extreme learning machine and a pre-trained random forest model; The determining of the health status evaluation value of the battery to be tested based on the current operating condition information, the pre-trained state evaluation model set, and the battery characteristic parameter set includes: Determining the operating condition type to which the current operating condition information belongs, wherein the operating condition types include dynamic operating conditions, static operating conditions, and resource-constrained scenarios; Selecting a target state assessment model that matches the operating condition type from the pre-trained state assessment model set, including: using the pre-trained lightweight long short-term memory network model for the dynamic operating condition, using the pre-trained random forest model for the static operating condition, and using the pre-trained extreme learning machine for the resource-constrained scenario; Inputting the battery characteristic parameter set into the target state assessment model to obtain a model output result as a health state assessment value of the battery to be tested; The method of determining the capacity attenuation prediction value based on the health status assessment value, the battery characteristic parameter set, and the pre-trained time convolutional network model includes: Determining a time series input matrix according to the health status assessment value and the dynamic characteristic parameters in the battery characteristic parameter set; Inputting the time series input matrix into a pre-trained temporal convolutional network model to determine a predicted value of capacity decay to be corrected under a set number of future cycles; Determine the historical error feedback value based on historical prediction results and historical actual results; A capacity decay prediction value is determined according to the historical error feedback value and the capacity decay prediction value to be corrected.

2. The method according to claim 1, characterized in that The extracting features based on the operating parameter information to determine a battery feature parameter set includes: Preprocessing the operating parameter information to obtain intermediate operating parameter information; Performing feature extraction on the intermediate operating parameter information to obtain intermediate parameter features; The intermediate parameter characteristics are screened based on a preset algorithm set to determine a battery characteristic parameter set, where the battery characteristic parameter set includes basic battery operation data and dynamic characteristic parameters.

3. The method according to claim 1, characterized in that The training steps of the pre-trained extreme learning machine include: Obtain the feature matrix of the sample to be input and the initial extreme learning machine; Randomly initializing the weight matrix of the initial extreme learning machine using a vulture search optimization algorithm to obtain an initial weight matrix set; Determining a prediction error of each initial weight matrix in the initial weight matrix set according to the initial extreme learning machine and the feature matrix of the sample to be input; Optimizing each of the initial weight matrices using the prediction error to obtain a candidate weight matrix set and performing iterative optimization until a termination condition is met to obtain a final weight matrix; A pre-trained extreme learning machine is determined according to the final weight matrix and the initial extreme learning machine.

4. The method according to claim 1, wherein After determining the capacity fade prediction value according to the health status assessment value and the battery characteristic parameter set, the method further includes: Adaptively optimize the pre-trained temporal convolutional network model to determine an updated pre-trained temporal convolutional network model.

5. The method according to claim 4, characterized in that Adaptively optimizing the pre-trained temporal convolutional network model to determine an updated pre-trained temporal convolutional network model includes: Obtain the capacity decay prediction value and actual value in the historical period to form the capacity decay prediction sequence and actual value sequence respectively; Extracting the global trend and local fluctuation of the capacity decay prediction sequence through discrete wavelet decomposition; Determining a fusion prediction sequence according to the global trend, the local fluctuation and the fusion weight; According to the fused prediction sequence, the true value sequence and the improved whale optimization algorithm, the model parameters of the pre-trained temporal convolutional network model are adjusted to obtain an updated pre-trained temporal convolutional network model.

6. A battery capacity attenuation prediction device, characterized in that: include: An information acquisition module is used to obtain operating parameter information and current operating condition information of the battery to be tested; A first determining module, configured to extract features based on the operating parameter information and determine a battery feature parameter set; A second determination module is used to determine the health status evaluation value of the battery to be tested based on the current operating condition information, the pre-trained state evaluation model set and the battery characteristic parameter set; a third determination module, configured to determine a capacity decay prediction value based on the health status assessment value, the battery characteristic parameter set, and a pre-trained temporal convolutional network model; The pre-trained state evaluation model set includes a pre-trained lightweight long short-term memory network model, a pre-trained extreme learning machine and a pre-trained random forest model; The second determining module is specifically configured to: Determining the operating condition type to which the current operating condition information belongs, wherein the operating condition types include dynamic operating conditions, static operating conditions, and resource-constrained scenarios; Selecting a target state assessment model that matches the operating condition type from the pre-trained state assessment model set, including: using the pre-trained lightweight long short-term memory network model for the dynamic operating condition, using the pre-trained random forest model for the static operating condition, and using the pre-trained extreme learning machine for the resource-constrained scenario; Inputting the battery characteristic parameter set into the target state assessment model to obtain a model output result as a health state assessment value of the battery to be tested; The third determining module is specifically configured to: Determining a time series input matrix according to the health status assessment value and the dynamic characteristic parameters in the battery characteristic parameter set; Inputting the time series input matrix into a pre-trained temporal convolutional network model to determine a predicted value of capacity decay to be corrected under a set number of future cycles; Determine the historical error feedback value based on historical prediction results and historical actual results; A capacity decay prediction value is determined according to the historical error feedback value and the capacity decay prediction value to be corrected.

7. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the battery capacity attenuation prediction method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the battery capacity attenuation prediction method according to any one of claims 1 to 5 when executed.

Images