Method and system for monitoring health state and estimating service life of battery of electric vehicle

Abstract

The invention provides an electric vehicle battery health state monitoring and service life estimation method and system, and the method comprises the steps: S1, obtaining a real-time data flow of a battery during the operation of a vehicle, extracting the original records of charging and discharging depth, temperature change, internal resistance parameters and capacity data from the real-time data flow, a dynamic monitoring sample is calculated through voltage, current and temperature values collected by a sensor; s3, calculating a capacity attenuation rule through the time sequence feature set, analyzing the decrease amplitude of the capacity after each charge-discharge cycle by adopting an exponential attenuation model, obtaining a long-term operation trend from historical data, and obtaining a dynamic curve of capacity attenuation; and S8, if the confidence interval of the final estimation probability distribution is greater than a preset threshold value, recollecting data with higher frequency for the dynamic monitoring sample, and performing iterative optimization on the life estimation model through the updated sample to obtain an adjusted life detection result.

Description

Technical Field

[0001] This invention relates to the field of new energy vehicle technology, and in particular to a method and system for monitoring the health status and estimating the lifespan of electric vehicle batteries. Background Technology

[0002] Battery life testing for new energy vehicles is a core area of ​​the new energy vehicle industry's development, directly impacting vehicle performance, safety, and user experience; its importance is self-evident. With the increasing prevalence of new energy vehicles, battery life not only affects driving range and operating costs but also has a profound impact on resource recycling and environmental protection. Therefore, accurate testing and evaluation of battery life has become a crucial aspect of the industry's development.

[0003] Currently, although battery life testing technology has made some progress, existing methods still have significant limitations. Many traditional solutions rely on a single indicator or static testing, which makes it difficult to comprehensively reflect the real state of the battery under complex usage scenarios, and they do not make full use of historical data, resulting in a large deviation between the assessment results and the actual lifespan.

[0004] Against this backdrop, the core challenges facing this field have become increasingly apparent. Among these, the dynamic monitoring of battery health status, the comprehensiveness of evaluation indicators, and the effective integration of historical data have become pressing technical factors that need to be addressed. Due to the lack of dynamic monitoring methods, it is difficult to accurately capture the degradation patterns of batteries under different operating conditions; the limited availability of evaluation indicators makes it impossible to analyze key parameters such as internal resistance and capacity in a coordinated manner; and insufficient integration of historical data restricts the assessment of the long-term impact of factors such as depth of charge / discharge and temperature changes. These unresolved technical factors collectively lead to low accuracy in lifespan detection, which in turn affects the reliability of subsequent predictions, forming a unique technical challenge.

[0005] Therefore, how to improve the accuracy and reliability of battery life testing through multi-indicator collaborative analysis and in-depth mining of historical data has become a key issue that this research urgently needs to address. Only by solving this problem can we provide a solid data foundation for the life prediction of new energy vehicle batteries and promote the industry towards a more efficient and sustainable development. Summary of the Invention

[0006] To address the technical problems raised in the background art, the first aspect of the present invention provides... A method for monitoring the health status and predicting the lifespan of electric vehicle batteries, the method comprising: S1 acquires real-time data streams of the battery during vehicle operation, extracts raw records of charge / discharge depth, temperature change, internal resistance parameters, and capacity data from them, and calculates dynamic monitoring samples using voltage, current, and temperature values ​​collected by sensors. S2. Preprocess the dynamic monitoring samples, use the sliding window method to segment the time series data, extract the short-term fluctuation features of charge and discharge depth and temperature changes from the segmentation results, and obtain the time series feature set of battery operating status. S3 calculates the capacity decay law through time series feature set, uses exponential decay model to analyze the capacity decrease after each charge-discharge cycle, obtains long-term operating trend from historical data, and obtains dynamic curve of capacity decay. S4. After obtaining the dynamic curve of capacity decay, combined with the analysis of internal resistance parameters, the rate of change of internal resistance over time is calculated by electrochemical impedance spectroscopy. Multi-index synergistic features are extracted from the dynamic curve and the rate of change of internal resistance to obtain a preliminary quantitative characterization of the health status. S5 integrates historical data for the preliminary quantitative characterization of health status, extracts charging and discharging depth and temperature change records of the vehicle's entire life cycle from the database, and sorts the feature importance of multidimensional data through the random forest algorithm to obtain the weight distribution of key influencing factors. S6. Based on the weight distribution of key influencing factors, data is deeply mined, and a long short-term memory network is used to model the time dependence of historical data. The degradation trend of the battery under different operating conditions is predicted from the modeling results, and an intermediate parameter set for life prediction is obtained. S7 uses a multi-index collaborative analysis based on the intermediate parameter set of life prediction, and employs a weighted average method to fuse the predicted values ​​of internal resistance parameters, capacity decay law, and the influence of temperature change, and calculates the final predicted probability distribution of battery life from the fusion results. S8. If the confidence interval of the final estimated probability distribution is greater than the preset confidence interval threshold, then data with a higher frequency will be collected again for the dynamic monitoring samples. The life prediction model will be iteratively optimized using the updated samples to obtain the adjusted life detection results. S9 generates a dynamic report on battery health status based on the adjusted life test results. It uses visualization technology to map capacity decay patterns, internal resistance changes, and life prediction reliability onto a time axis, resulting in a monitoring archive that reflects the entire battery life cycle.

[0007] Optionally, in step S1, the real-time data stream of the battery during vehicle operation is acquired, and the original records of charge / discharge depth, temperature change, internal resistance parameters, and capacity data are extracted from it. Dynamic monitoring samples are calculated using voltage, current, and temperature values ​​collected by sensors, including: Step S11: Collect the original values ​​of battery voltage, current and temperature during vehicle operation using sensors; Step S12: Calculate the original data of charge and discharge depth based on the time integral of the original voltage and current values, and generate a three-column time series data table by combining the original temperature value. Step S13: For the original data of charge / discharge depth and original data of capacity, the least squares method is used to fit the linear regression equation, the slope is used as the capacity decay trend, the intercept is used as the initial capacity, and the dynamic sample data table is output. Step S14: Extract instantaneous voltage and current values ​​from the dynamic sample data table, and use the DC internal resistance method to calculate the ratio of voltage difference to current difference in a 10-second window to obtain the internal resistance parameter sequence. Step S15: Calculate the Pearson correlation coefficient between the instantaneous temperature values ​​in the dynamic sample data table and the internal resistance parameter sequence. Step S16: If the absolute value of the correlation coefficient is greater than the correlation coefficient threshold, the internal resistance parameter sequence is corrected using the linear compensation formula to generate the corrected internal resistance parameter table. Step S17: When the number of data points set in the modified internal resistance parameter table exceeds the internal resistance parameter threshold, extract the capacity decay rate, internal resistance change rate and temperature slope from the dynamic sample data table as feature vectors, input them into the pre-trained support vector machine classification model, and output the health status label. Step S18: Based on the capacity decay curve corresponding to the health status label, calculate the percentage deviation between the actual capacity and the nominal capacity at the current charge / discharge depth, and update the capacity data column in the dynamic sample data table. Step S19: Perform a sliding window variance calculation on the updated capacity data column and the original temperature value. If the variance value is less than the variance threshold, it is determined to be in a stable state.

[0008] Optionally, in step S16, if the absolute value of the correlation coefficient is greater than the correlation coefficient threshold, the internal resistance parameter sequence is corrected using a linear compensation formula to generate a corrected internal resistance parameter table, wherein the correlation coefficient threshold is set to 0.7.

[0009] Optionally, in step S17, when the number of data points set in the modified internal resistance parameter table exceeds the internal resistance parameter threshold, the capacity decay rate, internal resistance change rate, and temperature slope are extracted from the dynamic sample data table as feature vectors, input into the pre-trained support vector machine classification model, and the set number of data points in the output health status label are 5 consecutive data points.

[0010] Optionally, the threshold value of the internal resistance parameter is set to 0.1 ohms.

[0011] Optionally, in step S19, the updated capacity data column and the original temperature value are subjected to a sliding window variance calculation. When the variance value is less than the variance threshold, it is determined to be in a stable state. The variance threshold is set to 0.01.

[0012] Optionally, in step S2, the dynamic monitoring samples are preprocessed by segmenting the time series data using a sliding window method. Short-term fluctuation features of charge / discharge depth and temperature changes are extracted from the segmentation results to obtain a time series feature set of the battery operating state, including: Step S21: The time series data of the dynamic monitoring sample is segmented using a sliding window with a fixed length and a fixed step size to obtain a segmented data sequence; Step S22: Extract the difference between the maximum and minimum voltage values ​​within each window from the segmented data sequence as the charge / discharge depth feature to form a charge / discharge depth sequence; Step S23: Extract the standard deviation of the temperature values ​​within each window from the segmented data sequence as the temperature change feature to form a temperature change sequence; Step S24: Align the charge / discharge depth sequence and the temperature change sequence by timestamp and then stitch them together horizontally to generate a short-term fluctuation feature set; Step S25: If the number of columns in the short-term fluctuation feature set exceeds the column number threshold, the feature set is first standardized by Z-score, and then principal component analysis is used to retain the components with a cumulative contribution rate of more than 95% to obtain the dimensionality-reduced feature set. Step S26: Based on the dimensionality reduction feature set, the K-means clustering algorithm is used to set the number of clusters to 3, and Euclidean distance is used for iterative calculation to obtain the state classification label; Step S27: Arrange the state classification labels in the original time order, map them to the timestamps of the segmented data sequence to generate a state encoding sequence, which serves as the final time series feature set.

[0013] Optionally, in step S21, a sliding window with a fixed length and a fixed step size is used to segment the time series data of the dynamic monitoring sample to obtain a segmented data sequence in which the fixed length is 30 sampling points and the fixed step size is 5 sampling points.

[0014] Optionally, in step S25, if the number of columns in the short-term fluctuation feature set exceeds the column number threshold, the feature set is first standardized by Z-score, and then principal component analysis is used to retain the components with a cumulative contribution rate of more than 95% to obtain the dimensionality-reduced feature set, wherein the column number threshold is set to 10.

[0015] A second aspect of the present invention provides a system for monitoring the health status and predicting the lifespan of an electric vehicle battery, which uses the method described above to monitor the health status and predict the lifespan of the electric vehicle battery. The system includes: The data acquisition module is used to acquire real-time data streams of the battery during vehicle operation, extract raw records of charge / discharge depth, temperature change, internal resistance parameters and capacity data from it, and calculate dynamic monitoring samples by using voltage, current and temperature values ​​collected by sensors. The preprocessing module is used to preprocess the dynamic monitoring samples. It uses a sliding window method to segment the time series data and extracts short-term fluctuation features of charge and discharge depth and temperature changes from the segmentation results to obtain the time series feature set of battery operating status. The capacity decay analysis module is used to calculate the capacity decay pattern through time series feature sets, analyze the capacity decrease after each charge-discharge cycle using an exponential decay model, obtain long-term operating trends from historical data, and obtain dynamic curves of capacity decay. The health status quantification module is used to obtain the dynamic curve of capacity decay, and then, in combination with internal resistance parameter analysis, calculate the rate of change of internal resistance over time using electrochemical impedance spectroscopy. It extracts multi-index synergistic features from the dynamic curve and the rate of change of internal resistance to obtain a preliminary quantitative characterization of the health status. The historical data integration module is used to integrate historical data for the preliminary quantitative characterization of health status. It extracts charging and discharging depth and temperature change records throughout the vehicle's life cycle from the database, and sorts the multidimensional data by feature importance using the random forest algorithm to obtain the weight distribution of key influencing factors. The lifespan prediction module is used to perform in-depth data mining based on the weight distribution of key influencing factors, and to model the time dependence of historical data using a long short-term memory network. From the modeling results, it predicts the battery degradation trend under different operating conditions and obtains an intermediate parameter set for lifespan prediction. The multi-index collaborative analysis module is used to perform multi-index collaborative analysis through the intermediate parameter set of life prediction. It uses a weighted average method to fuse the predicted values ​​of internal resistance parameters, capacity decay law and temperature change effects, and calculates the final predicted probability distribution of battery life from the fusion results. The model optimization module is used to re-collect more frequent data from the dynamic monitoring samples if the confidence interval of the final estimated probability distribution is greater than the preset confidence interval threshold. The updated samples are then used to iteratively optimize the life prediction model to obtain the adjusted life detection results. The report generation module is used to generate dynamic reports on battery health status based on the adjusted life test results. It uses visualization technology to map capacity decay patterns, internal resistance changes, and life prediction reliability onto a time axis, resulting in a monitoring archive that reflects the entire battery life cycle.

[0016] The technical solutions provided by the embodiments of the present invention have the following beneficial effects: This invention provides a method and system for monitoring the health status and predicting the lifespan of electric vehicle batteries. By acquiring real-time data streams of the battery during vehicle operation, key parameters such as charge / discharge depth and temperature changes are extracted. The time series data is segmented and features are extracted using the sliding window method. The capacity decay law is analyzed by combining the exponential decay model. The internal resistance change rate is calculated using the electrochemical impedance spectroscopy method. The feature importance of multidimensional data is ranked using the random forest algorithm. The historical data is modeled using a long short-term memory network to predict the decay trend under different operating conditions. Finally, the weighted average method is used to fuse multiple indicators to obtain the predicted probability distribution of battery life.

[0017] This invention can also dynamically adjust the sampling frequency based on the confidence interval, iteratively optimize the prediction model, and generate a dynamic health status report reflecting the entire life cycle of the battery, providing comprehensive and reliable technical support for electric vehicle battery management. Attached Figure Description

[0018] Figure 1 This is a flowchart of the electric vehicle battery health status monitoring and lifespan prediction method of the present invention.

[0019] Figure 2 This is a schematic diagram of the electric vehicle battery health status monitoring and life prediction system of the present invention. Detailed Implementation

[0020] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0021] like Figure 1 As shown, in a first aspect, the present invention provides a method for monitoring the health status and predicting the lifespan of an electric vehicle battery, the method comprising: S1 acquires real-time data streams of the battery during vehicle operation, extracts raw records of charge / discharge depth, temperature changes, internal resistance parameters, and capacity data from these streams, and calculates dynamic monitoring samples using voltage, current, and temperature values ​​collected by sensors.

[0022] Optionally, this step also includes: Step S11: Collect the raw values ​​of battery voltage, current, and temperature during vehicle operation using sensors. Step S12: Calculate the raw charge / discharge depth data based on the time integrals of the raw voltage and current values, and generate a three-column time-series data table by combining it with the raw temperature value. Step S13: Fit a linear regression equation to the raw charge / discharge depth data and raw capacity data using the least squares method. The slope represents the capacity decay trend, and the intercept represents the initial capacity. Output a dynamic sample data table. Step S14: Extract instantaneous voltage and current values ​​from the dynamic sample data table. Calculate the ratio of voltage difference to current difference using the DC internal resistance method, with a 10-second window, to obtain the internal resistance parameter sequence. Step S15: Calculate the Pearson correlation coefficient between the instantaneous temperature value in the dynamic sample data table and the internal resistance parameter sequence. Step S16: If the absolute value of the correlation coefficient is greater than the correlation coefficient threshold of 0.7, correct the internal resistance parameter sequence using a linear compensation formula, generating a corrected internal resistance parameter table. Step S17: When the number of data points in the corrected internal resistance parameter table exceeds the internal resistance parameter threshold of 0.1 ohms, extract the capacity decay rate, internal resistance change rate, and temperature slope as feature vectors from the dynamic sample data table, input them into the pre-trained support vector machine classification model, and output a health status label. Step S18: Based on the capacity decay curve corresponding to the health status label, calculate the percentage deviation between the actual capacity and the nominal capacity at the current charge / discharge depth, and update the capacity data column in the dynamic sample data table. Step S19: Perform a sliding window variance calculation on the updated capacity data column and the original temperature value. A stable state is determined when the variance value is less than the variance threshold of 0.01.

[0023] Preferably, the set number of data points is 5 consecutive data points.

[0024] Specifically, collecting raw values ​​of battery voltage, current, and temperature during vehicle operation via sensors is the foundation of battery management.

[0025] For example, when an electric vehicle is running, sensors record voltage (e.g., 13.5 volts), current (e.g., 20 amperes), and temperature (e.g., 25 degrees Celsius) per second, forming time-series data. This data reflects the real-time state of the battery, providing a basis for subsequent analysis.

[0026] It should be noted that the depth of charge and discharge is calculated based on the time integration of voltage and current. The principle is to convert the current accumulation over time into a change in electrical quantity, and then divide it by the nominal capacity.

[0027] For example, a battery with a nominal capacity of 100 Ah has an average current of 20 Amperes over 10 seconds and a depth of charge / discharge of approximately 3.33%. Combined with temperature, three data columns are generated to intuitively display the battery's dynamics.

[0028] In one possible implementation, a linear regression is fitted using the least squares method to the raw data on charge / discharge depth and capacity. Assuming that the charge / discharge depth decreases from 0% to 80% over a period of time, and the capacity decays from 100 AH to 95 AH, the fitted result shows a slope of -0.0625 representing the capacity decay trend, with an intercept of 100 AH representing the initial capacity. This method quantifies the decay pattern, making the dynamic sample data table more predictive.

[0029] Specifically, instantaneous voltage values, such as 13.5 volts and 13.3 volts, and instantaneous current values, such as 20 amps and 22 amps, are extracted from the table, and the internal resistance is calculated using a 10-second window. Dividing the voltage difference of 0.2 volts by the current difference of 2 amps yields an internal resistance of approximately 0.1 ohms, generating a sequence of internal resistance parameters. This DC internal resistance method is simple, efficient, and suitable for real-time monitoring.

[0030] For example, the instantaneous temperature value, such as 25 degrees Celsius to 30 degrees Celsius, is compared with the internal resistance sequence to calculate the Pearson correlation coefficient. If the absolute value of the correlation coefficient reaches 0.8, exceeding the preset correlation coefficient threshold of 0.7, it indicates that temperature significantly affects the internal resistance.

[0031] Preferably, a linear compensation formula is used, such as increasing the internal resistance by 0.005 ohms for every 1 degree Celsius increase in temperature, resulting in a more accurate internal resistance sequence after correction.

[0032] After the internal resistance parameter table is generated, if five consecutive data points (e.g., 0.12, 0.13, 0.14, 0.15, 0.16 ohms) exceed the internal resistance parameter threshold of 0.1 ohms, feature vectors are extracted from the dynamic table. These vectors include parameters such as capacity decay rate (0.5% / day), internal resistance change rate (0.01 ohms / day), and temperature slope (0.2 degrees Celsius / day). These vectors are then input into a support vector machine model, which outputs a "sub-healthy" label. This method improves the accuracy of state assessment.

[0033] In one embodiment, based on the capacity decay curve corresponding to the health status tag, assuming the current charge / discharge depth is 50%, the actual capacity is 96 Ah, and the nominal capacity is 100 Ah, then the percentage deviation between the actual capacity and the nominal capacity is 4%. After updating the capacity column of the dynamic table, the variance is calculated using a sliding window with the original temperature value.

[0034] For example, if the capacity is 96, 96.1, or 96.2 amp-hours, and the temperature is 25, 25.1, or 25.2 degrees Celsius, and the variance is less than the variance threshold of 0.01, it is considered to be in a stable state.

[0035] Understandably, this stability assessment helps optimize battery management strategies and extend battery life.

[0036] For example, under stable conditions, the charging strategy can be adjusted to reduce the risk of overcharging and significantly improve safety and economy.

[0037] S2. Preprocessing is performed on the dynamic monitoring samples. The time series data is segmented using the sliding window method. Short-term fluctuation features of charge and discharge depth and temperature changes are extracted from the segmentation results to obtain the time series feature set of battery operating status.

[0038] Optionally, this step also includes: Step S21: The time series data of the dynamic monitoring samples is segmented using a sliding window with a fixed length of 30 sampling points and a fixed step size of 5 sampling points to obtain a segmented data sequence. Step S22: The difference between the maximum and minimum voltage values ​​within each window is extracted from the segmented data sequence as a charge / discharge depth feature, forming a charge / discharge depth sequence. Step S23: The standard deviation of the temperature values ​​within each window is extracted from the segmented data sequence as a temperature change feature, forming a temperature change sequence. Step S24: The charge / discharge depth sequence and the temperature change sequence are aligned by timestamps and then horizontally concatenated to generate a short-term fluctuation feature set. Step S25: If the number of columns in the short-term fluctuation feature set exceeds the threshold of 10, the feature set is first standardized using Z-score, and then principal component analysis is used to retain components with a cumulative contribution rate exceeding 95%, resulting in a dimensionality-reduced feature set. Step S26: Based on the dimensionality-reduced feature set, the K-means clustering algorithm is used with a cluster number of 3, and Euclidean distance is used for iterative calculation to obtain the state classification label. Step S27: Arrange the state classification labels in the original time order, map them to the timestamps of the segmented data sequence to generate a state encoding sequence, which serves as the final time series feature set.

[0039] Specifically, by using a sliding window with a fixed length of 30 sampling points and a fixed step size of 5 sampling points to segment the time series data, continuous battery operation data can be decomposed into multiple segments.

[0040] For example, assuming a sensor records voltage, current, and temperature values ​​once per second, a 600-second data segment would be divided into multiple 30-second windows, sliding every 5 seconds to generate an overlapping sequence of segments. This method can capture short-term trends in data.

[0041] For example, if a window records 30 voltage values, ranging from 13.2 volts to 13.8 volts, it reflects the dynamic fluctuations of the battery during this period. Extracting the difference between the maximum and minimum voltage values ​​from the segmented data sequence as a charge / discharge depth feature is to quantify the range of charge variation within each window.

[0042] Specifically, a voltage drop of 0.6 volts from 13.8 volts to 13.2 volts within a window reflects the battery's charge and discharge amplitude during that period, forming a depth sequence. This method visually demonstrates the battery's operating intensity.

[0043] In one possible implementation, if the voltage difference in another window is only 0.3 volts, it indicates that the battery is operating relatively smoothly during that period, thus making the depth sequence more discriminative. Extracting the standard deviation of the temperature values ​​within each window as a temperature variation feature focuses on the degree of temperature fluctuation.

[0044] For example, within one window, the temperature fluctuates between 24 and 26 degrees Celsius, with a standard deviation of 0.5 degrees Celsius, while in another window, the temperature remains around 25 degrees Celsius, with a standard deviation close to 0. This series of temperature changes can reflect the non-uniformity of heat in the environment or inside the battery.

[0045] It should be noted that the magnitude of the standard deviation is directly related to the stability of battery operation and is helpful for subsequent classification.

[0046] Aligning and concatenating the charge / discharge depth sequence and temperature change sequence by timestamp to generate a short-term fluctuation feature set is to integrate multi-dimensional information. For example, at a timestamp, the charge / discharge depth value is 0.6 volts and the temperature standard deviation is 0.5 degrees Celsius. After concatenation, a two-dimensional vector is formed.

[0047] Preferably, if a certain data segment continuously displays high charge / discharge depths and high temperature fluctuations, the short-term feature set can highlight the abnormal patterns. This multi-dimensional perspective enhances the expressive power of the features. If the number of columns in the feature set exceeds the threshold of 10, Z-score standardization is performed first, followed by principal component analysis for dimensionality reduction. The aim is to reduce redundancy and retain key information.

[0048] In one embodiment, it is assumed that the feature set contains 12 columns including depth of charge / discharge, temperature standard deviation, and other parameters. After standardization, the first few components are retained by principal component analysis, so that the cumulative contribution rate reaches 95%.

[0049] For example, after dimensionality reduction, only 3 columns may remain, yet they can still represent most of the fluctuation information. This approach reduces computational complexity. By using the K-means clustering algorithm with a cluster number of 3 based on the dimensionality-reduced feature set, the battery state can be divided into different categories.

[0050] For example, clustering may yield three categories of labels: "stable", "fluctuating", and "abnormal".

[0051] Specifically, a battery with a low depth of charge / discharge and stable temperature is categorized as "stable," while a high depth of charge / discharge and large temperature fluctuations may be considered "abnormal." This classification intuitively reflects the battery's operating status. Mapping the status classification labels to timestamps to generate a status encoding sequence further integrates time-dimensional information.

[0052] It is understandable that if the label changes from "stable" to "abnormal" over a certain period of time, the encoding sequence can reflect the trend of state change.

[0053] For example, the appearance of an "abnormal" label after a series of "stable" labels may indicate an impending decline in battery performance. This sequential characteristic provides a basis for long-term monitoring.

[0054] S3 calculates the capacity decay pattern through time series feature sets, uses an exponential decay model to analyze the capacity decrease after each charge-discharge cycle, obtains the long-term operating trend from historical data, and obtains the dynamic curve of capacity decay.

[0055] Optionally, this step also includes: Step S31: Collect charge-discharge cycle records using time-series data to obtain a capacity change dataset. Step S32: Calculate the capacity difference for each charge-discharge cycle from the capacity change dataset to obtain an amplitude sequence. Step S33: Fit the amplitude sequence to an exponential decay model using the least squares method to obtain decay parameters. Step S34: Calculate the capacity decay value under long-term operation using linear extrapolation based on the decay parameters to obtain a predicted decay sequence. Step S35: If the predicted decay sequence exceeds a preset decay sequence threshold, adjust the analysis interval of the time series to obtain corrected trend data. Step S36: Use Matplotlib to plot a dynamic curve using the corrected trend data to obtain a visual representation of capacity decay. Step S37: Analyze the relationship between the number of cycles and decay based on the dynamic curve to obtain operational optimization suggestions.

[0056] Specifically, by collecting charge-discharge cycle records through time-series data, a capacity change dataset is obtained, which can be understood as obtaining continuous capacity information from battery operation.

[0057] For example, suppose a battery records a capacity of 1000 mAh after each charge, which drops to 950 mAh after discharge. After multiple cycles, this forms a capacity sequence that changes over time. This data acquisition method relies on sensors to record the start and end states of each cycle, intuitively reflecting the battery's capacity change trend. Calculating the capacity difference for each charge-discharge cycle from the capacity change dataset yields an amplitude sequence, which quantifies the capacity loss in each cycle.

[0058] Specifically, if the capacity decreases from 1000 mA / h to 950 mA / h in the first cycle, the difference is 50 mA / h; and in the second cycle, it decreases from 995 mA / h to 940 mA / h, the difference is 55 mA / h. Through multiple cycles, a difference sequence is generated, such as 50, 55, and 48 mA / h. This sequence visually reflects the capacity decay rate in each cycle. Using the least squares method to fit an exponential decay model to the magnitude sequence and obtain the decay parameters is a common trend analysis method.

[0059] For example, assuming the amplitude sequence is 50, 45, and 40 mA, gradually decreasing, the least squares method can determine an exponential decay curve through fitting, outputting a decay rate parameter. This parameter quantifies the rate at which capacity loss decreases with increasing cycle number, providing a basis for subsequent predictions. Using the decay parameter, linear extrapolation is used to calculate the capacity decay value under long-term operation, obtaining a predicted decay sequence; this is a trend-based extrapolation method.

[0060] In one possible implementation, if the decay parameter indicates a 5% capacity reduction every 100 cycles, and the current capacity is 900 mA, extrapolation suggests it might drop to 600 mA after 1000 cycles. Predicted sequences, such as 900, 850, and 800 mA, reflect the long-term capacity trend. If the predicted decay sequence exceeds a preset decay sequence threshold, the analysis interval of the time series is adjusted to obtain corrected trend data, thus improving prediction accuracy.

[0061] For example, if the decay sequence threshold is set to 700 mA, and the predicted value falls to 600 mA, exceeding the range, the data from the most recent 500 cycles can be re-analyzed. The corrected trend may show a more gradual decay. This adjustment better reflects the actual operating conditions. Using Matplotlib to plot a dynamic curve based on the corrected trend data, a visual representation of capacity decay can be obtained, facilitating an intuitive understanding of the decay process.

[0062] Specifically, the horizontal axis represents the number of cycles, and the vertical axis represents the capacity value, plotting a curve that gradually decreases from 1000 mA to 800 mA.

[0063] Preferably, the curve can also highlight key points, such as the number of cycles at which capacity decreases by 20%, helping to quickly identify degradation characteristics. Analyzing the relationship between the number of cycles and degradation based on the dynamic curve to obtain operational optimization suggestions is a process of extracting practical information from data.

[0064] In one embodiment, if the curve shows a 30% capacity decrease after 500 cycles, with slower initial degradation, it can be suggested to adjust the charging strategy at 400 cycles, for example, by lowering the upper charging voltage limit. This analysis can provide direct evidence for battery management.

[0065] Understandably, if the curve drops sharply in a certain segment, it may indicate that the operating conditions have deteriorated, and it is recommended to check the temperature or current settings.

[0066] It should be noted that the processing at each stage revolves around capacity changes, forming a complete time series analysis chain from data collection to prediction and optimization.

[0067] For example, the initial capacity difference calculation provides data support for subsequent fitting, while the predicted sequences and visualizations further transform the analysis results into actionable recommendations. This progressive approach ensures the rigor and practicality of the analysis.

[0068] S4. After obtaining the dynamic curve of capacity decay, combined with the analysis of internal resistance parameters, the rate of change of internal resistance over time is calculated using the electrochemical impedance spectroscopy method. Multi-index synergistic features are extracted from the dynamic curve and the rate of change of internal resistance to obtain a preliminary quantitative characterization of the health status.

[0069] Optionally, this step also includes: Step S41: Acquire battery operating data and generate a dynamic curve by recording the capacity decay over time. Step S42: Smooth the dynamic curve and extract feature points. Step S43: Calculate the internal resistance using Ohm's law to determine the trend of internal resistance change over time. Step S44: Process the internal resistance data using Fourier transform to obtain the rate of change reflected by the impedance spectrum curve. Step S45: Based on the rate of change and the dynamic curve, use principal component analysis to obtain a multi-index data set to determine the combination of synergistic features. Step S47: If the synergistic features exceed a preset synergistic feature threshold, determine the health status category using K-means clustering; otherwise, calculate the health status quantification value using standard deviation. Step S48: Generate preliminary health status characterization results using the LIBSVM tool based on the health status quantification value and synergistic features. Step S49: Determine the mapping relationship between capacity decay and health status by comparing the characterization results with pre-established historical data.

[0070] Specifically, acquiring battery operating data and generating dynamic curves by recording the change in capacity decay over time can be understood as collecting continuous data from the actual operation of the battery.

[0071] For example, a battery with an initial capacity of 1200 mAh, after multiple charge-discharge cycles, records the capacity value at each time point, such as 1150 mAh on day 10 and 1100 mAh on day 20, forming a curve showing a decrease over time. This method relies on real-time monitoring by sensors to intuitively reflect changes in battery performance. Smoothing the dynamic curve and extracting feature points is a method to optimize data presentation.

[0072] For example, if the curve fluctuates significantly due to noise, a moving average method can be used to smooth it, resulting in a clearer trend. Key points can then be identified, such as the moment when the capacity decreases by 10%. This smoothing highlights the main trend, facilitating subsequent analysis. Calculating internal resistance using Ohm's law and determining its trend over time is an important method for analyzing battery aging.

[0073] Specifically, the internal resistance is calculated by measuring the battery's voltage at a specific current. For example, if the initial internal resistance is 0.05 ohms, it may rise to 0.07 ohms after 100 cycles, reflecting a gradual increase in internal resistance over time. This change indicates battery performance degradation. Processing the internal resistance data using Fourier transform yields the rate of change reflected in the impedance spectrum curve, a frequency domain analysis method.

[0074] In one possible implementation, the internal resistance time series is converted into a frequency domain signal, and the rate of impedance change with frequency is observed. A significant increase in impedance at low frequencies indicates a slowdown in internal chemical reactions. This analysis can reveal deep-seated degradation mechanisms. Based on the rate of change and dynamic curves, principal component analysis is used to obtain a multi-indicator data set to determine the combination of synergistic features; this is a process of integrating multi-dimensional information.

[0075] For example, indicators such as capacity decay rate and internal resistance change rate are input into principal component analysis to extract the main influencing factors and form comprehensive features. If the collaborative features exceed the preset collaborative feature threshold, the health status category is determined by K-means clustering; otherwise, the standard deviation is used to calculate the quantitative value.

[0076] Preferably, if the eigenvalue exceeds the upper limit of the collaborative feature threshold, clustering can be used to classify it into three categories: "healthy," "sub-healthy," and "deteriorating." If it is normal, the standard deviation of the eigenvalue is calculated, such as 0.02, to quantify the degree of fluctuation. This classification or quantification facilitates state assessment. Using the LIBSVM tool to generate preliminary health status representations based on the quantified health status values ​​and collaborative features is a machine learning application.

[0077] In one embodiment, after inputting feature data, LIBSVM outputs a health score, such as 90 points indicating a good health condition, and 70 points suggesting the need for attention. This representation is intuitive and easy to operate. The core step of the analysis is to determine the mapping relationship between capacity decay and health status by comparing the representation results with pre-established historical data.

[0078] For example, if historical data shows a 15% capacity decrease when the health score is 80, and the current score of 75 corresponds to an 18% capacity decrease, it can be inferred that the battery is nearing critical condition. This mapping provides a basis for management decisions.

[0079] Understandably, long-term comparisons can also optimize the setting of collaborative feature thresholds and improve prediction accuracy.

[0080] S5 integrates historical data for the preliminary quantitative characterization of health status, extracts charging and discharging depth and temperature change records throughout the vehicle's life cycle from the database, and sorts the multidimensional data by feature importance using the random forest algorithm to obtain the weight distribution of key influencing factors.

[0081] Optionally, this step also includes: Step S51: Extract vehicle charge / discharge depth and temperature change records from the database to construct a historical dataset containing lifecycle information. Step S52: Preprocess the historical dataset, including missing value imputation and normalization. Step S53: Use a random forest algorithm to rank the preprocessed multidimensional data by feature importance, obtaining the weight distribution of each feature. Step S54: Determine key factors based on the weight distribution and generate a quantitative representation of health status using a weighted average method. Step S55: If the quantitative representation result exceeds a preset threshold, use a linear regression algorithm to predict the trend of key factors and obtain the change trend. Step S56: Obtain the boundary values ​​of charge / discharge depth and temperature changes affecting health status based on the change trend, and use an isolated forest algorithm to identify potential outliers. Step S57: Determine a dynamic adjustment scheme for key factors through comparative analysis of outliers and historical data, obtaining an optimized weight distribution. Step S58: Update the quantitative representation result using the optimized weight distribution to generate the final health status assessment data.

[0082] Specifically, extracting vehicle charge / discharge depth and temperature change records from the database to construct a historical data set containing lifecycle information forms the basis for health status analysis.

[0083] For example, data can be obtained from an electric vehicle battery management system to record the depth of charge and discharge during each charge, such as 20% and 80%, as well as the operating temperature, such as 25°C and 40°C, forming a data table with timestamps. This method can comprehensively reflect the battery's usage history.

[0084] Understandably, the more comprehensive the data, the more accurate the subsequent analysis. Preprocessing historical datasets, including imputing missing values ​​and normalization, can improve data quality.

[0085] Specifically, if a charge / discharge depth record is missing, it can be filled using the average of the preceding and following data, such as 50%; temperature data is filled in using linear interpolation. Normalization maps depth and temperature to the 0-1 range, for example, 80% charge / discharge depth is converted to 0.8, and 40°C is converted to 0.67, facilitating algorithm processing. This preprocessing ensures data consistency. The random forest algorithm is used to rank the feature importance of the preprocessed multidimensional data, obtaining the weight distribution of each feature, which intuitively reflects the influencing factors.

[0086] For example, the analysis results might show a weighting of depth of charge / discharge of 0.6, temperature change of 0.3, and cycle count of 0.1. This ranking highlights key variables.

[0087] It should be noted that random forests reduce bias and make the results more reliable by using multiple decision tree votes. Determining key factors based on weight distribution and using a weighted average method to generate a quantitative representation of health status is a concise and comprehensive approach.

[0088] In one possible implementation, if a depth weight of 0.6 corresponds to a value of 0.8, and a temperature weight of 0.3 corresponds to a value of 0.67, the weighted average yields 0.68 as the representation value. This method facilitates rapid state assessment. If the quantification result exceeds a preset quantification threshold, such as 0.7, a linear regression algorithm is used to predict the trend of key factors and obtain the changing trend.

[0089] For example, the predicted depth of charge / discharge (DDC) increases from 0.8 to 0.9, and the temperature increases from 0.67 to 0.75, reflecting the future degradation direction. This prediction provides a basis for early intervention. Based on the changing trends, the boundary values ​​of DDC and temperature changes affecting health status are obtained, and the isolated forest algorithm is used to identify potential anomalies for further refined analysis.

[0090] Preferably, a charge / discharge depth exceeding 0.85 or a temperature above 0.8 is marked as an anomaly, indicating potential overcharging or overheating. The Isolation Forest algorithm efficiently identifies risks by isolating anomalies. Comparative analysis of anomalies with historical data determines dynamic adjustment schemes for key factors, resulting in an optimized weight distribution, which is crucial for improving the assessment.

[0091] In one embodiment, if anomalies are concentrated at high temperatures, the temperature weight is adjusted from 0.3 to 0.4, and the depth of charge / discharge weight is reduced to 0.5. This adjustment better reflects actual operating conditions. The optimized weight distribution is used to update the quantitative characterization results, generating the final health status assessment data to ensure accuracy.

[0092] For example, the representation value under the new weights is updated from 0.68 to 0.65, reflecting a more realistic state. This update improves the adaptability of the evaluation.

[0093] S6. Based on the weight distribution of key influencing factors, data is deeply mined, and a long short-term memory network is used to model the time dependence of historical data. The degradation trend of the battery under different operating conditions is predicted from the modeling results, and an intermediate parameter set for life prediction is obtained.

[0094] Optionally, this step also includes: Step S61: Extract key factors from battery operation data using principal component analysis, determine weight distribution based on factor contribution, and obtain the core features affecting battery degradation to obtain a feature set. Step S62: Extract time dependencies from the feature set and historical operation data, and model the time series data using a Long Short-Term Memory (LSTM) network to obtain a time series model. Step S63: Analyze operation data at different charge / discharge rates using the time series model to obtain degradation trend prediction results. Step S64: Based on the degradation trend prediction results, if the degradation rate exceeds a preset degradation rate threshold of 0.1% per hour, adjust the charge / discharge rate parameters to obtain an optimized degradation curve. Step S65: Based on the optimized degradation curve and data at different charge / discharge rates, obtain an intermediate parameter set for lifespan estimation, including battery capacity, internal resistance, and cycle count. Step S66: Use a random forest algorithm to predict and analyze the intermediate parameter set, setting the number of decision trees to 100 and the maximum depth to 10, to obtain the final lifespan estimate. Step S67: By comparing the final lifespan estimate with historical data, the least squares method is used to calculate the deviation correction coefficient of the prediction result, and the adjusted lifespan prediction value is obtained.

[0095] Specifically, key factors are extracted from battery operating data through principal component analysis.

[0096] Understandably, this method can reduce the dimensionality of complex multidimensional data, highlighting the features that have the greatest impact on battery degradation.

[0097] For example, data such as battery voltage, current, temperature, and charge / discharge cycles can be analyzed to identify temperature and depth of charge / discharge as the primary factors. In a specific implementation, assuming temperature contributes 40% and depth of charge / discharge 35%, with other factors like current contributing less, the weight distribution can be determined accordingly, forming a feature set containing core characteristics. This extraction method focuses on key variables. Extracting time dependencies from the feature set and historical operational data, and using Long Short-Term Memory (LSTM) networks to model time-series data, is an effective approach for handling dynamically changing data.

[0098] For example, over the past year, the battery's temperature gradually increased from 20°C to 35°C, and its depth of charge / discharge changed from 50% to 70%. These changes exhibit a regular pattern over time. Long Short-Term Memory (LSTM) networks generate time-series models reflecting degradation by remembering long-term trends and short-term fluctuations.

[0099] It should be noted that the model can capture the characteristic of accelerated degradation as temperature increases. By analyzing operating data at different charge / discharge rates using time series models, the degradation trend can be predicted.

[0100] Specifically, at a 1C rate, the battery capacity decreases by 0.05% per month, while at a 2C rate it increases to 0.12%.

[0101] In one possible implementation, the model predicts a capacity decline trend over the next three months, highlighting the risks of operating at high rates.

[0102] Optionally, the decay rate threshold is set to 0.1% per hour. If the decay rate exceeds the decay rate threshold of 0.1% per hour, the multiplier is adjusted.

[0103] For example, reducing the rate from 2C to 1.5C flattens the degradation curve and extends battery life. Based on the optimized degradation curve and rate data, an intermediate set of parameters for lifespan estimation, such as capacity, internal resistance, and cycle count, is obtained.

[0104] Preferably, the capacity decreases from 100% to 80%, the internal resistance increases from 50 milliohms to 70 milliohms, and the cycle life reaches 500 cycles. These parameters reflect the battery status.

[0105] In one embodiment, the remaining lifetime is inferred from this data. A random forest algorithm is used to predict the intermediate parameter set, with 100 decision trees and a maximum depth of 10, to generate the final lifetime estimate.

[0106] For example, the prediction shows a remaining lifespan of 2 years. By comparing the lifespan estimate with historical data and using the least squares method to calculate the deviation correction factor, the prediction accuracy can be improved.

[0107] For example, if the predicted lifespan is 2 years, while the historical actual lifespan is 2.2 years, the deviation coefficient can be adjusted to 1.1. The adjusted lifespan prediction is closer to reality. This method ensures the reliability of the results through data correction.

[0108] In one embodiment, starting from the core solution, if further analysis is needed, secondary factors such as environmental humidity can be added to further refine the feature set and enrich the prediction dimensions. This multi-faceted approach makes the analysis logic rigorous and more adaptable.

[0109] S7 performs multi-index collaborative analysis using the intermediate parameter set for life prediction, and uses a weighted average method to fuse the predicted values ​​of internal resistance parameters, capacity decay law and temperature change effects, and calculates the final predicted probability distribution of battery life from the fusion results.

[0110] Optionally, this step also includes: Step S71: Construct an initial dataset by collecting data on internal resistance parameters, capacity decay, and temperature changes. Step S72: Fuse the predicted values ​​of internal resistance parameters, capacity decay patterns, and the impact of temperature changes using a weighted average method. The weighting coefficients are determined based on the statistical characteristics of historical data to obtain a preliminary fusion result. Step S73: Extract features from the preliminary fusion result. These features include the rate of change of internal resistance parameters, the rate of capacity decay, and the trend of temperature changes. Step S74: Calculate the intermediate parameter of the probability distribution, which is the standard deviation of the features. Step S75: If the intermediate parameter exceeds a preset feature standard deviation threshold, re-fuse the impact of internal resistance parameters and temperature changes by adjusting the weighting coefficients. The method for adjusting the weighting coefficients is to perform a linear adjustment based on the difference between the intermediate parameter and the feature standard deviation threshold to obtain an optimized fusion result. Step S76: Based on the optimized fusion result, apply a Bayesian method to calculate the probability distribution of battery life. The specific implementation of the Bayesian method involves using a Markov chain Monte Carlo method for parameter estimation. Step S77: Obtain the statistical characteristics of the probability distribution, including the mean and variance of the probability distribution. Step S78: Determine the estimated battery life range using a support vector machine (SVM). The SVM is trained using a radial basis function as the kernel function. Step S79: For the determined estimated range, match similar patterns from historical data. The similarity pattern matching method uses Euclidean distance to calculate similarity and determine the final estimated battery life.

[0111] Specifically, an initial dataset is constructed by collecting data on internal resistance parameters, capacity decay, and temperature changes.

[0112] Understandably, this approach laid the foundation for subsequent analysis.

[0113] For example, the operating records of a certain battery show that the internal resistance gradually increased from the initial 50 milliohms to 60 milliohms, the capacity decreased from 100% to 95%, and the temperature fluctuated from 25°C to 30°C. These data reflect the dynamic changes in the battery's state.

[0114] For example, the initial dataset can be organized by time series, covering multiple operating cycles. A weighted average method is used to fuse the predicted values ​​of internal resistance parameters, capacity decay patterns, and the effects of temperature changes. The weighting coefficients are determined based on the statistical characteristics of historical data to obtain preliminary fusion results.

[0115] Specifically, assuming that historical data analysis shows that temperature accounts for 40% of the decay, internal resistance accounts for 35%, and capacity decay accounts for 25%, then the weights are allocated accordingly.

[0116] In one possible implementation, a battery's predicted temperature is 32°C, predicted internal resistance is 62 milliohms, and predicted capacity decay is 4%. These are fused to obtain a comprehensive index for subsequent analysis. Features are extracted from the initial fusion results, including the rate of change of internal resistance, the rate of capacity decay, and the trend of temperature change.

[0117] For example, the rate of change of internal resistance can be calculated by the difference in internal resistance between adjacent time points, the rate of capacity decay is reflected as the percentage decrease per month, and the temperature trend reflects the direction of heating or cooling.

[0118] Preferably, if a battery's internal resistance changes at a rate of 2 milliohms per month, its capacity decays at a rate of 0.5%, and its temperature shows a slow upward trend, these characteristics can intuitively describe the decay process. The intermediate parameter of the probability distribution, i.e., the standard deviation of the characteristics, is then calculated.

[0119] It should be noted that this step provides a basis for quantifying uncertainty.

[0120] In one embodiment, if the standard deviation of the rate of change of internal resistance is 0.5 milliohms and the standard deviation of the rate of capacity decay is 0.1%, these values ​​reflect the degree of data fluctuation. If the intermediate parameters exceed a preset characteristic standard deviation threshold, for example, if the standard deviation exceeds 1, the weighting coefficients are linearly adjusted and the data is re-fused.

[0121] For example, adjusting the temperature weight from 40% to 45% and reducing the internal resistance weight to 30% results in a more realistic fusion outcome. Based on the optimized fusion results, the probability distribution of battery lifetime is calculated using a Bayesian method, and the parameters are estimated using a Markov chain Monte Carlo method.

[0122] Specifically, the parameters of the lifetime distribution are estimated through multiple iterative simulations.

[0123] For example, the lifespan distribution of a certain battery shows a mean of 3 years and a variance of 0.5 years. This method can effectively handle the randomness in the data. It obtains the statistical characteristics of the probability distribution, such as the mean and variance of the probability distribution.

[0124] In one embodiment, the mean of the probability distribution can be used as a reference point for lifetime estimation, while the variance of the probability distribution indicates the confidence range of the prediction.

[0125] For example, a probability distribution with a mean of 3 years and a variance of 0.5 years indicates that life expectancy may fluctuate between 2.5 and 3.5 years. The life expectancy prediction range is determined using a support vector machine, and the radial basis function is used as the kernel function to train the model.

[0126] For example, after training, the model outputs a battery life range of 2.8 to 3.2 years. For the estimated range, similar patterns are matched from historical data, and the similarity is calculated using Euclidean distance.

[0127] For example, finding the battery with the smallest Euclidean distance between its past operating mode and current characteristics, with a lifespan of 3.1 years, can be used as a reference value to determine the final prediction. This matching method improves the reliability of the prediction.

[0128] In one possible implementation, if further analysis is needed, secondary features such as the number of iterations can be added to refine the results and enhance adaptability.

[0129] S8. If the confidence interval of the final estimated probability distribution is greater than the preset confidence interval threshold, then more frequent data will be collected for the dynamic monitoring samples. The life prediction model will be iteratively optimized using the updated samples to obtain the adjusted life detection results.

[0130] Optionally, this step also includes: Step S81: If the confidence interval width of the probability distribution exceeds the confidence interval threshold, calculate the sampling frequency increment based on the coefficient of variation of the dynamic monitoring data to obtain adjusted high-frequency sample data. Step S82: Remove outliers from the high-frequency sample data using a box plot method to obtain cleaned data. Step S83: Extract the time-domain mean, time-domain variance, and frequency-domain FFT amplitude from the cleaned data to form a feature set. Update the parameters of the lifetime prediction model using Markov chain Monte Carlo sampling implemented in PyMC3 to obtain the updated model. Step S84: Recalculate the probability distribution using the updated model to determine if the confidence interval width still exceeds the confidence interval threshold. Step S85: If it does, iteratively optimize the model parameters using the Adam optimizer in TensorFlow with mean squared error as the objective function to obtain the optimized model. Step S86: Calculate the lifetime prediction value based on the optimized model parameters and combine it with the probability distribution to generate the final detection result. Step S87: By calculating the mean Bland-Altman difference between the initial and final results, the stability of the data is determined, and the verified lifetime test conclusion is output.

[0131] In one possible implementation, if the confidence interval width of the probability distribution exceeds a preset confidence interval threshold (e.g., the threshold is set to 0.8 years, but the actual calculated value is 1.2 years), the sampling frequency needs to be adjusted based on the coefficient of variation of the dynamic monitoring data. The coefficient of variation reflects the relative degree of data fluctuation. For example, if the mean of a battery's internal resistance is 60 milliohms, the standard deviation is 3 milliohms, and the coefficient of variation is 0.05, assuming the preset rule is that for every 0.01 increase in the coefficient of variation, the sampling frequency increment increases by 10%, then the frequency is adjusted from once per day to 1.1 times per day.

[0132] For example, after high-frequency acquisition, the internal resistance of the data record on a certain day changed from 61 milliohms to 62 milliohms, the capacity decreased from 94% to 93%, and the temperature increased from 31°C to 32°C. These data provide a denser sample for subsequent analysis.

[0133] Preferably, box plots are used to remove outliers from high-frequency sample data. Box plots identify outliers by their interquartile range. For example, if the median of internal resistance data is 61 milliohms and the interquartile range is 2 milliohms, with upper and lower limits of 58 milliohms and 64 milliohms respectively, then a record of 65 milliohms is considered an outlier and removed.

[0134] Specifically, the cleaned data retained records of internal resistance between 58 and 63 milliohms, capacity variation between 92% and 95%, and temperature between 30°C and 33°C, eliminating sudden anomalies such as 66 milliohms or 28°C, thus ensuring data quality.

[0135] It should be noted that when extracting feature sets from the cleaned data, the time-domain mean can be taken as 60.5 milliohms of internal resistance and 93.5% of capacitance, while the time-domain variance reflects the fluctuation amplitude. For example, the internal resistance variance is 1.5. The frequency domain FFT amplitude is used to extract the peak value of the periodic signal through Fourier transform. For example, the main frequency amplitude of temperature fluctuation is 0.8.

[0136] In one embodiment, these features are input into PyMC3, and the model parameters are updated through Markov chain Monte Carlo sampling. For example, the mean was previously estimated to be 3 years, and after the update, it is adjusted to 2.9 years, and the time-domain variance is reduced from 0.5 years to 0.4 years, reflecting a more accurate parameter estimate.

[0137] For example, if the confidence interval shrinks from 1.2 years to 0.9 years after recalculating the probability distribution, but still exceeds the confidence interval threshold of 0.8 years, further optimization is needed.

[0138] Understandably, the Adam optimizer based on TensorFlow iteratively adjusts the model parameters with the mean squared error as the target.

[0139] For example, the initial model predicted a lifetime of 2.8 years, while the actual observed lifetime was 3 years, with a mean squared error of 0.04. After multiple iterations, the mean squared error was reduced to 0.01, and the predicted value was adjusted to 2.95 years. The optimized model is closer to the real data.

[0140] In one embodiment, the lifetime estimate is calculated based on the optimized model, for example, an output of 2.95 years. Combined with the probability distribution, the 90% confidence interval is 2.7 to 3.2 years, forming the final detection result.

[0141] Specifically, the final result not only provides a point estimate but also a range reference to facilitate subsequent decision-making. One possible implementation is to verify stability using the Bland-Altman mean difference. For example, if the initial result is 3 years and the final result is 2.95 years, the mean difference is 0.05 years, which is below the mean difference threshold of 0.1 years, indicating high data consistency and reliable conclusions.

[0142] Preferably, if further expansion is needed, cycle count features can be incorporated. For example, if a battery's internal resistance increases to 63 milliohms after 500 cycles, the model can be further refined by combining time-domain and frequency-domain features. This approach improves the robustness of predictions through multi-dimensional data fusion.

[0143] It should be noted that the verified conclusions can be directly applied to battery maintenance plans to extend the battery's lifespan.

[0144] S9 generates a dynamic report on battery health status based on the adjusted life test results. It uses visualization technology to map capacity decay patterns, internal resistance changes, and life prediction reliability onto a time axis, resulting in a monitoring archive that reflects the entire battery life cycle.

[0145] Optionally, this step also includes: Step S91: Obtain battery data through lifespan testing results, clean the data by handling missing values ​​and removing outliers to obtain initial health status values. Step S92: Extract capacity decay characteristics and internal resistance change characteristics from the initial health status values, and use linear regression to determine if the characteristics exceed preset health status thresholds, identifying outlier distributions. Step S93: For the outlier distributions, obtain the changing trends on the time axis, and use Matplotlib visualization tools to map capacity decay and internal resistance changes to obtain a dynamic monitoring sequence. Step S94: Based on the dynamic monitoring sequence, use the random forest algorithm to predict the reliability of lifespan estimation, obtaining trend data for the entire lifespan. Step S95: Extract key time nodes from the trend data. If the capacity decay corresponding to a node exceeds the capacity decay threshold, adjust the lifespan estimation parameters to obtain optimized predicted values. Step S96: Using the optimized predicted values, use regression analysis to obtain the mapping relationship, and use Matplotlib to generate monitoring archives on the time axis to obtain a dynamic reflection of the entire lifespan. Step S97: Based on the dynamic reflection, determine whether the health status meets expectations, use a t-test to analyze deviations, and obtain the final monitoring results.

[0146] Specifically, after obtaining battery data through lifespan testing, the data needs to be cleaned first.

[0147] For example, suppose a battery dataset contains records of internal resistance, capacity, and temperature, but some data is missing due to sensor malfunction, such as a missing capacity value. Mean imputation can be used to fill the missing value with the average capacity of the previous and subsequent days. For example, if the previous day's value was 95% and the next day's was 94%, the imputation value would be 94.5%. For outlier removal, a simple threshold method can be used. If the internal resistance suddenly jumps to 100 milliohms, while the normal range is between 55 and 65 milliohms, this record is removed. After cleaning, the initial healthy state values ​​can be set at 94% capacity and 60 milliohms internal resistance, serving as the basis for subsequent analysis. When extracting features from the initial healthy state values, the capacity decay feature can be obtained by subtracting the current value from 100% initial capacity; for example, a decay to 94% would be 6%. The internal resistance change feature is observed based on the increase; for example, an increase from 50 milliohms to 60 milliohms is 20%. When using linear regression to determine if the healthy state thresholds are exceeded, assuming a capacity decay threshold of 5% and an internal resistance increase threshold of 15%, both 6% and 20% exceed the limits, indicating the presence of outliers.

[0148] Specifically, the trend can be determined by the regression slope. If the slope shows that the internal resistance increases rapidly over time, the distribution of outliers is more obvious. For the distribution of outliers, the trend of change on the time axis needs to be visualized.

[0149] In one possible implementation, a line graph is plotted using Matplotlib, with the horizontal axis representing time and the vertical axis representing capacity and internal resistance.

[0150] For example, as capacity decreases from 100% on day 1 to 94% on day 30, and internal resistance increases from 50 milliohms to 60 milliohms, inflection points or abrupt change regions can be observed in the graph. This dynamic monitoring sequence visually reflects the battery degradation process, facilitating subsequent analysis. Based on the dynamic monitoring sequence, the random forest algorithm can predict lifetime reliability.

[0151] Preferably, the input features include capacity decay rate, internal resistance increase and temperature change, and the output is the remaining lifetime probability.

[0152] For example, a battery forecast shows 2 years remaining in its full lifespan, with a confidence level of 85%. This trend data covers a complete forecast from the current state to the end of its lifespan, supporting long-term planning. When extracting key time points from trend data, if the capacity degradation at a certain point exceeds 10% (e.g., drops to 90%), the lifespan parameters need to be adjusted.

[0153] Understandably, adjustments can be based on comparisons with historical data. For example, if a similar battery has an average remaining lifespan of 1.5 years at 90% capacity, the predicted value could be adjusted from 2 years to 1.5 years. This optimized prediction is closer to reality. Regression analysis can then generate a mapping relationship using the optimized prediction.

[0154] For example, internal resistance is negatively correlated with lifespan; the regression curve shows that for every 5 milliohms increase in internal resistance, lifespan decreases by 0.2 years. After plotting a timeline using Matplotlib, for example from day 1 to day 730, capacity decreases from 100% to 85%, while internal resistance increases from 50 milliohms to 70 milliohms, forming a dynamic reflection of the entire life cycle. When judging whether the health status meets expectations based on this dynamic reflection, the t-test can be used to analyze deviations in health status.

[0155] In one embodiment, the predicted lifespan was 1.5 years, while the actual observed mean was 1.6 years. A t-test showed no significant deviation in health status, indicating the reliability of the results. These final monitoring results provide a basis for battery maintenance, such as early replacement or adjustment of usage strategies, to extend the overall lifespan.

[0156] It should be noted that if the extended scheme incorporates cycle count characteristics, such as the capacity decreasing to 92% after 300 cycles, the trend prediction can be further refined. This multi-dimensional fusion approach enhances the comprehensiveness of the analysis and provides more reliable support for battery management.

[0157] like Figure 2 As shown, in a second aspect, the present invention provides a system for monitoring the health status and predicting the lifespan of an electric vehicle battery, which uses the method described above to monitor the health status and predict the lifespan of the electric vehicle battery. The system includes: The data acquisition module is used to acquire real-time data streams of the battery during vehicle operation, extract raw records of charge / discharge depth, temperature change, internal resistance parameters and capacity data from it, and calculate dynamic monitoring samples by using voltage, current and temperature values ​​collected by sensors. The preprocessing module is used to preprocess the dynamic monitoring samples. It uses a sliding window method to segment the time series data and extracts short-term fluctuation features of charge and discharge depth and temperature changes from the segmentation results to obtain the time series feature set of battery operating status. The capacity decay analysis module is used to calculate the capacity decay pattern through time series feature sets, analyze the capacity decrease after each charge-discharge cycle using an exponential decay model, obtain long-term operating trends from historical data, and obtain dynamic curves of capacity decay. The health status quantification module is used to obtain the dynamic curve of capacity decay, and then, in combination with internal resistance parameter analysis, calculate the rate of change of internal resistance over time using electrochemical impedance spectroscopy. It extracts multi-index synergistic features from the dynamic curve and the rate of change of internal resistance to obtain a preliminary quantitative characterization of the health status. The historical data integration module is used to integrate historical data for the preliminary quantitative characterization of health status. It extracts charging and discharging depth and temperature change records throughout the vehicle's life cycle from the database, and sorts the multidimensional data by feature importance using the random forest algorithm to obtain the weight distribution of key influencing factors. The lifespan prediction module is used to perform in-depth data mining based on the weight distribution of key influencing factors, and to model the time dependence of historical data using a long short-term memory network. From the modeling results, it predicts the battery degradation trend under different operating conditions and obtains an intermediate parameter set for lifespan prediction. The multi-index collaborative analysis module is used to perform multi-index collaborative analysis through the intermediate parameter set of life prediction. It uses a weighted average method to fuse the predicted values ​​of internal resistance parameters, capacity decay law and temperature change effects, and calculates the final predicted probability distribution of battery life from the fusion results. The model optimization module is used to re-collect more frequent data from the dynamic monitoring samples if the confidence interval of the final estimated probability distribution is greater than the preset confidence interval threshold. The updated samples are then used to iteratively optimize the life prediction model to obtain the adjusted life detection results. The report generation module is used to generate dynamic reports on battery health status based on the adjusted life test results. It uses visualization technology to map capacity decay patterns, internal resistance changes, and life prediction reliability onto a time axis, resulting in a monitoring archive that reflects the entire battery life cycle.

[0158] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for monitoring the health status and predicting the lifespan of electric vehicle batteries, characterized in that, The method includes: S1 acquires real-time data streams of the battery during vehicle operation, extracts raw records of charge / discharge depth, temperature change, internal resistance parameters, and capacity data from them, and calculates dynamic monitoring samples using voltage, current, and temperature values ​​collected by sensors. S2. Preprocess the dynamic monitoring samples, use the sliding window method to segment the time series data, extract the short-term fluctuation features of charge and discharge depth and temperature changes from the segmentation results, and obtain the time series feature set of battery operating status. S3 calculates the capacity decay law through time series feature set, uses exponential decay model to analyze the capacity decrease after each charge-discharge cycle, obtains long-term operating trend from historical data, and obtains dynamic curve of capacity decay. S4. After obtaining the dynamic curve of capacity decay, combined with the analysis of internal resistance parameters, the rate of change of internal resistance over time is calculated by electrochemical impedance spectroscopy. Multi-index synergistic features are extracted from the dynamic curve and the rate of change of internal resistance to obtain a preliminary quantitative characterization of the health status. S5 integrates historical data for the preliminary quantitative characterization of health status, extracts charging and discharging depth and temperature change records of the vehicle's entire life cycle from the database, and sorts the feature importance of multidimensional data through the random forest algorithm to obtain the weight distribution of key influencing factors. S6. Based on the weight distribution of key influencing factors, data is deeply mined, and a long short-term memory network is used to model the time dependence of historical data. The degradation trend of the battery under different operating conditions is predicted from the modeling results, and an intermediate parameter set for life prediction is obtained. S7 uses a multi-index collaborative analysis based on the intermediate parameter set of life prediction, and employs a weighted average method to fuse the predicted values ​​of internal resistance parameters, capacity decay law, and the influence of temperature change, and calculates the final predicted probability distribution of battery life from the fusion results. S8. If the confidence interval of the final estimated probability distribution is greater than the preset confidence interval threshold, then data with a higher frequency will be collected again for the dynamic monitoring samples. The life prediction model will be iteratively optimized using the updated samples to obtain the adjusted life detection results. S9 generates a dynamic report on battery health status based on the adjusted life test results. It uses visualization technology to map capacity decay patterns, internal resistance changes, and life prediction reliability onto a time axis, resulting in a monitoring archive that reflects the entire battery life cycle.

2. The method according to claim 1, characterized in that, Step S1 involves acquiring real-time data streams of the battery during vehicle operation, extracting raw records of charge / discharge depth, temperature changes, internal resistance parameters, and capacity data, and calculating dynamic monitoring samples using voltage, current, and temperature values ​​collected by sensors, including: Step S11: Collect the original values ​​of battery voltage, current and temperature during vehicle operation using sensors; Step S12: Calculate the original data of charge and discharge depth based on the time integral of the original voltage and current values, and generate a three-column time series data table by combining the original temperature value. Step S13: For the original data of charge / discharge depth and original data of capacity, the least squares method is used to fit the linear regression equation, the slope is used as the capacity decay trend, the intercept is used as the initial capacity, and the dynamic sample data table is output. Step S14: Extract instantaneous voltage and current values ​​from the dynamic sample data table, and use the DC internal resistance method to calculate the ratio of voltage difference to current difference in a 10-second window to obtain the internal resistance parameter sequence. Step S15: Calculate the Pearson correlation coefficient between the instantaneous temperature values ​​in the dynamic sample data table and the internal resistance parameter sequence. Step S16: If the absolute value of the correlation coefficient is greater than the correlation coefficient threshold, the internal resistance parameter sequence is corrected using the linear compensation formula to generate the corrected internal resistance parameter table. Step S17: When the number of data points set in the modified internal resistance parameter table exceeds the internal resistance parameter threshold, extract the capacity decay rate, internal resistance change rate and temperature slope from the dynamic sample data table as feature vectors, input them into the pre-trained support vector machine classification model, and output the health status label. Step S18: Based on the capacity decay curve corresponding to the health status label, calculate the percentage deviation between the actual capacity and the nominal capacity at the current charge / discharge depth, and update the capacity data column in the dynamic sample data table. Step S19: Perform a sliding window variance calculation on the updated capacity data column and the original temperature value. If the variance value is less than the variance threshold, it is determined to be in a stable state.

3. The method according to claim 2, characterized in that, In step S16, if the absolute value of the correlation coefficient is greater than the correlation coefficient threshold, the internal resistance parameter sequence is corrected using a linear compensation formula to generate a corrected internal resistance parameter table, wherein the correlation coefficient threshold is set to 0.

7.

4. The method according to claim 3, characterized in that, In step S17, when the number of data points set in the modified internal resistance parameter table exceeds the internal resistance parameter threshold, the capacity decay rate, internal resistance change rate, and temperature slope are extracted from the dynamic sample data table as feature vectors, input into the pre-trained support vector machine classification model, and the set number of data points in the output health status label are 5 consecutive data points.

5. The method according to claim 4, characterized in that, The threshold value for the internal resistance parameter is set to 0.1 ohms.

6. The method according to claim 5, characterized in that, In step S19, the updated capacity data column and the original temperature value are used to perform a sliding window variance calculation. When the variance value is less than the variance threshold, it is determined to be in a stable state. The variance threshold is set to 0.

01.

7. The method according to claim 1, characterized in that, Step S2 involves preprocessing the dynamic monitoring samples by segmenting the time-series data using a sliding window method. Short-term fluctuation features of charge / discharge depth and temperature changes are extracted from the segmentation results to obtain a time-series feature set of the battery's operating state, including: Step S21: The time series data of the dynamic monitoring sample is segmented using a sliding window with a fixed length and a fixed step size to obtain a segmented data sequence; Step S22: Extract the difference between the maximum and minimum voltage values ​​within each window from the segmented data sequence as the charge / discharge depth feature to form a charge / discharge depth sequence; Step S23: Extract the standard deviation of the temperature values ​​within each window from the segmented data sequence as the temperature change feature to form a temperature change sequence; Step S24: Align the charge / discharge depth sequence and the temperature change sequence by timestamp and then stitch them together horizontally to generate a short-term fluctuation feature set; Step S25: If the number of columns in the short-term fluctuation feature set exceeds the column number threshold, the feature set is first standardized by Z-score, and then principal component analysis is used to retain the components with a cumulative contribution rate of more than 95% to obtain the dimensionality-reduced feature set. Step S26: Based on the dimensionality reduction feature set, the K-means clustering algorithm is used to set the number of clusters to 3, and Euclidean distance is used for iterative calculation to obtain the state classification label; Step S27: Arrange the state classification labels in the original time order, map them to the timestamps of the segmented data sequence to generate a state encoding sequence, which serves as the final time series feature set.

8. The method according to claim 7, characterized in that, In step S21, a sliding window with a fixed length and a fixed step size is used to segment the time series data of the dynamic monitoring sample to obtain a segmented data sequence in which the fixed length is 30 sampling points and the fixed step size is 5 sampling points.

9. The method according to claim 8, characterized in that, In step S25, if the number of columns in the short-term fluctuation feature set exceeds the column number threshold, the feature set is first standardized by Z-score, and then principal component analysis is used to retain the components with a cumulative contribution rate of more than 95% to obtain the dimensionality-reduced feature set, where the column number threshold is set to 10.

10. A system for monitoring the health status and predicting the lifespan of electric vehicle batteries, characterized in that, The system for monitoring the health status and predicting the lifespan of an electric vehicle battery using the method described in any one of claims 1-9, the system comprising: The data acquisition module is used to acquire real-time data streams of the battery during vehicle operation, extract raw records of charge / discharge depth, temperature change, internal resistance parameters and capacity data from it, and calculate dynamic monitoring samples by using voltage, current and temperature values ​​collected by sensors. The preprocessing module is used to preprocess the dynamic monitoring samples. It uses a sliding window method to segment the time series data and extracts short-term fluctuation features of charge and discharge depth and temperature changes from the segmentation results to obtain the time series feature set of battery operating status. The capacity decay analysis module is used to calculate the capacity decay pattern through time series feature sets, analyze the capacity decrease after each charge-discharge cycle using an exponential decay model, obtain long-term operating trends from historical data, and obtain dynamic curves of capacity decay. The health status quantification module is used to obtain the dynamic curve of capacity decay, and then, in combination with internal resistance parameter analysis, calculate the rate of change of internal resistance over time using electrochemical impedance spectroscopy. It extracts multi-index synergistic features from the dynamic curve and the rate of change of internal resistance to obtain a preliminary quantitative characterization of the health status. The historical data integration module is used to integrate historical data for the preliminary quantitative characterization of health status. It extracts charging and discharging depth and temperature change records throughout the vehicle's life cycle from the database, and sorts the multidimensional data by feature importance using the random forest algorithm to obtain the weight distribution of key influencing factors. The lifespan prediction module is used to perform in-depth data mining based on the weight distribution of key influencing factors, and to model the time dependence of historical data using a long short-term memory network. From the modeling results, it predicts the battery degradation trend under different operating conditions and obtains an intermediate parameter set for lifespan prediction. The multi-index collaborative analysis module is used to perform multi-index collaborative analysis through the intermediate parameter set of life prediction. It uses a weighted average method to fuse the predicted values ​​of internal resistance parameters, capacity decay law and temperature change effects, and calculates the final predicted probability distribution of battery life from the fusion results. The model optimization module is used to re-collect more frequent data from the dynamic monitoring samples if the confidence interval of the final estimated probability distribution is greater than the preset confidence interval threshold. The updated samples are then used to iteratively optimize the life prediction model to obtain the adjusted life detection results. The report generation module is used to generate dynamic reports on battery health status based on the adjusted life test results. It uses visualization technology to map capacity decay patterns, internal resistance changes, and life prediction reliability onto a time axis, resulting in a monitoring archive that reflects the entire battery life cycle.

Images