Method and system for monitoring health condition of power battery of electric vehicle in real time

Abstract

The invention belongs to the technical field of battery monitoring, and discloses an electric vehicle power battery health condition real-time monitoring method and system, and the method comprises the steps: obtaining multi-parameter data such as voltage, current, temperature, internal resistance and the like through a battery management system, and carrying out the synchronous collection of the data through a preset sampling frequency, and obtaining a multi-dimensional time series data set; aiming at the multi-dimensional time sequence data set, adopting a feature extraction algorithm to separate dynamic change features of voltage, current, temperature and internal resistance to obtain a multi-parameter feature set; if the deviation of the battery health state output by the dynamic mapping relation model exceeds a preset threshold value, analyzing a fluctuation mode in the multi-parameter feature set through an anomaly detection algorithm, and judging whether a potential degradation risk exists or not; and according to the comprehensive representation of the battery health state, a time sequence prediction algorithm is adopted to analyze the future change trend of the multi-parameter feature set, and a short-term prediction value of the battery health state is generated.

Description

Technical Field

[0001] The present invention belongs to the technical field of battery monitoring, and in particular relates to a method and system for real-time monitoring of the health status of a power battery of an electric vehicle. Background Art

[0002] As the core of new energy vehicles, the performance of the power battery of electric vehicles is directly related to the safety, endurance and service life of the vehicle. Real-time monitoring of the health status of power batteries is a key technical field to ensure the efficient operation of electric vehicles, and is of great significance to improving user experience and promoting the sustainable development of the industry. With the widespread application of electric vehicles, the demand for health management of power batteries has become increasingly prominent, and research on how to accurately and in real time evaluate the battery status has become the focus of industry attention. However, many current monitoring methods have significant limitations. Some solutions rely on single parameter analysis, which is difficult to fully reflect the complex degradation mechanism of the battery; although other methods combine multiple parameters, the data processing efficiency is low and it is difficult to meet the real-time requirements. In addition, the existing technology often suffers from a decrease in monitoring accuracy due to drastic parameter fluctuations under complex working conditions, such as high temperature, low temperature or high load scenarios. These limitations make it difficult to accurately assess the health of the battery.

[0003] In this field, the core challenges focus on how to effectively use key parameters in the battery management system, such as voltage, current, temperature and internal resistance, to achieve dynamic assessment of health status. First, the interactions between parameters are complex, and changes in a single parameter are difficult to accurately characterize the overall state of the battery. It is necessary to establish a comprehensive mapping relationship between multiple parameters. Secondly, real-time requirements pose higher challenges to data processing speed, especially when parameters fluctuate abnormally, how to quickly identify and warn of potential risks. Third, the degradation patterns of batteries under different operating conditions vary significantly, and monitoring methods need to adapt to diverse scenarios to ensure robustness. These technical factors have not been fully addressed, resulting in insufficient accuracy and reliability of health monitoring, which in turn affects the safety and efficiency of electric vehicles.

[0004] Therefore, how to analyze the voltage, current, temperature, internal resistance and other parameters in the battery management system, establish a dynamic mapping relationship between them and the health status, and achieve efficient and accurate real-time monitoring has become a key issue in this study. Summary of the invention

[0005] In order to solve the problems existing in the prior art, the present invention provides a method and system for real-time monitoring of the health status of electric vehicle power batteries, which realizes real-time monitoring, anomaly detection and trend prediction of battery health status, and provides comprehensive technical support for battery management and maintenance.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A real-time monitoring method for the health status of an electric vehicle's power battery, the method comprising:

[0008] Synchronously collect battery multi-parameter data at a preset sampling frequency to obtain a multi-dimensional time series data set; wherein, the battery multi-parameter data includes: voltage, current, temperature, and internal resistance;

[0009] For the multi-dimensional time series data set, use a feature extraction algorithm to separate the dynamic change characteristics of voltage, current, temperature, and internal resistance respectively to obtain a multi-parameter feature set;

[0010] According to the multi-parameter feature set, use a deep learning model to model the interaction between voltage, current, temperature, and internal resistance, generate a dynamic mapping relationship model, and determine the comprehensive characterization of the battery health state;

[0011] If the deviation of the battery health state output by the dynamic mapping relationship model exceeds a preset threshold, analyze the fluctuation pattern in the multi-parameter feature set through an anomaly detection algorithm to determine whether there is a potential degradation risk;

[0012] According to the comprehensive characterization of the battery health state, use a time series prediction algorithm to analyze the future change trend of the multi-parameter feature set and generate a short-term prediction value of the battery health state.

[0013] Preferably, synchronously collect battery multi-parameter data at a preset sampling frequency to obtain a multi-dimensional time series data set, including:

[0014] Collect voltage, current, temperature, and internal resistance data through the battery management system at a preset sampling frequency, synchronously generate a multi-dimensional time series data set, and obtain an initial data set;

[0015] If there are missing values in the initial data set, complete the missing data by linear interpolation to generate a complete data set;

[0016] Use the principal component analysis method to perform dimensionality reduction processing on the complete data set, extract the main feature vectors, and obtain a dimensionality-reduced data set.

[0017] Preferably, for the multi-dimensional time series data set, use a feature extraction algorithm to separate the dynamic change characteristics of voltage, current, temperature, and internal resistance respectively to obtain a multi-parameter feature set, including:

[0018] Use the sliding window method to perform time series segmentation on the initial feature set to obtain the dynamic change trend of each parameter;

[0019] If the variance of the dynamic change trend exceeds a preset threshold, decompose the frequency components of each parameter through the fast Fourier transform algorithm to determine the periodic characteristics;

[0020] Separate high-frequency and low-frequency signals from voltage, current, temperature, and internal resistance according to periodic characteristics to obtain a refined feature set;

[0021] Obtain a standardized feature set by normalizing the refined feature set;

[0022] If the feature dimension of the standardized feature set exceeds a preset threshold, perform dimensionality reduction using the principal component analysis algorithm to obtain an optimized feature set;

[0023] Generate a multi-parameter feature set based on the optimized feature set.

[0024] Preferably, based on the multi-parameter feature set, use a deep learning model to model the interaction between voltage, current, temperature, and internal resistance, generate a dynamic mapping relationship model, and determine a comprehensive representation of the battery health state, including:

[0025] Obtain a multi-parameter feature set, extract time-series data from voltage, current, temperature, and internal resistance to obtain an original feature set;

[0026] Use a preprocessing algorithm to denoise and standardize the original feature set to obtain a normalized feature set;

[0027] Through a deep learning model, input the normalized feature set, model the interaction between voltage, current, temperature, and internal resistance to obtain a dynamic mapping relationship;

[0028] According to the dynamic mapping relationship, extract the non-linear correlation pattern between features to determine the interaction feature set;

[0029] If the complexity of the interaction feature set is higher than a preset threshold, use a dimensionality reduction algorithm to process the interaction feature set to obtain a simplified feature set;

[0030] Through a pre-trained model, input the simplified feature set, generate a comprehensive representation of the battery health state, and judge the health state level;

[0031] According to the health state level, map it to a preset state evaluation system to obtain the final state evaluation result.

[0032] Preferably, if the deviation of the battery health state output by the dynamic mapping relationship model exceeds a preset threshold, analyze the fluctuation pattern in the multi-parameter feature set through an anomaly detection algorithm to determine whether there is a potential degradation risk, including:

[0033] Calculate the battery health state through the dynamic mapping model to obtain the health state deviation;

[0034] If the health state deviation exceeds the preset threshold, obtain feature data from the multi-parameter feature set to determine the fluctuation pattern;

[0035] Analyze the fluctuation pattern using an anomaly detection algorithm to judge the potential degradation risk;

[0036] Extract key features from the multi-parameter feature set according to the potential degradation risk to obtain the degradation trend;

[0037] Process the degradation trend through a regression analysis algorithm to determine the degradation rate;

[0038] If the degradation rate exceeds the safe range, obtain relevant parameters from historical data to judge the degradation type;

[0039] Adjust the parameters of the dynamic mapping model according to the degradation type to optimize the health state deviation.

[0040] Preferably, according to the comprehensive characterization of the battery health state, use a time series prediction algorithm to analyze the future change trend of the multi-parameter feature set and generate short-term prediction values of the battery health state, including:

[0041] Obtain the data sequence of the multi-parameter feature set and extract the parameter set from the real-time monitoring of the battery health state;

[0042] Process the parameter set using a feature analysis method. If the correlation between parameters exceeds the preset threshold, eliminate redundant features to obtain a refined feature set;

[0043] Build a model for the refined feature set through a time series prediction algorithm to generate a change trend model;

[0044] According to the change trend model, process the data sequence using a sliding window method to determine the input sequence for short-term prediction;

[0045] If the integrity of the input sequence meets the preset threshold, generate short-term health state prediction values through a prediction algorithm;

[0046] For the short-term health state prediction values, use a trend analysis method to judge the deviation between the prediction values and historical data to obtain a corrected prediction result;

[0047] Generate a short-term prediction output of the battery health state through the corrected prediction result.

[0048] The present invention also provides a real-time monitoring system for the health status of an electric vehicle power battery. The system is used to implement the foregoing method. The system includes: a data acquisition module, a feature extraction module, a dynamic modeling module, an anomaly detection module, and a prediction analysis module;

[0049] The data acquisition module is used to synchronously collect battery multi-parameter data at a preset sampling frequency to obtain a multi-dimensional time series data set; wherein, the battery multi-parameter data includes: voltage, current, temperature, internal resistance;

[0050] The feature extraction module is used to separate the dynamic change features of voltage, current, temperature, and internal resistance respectively from the multi-dimensional time series data set by using a feature extraction algorithm, and obtain a multi-parameter feature set;

[0051] The dynamic modeling module is used to model the interaction between voltage, current, temperature, and internal resistance according to the multi-parameter feature set by using a deep learning model, generate a dynamic mapping relationship model, and determine the comprehensive representation of the battery health state;

[0052] The anomaly detection module is used to analyze the fluctuation pattern in the multi-parameter feature set through an anomaly detection algorithm and determine whether there is a potential degradation risk if the deviation of the battery health state output by the dynamic mapping relationship model exceeds a preset threshold;

[0053] The prediction analysis module is used to analyze the future change trend of the multi-parameter feature set by using a time series prediction algorithm according to the comprehensive representation of the battery health state, and generate a short-term prediction value of the battery health state.

[0054] Compared with the prior art, the beneficial effects of the present invention are:

[0055] The present invention discloses a method and system for real-time monitoring of the health status of an electric vehicle power battery. Multi-parameter data is collected through a battery management system, and synchronous sampling of voltage, current, temperature, internal resistance, etc. is performed to obtain a multi-dimensional time series data set. A feature extraction algorithm is used to separate the dynamic change features of each parameter, and then a deep learning model is used to establish a dynamic mapping relationship between the parameters, so as to determine the comprehensive representation of the battery health state. When the deviation of the health state exceeds the threshold, the present invention analyzes the parameter fluctuation pattern through an anomaly detection algorithm to determine the potential degradation risk. Finally, based on the health state representation, a time series prediction algorithm is used to analyze the future trend of the parameter features and generate a short-term prediction value. The present invention realizes the real-time monitoring, anomaly detection, and trend prediction of the battery health state, and provides comprehensive technical support for battery management and maintenance. Description of the Drawings

[0056] In order to more clearly illustrate the technical solutions of the present invention, the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0057] Figure 1 It is a schematic flowchart of a method for real-time monitoring of the health status of an electric vehicle power battery according to an embodiment of the present invention;

[0058] Figure 2 It is a schematic structural diagram of a system for real-time monitoring of the health status of an electric vehicle power battery according to an embodiment of the present invention. Specific Embodiments

[0059] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0060] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0061] Embodiment 1

[0062] As Figure 1 shown, this embodiment provides a method for real-time monitoring of the health status of an electric vehicle power battery, which may specifically include:

[0063] S101. Obtain multi-parameter data such as voltage, current, temperature, and internal resistance through the battery management system, and synchronously collect the data at a preset sampling frequency to obtain a multi-dimensional time series data set;

[0064] S102. For the multi-dimensional time series data set, use a feature extraction algorithm to separate the dynamic change characteristics of voltage, current, temperature, and internal resistance respectively to obtain a multi-parameter feature set;

[0065] S103. According to the multi-parameter feature set, use a deep learning model to model the interaction between voltage, current, temperature, and internal resistance, generate a dynamic mapping relationship model, and determine the comprehensive representation of the battery health status;

[0066] S104. If the deviation of the battery health status output by the dynamic mapping relationship model exceeds the preset threshold, analyze the fluctuation pattern in the multi-parameter feature set through an anomaly detection algorithm to determine whether there is a potential degradation risk;

[0067] S105. According to the comprehensive representation of the battery health status, use a time series prediction algorithm to analyze the future change trend of the multi-parameter feature set and generate a short-term prediction value of the battery health status.

[0068] In this embodiment, S101. Obtain multi-parameter data such as voltage, current, temperature, and internal resistance through the battery management system, and synchronously collect the data at a preset sampling frequency to obtain a multi-dimensional time series data set, including:

[0069] The battery management system collects voltage, current, temperature, and internal resistance data at a preset sampling frequency, synchronously generates a multi-dimensional time series data set, and obtains an initial data set. If there are missing values in the initial data set, linear interpolation is used to complete the missing data and generate a complete data set. The principal component analysis method is used to reduce the dimension of the complete data set, extract the main feature vectors, and obtain a reduced-dimension data set. The clustering analysis method is used to group the reduced-dimension data set, determine the similarity pattern between data points, and obtain a classified data set. If a group of data points in the classified data set deviates from the preset threshold, it is marked as an abnormal state, and an abnormal mark data set is generated. Based on the abnormal mark data set, time series analysis is used to model the time distribution of abnormal points, and an abnormal trend model is obtained. The potential abnormal points in the future time period are predicted through the abnormal trend model, and a prediction result is generated.

[0070] In this embodiment, in S102, for the multi-dimensional time series data set, a feature extraction algorithm is used to separate the dynamic change characteristics of voltage, current, temperature, and internal resistance respectively, and a multi-parameter feature set is obtained, including:

[0071] The principal component analysis algorithm is used to extract the preliminary characteristics of voltage, current, temperature, and internal resistance from the multi-dimensional time series data set, and an initial feature set is obtained. The sliding window method is used to perform time series segmentation on the initial feature set to obtain the dynamic change trend of each parameter. If the variance of the dynamic change trend exceeds the preset threshold, the frequency components of each parameter are decomposed by the fast Fourier transform algorithm to determine the periodic characteristics. According to the periodic characteristics, high-frequency and low-frequency signals are separated from voltage, current, temperature, and internal resistance, and a refined feature set is obtained. By normalizing the refined feature set, a standardized feature set is obtained. If the feature dimension of the standardized feature set exceeds the preset threshold, the principal component analysis algorithm is used for dimension reduction processing to obtain an optimized feature set. According to the optimized feature set, a multi-parameter feature set is generated.

[0072] Preferably, when separating high-frequency and low-frequency signals from parameters such as voltage and current, the cut-off frequency can be set to 0.1 Hz. Assuming that the current data contains high-frequency signals with rapid fluctuations and low-frequency signals with slow changes, after separation, the high-frequency part can reflect the instantaneous load change, and the low-frequency part can reflect the long-term trend. This refined feature set is more suitable for in-depth analysis of battery performance.

[0073] It should be noted that if the feature dimension is too high, such as exceeding 10 dimensions, the principal component analysis is used again for dimension reduction. Assuming that the original feature set contains 12 dimensions, the first 3 principal components are retained after dimension reduction, covering 95% of the data variance, and an optimized feature set is generated. This method reduces the computational burden while ensuring information integrity.

[0074] It can be understood that the finally generated multi-parameter feature set can be used for battery state assessment.

[0075] For example, the optimized feature set may show that the voltage and internal resistance characteristics of a certain battery deviate from the normal range, indicating a potential aging risk. Such a feature set provides unified basic data for subsequent analysis.

[0076] For example, in practical applications, the feature set of a certain battery pack may reveal a pattern of abnormal temperature increase. Combining with the current characteristics, it can be judged whether it is caused by overload. This multi-parameter analysis improves the comprehensiveness of diagnosis through comprehensive features.

[0077] In this embodiment, S103: According to the multi-parameter feature set, a deep learning model is used to model the interaction between voltage, current, temperature, and internal resistance, generate a dynamic mapping relationship model, and determine the comprehensive characterization of the battery health state, including:

[0078] Obtain the multi-parameter feature set, extract time-series data from voltage, current, temperature, and internal resistance to obtain the original feature set. Use a preprocessing algorithm to denoise and standardize the original feature set to obtain a normalized feature set. Through the deep learning model, input the normalized feature set to model the interaction between voltage, current, temperature, and internal resistance to obtain a dynamic mapping relationship. According to the dynamic mapping relationship, extract the non-linear association pattern between features to determine the interaction feature set. If the complexity of the interaction feature set is higher than the preset threshold, use a dimensionality reduction algorithm to process the interaction feature set to obtain a simplified feature set. Through the pre-trained model, input the simplified feature set to generate the comprehensive characterization of the battery health state and judge the health state level. According to the health state level, map it to the preset state evaluation system to obtain the final state evaluation result.

[0079] Specifically, the deep learning model models the interaction between voltage, current, temperature, and internal resistance to generate a dynamic mapping relationship model, including: implemented by the WOA-BP neural network prediction model.

[0080] (1) Hunting for prey: The search range of the WOA algorithm is the entire solution space. At the beginning of the algorithm, it is assumed that the current position of the prey is the global optimal solution position. Then, the humpback whales in the population will gather towards the prey. When the humpback whales gather towards the prey, their positions will change. At this time, the mathematical model of the position of the humpback whales is:

[0081] D = |C · X * (t) - X(t)| (1)

[0082] X(t + 1) = X * (t) - A · D (2)

[0083] Where D represents the distance between the humpback whale and the prey; X*(t) represents the position vector of the current optimal solution; X(t) represents the current position of the humpback whale; t represents the current iteration number; A and C represent coefficient vectors, and their expressions are as follows:

[0084] A = 2a * r1 - a (3)

[0085] C = 2 * r2 (4)

[0086] Where r1 and r2 represent random vectors in the interval [0, 1]; a represents the linear convergence factor, which linearly decreases from 2 to 0 during the iteration process.

[0087] (2) Bubble-net predation: The predation of humpback whales mainly has two mechanisms: surrounding the prey and bubble-net attack. When hunting, the humpback whale approaches the prey in a spiral upward manner and then preys on the prey. The position update between the humpback whale and the prey is represented by a logarithmic spiral equation, and its mathematical model is as follows:

[0088] X(t + 1) = D′ * e bl * cos(2πl) + X * (t) (5)

[0089] D′ = |X * (t) - X(t)| (6)

[0090] Where D' represents the distance between the current humpback whale and the prey; b represents the spiral shape parameter; l represents a random number in the interval [0, 1].

[0091] There are two predation behaviors of the humpback whale during the process of approaching the prey. Here, it is assumed that the probability of the humpback whale randomly selecting the two predation behaviors is 50% each. The mathematical model of the updated position of the humpback whale can be obtained as follows:

[0092]

[0093] Where p represents the probability of the predation mechanism, which is a random number in the interval [0, 1].

[0094] (3) Randomly search for prey: As the number of iterations t increases, the parameters A and the linear convergence factor a in Equation (3) will gradually decrease. When |A| < 1, each humpback whale will gradually surround the current optimal solution, and at this time, the algorithm is in the local optimization stage. To ensure that all humpback whales can search sufficiently in the solution space, the algorithm will update the positions according to the distances between each humpback whale, so as to achieve the purpose of random search. Therefore, when |A| ≥ 1, the current position of a randomly selected humpback whale will be used as a reference to update the positions of other humpback whales, enabling the humpback whale population to conduct global search, thereby avoiding being in the local optimization state. The mathematical model for the positions of humpback whales is as follows:

[0095] D″ = |C * X rand (t) - X(t)| (8)

[0096] X(t + 1) = X rand (t) - A * D (9)

[0097] In the formula, D" represents the distance between the currently searching humpback whale and the random humpback whale; Xrana(t) represents the position of a randomly selected humpback whale in the current population.

[0098] Aiming at the defects of the BP neural network in the initial weights and thresholds, WOA is used to optimize it, and the implementation process is as follows: (1) Divide the input sample data of lithium-ion batteries (such as battery voltage, current, etc.) into an input set and an output set, and normalize the data in the input set; (2) Determine the topological structure of the BP neural network, such as the number of layers and the number of neurons in each layer, and initialize the weights and thresholds of the BP neural network; (3) Initialize the parameters of the WOA algorithm, set the number of iterations, the population size, initialize the position information, and select the mean square error of BP neural network training as the fitness function of WOA; (4) Calculate the fitness of each humpback whale, find and record the position of the current optimal humpback whale and save it as the optimal position; (5) When the number of evolutionary generations t < the maximum number of evolutionary generations t max When, calculate and update the parameters A, C, 1, p, and a in the WOA algorithm; (6) When p < 0.5, if |A| < 1, update the position of the humpback whale according to Equation (2); if |A| ≥ 1, update the position of the humpback whale according to Equation (9); (7) When p ≥ 0.5, update the position of the humpback whale according to Equation (5); (8) Calculate the fitness of each humpback whale in the current population, compare it with the previously saved optimal humpback whale position, select the optimal position at this time and save it. At the same time, judge whether t > t maxWhether it holds. If so, proceed to the next step; otherwise, set \(t = t + 1\) and repeat steps (5) - (8); (9) When the preset number of iterations is reached or other termination conditions are met, output the optimal humpback whale position and record the optimal weight and threshold parameters; (10) The BP neural network obtains the optimal parameters found by the WOA algorithm for network training; (11) Perform simulation prediction on the trained BP neural network; (12) Output the SOC prediction value.

[0099] As the number of iterations increases, we gradually decrease the value of the weight to improve the optimization accuracy in the later stage of iteration. At this stage, the algorithm pays more attention to local details and tries to conduct a more in - depth search near the potential solutions that have been discovered to further improve the quality of the solution. By gradually decreasing the value of the weight, we can make the algorithm focus more on the local area, more finely adjust the position of the solution, and thus improve the local optimization ability of the algorithm. The mathematical model for introducing the adaptive weight is:

[0100]

[0101] In the formula, \(\omega\) min represents the minimum weight; \(\omega\) max represents the maximum weight; \(m\) represents a random number between \([0, 1]\); \(t\) represents the current iteration number; \(gen\) max represents the maximum number of iterations.

[0102] After introducing the adaptive weight, new position update formulas in the two stages of bubble - net predation and random prey search can be obtained. The formulas are as follows:

[0103]

[0104] \(X(t + 1)=\omega*X\) rand (t)-A*D (12)

[0105] After introducing the adaptive weight, the value range of the weight can be restricted to ensure that the weight value is within a reasonable range and can adaptively adjust the range to adapt to the optimization process. This means that during the search process, the influence of some individuals may be strengthened while the influence of other individuals may be weakened to better balance the trade - off between global search and local search. This dynamic adjustment helps to maintain the diversity of the population and prevent premature convergence to local optimal solutions. At the same time, it allows the algorithm to more flexibly adjust the search strategy to adapt to different environments, thereby improving the adaptability of the algorithm.

[0106] Specifically, the average value is taken for every 10 data points to smooth out abnormal fluctuations. Standardization maps each parameter to the range from 0 to 1. For example, the voltage range is linearly scaled from 3.0V to 4.2V to ensure that parameters with different dimensions are comparable. In this way, the normalized feature set provides a unified basis for subsequent modeling.

[0107] Exemplarily, the model maps the feature set to a quantitative characterization of the battery health state and outputs a health score from 0 to 100. A score above 80 indicates good, 60 to 80 is medium, and below 60 is poor.

[0108] It can be understood that the health state level needs to be mapped to a preset evaluation system to output the final result.

[0109] For example, the good level corresponds to normal operation and it is recommended to continue monitoring; the medium level indicates that attention is needed and it may be necessary to adjust the charge and discharge strategy; the poor level recommends replacing the battery.

[0110] In this embodiment, S104, if the deviation of the battery health state output by the dynamic mapping relationship model exceeds the preset threshold, the fluctuation pattern in the multi-parameter feature set is analyzed through an anomaly detection algorithm to determine whether there is a potential degradation risk, including:

[0111] Calculate the battery health state through the dynamic mapping model to obtain the health state deviation. If the health state deviation exceeds the preset threshold, obtain the feature data from the multi-parameter feature set to determine the fluctuation pattern. Use the anomaly detection algorithm to analyze the fluctuation pattern to judge the potential degradation risk. According to the potential degradation risk, extract the key features from the multi-parameter feature set to obtain the degradation trend. Process the degradation trend through the regression analysis algorithm to determine the degradation rate. If the degradation rate exceeds the safe range, obtain the relevant parameters from the historical data to judge the degradation type. According to the degradation type, adjust the parameters of the dynamic mapping model to optimize the health state deviation.

[0112] For example, extract the voltage fluctuation data within a week from the battery management system and find that the voltage shows a periodic decrease under high load. Combining with the temperature data, it is observed that the voltage drops more significantly at high temperatures. This fluctuation pattern indicates that there may be performance degradation caused by the thermal effect.

[0113] Specifically, when analyzing, the temporal correlation between voltage and temperature can be extracted to generate a fluctuation curve, and then the abnormal points can be identified.

[0114] Preferably, when using the anomaly detection algorithm to analyze the fluctuation pattern, a statistical outlier detection method can be used. Assume that for a certain battery under specific working conditions, there are multiple abnormal points where the internal resistance data is higher than twice the mean. Combining with the current data analysis, it is confirmed that these abnormal points appear in the fast charge and discharge stage. The anomaly detection algorithm marks these points as potential degradation risks by setting a dynamic threshold.

[0115] It should be noted that the advantage of this method is that it can quickly locate anomalies without complex modeling.

[0116] In one embodiment, key features are extracted from the multi-parameter feature set to determine the degradation trend, which can focus on the internal resistance and temperature.

[0117] For example, analysis finds that the internal resistance has slowly increased from 0.1 ohm to 0.15 ohm in the past month, and the temperature often exceeds 45 degrees Celsius under high load. After extracting the key features, a trend graph of the internal resistance changing with time can be drawn to confirm that the degradation trend shows a linear increase. Such analysis helps to clarify the degradation direction and facilitate subsequent processing.

[0118] For example, when using the regression analysis algorithm to process the degradation trend to determine the degradation rate, a linear regression method can be adopted. Suppose by analyzing the relationship between the internal resistance and the number of cycles, it is confirmed that the internal resistance increases by 0.02 ohm per 100 cycles, and the degradation rate is about 0.0002 ohm / cycle. If the safety range is 0.0001 ohm / cycle, the current rate exceeds the standard. This indicates that the degradation may accelerate and further inspection is required. The advantage of regression analysis is that it can quantify the trend and facilitate risk assessment.

[0119] Specifically, when obtaining relevant parameters from historical data to judge the degradation type, the number of cycles, the charging rate, and the ambient temperature can be analyzed.

[0120] For example, historical data shows that the internal resistance of the battery grows faster at high charging rates, and the capacity decays significantly under high-temperature environments. It is judged that the degradation type is the aging of the electrode material. This analysis depends on the integrity of historical data and can effectively distinguish thermal degradation or mechanical degradation.

[0121] In a possible implementation, adjusting the parameters of the dynamic mapping model to optimize the health state deviation can be achieved by updating the weight matrix. Suppose the initial deviation of the model is 8%. By analyzing the degradation type, the weights of temperature and internal resistance are increased, and the deviation drops to 4% after recalculation, which is closer to the actual state. The significance of this adjustment is to improve the sensitivity of the model to key parameters and optimize the prediction accuracy.

[0122] It can be understood that each step of the above method focuses on the accurate assessment of the battery health state, from deviation calculation to model optimization, forming a closed-loop analysis. The implementation of each link depends on the dynamic interaction of multi-parameter features to ensure the reliability of the analysis results.

[0123] In this embodiment, S105. According to the comprehensive characterization of the battery health state, a time series prediction algorithm is used to analyze the future change trend of the multi-parameter feature set and generate a short-term prediction value of the battery health state, including:

[0124] Obtain the data sequence of the multi-parameter feature set, and extract the parameter set from the real-time monitoring of the battery health state. Process the parameter set using a feature analysis method. If the correlation between parameters exceeds a preset threshold, redundant features are removed to obtain a refined feature set. Model the refined feature set using a time series prediction algorithm to generate a change trend model. According to the change trend model, use a sliding window method to process the data sequence to determine the input sequence for short-term prediction. If the integrity of the input sequence meets the preset threshold, generate a short-term health state prediction value through a prediction algorithm. For the short-term health state prediction value, use a trend analysis method to judge the deviation between the prediction value and historical data to obtain a corrected prediction result. Generate the short-term prediction output of the battery health state through the corrected prediction result.

[0125] In one embodiment, an ARIMA model can be used to analyze the change trend of voltage over time.

[0126] For example, based on the voltage data of the past 1000 cycles, the model identifies that the voltage shows a slow downward trend, and predicts that the voltage change range in the next 100 cycles is from 3.6V to 3.65V.

[0127] Exemplarily, this modeling method can capture the laws in long-term operation and provide a reliable basis for subsequent predictions. According to the change trend model, use a sliding window method to process the data sequence to determine the input sequence for short-term prediction.

[0128] It can be understood that the sliding window extracts data through a fixed time period. For example, with a 10-minute window, it contains 600 data points.

[0129] Preferably, the window sliding step size is set to 1 minute to smooth the data.

[0130] For example, the voltage within a certain window fluctuates from 3.7V to 3.68V. The input sequence contains these change points to ensure the continuity of the prediction. If the integrity of the input sequence meets the preset threshold, generate a short-term health state prediction value through a prediction algorithm.

[0131] Embodiment 2

[0132] As Figure 2 shown, the present invention provides a real-time monitoring system for the health status of an electric vehicle power battery, mainly including:

[0133] A data acquisition module, configured to obtain multi-parameter data such as voltage, current, temperature, and internal resistance through a battery management system, and synchronously acquire the data at a preset sampling frequency to obtain a multi-dimensional time series data set;

[0134] A feature extraction module, which is used for a multi-dimensional time series data set, and adopts a feature extraction algorithm to separate the respective dynamic change features of voltage, current, temperature, and internal resistance, so as to obtain a multi-parameter feature set;

[0135] A dynamic modeling module, which is used for based on the multi-parameter feature set, adopts a deep learning model to model the interaction between voltage, current, temperature, and internal resistance, generates a dynamic mapping relationship model, and determines a comprehensive characterization of the battery health state;

[0136] An anomaly detection module, which is used for if the deviation of the battery health state output by the dynamic mapping relationship model exceeds a preset threshold, then analyzes the fluctuation pattern in the multi-parameter feature set through an anomaly detection algorithm to determine whether there is a potential degradation risk;

[0137] A prediction analysis module, which is used for based on the comprehensive characterization of the battery health state, adopts a time series prediction algorithm to analyze the future change trend of the multi-parameter feature set, and generates a short-term prediction value of the battery health state.

[0138] The above embodiments are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A real-time monitoring method for the health status of an electric vehicle power battery, characterized in that, The method includes: Synchronously collecting battery multi-parameter data at a preset sampling frequency to obtain a multi-dimensional time series data set; wherein, the battery multi-parameter data includes: voltage, current, temperature, and internal resistance; For the multi-dimensional time series data set, using a feature extraction algorithm to separate the dynamic change characteristics of voltage, current, temperature, and internal resistance respectively to obtain a multi-parameter feature set; According to the multi-parameter feature set, using a deep learning model to model the interaction between voltage, current, temperature, and internal resistance, generating a dynamic mapping relationship model, and determining a comprehensive representation of the battery health state; If the deviation of the battery health state output by the dynamic mapping relationship model exceeds a preset threshold, then analyze the fluctuation pattern in the multi-parameter feature set through an anomaly detection algorithm to determine whether there is a potential degradation risk; According to the comprehensive representation of the battery health state, using a time series prediction algorithm to analyze the future change trend of the multi-parameter feature set and generate a short-term prediction value of the battery health state.

2. The method according to claim 1, wherein Synchronously collecting battery multi-parameter data at a preset sampling frequency to obtain a multi-dimensional time series data set, including: Collecting voltage, current, temperature, and internal resistance data through a battery management system at a preset sampling frequency, synchronously generating a multi-dimensional time series data set to obtain an initial data set; If there are missing values in the initial data set, then complete the missing data through linear interpolation to generate a complete data set; Using the principal component analysis method to perform dimensionality reduction processing on the complete data set, extracting the main feature vectors to obtain a dimensionality-reduced data set.

3. The method according to claim 1, wherein For the multi-dimensional time series data set, using a feature extraction algorithm to separate the dynamic change characteristics of voltage, current, temperature, and internal resistance respectively to obtain a multi-parameter feature set, including: Using a sliding window method to perform time series segmentation on the initial feature set to obtain the dynamic change trends of each parameter; If the variance of the dynamic change trend exceeds a preset threshold, then decompose the frequency components of each parameter through a fast Fourier transform algorithm to determine the periodic characteristics; According to the periodic characteristics, separate the high-frequency and low-frequency signals from voltage, current, temperature, and internal resistance to obtain a refined feature set; By performing normalization processing on the refined feature set, obtaining a standardized feature set; If the feature dimension of the standardized feature set exceeds a preset threshold, then use the principal component analysis algorithm for dimensionality reduction processing to obtain an optimized feature set; Generate a multi-parameter feature set according to the optimized feature set.

4. The method according to claim 1, wherein According to the multi-parameter feature set, using a deep learning model to model the interaction between voltage, current, temperature, and internal resistance, generating a dynamic mapping relationship model, and determining a comprehensive representation of the battery health state, including: Obtain the multi-parameter feature set, extract time series data from voltage, current, temperature, and internal resistance to obtain an original feature set; Using a preprocessing algorithm to perform denoising and standardization on the original feature set to obtain a normalized feature set; Through a deep learning model, input the normalized feature set, model the interaction between voltage, current, temperature, and internal resistance to obtain a dynamic mapping relationship; According to the dynamic mapping relationship, extract the non-linear association patterns between features to determine an interaction feature set; If the complexity of the interaction feature set is higher than a preset threshold, then use a dimensionality reduction algorithm to process the interaction feature set to obtain a simplified feature set; Using a pre-trained model, input a simplified feature set to generate a comprehensive representation of the battery health state and determine the health state level; According to the health state level, map it to a preset state evaluation system to obtain the final state evaluation result.

5. The method according to claim 1, wherein If the deviation of the battery health state output by the dynamic mapping relationship model exceeds the preset threshold, analyze the fluctuation pattern in the multi-parameter feature set through an anomaly detection algorithm to determine whether there is a potential degradation risk, including: Calculate the battery health state through the dynamic mapping model to obtain the health state deviation; If the health state deviation exceeds the preset threshold, obtain the feature data from the multi-parameter feature set to determine the fluctuation pattern; Use the anomaly detection algorithm to analyze the fluctuation pattern to judge the potential degradation risk; Extract key features from the multi-parameter feature set according to the potential degradation risk to obtain the degradation trend; Process the degradation trend through the regression analysis algorithm to determine the degradation rate; If the degradation rate exceeds the safe range, obtain relevant parameters from the historical data to judge the degradation type; According to the degradation type, adjust the parameters of the dynamic mapping model to optimize the health state deviation.

6. The method according to claim 1, wherein According to the comprehensive representation of the battery health state, use the time series prediction algorithm to analyze the future change trend of the multi-parameter feature set and generate the short-term prediction value of the battery health state, including: Obtain the data sequence of the multi-parameter feature set and extract the parameter set from the real-time monitoring of the battery health state; Use the feature analysis method to process the parameter set. If the correlation between parameters exceeds the preset threshold, remove redundant features to obtain a refined feature set; Model the refined feature set through the time series prediction algorithm to generate a change trend model; According to the change trend model, use the sliding window method to process the data sequence to determine the input sequence for short-term prediction; If the integrity of the input sequence meets the preset threshold, generate the short-term health state prediction value through the prediction algorithm; For the short-term health state prediction value, use the trend analysis method to judge the deviation between the prediction value and the historical data to obtain the corrected prediction result; Generate the short-term prediction output of the battery health state through the corrected prediction result.

7. A real-time monitoring system for the health status of an electric vehicle power battery, the system being used to implement the method described in any one of claims 1-6, characterized in that, The system includes: a data acquisition module, a feature extraction module, a dynamic modeling module, an anomaly detection module, and a prediction analysis module; The data acquisition module is used to synchronously collect battery multi-parameter data at a preset sampling frequency to obtain a multi-dimensional time series data set; among them, the battery multi-parameter data includes: voltage, current, temperature, internal resistance; The feature extraction module is used to separate the dynamic change characteristics of voltage, current, temperature, and internal resistance respectively from the multi-dimensional time series data set through a feature extraction algorithm to obtain a multi-parameter feature set; The dynamic modeling module is used to model the interaction between voltage, current, temperature, and internal resistance according to the multi-parameter feature set by using a deep learning model, generate a dynamic mapping relationship model, and determine the comprehensive representation of the battery health state; The anomaly detection module is used to, if the deviation of the battery health state output by the dynamic mapping relationship model exceeds the preset threshold, analyze the fluctuation pattern in the multi-parameter feature set through an anomaly detection algorithm to determine whether there is a potential degradation risk; The prediction analysis module is used to analyze the future change trend of the multi-parameter feature set by using a time series prediction algorithm according to the comprehensive characterization of the battery health state, and generate a short-term prediction value of the battery health state.

Images