An electric vehicle power battery on-line safety management and early warning method

Abstract

The application discloses an online safety management and early warning method for power batteries of electric vehicles, comprising the following steps: obtaining multi-dimensional operation data of the power batteries and preprocessing the data to obtain a standardized data set; obtaining dynamic working conditions and time sequence characteristics based on the standardized data set, generating a battery state feature set and judging whether the battery state feature set meets preset conditions, if not, optimizing the battery state feature set to obtain an optimized feature set; constructing a risk assessment model, combining the optimized feature set to obtain a battery safety risk probability and judging a risk level; if the risk level exceeds a threshold value, obtaining a personalized management strategy generated by a cloud and sent to a vehicle-side controller based on the optimized feature set; and through an online learning algorithm, fusing real-time data of the vehicle side and battery aging data, dynamically adjusting parameters of the risk assessment model, generating an adaptive model and outputting a graded early warning signal. The application effectively evaluates the safety risk of the power batteries under complex working conditions, and provides real-time early warning and personalized management strategies.

Description

Technical Field

[0001] This invention belongs to the field of information technology, and in particular relates to an online safety management and early warning method for electric vehicle power batteries. Background Technology

[0002] As a core direction of new energy vehicles, the safety of electric vehicles' power batteries directly affects the reliability of vehicle operation and the safety of users' lives and property. Power battery management is not only a key pillar in the development of electric vehicle technology but also an important guarantee for promoting green transportation and energy transition. However, current power battery safety management methods have significant limitations in practical applications. Many solutions rely too heavily on single models, making it difficult to comprehensively capture the dynamic characteristics of batteries under complex operating conditions. Furthermore, existing methods struggle to balance real-time performance and accuracy, easily leading to delayed or misjudged risk assessments, especially under battery aging or extreme conditions. In addition, insufficient collaboration between the vehicle and cloud platforms limits the depth of data analysis and the ability for personalized management.

[0003] These limitations reflect the core challenges in the field of power battery safety management. First, the complex correlations of battery state parameters are difficult to model accurately, and the interaction between electrochemical characteristics and external operating conditions increases the difficulty of prediction. Second, the accuracy of safety feature extraction and risk assessment is insufficient; existing methods are prone to losing key information when processing multidimensional data, resulting in untimely and ineffective early warning strategies. Finally, the adaptability of models is a significant issue; battery aging or changes in operating conditions reduce the predictive ability of models, making long-term reliable online management difficult. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes an online safety management and early warning method for electric vehicle power batteries, thereby resolving the issues present in the prior art.

[0005] To achieve the above objectives, the present invention provides an online safety management and early warning method for electric vehicle power batteries, comprising:

[0006] Acquire multidimensional operational data of the power battery and preprocess it to obtain a standardized dataset;

[0007] Based on the standardized dataset, dynamic operating conditions and time series features are obtained, and a battery status feature set is generated. If the battery status feature set does not meet the preset conditions, the battery status feature set is optimized to obtain an optimized feature set.

[0008] A risk assessment model is constructed, and the battery safety risk probability is obtained by combining the optimized feature set. The risk level is then determined based on the battery safety risk probability.

[0009] If the risk level exceeds the threshold value, a cloud-generated personalized management strategy is obtained based on the optimized feature set and is delivered to the vehicle-side controller.

[0010] The risk assessment model parameters are dynamically adjusted by fusing the real-time data on the vehicle side and the battery aging data through an online learning algorithm, an adaptive model is generated, and a graded early warning signal is output.

[0011] Optionally, the process of obtaining the standardized data set comprises:

[0012] The multi-dimensional operation data including voltage, current, temperature and state of charge are denoised by using a filtering technique to generate a denoised data set, the abnormal data in the denoised data set exceeding a preset threshold value are marked, and the marked data are standardized to generate a standardized data set.

[0013] Optionally, the process of generating the battery state feature set comprises:

[0014] Based on the standardized data set, multi-dimensional features are obtained, a convolutional neural network is used to extract dynamic features under complex working conditions to obtain an initial feature set, a long short-term memory network is used to analyze the time sequence characteristics of the initial feature set to obtain a time sequence feature set, when the time sequence feature set does not match a preset working condition threshold value, the time sequence feature set is re-clustered and the dynamic features and the time sequence features are fused to generate a comprehensive feature set, according to the comprehensive feature set, a convolutional neural network is used to deeply analyze the feature set to extract hidden working condition change features to obtain a deep feature set, if the correlation between the features in the deep feature set is lower than a preset threshold value, the feature set is optimized by dimension reduction technology to obtain an optimized feature set, and a long short-term memory network is used to model the long-term dependence relationship of the time sequence based on the optimized feature set to generate a battery state feature set.

[0015] Optionally, the process of obtaining the optimized feature set comprises:

[0016] If the voltage fluctuation in the battery state feature set exceeds a preset threshold value, the feature set is smoothed by a Kalman filtering algorithm to obtain the optimized feature set.

[0017] Optionally, the process of constructing the risk assessment model, obtaining the battery safety risk probability in combination with the optimized feature set, and judging the risk level based on the battery safety risk probability comprises:

[0018] The optimization feature set is standardized to extract dynamic change features in the time sequence, and a dynamic feature set is obtained; if the signal fluctuation in the dynamic feature set exceeds a preset threshold, the data is segmented by a sliding window technique and statistical distribution characteristics are calculated to obtain a distribution feature set; a classification model is trained using a random forest algorithm, and a risk probability value is output; when the risk probability value exceeds a threshold, the probability value is calibrated by logistic regression, and is mapped to a risk level.

[0019] Optionally, based on the optimization feature set, a cloud-generated personalized management strategy is obtained and delivered to the vehicle-side controller, and the process includes:

[0020] The optimization feature set is compressed by the vehicle-side controller to obtain a compressed feature set; the compressed feature set is encrypted and transmitted to the cloud through data encryption technology, a cloud confirmation signal is obtained, and an upload confirmation state is obtained; if the upload confirmation state is successful, the encrypted feature set is decrypted and analyzed by a pre-trained deep learning model to obtain an analysis feature set; according to the analysis feature set, a dynamic change trend is extracted to obtain a trend feature set; a personalized management strategy is generated through the trend feature set, and a strategy execution parameter is determined; the strategy execution parameter is delivered to the vehicle-side controller through the cloud and an execution confirmation state is obtained, and it is determined whether the strategy deployment is completed.

[0021] Optionally, the process of fusing vehicle-side real-time data and battery aging data through an online learning algorithm, dynamically adjusting risk assessment model parameters, generating an adaptive model and outputting a graded early warning signal includes:

[0022] Battery operation data is obtained by vehicle-side sensors, and data fusion technology is used to integrate the battery operation data to obtain a fused data set; if the fused data set meets a preset condition, the parameters of the risk assessment model are updated using an online learning algorithm, the model weights are adjusted in combination with battery aging data to obtain an adaptive model; according to the adaptive model, real-time dynamic characteristics are extracted from multi-dimensional operation data to determine the safety state of the battery under complex working conditions, and a warning signal is output.

[0023] Optionally, it also includes: obtaining cloud historical operation data, combining real-time dynamic characteristics, using a clustering algorithm to analyze battery aging trends, and generating an aging feature set; according to the aging feature set, long-term prediction parameters of the adaptive model are updated to predict risk probability under future working conditions, and dynamic management instructions are output.

[0024] The application also provides a computer device, including a memory, a processor to store a computer program on the memory and run the computer program on the processor, and the processor executes the computer program to realize the steps of the above method.

[0025] The application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the method.

[0026] Compared with the prior art, the application has the following advantages and technical effects:

[0027] The application discloses an online safety management and early warning method for power batteries of electric vehicles, which comprises the following steps: obtaining multi-dimensional operation data of the batteries, extracting dynamic characteristics under complex working conditions by using data preprocessing and deep learning algorithms, and constructing a risk assessment model.When the risk level is high, the application transmits the optimized feature set to the cloud to obtain a personalized management strategy, and updates the model parameters at the vehicle end by using an online learning algorithm, adjusts the weight in combination with the battery aging data, and realizes self-adaptive optimization. The application also analyzes the battery aging trend by using a clustering algorithm, updates the long-term prediction parameters, realizes dynamic prediction and management of the risk probability under future working conditions, and effectively evaluates the safety risk of the power batteries under complex working conditions, provides real-time early warning and personalized management strategies, and improves the safety and service life of the batteries. BRIEF DESCRIPTION OF DRAWINGS

[0028] The accompanying drawings, which form a part of the present application, are intended to provide further understanding of the present application, and the illustrative embodiments of the present application and their description serve the purpose of explaining the present application. The accompanying drawings should not be construed as an inappropriate limitation on the present application. In the drawings:

[0029] Figure 1 The method flowchart of the embodiments of the application. DETAILED DESCRIPTION

[0030] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0031] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a group of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0032] Embodiment one

[0033] As shown in the figure, the embodiment provides an online safety management and early warning method for power batteries of electric vehicles, which comprises the following steps: Figure 1

[0034] Obtaining multi-dimensional operation data of the power batteries and preprocessing the data to obtain a standardized data set;

[0035] As a specific implementation, the process of obtaining the standardized data set comprises the following steps:

[0036] The multi-dimensional running data is denoised by using a filtering technique to generate a denoised data set; abnormal data exceeding a preset threshold in the denoised data set is marked; and the marked data is normalized to generate a normalized data set; wherein the multi-dimensional running data includes voltage, current, temperature and state of charge.

[0037] Specifically, multi-dimensional running data including voltage data, current data, temperature data and state of charge is obtained from the power battery and stored as an initial data set. Noise is removed from the initial data set by using a filtering technique to generate a denoised data set. If the voltage data or current data in the denoised data set exceeds a preset threshold, the abnormal data is marked to obtain a marked data set. According to the marked data set, the voltage data, current data, temperature data and state of charge are normalized by using a normalization technique to generate a normalized data set.

[0038] Exemplarily, the initial data set is collected by obtaining voltage, current, temperature and state of charge data from the power battery management system. Assuming that a battery pack of an electric vehicle includes 100 single batteries, the data is recorded once per second, the voltage range is 3.0-4.2V, the current is -50 to 200A, the temperature is -20 to 60℃, and the state of charge is 0-100%. These data are transmitted to the cloud in real time through sensors to form a high-frequency initial data set, laying a foundation for subsequent analysis. The collected high-frequency characteristics ensure the comprehensiveness of the data, which helps to capture the subtle changes in the battery operation.

[0039] The filtering technique is used to remove noise to generate a denoised data set. Commonly used median filtering can effectively smooth the sudden noise in voltage and current. For example, a certain battery voltage sequence is 3.6V, 3.61V, 3.8V, 3.62V, and obviously 3.8V is an abnormal spike. The median filtering replaces the spike with the middle value 3.61V, retaining the trend while eliminating noise. The denoised data set is smoother, which can truly reflect the battery operating state and avoid noise interference in subsequent analysis.

[0040] The abnormal data marking is directed to the voltage and current in the denoised data set. The voltage threshold is set to 3.2-4.0V and the current threshold is set to -30 to 180A. If the voltage is 4.1V at a certain time, it is marked as overvoltage abnormality; if the current is 190A, it is marked as overcurrent abnormality. These labels form a marked data set, which is convenient for tracing the root cause of the problem. The abnormal marking improves the data quality and provides a warning basis for battery safety management.

[0041] The standardization technology normalizes four types of data in the labeled dataset. Voltage, current, temperature, and state of charge need to be mapped to the 0-1 interval due to different dimensions. For example, voltage 3.6V in the range of 3.2-4.0V is normalized to (3.6-3.2) / (4.0-3.2)=0.5. The standardized dataset eliminates the dimension effect, making different data comparable and providing a consistent basis for feature extraction.

[0042] Based on the standardized dataset, dynamic working conditions and time series features are obtained to generate a battery state feature set. If the battery state feature set does not meet the preset conditions, the battery state feature set is optimized to obtain an optimized feature set.

[0043] As a specific implementation, the process of generating the battery state feature set includes:

[0044] Based on the standardized dataset, multi-dimensional features are obtained, and convolutional neural networks are used to extract dynamic features under complex working conditions to obtain an initial feature set. Long short-term memory networks are used to analyze the time series characteristics of the initial feature set to obtain a time series feature set. When the time series feature set does not match the preset working condition threshold, the time series feature set is re-clustered and the dynamic features and time series features are fused to generate a comprehensive feature set. According to the comprehensive feature set, convolutional neural networks are used to perform deep analysis on the feature set to extract hidden working condition change features to obtain a deep feature set. If the correlation between the features in the deep feature set is lower than the preset threshold, the feature set is optimized through dimension reduction technology to obtain an optimized feature set. Through the optimized feature set, long short-term memory networks are used to model the long-term dependence of time series to generate a battery state feature set.

[0045] For example, when obtaining multi-dimensional features from the standardized dataset, the diversity and consistency of the data need to be ensured. In the power battery operation scenario, the standardized dataset usually includes voltage, current, temperature, and state of charge dimensions. Assuming that a battery pack of an electric vehicle collects data every second, the voltage range is 3.0-4.2V, the current is -50 to 200A, the temperature is -20 to 60℃, and the state of charge is 0-100%. Key dimensions such as voltage and state of charge are extracted through principal component analysis to generate a multi-dimensional feature set containing main information. This process preserves the core characteristics of the data and lays a foundation for subsequent analysis.

[0046] Convolutional neural networks are used to extract dynamic features under complex working conditions. Convolutional neural networks capture local change patterns of voltage and current through convolution kernels. For example, under acceleration working conditions, the current of a certain battery pack quickly rises from 50A to 150A, and the convolutional neural network identifies this mutation pattern to generate an initial feature set containing dynamic features such as current change rate. This method can effectively capture short-term fluctuations during working condition switching.

[0047] The long short-term memory network analyzes the time sequence characteristics of the initial feature set. The long short-term memory network is good at processing long-term dependencies of time series. For example, a certain battery is in high-temperature working conditions for three consecutive days, the temperature is maintained at 45℃, and the state of charge gradually decreases to 30%. The long short-term memory network extracts the time sequence correlation between temperature and state of charge through the memory gate mechanism, forms a time sequence feature set, and reflects the gradual change trend of the battery.

[0048] If the time sequence feature set does not match the preset working condition threshold, it is re-clustered. For example, the preset high-temperature working condition is temperature higher than 40℃ and state of charge lower than 50%, but a certain feature set shows temperature of 38℃ and state of charge of 20%. Through K-means clustering, the data is divided into high-temperature low-charge and other categories, generating an adjusted feature set. This adjustment ensures that the feature set is more consistent with the actual working condition.

[0049] The feature fusion technology integrates dynamic features and time sequence features. A weighted fusion method is used, with dynamic features such as current rate of change accounting for 40% of the weight, and time sequence features such as temperature trend accounting for 60% of the weight. For example, when a certain battery is accelerating, the current rate of change is high, and the integrated feature set after fusion reflects both transient fluctuations and long-term trends, providing a comprehensive perspective for subsequent analysis.

[0050] The convolutional neural network performs deep analysis on the integrated feature set and extracts hidden working condition change features. For example, when a certain battery starts at low temperature, the voltage drop speed is abnormal, and the convolutional neural network extracts the interaction mode between voltage and temperature through deep convolution, generating a deep feature set and revealing potential working condition abnormalities.

[0051] If the correlation of the deep feature set is lower than the threshold, such as a correlation coefficient less than 0.7, it is optimized through dimensionality reduction technology. Principal component analysis can compress high-dimensional features such as voltage and current to main components. For example, 80% of the variance of the features is retained to generate an optimized feature set, reducing the computational burden of subsequent calculations. For example, the long short-term memory network models the long-term dependencies of time series on the optimized feature set. The time required for a certain battery to cycle from 80% to 20% state of charge gradually shortens, and the long short-term memory network captures this trend to generate a battery state feature set, reflecting the capacity degradation characteristics. This feature set provides accurate basis for battery health management.

[0052] As a specific implementation, the process of obtaining the optimized feature set includes:

[0053] If the voltage fluctuation in the battery state feature set exceeds the preset threshold, the feature set is smoothed through the Kalman filter algorithm to obtain the optimized feature set;

[0054] A risk assessment model is constructed, and a battery safety risk probability is obtained in combination with the optimized feature set, and a risk level is determined based on the battery safety risk probability, the process including:

[0055] The optimization feature set is standardized to extract dynamic change features in the time series to obtain a dynamic feature set. If the signal fluctuation in the dynamic feature set exceeds a preset threshold, the data is segmented by a sliding window technique and statistical distribution characteristics are calculated to obtain a distribution feature set. A classification model is trained using a random forest algorithm to output a risk probability value. When the risk probability value exceeds a threshold, the probability value is calibrated by logistic regression and mapped to a risk level.

[0056] Specifically, according to the optimization feature set, the data is preprocessed using standardization techniques to obtain a standardized feature set. According to the standardized feature set, dynamic change features in the time series are extracted to obtain a dynamic feature set. If the signal fluctuation in the dynamic feature set exceeds a preset threshold, the data is segmented by a sliding window technique to obtain a segmented feature set. According to the segmented feature set, the statistical distribution characteristics of each segment are calculated to obtain a distribution feature set. According to the distribution feature set, a classification model is trained using a random forest algorithm to output a risk probability value to obtain a probability feature set. If the risk probability in the probability feature set exceeds a preset threshold, the probability value is calibrated by a logistic regression algorithm to obtain a calibrated feature set. According to the calibrated feature set, the probability value is mapped to a risk level to determine the final risk level.

[0057] For example, assume that the voltage data range of a certain battery is 3.2 to 4.2 volts, and the current range is 0.5 to 5 amperes. After standardization, the voltage data may be converted to a range of -1.5 to 1.5, and the current data is similarly processed. This approach can eliminate dimensional effects and facilitate subsequent analysis. According to the standardized feature set, dynamic change features in the time series are extracted to obtain a dynamic feature set.

[0058] Specifically, dynamic features can include the rate of change of voltage over time or the abrupt change point of current.

[0059] Suppose a certain battery pack is being monitored, and the voltage change rate is extracted. It is observed that the voltage changes by about 0.01 volts per minute, but at a certain time it increases to 0.05 volts, which is recorded as a dynamic feature. This step focuses on capturing short-term fluctuations and trends in the data. If the signal fluctuation in the dynamic feature set exceeds a preset threshold, the data is segmented by a sliding window technique to obtain a segmented feature set.

[0060] The sliding window technique can divide the time series into fixed-length segments, for example, with a 10-second window and a 2-second slide. Suppose the dynamic features of a certain battery show that the voltage change rate exceeds the threshold of 0.03 volts multiple times. By using the sliding window technique, the data for a day can be divided into hundreds of segments, each containing information about the fluctuations of voltage and current. This segmentation facilitates subsequent detailed analysis. According to the segmented feature set, the statistical distribution characteristics of each segment are calculated to obtain a distribution feature set.

[0061] The statistical distribution characteristics can include mean, variance, and skewness. For example, the mean voltage in a certain segment of data is 3.7 volts, the variance is 0.02, and the skewness is 0.1, indicating that the data distribution is relatively concentrated and slightly skewed to the right. These statistical indicators reflect the regularity of each segment of data and provide a basis for model training. According to the distribution feature set, a classification model is trained using the random forest algorithm, and a risk probability value is output, obtaining a probability feature set.

[0062] The random forest determines the battery state risk through multiple decision trees. Assuming that a certain segment of distribution characteristics is input, the model may output a risk probability of 0.75, indicating that the segment of data corresponds to a high-risk state. The advantage of random forest lies in its modeling capability for complex feature relationships. If the risk probability in the probability feature set exceeds a preset threshold, the probability value is calibrated through a logistic regression algorithm to obtain a calibrated feature set.

[0063] The logistic regression can fine-tune the probability output by the random forest. For example, assuming that the random forest outputs a risk probability of 0.78, which exceeds the threshold of 0.7, the logistic regression may calibrate it to 0.73 based on historical data to improve accuracy. This calibration can optimize the reliability of the probability value. According to the calibrated feature set, the probability value is mapped to a risk level to determine the final risk level.

[0064] Set the probability from 0 to 0.3 as low risk, 0.3 to 0.6 as medium risk, and above 0.6 as high risk. Assuming that the calibrated probability is 0.73, it is mapped to high risk, prompting the need to check the battery. This mapping method is intuitive and easy to apply to actual monitoring scenarios.

[0065] If the risk level exceeds the threshold, an individualized management strategy generated by the cloud is obtained based on the optimized feature set and delivered to the vehicle-side controller.

[0066] As a specific implementation, the process of obtaining an individualized management strategy generated by the cloud based on the optimized feature set and delivering it to the vehicle-side controller includes:

[0067] The vehicle-side controller compresses the optimized feature set to obtain a compressed feature set. The compressed feature set is encrypted and transmitted to the cloud through data encryption technology, and a cloud confirmation signal is obtained to determine the upload confirmation status. If the upload confirmation status is successful, the encrypted feature set is decrypted and analyzed through a pre-trained deep learning model to obtain an analysis feature set. According to the analysis feature set, a dynamic change trend is extracted to obtain a trend feature set. Through the trend feature set, an individualized management strategy is generated, and strategy execution parameters are determined. The strategy execution parameters are delivered to the vehicle-side controller through the cloud, and an execution confirmation status is obtained to determine whether the strategy deployment is complete.

[0068] In one possible implementation, if the risk level exceeds a preset threshold, the vehicle-side controller compresses the optimized feature set to reduce the data transmission burden. The generation of the compressed feature set can be achieved through principal component analysis technology, which retains the main features while reducing the data dimension.

[0069] For example, the original feature set contains more than ten dimensions such as battery temperature, voltage, and current. After compression, only three main components may be retained, and the data volume is reduced by about 70%, which facilitates subsequent processing.

[0070] The compression process needs to ensure that key information is not lost to support the accuracy of subsequent analysis. Specifically, based on the compressed feature set, data encryption technology is used to protect data security, and the AES-256 encryption algorithm can be used to encrypt data. For example, the compressed feature set is converted into an encrypted data packet, and the vehicle-side controller adds a unique identifier to each data packet before transmission to ensure that the data is not tampered with during transmission. This encryption method is particularly important in the Internet of Vehicles scenario, which can effectively prevent data leakage and improve system security.

[0071] The vehicle-side controller uploads the encrypted feature set to the cloud and waits for a confirmation signal from the cloud. For example, the controller sends the data packet to the cloud server through the 4G network, and the server returns an acknowledgement code after receiving it, marked as "upload success". If the network is unstable, the controller can start a retransmission mechanism until it receives a confirmation signal. This mechanism ensures data integrity and provides a reliable basis for subsequent analysis.

[0072] After confirming the success of the upload, the cloud uses a deep learning model to decrypt and analyze the encrypted feature set to generate an analysis feature set. One possible implementation is to use a convolutional neural network model to identify potential patterns in the data. For example, the model can analyze the change of battery voltage over time and detect abnormal fluctuations. The analysis feature set may include indicators such as voltage mutation frequency, providing data support for subsequent trend extraction.

[0073] Based on the analysis feature set, dynamic change trends are extracted to generate a trend feature set. Time series analysis techniques can be used to identify the change rules of battery performance over time. For example, the trend feature set may show that the temperature of a certain battery pack has been rising over the past week, indicating a potential risk. This trend extraction helps to develop more accurate response measures.

[0074] The trend feature set is used to generate individualized management strategies and determine strategy execution parameters. For example, in response to the trend of rising temperature, the strategy may include reducing the charging power or increasing the running time of the cooling system. The parameters can be set to reduce the charging power to 80% and increase the running frequency of the cooling system to once an hour. This strategy can effectively extend the battery life.

[0075] The policy execution parameters are issued to the vehicle-side controller through the cloud and the execution confirmation status is obtained. For example, the cloud encapsulates the parameters as an instruction package and sends it to the controller through an encrypted channel. The controller returns a "deployment complete" status after execution to ensure that the policy is implemented. It should be noted that the controller will verify the parameters before execution to avoid system abnormalities caused by parameter errors.

[0076] The judgment of the execution confirmation status can be achieved through log recording. For example, the controller generates an execution log to record the time and effect of policy execution and uploads it to the cloud for subsequent auditing. This mechanism helps to track the effect of the policy and ensures the stability and reliability of the system.

[0077] Through online learning algorithm, real-time data and battery aging data are fused to dynamically adjust the risk assessment model parameters, generate adaptive model and output graded warning signals.

[0078] As a specific implementation, the process of fusing real-time data and battery aging data through online learning algorithm to dynamically adjust the risk assessment model parameters, generate adaptive model and output graded warning signals includes:

[0079] The battery operation data is obtained through the vehicle-side sensor, and the data fusion technology is used to integrate the battery operation data to obtain a fusion data set. If the fusion data set meets the preset condition, the parameters of the risk assessment model are updated using the online learning algorithm, the model weight is adjusted in combination with the battery aging data, and an adaptive model is obtained. According to the adaptive model, the real-time dynamic characteristics are extracted from the multi-dimensional operation data to judge the safety state of the battery under complex working conditions, and the warning signal is output.

[0080] Specifically, the battery operation data is obtained through the vehicle-side sensor, and the data fusion technology is used to integrate the data to obtain a fusion data set. If the fusion data set meets the preset condition, the parameters of the risk assessment model are updated using the online learning algorithm to obtain an updated parameter set. According to the updated parameter set, the risk assessment model weight is adjusted, and in combination with the battery aging data, an adaptive weight set is obtained. The adaptive weight set is used to optimize the risk assessment model to generate an optimized risk model. Through the optimized risk model, the fusion data set is analyzed to determine the potential risk level to obtain risk level data. If the risk level data is higher than the preset threshold, a dynamic adjustment strategy is generated through the vehicle-side controller to determine the strategy parameters. According to the strategy parameters, the vehicle-side control logic is updated to obtain an adjusted control state. According to the adaptive model, the real-time dynamic characteristics are extracted from the multi-dimensional operation data to judge the safety state of the battery under complex working conditions, and the warning signal is output.

[0081] Further, multi-dimensional operation data is collected by sensors, and a clean data set is generated by data cleaning techniques. If the clean data set meets preset conditions, a dynamic feature set is obtained by processing the clean data set through feature extraction techniques. According to the dynamic feature set, a cluster algorithm is used to analyze the feature distribution to determine the working condition classification data. If the working condition classification data indicates a complex working condition, a safety state data is obtained by processing the dynamic feature set through a state evaluation algorithm. According to the safety state data, a threshold comparison technique is used to determine whether to trigger a warning, and a warning signal is obtained. The control parameter set is updated through the warning signal to determine the adjusted control state.

[0082] For example, the sensors can include temperature sensors, voltage sensors, and current sensors, which collect the operation state data of the battery pack in real time. For example, the temperature sensor records the battery surface temperature, the voltage sensor monitors the single battery voltage, and the current sensor captures the charging and discharging current. These data form a multi-dimensional data set, providing comprehensive information for subsequent analysis. The collection of multi-dimensional data needs to ensure time synchronization to ensure data consistency. The key step is to generate a clean data set by using data cleaning techniques. Data cleaning aims to remove noise and outliers.

[0083] Temperature data can be processed by a sliding window average method to remove sudden values caused by sensor jitter. For example, it is found that the temperature data jumps from 25 degrees to 50 degrees in a short time, and the cleaning algorithm identifies it as an outlier, which is replaced by the average value of the previous and subsequent data. In addition, missing values can be supplemented by linear interpolation. This way ensures the integrity and accuracy of the clean data set, laying a foundation for subsequent feature extraction. If the clean data set meets the preset conditions, a dynamic feature set is generated by feature extraction techniques.

[0084] The preset condition can be that the data integrity reaches more than 95%. The feature extraction technique can select the principal component analysis method to reduce the multi-dimensional data to key features. For example, the cycle number and internal resistance change rate are extracted from the voltage, current, and temperature data as dynamic features. These features reflect the running state and potential degradation trend of the battery.

[0085] Feature extraction can also combine time series analysis to capture the changing rules of data over time. The use of a clustering algorithm to analyze the feature distribution to determine the working condition classification data is a further refinement of the analysis process. The K-means clustering algorithm can be used to divide the dynamic feature set into normal working conditions, light load working conditions, and heavy load working conditions. For example, a battery pack with low internal resistance change rate and stable temperature is classified as normal working condition; while the internal resistance rises rapidly and the temperature is high, it is classified as heavy load working condition. The clustering results help to identify the complexity of the vehicle operating environment and provide a basis for state evaluation. If the working condition classification data indicates a complex working condition, a safety state data is generated by processing the dynamic feature set through a state evaluation algorithm.

[0086] The state evaluation algorithm can be based on fuzzy logic, integrating features such as internal resistance, temperature, and cycle number, to output a safety state score. For example, a score below 60 indicates potential risk. Such a scoring mechanism intuitively reflects the battery health status, facilitating subsequent decision-making. The core of the decision-making process is to determine whether to trigger a warning using threshold comparison techniques. For example, a safety state score below 60 or a temperature exceeding 45 degrees triggers a warning signal.

[0087] The warning signal can be hierarchical, with a score of 50 to 60 indicating a low-level warning and a score below 50 indicating a high-level warning. This hierarchical design facilitates the implementation of differentiated response measures. The final step is to update the control parameter set and determine the adjusted control state through the warning signal.

[0088] The control parameter set includes the charging rate and the upper limit of power output. For example, when a high-level warning is triggered, the charging rate is reduced to 0.5C and the upper limit of power output is reduced by 20%. Such adjustments can respond to potential risks in a timely manner and ensure stable vehicle operation.

[0089] As a specific implementation, it also includes obtaining cloud historical operation data, combining real-time dynamic characteristics, using clustering algorithm to analyze battery aging trend, and generating aging feature set; updating the long-term prediction parameters of the adaptive model according to the aging feature set, predicting the risk probability under future working conditions, and outputting dynamic management instructions.

[0090] Specifically, the historical operation data of the battery is obtained from the cloud, and the historical operation data is processed using data cleaning techniques to obtain a cleaned data set. According to the cleaned data set, real-time dynamic characteristics are obtained, and the cleaned data set and dynamic characteristics are processed using feature fusion techniques to obtain a fusion feature set. If the fusion feature set meets the preset conditions, the clustering algorithm is used to analyze the fusion feature set to obtain aging trend data. According to the aging trend data, the feature extraction technique is used to process the aging trend data to obtain the aging feature set. If the aging feature set exceeds the range after comparison with the preset threshold, the state evaluation algorithm is used to process the aging feature set to obtain battery state data. According to the battery state data, the threshold comparison technique is used to determine the battery state data, and the control parameter adjustment set is determined. According to the control parameter adjustment set, the operation control strategy is updated, and the adjusted control state is generated.

[0091] The feature data is acquired from the aging feature set, the aging feature set is processed by using a feature decomposition technique to obtain a decomposition feature set. If the decomposition feature set meets a preset condition, the decomposition feature set is processed by using a parameter adjustment technique to obtain updated model parameters. Future working condition data is acquired according to the updated model parameters, the updated model parameters and the future working condition data are processed by using a probability prediction technique to obtain a risk probability distribution. If the risk probability distribution exceeds a preset threshold, the risk probability distribution is processed by using an instruction generation technique to obtain a dynamic management instruction set. The dynamic management instruction set is processed by using an instruction optimization technique to obtain an optimized instruction set according to the dynamic management instruction set. The running control strategy is updated by using the optimized instruction set to obtain an adjusted control state. The adjusted control state is processed by using a state feedback technique to obtain a feedback feature set according to the adjusted control state.

[0092] The embodiment further provides a computer device, comprising a memory, a processor to store a computer program on the memory and run the computer program on the processor, and the processor executes the computer program to implement the steps of the above method.

[0093] The embodiment further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the above method.

[0094] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An online safety management and early warning method for electric vehicle power batteries, characterized in that, The method comprises the following steps: acquiring multi-dimensional operation data of the power battery and preprocessing the data to obtain a standardized data set; obtaining dynamic working conditions and time sequence characteristics based on the standardized data set, generating a battery state feature set, and optimizing the battery state feature set if the battery state feature set does not meet preset conditions to obtain an optimized feature set; constructing a risk assessment model, combining the optimized feature set to obtain a battery safety risk probability, and judging a risk level based on the battery safety risk probability; if the risk level exceeds a threshold, obtaining a cloud-generated personalized management strategy based on the optimized feature set and delivering the strategy to a vehicle-side controller; fusing vehicle-side real-time data and battery aging data through an online learning algorithm, dynamically adjusting parameters of the risk assessment model, generating an adaptive model, and outputting a graded early warning signal.

2. The online safety management and early warning method for the power battery of the electric vehicle according to claim 1, wherein the process of obtaining the standardized data set comprises: using a filtering technique to denoise the multi-dimensional operation data to generate a denoised data set, marking abnormal data in the denoised data set that exceeds a preset threshold, and standardizing the marked data to generate the standardized data set; wherein the multi-dimensional operation data comprises voltage, current, temperature and state of charge.

3. The online safety management and early warning method for the power battery of the electric vehicle according to claim 1, wherein the process of generating the battery state feature set comprises: obtaining multi-dimensional features based on the standardized data set, using a convolutional neural network to extract dynamic features under complex working conditions to obtain an initial feature set, using a long short-term memory network to analyze time sequence characteristics of the initial feature set to obtain a time sequence feature set, re-clustering the time sequence feature set and fusing dynamic features and time sequence features when the time sequence feature set does not match a preset working condition threshold to generate a comprehensive feature set, using a convolutional neural network to deeply analyze the comprehensive feature set to extract hidden working condition change features to obtain a deep feature set, and optimizing the feature set through dimension reduction technology if the correlation between features in the deep feature set is lower than a preset threshold to obtain the optimized feature set, and using a long short-term memory network to model long-term dependence of time sequence based on the optimized feature set to generate the battery state feature set.

4. The online safety management and early warning method for the power battery of the electric vehicle according to claim 2, wherein the process of obtaining the optimized feature set comprises: if voltage fluctuation in the battery state feature set exceeds a preset threshold, smoothing the feature set through a Kalman filtering algorithm to obtain the optimized feature set.

5. The online safety management and early warning method for the power battery of the electric vehicle according to claim 1, wherein the process of constructing the risk assessment model, combining the optimized feature set to obtain the battery safety risk probability, and judging the risk level based on the battery safety risk probability comprises: ​ ​ ​ ​ The optimization feature set is standardized to extract dynamic change features in the time series to obtain a dynamic feature set; if the signal fluctuation in the dynamic feature set exceeds a preset threshold, the data is segmented by a sliding window technique and statistical distribution characteristics are calculated to obtain a distribution feature set; a classification model is trained using a random forest algorithm to output a risk probability value; when the risk probability value exceeds a threshold, the probability value is calibrated by logistic regression and mapped to a risk level.

6. The online safety management and early warning method for electric vehicle power batteries according to claim 1, characterized in that, The process of obtaining a cloud-generated personalized management strategy based on the optimization feature set and delivering it to the vehicle-side controller includes: The optimization feature set is compressed by the vehicle-side controller to obtain a compressed feature set; the compressed feature set is encrypted and transmitted to the cloud through data encryption technology to obtain a cloud confirmation signal and an upload confirmation status; if the upload confirmation status is successful, the encrypted feature set is decrypted and analyzed by a pre-trained deep learning model to obtain an analysis feature set; the dynamic change trend is extracted from the analysis feature set to obtain a trend feature set; the personalized management strategy is generated based on the trend feature set, and the strategy execution parameters are determined; the strategy execution parameters are delivered to the vehicle-side controller through the cloud and an execution confirmation status is obtained to determine whether the strategy deployment is complete.

7. The online safety management and early warning method for electric vehicle power batteries according to claim 1, characterized in that, The process of dynamically adjusting the risk assessment model parameters by fusing the vehicle-side real-time data and battery aging data through an online learning algorithm, generating an adaptive model, and outputting a graded early warning signal includes: The battery operation data is obtained through vehicle-side sensors, and the data fusion technology is used to integrate the battery operation data to obtain a fused data set; if the fused data set meets the preset conditions, the parameters of the risk assessment model are updated using the online learning algorithm, the model weights are adjusted in combination with the battery aging data to obtain an adaptive model; the real-time dynamic characteristics are extracted from the multi-dimensional operation data based on the adaptive model to determine the safety state of the battery under complex working conditions and output the early warning signal.

8. The method of claim 1, wherein, Further comprising: Obtaining cloud historical operation data, combining real-time dynamic characteristics, using clustering algorithm to analyze battery aging trend, generating aging feature set; Updating the long-term prediction parameters of the adaptive model based on the aging feature set, predicting the risk probability under future working conditions, and outputting dynamic management instructions.

9. A computer apparatus comprising: A memory and a processor to store a computer program stored on the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of the method of any one of claims 1-8.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1-8.

Images