Intelligent automobile-oriented driving behavior and working condition coupling durability health management system

Abstract

The invention relates to a driving behavior and working condition coupling durability health management system for an intelligent automobile, and belongs to the technical field of durability management of a power system of the intelligent automobile. The system comprises a coupling scene modal division unit which is used for acquiring driving behaviors and working condition characteristics, dynamically dividing coupling modals through preprocessing and clustering analysis, and presetting an attenuation incidence relation; the multi-source data lightweight fusion unit outputs a high-quality data set through feature-level pre-fusion and decision-level lightweight fusion; the coupling factor decoupling and nonlinear prediction unit peels off independent influence factors, retains coupling residual factors, and calculates and dynamically updates a life prediction value by means of a nonlinear model; and the prediction control linkage health management unit divides health levels, outputs adaptive control strategies and corrects and optimizes parameters and strategies through feedback. The method achieves the precise recognition of a driving behavior and working condition coupling scene, improves the multi-source data fusion efficiency and the durability prediction precision, and guarantees the operation reliability and safety of a power system.

Description

Technical Field

[0001] This invention belongs to the field of intelligent vehicle powertrain durability management technology, specifically relating to a durability and health management system that couples driving behavior and operating conditions for intelligent vehicles. Background Technology

[0002] As the automotive industry undergoes a profound transformation towards electrification and intelligence, intelligent electric vehicles have become the mainstream of global transportation development. The durability of their core power system (mainly including the battery pack and electric drive system) directly determines the reliability, safety, and total life-cycle cost of the vehicle, and is one of the key indicators for measuring the core competitiveness of a vehicle. Compared with traditional fuel vehicles, the operating status of the power components of intelligent electric vehicles is more susceptible to the coupling effects of complex environments and dynamic operating conditions. In particular, the coupling effect formed by the randomness of driving behavior (such as rapid acceleration, rapid deceleration, and frequent start-stop) and the diversity of driving conditions (such as urban traffic congestion, highway cruising, hill driving, and extreme temperature and humidity environments) will significantly aggravate the battery pack charge-discharge cycle losses, electric drive system mechanical wear and thermal degradation, resulting in a strong nonlinear characteristic of component degradation rate, which poses a severe challenge to the durability management of the power system.

[0003] Currently, the automotive industry's demand for powertrain durability management has shifted from the traditional passive maintenance model to a proactive predictive maintenance model, aiming to mitigate component failure risks in advance through real-time monitoring and accurate prediction. To this end, related technical fields have conducted a series of explorations and proposed durability management solutions based on sensor monitoring, data analysis, and model prediction. Existing technologies generally collect sensor data (such as battery voltage, current, and temperature, electric drive system speed, torque, and vibration), driving behavior-related data, and basic operating condition data during vehicle operation, and combine them with durability prediction models to assess the remaining life of components and formulate maintenance or control strategies accordingly.

[0004] However, existing technologies still have many technical shortcomings that need to be addressed when dealing with the durability management requirements of scenarios where driving behavior and operating conditions are coupled. These shortcomings are manifested in the following aspects: Firstly, existing solutions mostly consider the impact of driving behavior or operating conditions on durability performance separately, failing to fully explore the characteristic correlation under the coupling effect of the two, resulting in a bias in the understanding of component degradation mechanism, which in turn affects the accuracy of durability prediction. Studies have shown that different coupling combinations of driving behavior and operating conditions will produce different component degradation effects. For example, the impact of rapid acceleration behavior under high-speed conditions on battery pack degradation is much greater than that of smooth driving under urban congestion conditions, and existing technologies lack targeted analysis for such coupling scenarios. Secondly, there are obvious shortcomings in the multi-source data fusion process. On the one hand, it fails to effectively filter redundant information and noise interference in sensor data, resulting in inconsistent data quality. On the other hand, the fusion process often uses complex algorithms, which are difficult to adapt to the lightweight requirements of real-time vehicle monitoring. Furthermore, the fused dataset does not fully integrate the coupling characteristics of driving behavior and operating conditions, and cannot provide comprehensive data support for subsequent durability analysis.

[0005] In summary, current durability management technologies for intelligent electric vehicle power systems are insufficient to effectively address the complex challenges arising from the coupling of driving behavior and operating conditions. These technologies suffer from issues such as inaccurate data fusion, inadequate mining of coupling features, poor adaptability of prediction models, and imperfect health management closed loops. Consequently, these technologies result in low durability prediction accuracy and weak targeting of control strategies, failing to meet the high durability performance requirements of intelligent electric vehicles for their power systems. Summary of the Invention

[0006] To address the aforementioned problems in the existing technology, this invention provides a driving behavior and operating condition coupled durability and health management system for intelligent vehicles. The objective of this invention can be achieved through the following technical solutions: include: The coupled scenario modal segmentation unit acquires driving behavior features and operating condition features, integrates the two types of features to form a multi-dimensional feature vector; performs cluster analysis on the multi-dimensional feature vector, dynamically identifies and segments coupled scenario modes based on the clustering results; for each coupled scenario mode, presets the correlation between coupled features and the attenuation rate of battery pack and electric drive system components to characterize the durability impact law under different coupled scenarios. The multi-source data lightweight fusion unit performs hierarchical fusion processing on the acquired real-time sensor data, driving behavior data, and operating condition data. First, feature-level pre-fusion is performed to filter redundant information and suppress noise in the sensor data and extract key features. Then, decision-level lightweight fusion is performed to integrate the pre-processed sensor data with the driving behavior data and operating condition data and output a fused dataset. The coupling factor decoupling and nonlinear prediction unit, based on the fused dataset, separates the independent influencing factors of each factor on the durability degradation of the component, and retains the coupling residual factors of multiple factors; constructs a nonlinear durability prediction model, and calculates the life prediction values ​​of the battery pack and electric drive system based on the independent influencing factors and coupling residual factors; and dynamically updates the input factors and adjusts the life prediction values ​​according to the real-time fused data. The predictive control linkage health management unit classifies the health levels of the battery pack and electric drive system based on the predicted lifespan value and preset health judgment rules; outputs corresponding control strategies for different health levels; acquires component status data after the control strategy is executed, uses it as feedback information to correct relevant operating parameters, and optimizes subsequent control strategies.

[0007] Specifically, the process of integrating the two types of features to form a multi-dimensional feature vector is as follows: First, outlier removal and standardization preprocessing are performed on the acquired driving behavior features and operating condition features, respectively. Unify the numerical range and time dimension benchmark of the two types of features, and sort them according to the preset feature priority rules; The two types of preprocessed features are sequentially concatenated dimensionally to form a multidimensional feature vector that can characterize the coupling characteristics between driving behavior and operating conditions.

[0008] Specifically, the process of dynamically identifying and classifying coupled scene modalities based on clustering results is as follows: After statistical clustering, the commonalities and dispersion of the feature distribution of each data cluster are analyzed, and abnormal data clusters are removed by verifying the feature similarity within the cluster. Based on the actual driving scenario characteristics of intelligent vehicles, a matching rule for scenario type and data cluster characteristics is constructed; Based on the matching results, the coupling scenario type corresponding to each data cluster is identified, and finally the coupling scenario modality is dynamically divided according to the differences in scenario type.

[0009] Specifically, the process of establishing the correlation between the preset coupling characteristics and the attenuation rate of the battery pack and electric drive system components is as follows: based on the historical operating data and attenuation records of the key power components throughout their entire life cycle, the inherent correlation between coupling characteristics and component attenuation rate under different coupling scenario modes is explored through correlation analysis; based on the correlation rules, correlation thresholds and corresponding rules are set to establish a correlation correspondence system between coupling characteristics and component attenuation rate.

[0010] Specifically, the process of performing feature-level pre-fusion is as follows: the acquired real-time sensor data is filtered for validity, and invalid data that exceeds the normal range is removed; an adaptive noise reduction algorithm is used to suppress noise interference in the data; based on the component durability impact factor analysis, core sensitive features are extracted from the processed sensor data to form a feature-level pre-fusion result.

[0011] Specifically, the process of performing decision-level lightweight fusion is as follows: the feature-level pre-fusion results, driving behavior data and operating condition data are time-stamp aligned to eliminate data time sequence deviations; a data importance assessment system is established, and the weight ratio of various types of data to the vehicle durability analysis is preset; according to the weight ratio, a lightweight fusion algorithm is used to integrate various types of data in sequence to generate a fusion dataset that can reflect the vehicle's operating status.

[0012] Specifically, the process of separating the independent influencing factors of each factor on component durability degradation and retaining the coupled residual factors of multiple factors is as follows: The fused dataset is processed by a multi-factor decoupling algorithm to directly separate the independent influencing factors of the three core factors of driving behavior, working conditions, and operating conditions on the durability degradation of components. Simultaneously extract the superimposed impact of the interaction of three core factors; The superimposed influence portion is retained as a coupling residual factor of multiple factors.

[0013] Specifically, the process of constructing the nonlinear durability prediction model is as follows: It includes an input layer, a non-linear mapping layer, and an output layer; The input consists of historical independent influencing factors, coupled residual factors, and corresponding key power component attenuation data; the output is the time-series prediction results of the attenuation rate of key power components. The input layer performs dimension alignment and standardization preprocessing on the received feature data; the nonlinear mapping layer adopts a multilayer perceptron architecture and embeds nonlinear activation functions to perform layer-by-layer iterative transformation on the preprocessed features; the output layer integrates and operates on the mapped features to generate temporal prediction results.

[0014] Specifically, the process for calculating the predicted lifespan of the battery pack and electric drive system is as follows: The independent influencing factors and coupled residual factors obtained in real time are input into the nonlinear durability prediction model, and the attenuation trend data of key power components are calculated. By combining the initial design life parameters of the components with historical degradation data, the remaining life prediction value of the key power components is calculated by using the trend extrapolation method.

[0015] Specifically, the process of classifying the health levels of the battery pack and electric drive system is as follows: based on the preset health judgment rules and combined with the performance parameter thresholds of key power components, multiple gradient health level ranges and corresponding judgment thresholds are set. The calculated life prediction values ​​are compared with the judgment thresholds of each level one by one, and the health level of the key power components is determined based on the comparison results.

[0016] Specifically, the process of outputting control strategies corresponding to different health levels is as follows: a matching strategy library of health levels and control strategies is pre-built. The matching strategy library contains strategy types and basic parameters corresponding to each health level. Based on the health level of the key power components, the corresponding matching control strategy is retrieved, and the basic parameters of the strategy are refined and calibrated in combination with real-time operating data.

[0017] Specifically, the process of dynamically updating the input factors based on real-time fusion data is as follows: acquire the newly generated fusion dataset in real time, establish a factor change monitoring mechanism, and track the numerical changes of independent influencing factors and coupled residual factors in real time; when the factor numerical change exceeds the preset fluctuation range, replace the old factor data with the factor data corresponding to the newly generated fusion dataset, and synchronously input the updated factors into the nonlinear durability prediction model.

[0018] The beneficial effects of this invention are as follows: (1) By setting up a coupled scenario modality division unit and a multi-source data lightweight fusion unit, it is possible to accurately mine the coupled features of driving behavior and working conditions and dynamically divide the coupled scenario modality. At the same time, it can realize the hierarchical lightweight fusion of multi-source data, effectively solving the problems of insufficient coupled feature mining, inaccurate data fusion and difficulty in adapting to real-time monitoring requirements in the existing technology. On the one hand, by standardizing the preprocessing and dimension splicing of the two types of features, combined with cluster analysis, the dynamic identification of coupled scenarios is realized, ensuring the pertinence and rationality of scenario division. On the other hand, by using feature-level pre-fusion for redundant filtering, noise suppression and decision-level weighted lightweight fusion, the data quality and fusion efficiency are improved, providing comprehensive and high-quality data support for subsequent durability prediction, thereby improving the basic data reliability of the entire durability management system. (2) By setting up a coupling factor decoupling and nonlinear prediction unit and a predictive control linkage health management unit, the accurate prediction and closed-loop health management of component durability decay are realized, which effectively makes up for the defects of poor adaptability of prediction model and lack of specificity and closed-loop in health management in the existing technology; the coupling factor decoupling algorithm can separate independent influencing factors and coupling residual factors, and with the three-layer nonlinear prediction model, it can adapt to the nonlinear decay law under coupling scenario, which significantly improves the accuracy of life prediction; at the same time, the gradient health level classification and matching control strategy output based on the prediction results, combined with the real-time feedback correction mechanism, form a complete closed loop of "monitoring-prediction-control-correction", which can output accurate and adaptable control strategies for different health states and coupling scenarios, ensuring the reliability and safety of the power system operation and reducing the cost of use throughout the entire life cycle. Attached Figure Description

[0019] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0020] Figure 1 This is a system architecture diagram of a durability and health management system for driving behavior and operating conditions coupled in intelligent vehicles according to the present invention. Figure 2 This is a data flow diagram of a durability and health management system for intelligent vehicles that couples driving behavior and operating conditions, according to the present invention. Detailed Implementation

[0021] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0022] Please see Figure 1-2 A durability and health management system that couples driving behavior and operating conditions for intelligent vehicles; include: The coupled scenario modal segmentation unit acquires driving behavior features and operating condition features, integrates the two types of features to form a multi-dimensional feature vector; performs cluster analysis on the multi-dimensional feature vector, dynamically identifies and segments coupled scenario modes based on the clustering results; for each coupled scenario mode, presets the correlation between coupled features and the attenuation rate of battery pack and electric drive system components to characterize the durability impact law under different coupled scenarios. The multi-source data lightweight fusion unit performs hierarchical fusion processing on the acquired real-time sensor data, driving behavior data, and operating condition data. First, feature-level pre-fusion is performed to filter redundant information and suppress noise in the sensor data and extract key features. Then, decision-level lightweight fusion is performed to integrate the pre-processed sensor data with the driving behavior data and operating condition data and output a fused dataset. The coupling factor decoupling and nonlinear prediction unit, based on the fused dataset, separates the independent influencing factors of each factor on the durability degradation of the component, and retains the coupling residual factors of multiple factors; constructs a nonlinear durability prediction model, and calculates the life prediction values ​​of the battery pack and electric drive system based on the independent influencing factors and coupling residual factors; and dynamically updates the input factors and adjusts the life prediction values ​​according to the real-time fused data. The predictive control linkage health management unit classifies the health levels of the battery pack and electric drive system based on the predicted lifespan value and preset health judgment rules; outputs corresponding control strategies for different health levels; acquires component status data after the control strategy is executed, uses it as feedback information to correct relevant operating parameters, and optimizes subsequent control strategies.

[0023] In this embodiment, driving behavior characteristics directly reflect the quantitative characteristics of the driver's operating habits and behaviors, including the rate of change of accelerator pedal opening, brake pedal trigger frequency, steering operation amplitude and frequency, parking operation frequency, energy recovery mode selection preference, and parameters related to the smoothness of vehicle speed adjustment; operating condition characteristics reflect the characteristics of the vehicle's driving environment and operating status, including road condition type, ambient temperature and humidity, altitude, road gradient, vehicle load rate, energy consumption status corresponding to remaining range, charging frequency and charging time.

[0024] In this embodiment, the independent influencing factors specifically include three categories: independent influencing factors related to driving behavior, such as the quantified independent impact of frequent deep pressing of the accelerator pedal on the battery charge / discharge rate and the quantified independent impact of frequent braking on the torque mutation of the electric drive system; independent influencing factors related to operating conditions, such as the quantified independent impact of high temperature environment on battery thermal degradation and the quantified independent impact of steep slope conditions on the load of the electric drive system; and independent influencing factors related to vehicle condition, such as the quantified independent impact of initial battery capacity on the degradation rate and the quantified independent impact of gear wear degree of the electric drive system on operating resistance.

[0025] In this embodiment, the coupling residual factors specifically include: the quantified value of the superimposed impact on battery pack cycle degradation under the coupling effect of rapid acceleration driving behavior and high temperature conditions; the quantified value of the superimposed impact on mechanical wear of the electric drive system under the coupling effect of frequent braking behavior and steep slope conditions; and the quantified value of the superimposed impact on battery activity and electric drive system efficiency under the coupling effect of high load vehicle conditions and low temperature conditions.

[0026] In this embodiment, the operation-related parameters are the core parameters that support the closed-loop operation of the system of the present invention and need to be dynamically corrected according to the feedback data. Specifically, they include two categories: model operation parameters: the network weights of each layer of the nonlinear durability prediction model, the factor change fluctuation threshold, the health level determination threshold, and the correlation threshold between coupling characteristics and component decay rate; control strategy execution parameters: the battery charge and discharge rate limit value corresponding to different health levels, the upper limit value of electric drive system torque output, energy recovery intensity adjustment parameters, and temperature control system start and stop threshold.

[0027] In this embodiment, the core sensitive features specifically include the following three categories: electrical signal core sensitive features: standard deviation of individual cell voltage of the battery pack, charging and discharging current fluctuation value, voltage balance parameter; speed fluctuation value of electric drive system, output torque change rate, three-phase current imbalance; physical signal core sensitive features: temperature distribution uniformity parameter of battery pack, temperature change rate; vibration amplitude and frequency characteristics of electric drive system, temperature gradient of shell surface; operating condition related core sensitive features: battery voltage stability parameter under extreme temperature and humidity environment, load fluctuation characteristics of electric drive system under high speed / congestion conditions, transient response parameters of power components under rapid acceleration and deceleration scenarios.

[0028] Specifically, the process of integrating the two types of features to form a multi-dimensional feature vector is as follows: First, outlier removal and standardization preprocessing are performed on the acquired driving behavior features and operating condition features, respectively. Unify the numerical range and time dimension benchmark of the two types of features, and sort them according to the preset feature priority rules; The two types of preprocessed features are sequentially concatenated dimensionally to form a multidimensional feature vector that can characterize the coupling characteristics between driving behavior and operating conditions.

[0029] Specifically, the process of dynamically identifying and classifying coupled scene modalities based on clustering results is as follows: After statistical clustering, the commonalities and dispersion of the feature distribution of each data cluster are analyzed, and abnormal data clusters are removed by verifying the feature similarity within the cluster. Based on the actual driving scenario characteristics of intelligent vehicles, a matching rule for scenario type and data cluster characteristics is constructed; Based on the matching results, the coupling scenario type corresponding to each data cluster is identified, and finally the coupling scenario modality is dynamically divided according to the differences in scenario type.

[0030] Specifically, the process of establishing the correlation between the preset coupling characteristics and the attenuation rate of the battery pack and electric drive system components is as follows: based on the historical operating data and attenuation records of the key power components throughout their entire life cycle, the inherent correlation between coupling characteristics and component attenuation rate under different coupling scenario modes is explored through correlation analysis; based on the correlation rules, correlation thresholds and corresponding rules are set to establish a correlation correspondence system between coupling characteristics and component attenuation rate.

[0031] Specifically, the process of performing feature-level pre-fusion is as follows: the acquired real-time sensor data is filtered for validity, and invalid data that exceeds the normal range is removed; an adaptive noise reduction algorithm is used to suppress noise interference in the data; based on the component durability impact factor analysis, core sensitive features are extracted from the processed sensor data to form a feature-level pre-fusion result.

[0032] Specifically, the process of performing decision-level lightweight fusion is as follows: the feature-level pre-fusion results, driving behavior data and operating condition data are time-stamp aligned to eliminate data time sequence deviations; a data importance assessment system is established, and the weight ratio of various types of data to the vehicle durability analysis is preset; according to the weight ratio, a lightweight fusion algorithm is used to integrate various types of data in sequence to generate a fusion dataset that can reflect the vehicle's operating status.

[0033] Specifically, the process of separating the independent influencing factors of each factor on component durability degradation and retaining the coupled residual factors of multiple factors is as follows: The fused dataset is processed by a multi-factor decoupling algorithm to directly separate the independent influencing factors of the three core factors of driving behavior, working conditions, and operating conditions on the durability degradation of components. Simultaneously extract the superimposed impact of the interaction of three core factors; The superimposed influence portion is retained as a coupling residual factor of multiple factors.

[0034] Specifically, the process of constructing the nonlinear durability prediction model is as follows: It includes an input layer, a non-linear mapping layer, and an output layer; The input consists of historical independent influencing factors, coupled residual factors, and corresponding key power component attenuation data; the output is the time-series prediction results of the attenuation rate of key power components. The input layer performs dimension alignment and standardization preprocessing on the received feature data; the nonlinear mapping layer adopts a multilayer perceptron architecture and embeds nonlinear activation functions to perform layer-by-layer iterative transformation on the preprocessed features; the output layer integrates and operates on the mapped features to generate temporal prediction results.

[0035] Specifically, the process for calculating the predicted lifespan of the battery pack and electric drive system is as follows: The independent influencing factors and coupled residual factors obtained in real time are input into the nonlinear durability prediction model, and the attenuation trend data of key power components are calculated. By combining the initial design life parameters of the components with historical degradation data, the remaining life prediction value of the key power components is calculated by using the trend extrapolation method.

[0036] Specifically, the process of classifying the health levels of the battery pack and electric drive system is as follows: based on the preset health judgment rules and combined with the performance parameter thresholds of key power components, multiple gradient health level ranges and corresponding judgment thresholds are set. The calculated life prediction values ​​are compared with the judgment thresholds of each level one by one, and the health level of the key power components is determined based on the comparison results.

[0037] Specifically, the process of outputting control strategies corresponding to different health levels is as follows: a matching strategy library of health levels and control strategies is pre-built. The matching strategy library contains strategy types and basic parameters corresponding to each health level. Based on the health level of the key power components, the corresponding matching control strategy is retrieved, and the basic parameters of the strategy are refined and calibrated in combination with real-time operating data.

[0038] Specifically, the process of dynamically updating the input factors based on real-time fusion data is as follows: acquire the newly generated fusion dataset in real time, establish a factor change monitoring mechanism, and track the numerical changes of independent influencing factors and coupled residual factors in real time; when the factor numerical change exceeds the preset fluctuation range, replace the old factor data with the factor data corresponding to the newly generated fusion dataset, and synchronously input the updated factors into the nonlinear durability prediction model.

[0039] In this embodiment, the specific operations of the hierarchical fusion processing are as follows: In the feature-level pre-fusion stage, an adaptive Kalman filter algorithm is used to suppress noise in real-time sensor data. The redundancy information filtering threshold is set to ±3 times the standard deviation, and invalid data such as voltage, current, and temperature that exceed this range are removed. Then, the feature importance is calculated using the random forest algorithm, and core sensitive features with an importance score ≥0.8 (such as the standard deviation of battery pack cell voltage, the speed fluctuation value of electric drive system, etc.) are extracted to form the feature-level pre-fusion result. In the decision-level lightweight fusion stage, a synchronization algorithm with a timestamp alignment accuracy of ±10ms is used to eliminate the time sequence deviation of the three types of data. A data importance evaluation system is established, with a preset weighting of 60% for sensor data, 25% for driving behavior data, and 15% for vehicle condition data. The data is integrated through a weighted average lightweight fusion algorithm to output a fused dataset. The delay of the entire fusion process is controlled within a preset time to meet the real-time monitoring requirements.

[0040] In this embodiment, the specific steps for constructing the nonlinear durability prediction model are as follows: The input layer receives 32-dimensional feature data (including 12-dimensional historical independent influencing factors, 8-dimensional coupling residual factors, and 12-dimensional corresponding key power component decay data), and uses Z-score standardization to map the feature data to the [-1,1] interval to complete dimension alignment; the nonlinear mapping layer sets two hidden layers, with 64 neurons in the first layer and 32 neurons in the second layer, both embedded with the ReLU activation function, using He normality to initialize the weights, and using a Dropout layer (dropout probability 0.2) to suppress overfitting; the output layer uses a linear activation function to output the predicted component decay rate values ​​for the next 10 time series points; during model training, 100,000 historical running data are selected to form a sample set, and data augmentation techniques (random translation and scaling) are used to expand the sample to 150,000; the Adam optimizer is used, with the initial learning rate set to 0.001, decaying by 10% every 100 rounds, and iterative training for 500 rounds. Training stops when the mean squared error (MSE) of the validation set is ≤0.005.

[0041] In this embodiment, a four-gradient classification standard is adopted for classifying the health levels of the battery pack and electric drive system. The specific process is as follows: Based on the design specifications and industry standards of key power components, preset health level judgment thresholds are set, with the remaining capacity of the battery pack as the core indicator and the power decay rate of the electric drive system as the core indicator; Health Level 1 (Optimal): Battery pack remaining capacity ≥ 85%, electric drive system power decay rate ≤ 10%; Health Level 2 (Good): Battery pack remaining capacity 70%-85%, electric drive system power decay rate 10%-18%; Health Level 3 (Decayed): Battery pack remaining capacity 55%-70%, electric drive system power decay rate 18%-25%; Health Level 4 (Severe Decayed): Battery pack remaining capacity < 55%, electric drive system power decay rate > 25%; The calculated life prediction value is converted into the corresponding core indicator value and compared with the threshold of each level one by one to determine the health level of the component, with the comparison accuracy controlled within ±2%.

[0042] In this embodiment, the control strategies corresponding to different health levels are output as follows: a strategy matching library is pre-built, and differentiated parameters are set according to the characteristics of each level; at health level one, a conventional control strategy is output, with the battery charge / discharge rate limited to ≤1C and the upper limit of the electric drive system torque output being 100% of the rated value; at health level two, a mild protection strategy is output, with the battery charge / discharge rate limited to ≤0.8C, the upper limit of the electric drive system torque output reduced to 90% of the rated value, and the energy recovery intensity adjusted to 30%-40%; at health level three, a moderate protection strategy is output, activating the battery temperature control priority mode to control the battery operating temperature at 25-40℃, limiting the electric drive system to continuous high-load operation for ≤10 minutes, and the charge / discharge rate to ≤0.6C; at health level four, an emergency protection strategy is output, limiting the vehicle's maximum speed to ≤60km / h, prohibiting fast charging, triggering in-vehicle alarm prompts and pushing maintenance information, while limiting the electric drive system power to within 80% of the rated value to ensure driving safety.

[0043] In this embodiment, taking the daily travel scenario of a driver commuting in a smart car through "congested urban roads - cruising on suburban highways - returning on smooth urban roads" as an example, the specific implementation process is as follows: Coupled Scenario Data Acquisition and Modal Classification: After vehicle startup, the system's coupled scenario modal classification unit immediately starts, collecting two types of core data in real time: first, driving behavior characteristics (such as brake pedal trigger frequency and accelerator pedal opening change rate in congested urban areas, and speed regulation stability parameters on highways); second, operating condition characteristics (such as road condition type, ambient temperature and humidity in urban areas, and altitude and vehicle load rate on highways). Subsequently, outlier removal and standardization preprocessing are performed on the two types of features in sequence. After unifying the numerical range and time series benchmark, the dimensions are spliced ​​to form a multi-dimensional feature vector. Cluster analysis is performed on the vector to statistically analyze the commonalities in the feature distribution of each data cluster. After removing outlier data clusters, the scenario type is matched in combination with the characteristics of the actual driving scenario. Finally, three core modes are dynamically classified: "urban congestion - frequent acceleration and deceleration coupled mode", "suburban highway - smooth cruise coupled mode", and "urban smooth - constant speed driving coupled mode". The preset correlation between the coupled features and the attenuation rate of key power components under each mode is invoked. Multi-source data hierarchical lightweight fusion: Throughout the vehicle's journey, the multi-source data lightweight fusion unit works synchronously: First, feature-level pre-fusion is performed, filtering the validity of real-time data such as battery pack voltage / current and electric drive system speed / vibration collected by sensors, removing invalid data exceeding ±a times the standard deviation, suppressing data noise, and extracting core sensitive features to form a pre-fusion result; then, decision-level lightweight fusion is performed, aligning the pre-fusion result with the aforementioned driving behavior feature data and vehicle condition data (such as initial battery capacity and cumulative electric drive system runtime) with timestamps, and integrating them according to preset weight ratios (sensor data b%, driving behavior data c%, vehicle condition data d%) through a lightweight algorithm, continuously outputting a high-quality fused dataset to ensure data real-time performance and integrity; Coupling Factor Decoupling and Durability Prediction: Based on the real-time output fusion dataset, the coupling factor decoupling and nonlinear prediction unit uses a multi-factor decoupling algorithm to remove three types of independent influencing factors (such as the independent impact of frequent braking on the electric drive system due to driving behavior, the independent impact of high-speed environment on the battery due to operating conditions, and the independent impact of the battery's initial state on degradation due to vehicle condition). At the same time, it retains coupling residual factors (such as the superimposed impact of "frequent acceleration and deceleration + low temperature environment" on battery degradation in congested urban areas). The two types of factors are input into the constructed nonlinear durability prediction model to calculate the remaining life prediction values ​​of the battery pack and electric drive system in real time. During scene switching (such as from highway to urban area), the system dynamically updates the input factors according to the newly generated fusion dataset and adjusts the life prediction values ​​synchronously to ensure that the prediction results adapt to scene changes. Health Level Determination and Closed-Loop Control: The predictive control-linked health management unit continuously determines the health status of key power components based on real-time updated lifespan predictions and preset four-gradient health level determination thresholds. At the initial stage of driving, when components are at health level one, the system outputs a standard control strategy to ensure normal power output. Midway through highway driving, the battery pack experiences slight degradation due to continuous high load, causing the health level to drop to level two. The system immediately invokes a mild protection strategy, adjusting battery charge / discharge rate limits and energy recovery intensity parameters. During the return journey, the system collects component status data (such as battery temperature and electric drive system power output) in real time after the control strategy is executed, using this data as feedback to correct model operating parameters and health level determination thresholds, while simultaneously optimizing subsequent control strategy parameters. Throughout the journey, if a component's health level abnormally declines (e.g., due to sudden extreme conditions causing a drop to level three or four), the system immediately outputs the corresponding protection strategy and triggers an alarm. After the vehicle is turned off, the system automatically saves the modal segmentation results, fusion dataset, lifetime prediction data, and control strategy execution records for the entire driving process, providing data support for model optimization and strategy improvement in subsequent similar scenarios.

[0044] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A durability and health management system coupling driving behavior and operating conditions for intelligent vehicles, characterized in that, include: Couple the scene modality segmentation unit to obtain driving behavior features and working condition features, and integrate the two types of features to form a multi-dimensional feature vector; Cluster analysis is performed on multidimensional feature vectors, and coupled scenario modes are dynamically identified and classified based on the clustering results. For each coupled scenario mode, the correlation between coupling features and the attenuation rate of battery pack and electric drive system components is preset to characterize the durability impact law under different coupled scenarios. The multi-source data lightweight fusion unit performs hierarchical fusion processing on the acquired real-time sensor data, driving behavior data, and operating condition data; it first performs feature-level pre-fusion, filters redundant information and suppresses noise in the sensor data, and extracts key features. Then, a decision-level lightweight fusion is performed to integrate the preprocessed sensor data with driving behavior data and operating condition data, and output a fused dataset. The coupling factor decoupling and nonlinear prediction unit, based on the fused dataset, separates the independent influencing factors of each factor on the durability degradation of the component, and retains the coupling residual factors of multiple factors; constructs a nonlinear durability prediction model, and calculates the life prediction values ​​of the battery pack and electric drive system based on the independent influencing factors and coupling residual factors; and dynamically updates the input factors and adjusts the life prediction values ​​according to the real-time fused data. The predictive control linkage health management unit classifies the health levels of the battery pack and electric drive system based on the predicted lifespan value and preset health judgment rules. Output corresponding control strategies for different health levels; The system acquires component status data after the control strategy is executed, uses this data as feedback to correct relevant operating parameters, and optimizes subsequent control strategies.

2. The system according to claim 1, characterized in that, The specific process of integrating the two types of features to form a multi-dimensional feature vector is as follows: First, outlier removal and standardization preprocessing are performed on the acquired driving behavior features and operating condition features, respectively. Unify the numerical range and time dimension benchmark of the two types of features, and sort them according to the preset feature priority rules; The two types of preprocessed features are sequentially concatenated dimensionally to form a multidimensional feature vector that can characterize the coupling characteristics between driving behavior and operating conditions.

3. The system according to claim 1, characterized in that, The specific process of dynamically identifying and classifying coupled scene modes based on clustering results is as follows: After statistical clustering, the commonalities and dispersion of the feature distribution of each data cluster are analyzed, and abnormal data clusters are removed by verifying the feature similarity within the cluster. Based on the actual driving scenario characteristics of intelligent vehicles, a matching rule for scenario type and data cluster characteristics is constructed; Based on the matching results, the coupling scenario type corresponding to each data cluster is identified, and finally the coupling scenario modality is dynamically divided according to the differences in scenario type.

4. The system according to claim 1, characterized in that, The specific process of establishing the correlation between the preset coupling characteristics and the attenuation rate of the battery pack and electric drive system components is as follows: Based on the full life cycle historical operation data and attenuation records of key power components, the inherent correlation between coupling characteristics and component attenuation rate under different coupling scenario modes is explored through correlation analysis. Based on the correlation rules, correlation thresholds and corresponding rules are set to establish a correlation correspondence system between coupling characteristics and component attenuation rate.

5. The system according to claim 1, characterized in that, The specific process of feature-level pre-fusion is as follows: the acquired real-time sensor data is filtered for validity, and invalid data that exceeds the normal range is removed; an adaptive noise reduction algorithm is used to suppress noise interference in the data; Based on component durability impact factor analysis, core sensitive features are extracted from the processed sensor data to form feature-level pre-fusion results.

6. The system according to claim 1, characterized in that, The specific process of performing decision-level lightweight fusion is as follows: performing time stamp alignment processing on feature-level pre-fusion results, driving behavior data, and operating condition data to eliminate data timing bias; establishing a data importance assessment system and presetting the weight ratio of various types of data for vehicle durability analysis; Based on the weight ratios, a lightweight fusion algorithm is used to sequentially integrate various types of data to generate a fusion dataset that reflects the overall vehicle operating status.

7. The system according to claim 1, characterized in that, The specific process of separating the independent influencing factors of each factor on component durability degradation and retaining the coupled residual factors of multiple factors is as follows: The fused dataset is processed by a multi-factor decoupling algorithm to directly separate the independent influencing factors of the three core factors of driving behavior, working conditions, and operating conditions on the durability degradation of components. Simultaneously extract the superimposed impact of the interaction of three core factors; The superimposed influence portion is retained as a coupling residual factor of multiple factors.

8. The system according to claim 1, characterized in that, The specific process for constructing the nonlinear durability prediction model is as follows: It includes an input layer, a non-linear mapping layer, and an output layer; The input consists of historical independent influencing factors, coupled residual factors, and corresponding key power component attenuation data; the output is the time-series prediction results of the attenuation rate of key power components. The input layer performs dimension alignment and standardization preprocessing on the received feature data; The nonlinear mapping layer adopts a multilayer perceptron architecture, embedding nonlinear activation functions to iteratively transform the preprocessed features layer by layer; the output layer integrates the mapped features to generate temporal prediction results.

9. The system according to claim 1, characterized in that, The specific process for calculating the predicted lifespan of the battery pack and electric drive system is as follows: The independent influencing factors and coupled residual factors obtained in real time are input into the nonlinear durability prediction model, and the attenuation trend data of key power components are calculated. By combining the initial design life parameters of the components with historical degradation data, the remaining life prediction value of the key power components is calculated by using the trend extrapolation method.

10. The system according to claim 1, characterized in that, The specific process for classifying the health levels of battery packs and electric drive systems is as follows: based on preset health judgment rules and combined with the performance parameter thresholds of key power components, multiple gradient health level ranges and corresponding judgment thresholds are set. The calculated life prediction values ​​are compared with the judgment thresholds of each level one by one, and the health level of the key power components is determined based on the comparison results.

11. The system according to claim 1, characterized in that, The specific process of outputting control strategies corresponding to different health levels is as follows: a matching strategy library of health levels and control strategies is pre-built. The matching strategy library contains strategy types and basic parameters corresponding to each health level. Based on the health level of the key power components, the corresponding matching control strategy is retrieved, and the basic parameters of the strategy are refined and calibrated in combination with real-time operating data.

12. The system according to claim 1, characterized in that, The specific process of dynamically updating the input factors based on real-time fusion data is as follows: acquire the newly generated fusion dataset in real time, establish a factor change monitoring mechanism, and track the numerical changes of independent influencing factors and coupled residual factors in real time; when the factor numerical change exceeds the preset fluctuation range, replace the old factor data with the factor data corresponding to the newly generated fusion dataset, and synchronously input the updated factors into the nonlinear durability prediction model.

Images