Motor vehicle safety performance classification method and system based on multi-factor coupling effect

Abstract

The application relates to the technical field of data processing, and discloses a motor vehicle safety performance classification method and system based on multi-factor coupling influence. The method comprises the following steps: collecting multi-dimensional safety data of a motor vehicle, calculating mutual information values and transmission entropy to generate a coupling relationship network diagram, constructing a three-layer safety index system to form a comprehensive score, obtaining low-dimensional features through phase space mapping, determining a dynamic boundary value by applying density clustering, and classifying a real-time state to output risk warning information. The technical method realizes accurate classification and warning, overcomes defects such as neglecting factor coupling relationship, lacking scene adaptability and insufficient dynamic prediction capability in the prior art, and improves the accuracy and practicality of motor vehicle safety performance evaluation and warning.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and system for classifying motor vehicle safety performance based on the coupling effects of multiple factors. Background Technology

[0002] Traditional methods for assessing vehicle safety performance primarily rely on static testing and single-parameter analysis, such as crash safety testing, braking distance testing, and sideslip testing. These methods typically evaluate a specific aspect of a vehicle's performance under particular conditions in isolation, failing to comprehensively reflect the complex interplay of multiple factors in real-world road conditions. In recent years, with the development of sensor technology and onboard electronic systems, some advanced safety performance assessment systems have begun to employ multi-parameter fusion methods, such as Electronic Stability Control (ESC), Anti-lock Braking System (ABS), and Adaptive Cruise Control (ACC). These systems can monitor multiple vehicle parameters in real time and make corresponding controls; however, their assessment and warning mechanisms still rely on preset fixed thresholds and simple linear relationships.

[0003] However, existing technologies have significant shortcomings in assessing and classifying motor vehicle safety performance. First, most methods neglect the coupling effects between various safety-influencing factors, simply superimposing or averaging different parameters, failing to accurately reflect the nonlinear interactions between these factors. Second, traditional methods mostly use fixed thresholds for safety status judgment, lacking adaptability to different driving scenarios and environmental conditions, leading to false alarms or missed alarms under complex conditions. Third, most existing systems focus on real-time status assessment, lacking the ability to analyze the dynamic evolution of safety status over time, making it difficult to predict potential risks in advance. Finally, most methods emphasize single-dimensional safety performance evaluation, such as braking performance or lateral stability, lacking a comprehensive assessment framework that considers multiple safety factors, thus failing to fully reflect the actual safety status of motor vehicles. Summary of the Invention

[0004] This application provides a method and system for classifying motor vehicle safety performance based on the coupling effect of multiple factors. This method is used to achieve accurate classification and early warning, so as to overcome the defects of existing technologies such as ignoring the coupling relationship of factors, lacking scenario adaptability and insufficient dynamic prediction ability, and improve the accuracy and practicality of motor vehicle safety performance assessment and early warning.

[0005] Firstly, this application provides a method for classifying motor vehicle safety performance based on the coupling influence of multiple factors. This method includes: collecting dynamic parameters of the motor vehicle, environmental conditions, and driving behavior data through a distributed sensor network to obtain raw data of multidimensional safety influencing factors; calculating mutual information values ​​and transfer entropy on the raw data of the multidimensional safety influencing factors to generate a nonlinear factor coupling relationship network diagram; constructing a three-layer structured safety index system based on the nonlinear factor coupling relationship network diagram to form a comprehensive motor vehicle safety performance score; transforming the comprehensive motor vehicle safety performance score through phase space mapping and extracting key features to obtain a low-dimensional discriminant feature set; applying an adaptive density clustering method to the low-dimensional discriminant feature set to determine dynamic safety level boundary values; and using the dynamic safety level boundary values ​​to perform multi-scenario classification of the real-time collected motor vehicle status and output graded safety risk warning information.

[0006] Secondly, this application provides a vehicle safety performance classification system based on the coupling effect of multiple factors, wherein the vehicle safety performance classification system based on the coupling effect of multiple factors includes:

[0007] The data acquisition module is used to collect dynamic parameters of motor vehicles, environmental conditions and driving behavior data through a distributed sensor network to obtain raw data on multi-dimensional safety influencing factors.

[0008] The calculation module is used to calculate the mutual information value and transfer entropy of the original data of the multidimensional security influencing factors, and generate a nonlinear factor coupling relationship network diagram.

[0009] The construction module is used to construct a three-layer safety index system based on the nonlinear factor coupling relationship network diagram to form a comprehensive score for motor vehicle safety performance;

[0010] The conversion module is used to convert the comprehensive safety performance score of the motor vehicle through phase space mapping and extract key features to obtain a low-dimensional discriminative feature set;

[0011] The clustering module is used to apply an adaptive density clustering method to the low-dimensional discriminative feature set to determine the dynamic security level boundary value;

[0012] The classification module is used to classify the real-time collected vehicle status using the dynamic safety level boundary value and output graded safety risk warning information.

[0013] Thirdly, a vehicle safety performance classification device based on the coupling effect of multiple factors is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the vehicle safety performance classification device based on the coupling effect of multiple factors to execute the above-mentioned vehicle safety performance classification method based on the coupling effect of multiple factors.

[0014] Fourthly, a computer-readable storage medium is provided, wherein instructions are stored therein, which, when executed on a computer, cause the computer to perform the above-described method for classifying motor vehicle safety performance based on the coupling effects of multiple factors.

[0015] The technical solution provided in this application collects dynamic parameters of motor vehicles, environmental conditions, and driving behavior data through a distributed sensor network to obtain raw data on multi-dimensional safety influencing factors. This enables comprehensive and multi-faceted perception of the safety status of motor vehicles, providing a rich foundational data source for safety performance classification. By calculating mutual information values ​​and transfer entropy from the raw data of multi-dimensional safety influencing factors, a nonlinear factor coupling relationship network diagram is generated. This accurately captures and quantifies the complex nonlinear interactions between different safety factors, overcoming the limitation of traditional methods that ignore factor coupling. Based on the nonlinear factor coupling relationship network diagram, a three-layer safety index system is constructed to form a comprehensive score for motor vehicle safety performance, establishing a system from raw data to final evaluation. A complete indicator chain enables hierarchical interpretability of the evaluation results. The comprehensive vehicle safety performance score is transformed through phase space mapping and key features are extracted to obtain a low-dimensional discriminative feature set. Nonlinear dynamic analysis is applied to solve the problem of effectively representing high-dimensional data, improving the computational efficiency and accuracy of classification. Adaptive density clustering is applied to the low-dimensional discriminative feature set to determine dynamic safety level boundary values. Adaptive classification of safety states is achieved through an artificial intelligence density clustering algorithm, overcoming the rigidity of traditional fixed threshold classification. The dynamic safety level boundary values ​​are used to classify real-time collected vehicle states into multiple scenarios, outputting graded safety risk warning information, achieving accurate assessment and timely warning of safety risks. The artificial intelligence algorithm features in this scheme, especially mutual information and transfer entropy calculation, phase space mapping, and adaptive density clustering, make key contributions to the scheme: mutual information and transfer entropy algorithms can discover and quantify nonlinear correlation patterns from large amounts of raw data, breaking through the limitations of traditional linear analysis; phase space mapping and nonlinear dynamic feature extraction algorithms can capture the dynamic evolution of vehicle safety states, enhancing the system's predictive ability; the adaptive density clustering algorithm enables dynamic adjustment of safety state boundaries, improving the adaptability and accuracy of classification in different scenarios. The organic combination of these algorithms forms a complete intelligent analysis framework, enabling the motor vehicle safety performance classification method to have comprehensive capabilities of multi-dimensional perception, coupled analysis, dynamic adaptation and precise early warning, which significantly improves the comprehensiveness and accuracy of motor vehicle safety status assessment. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of an embodiment of the vehicle safety performance classification method based on the coupling effect of multiple factors in this application.

[0018] Figure 2 This is a schematic diagram of an embodiment of the motor vehicle safety performance classification system based on the coupling effect of multiple factors in this application.

[0019] Figure 3 This is a schematic block diagram of the structure of a motor vehicle safety performance classification device based on the coupling effect of multiple factors in an embodiment of the present invention. Detailed Implementation

[0020] This application provides a method and system for classifying motor vehicle safety performance based on the coupling effects of multiple factors. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0021] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the vehicle safety performance classification method based on the coupling effect of multiple factors in this application includes:

[0022] Step S101: Collect dynamic parameters of motor vehicles, environmental conditions and driving behavior data through a distributed sensor network to obtain raw data of multi-dimensional safety influencing factors;

[0023] Step S102: Calculate the mutual information value and transfer entropy of the original data of multidimensional security influencing factors, and generate a nonlinear factor coupling relationship network diagram;

[0024] Step S103: Construct a three-layer safety index system based on the nonlinear factor coupling relationship network diagram to form a comprehensive score for motor vehicle safety performance;

[0025] Step S104: Transform the comprehensive safety performance score of motor vehicles through phase space mapping and extract key features to obtain a low-dimensional discriminative feature set;

[0026] Step S105: Apply adaptive density clustering to the low-dimensional discriminative feature set to determine the dynamic security level boundary value;

[0027] Step S106: Use dynamic safety level boundary values ​​to classify the real-time collected vehicle status into multiple scenarios and output graded safety risk warning information.

[0028] It is understood that the executing entity of this application can be a motor vehicle safety performance classification system based on the coupling influence of multiple factors, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiments use a server as an example for illustration.

[0029] In this embodiment, a distributed sensor network is used to collect dynamic parameters of the vehicle, environmental conditions, and driving behavior data to obtain raw data on multidimensional safety influencing factors. Specifically, vehicle dynamic parameter sensors collect acceleration, yaw rate, and sideslip angle; environmental monitoring sensors collect rainfall, light intensity, and road friction coefficient; and driving behavior monitoring devices collect steering wheel angle, pedal pressure, and driver attention distribution. This data is transmitted to the central processing unit via the vehicle's high-speed CAN bus network and undergoes sensor drift compensation and system error correction using a real-time data calibration algorithm. During sensor drift compensation, a baseline value is set based on the stability characteristics of different sensors. When sensor readings deviate over time due to non-physical reasons, the drift is calculated by comparing the current value with the historical baseline value and automatically compensated for in the measured value. Simultaneously, the system also employs an adaptive sampling frequency adjustment mechanism to optimize data acquisition based on the changing characteristics of different parameters. For example, when the vehicle is in a high-speed turning state, the sampling frequency of yaw rate is automatically increased to above 100Hz to capture rapidly changing dynamic characteristics. Mutual information values ​​and transfer entropy are calculated from the raw data of multidimensional safety influencing factors to generate a nonlinear factor coupling relationship network diagram. This step first performs outlier detection and smoothing to obtain a preprocessed dataset. Then, for the factor pairs in the preprocessed dataset, the joint probability distribution is calculated using kernel density estimation to obtain the mutual information matrix between factors. The mutual information value characterizes the degree of interdependence between different security factors; a higher value indicates a stronger correlation between factors. Subsequently, the preprocessed dataset is segmented based on a sliding time window, and the transfer entropy between factors is calculated through symbolic transformation to form a directional coupling strength matrix. Transfer entropy describes the directionality of information flow and reveals the causal relationship between factors. After fusing the mutual information matrix and the directional coupling strength matrix, a multi-level factor coupling network topology is constructed. Redundant connections are pruned using a sparse representation learning method, retaining key coupling paths. Finally, the characteristics of the multi-level factor coupling network topology are analyzed at three scales: micro, meso, and macro, generating a nonlinear factor coupling relationship network graph.

[0030] A three-layer safety index system is constructed based on the nonlinear factor coupling relationship network diagram to form a comprehensive vehicle safety performance score. This step extracts basic safety indicators from the original data of multidimensional safety influencing factors to construct the bottom layer structure of the index system. Basic safety indicators directly reflect the vehicle's basic safety status, such as braking distance and lateral stability. Subsequently, based on the coupling path strength in the nonlinear factor coupling relationship network diagram, key coupling factor combinations are identified to generate mid-level coupling indicators, such as composite indicators like the "braking-steering coupling stability index." Information entropy is calculated for the mid-level coupling indicators to obtain indicator weight allocation coefficients; indicators with higher information entropy have lower weights because they contain more uncertain information. The mid-level coupling indicators are combined with the indicator weight allocation coefficients and fuzzy integral operations are used to form the top-level comprehensive safety performance index. The top-level comprehensive safety performance index is calibrated according to different driving scenario characteristics to obtain scenario-adaptive safety index values, which are then standardized to form the comprehensive vehicle safety performance score. The comprehensive vehicle safety performance score is transformed through phase space mapping and key features are extracted to obtain a low-dimensional discriminative feature set. First, a time series data matrix is ​​constructed for the comprehensive vehicle safety performance score, and the optimal time delay parameter is determined using the mutual information method. The mutual information method calculates the statistical dependency between a sequence and its delayed versions under different time delays, selecting the delay corresponding to the first local minimum of the mutual information value as the optimal value. Then, the pseudo-nearest neighbor method is used to analyze the time series data matrix to determine the optimal embedding dimension. The pseudo-nearest neighbor method examines the changes in nearest neighbors under increasing embedding dimensions; the dimension at which the proportion of pseudo-nearest neighbors falls below a threshold is the optimal embedding dimension. Based on the optimal time delay parameter and the optimal embedding dimension, the comprehensive score of motor vehicle safety performance is reconstructed into a high-dimensional phase space trajectory. Subsequently, the nonlinear dynamic characteristic parameters of the high-dimensional phase space trajectory are calculated, including the maximum Lyapunov exponent, correlation dimension, and approximate entropy value. Complex network features are extracted from the high-dimensional phase space trajectory to obtain a network topology feature quantification index. Combining the nonlinear dynamic characteristic parameters and the network topology feature quantification index, a low-dimensional discriminative feature set is obtained through nonlinear mapping transformation.

[0031] An adaptive density clustering method is applied to a low-dimensional discriminative feature set to determine dynamic security level boundary values. The specific process includes calculating the local density value and distance factor of each feature point in the low-dimensional discriminative feature set, constructing a density-distance decision map, and identifying initial cluster centers from the density-distance decision map using a density peak detection method. Clustering is performed based on the distribution characteristics of the initial cluster centers to obtain initial security level clusters. The effectiveness of the initial security level clusters is evaluated, and the optimal number of clusters is determined by calculating the silhouette coefficient. A closed decision surface is constructed around the boundary of the security level clusters corresponding to the optimal number of clusters, forming a static security level boundary. Based on historical data analysis of the temporal variation characteristics of the security level clusters, a dynamic boundary update rule is established to generate dynamic security level boundary values.

[0032] This process utilizes dynamic safety level boundary values ​​to classify real-time vehicle states across multiple scenarios and outputs tiered safety risk warning information. First, it receives dynamic vehicle parameters, environmental conditions, and driving behavior data collected by a distributed sensor network via a real-time data acquisition interface, constructing a real-time state data matrix. A fast feature extraction algorithm is applied to the real-time state data matrix to obtain real-time feature vectors. These real-time feature vectors are then projected onto the feature space containing the low-dimensional discriminant feature set using a phase space mapping transformation method, generating real-time low-dimensional discriminant points. The positional relationship between these real-time low-dimensional discriminant points and the dynamic safety level boundary values ​​is calculated to determine the current vehicle safety state category. Based on the current vehicle safety state category and the dynamic changing trend of the real-time low-dimensional discriminant points, a safety risk level is constructed. Finally, based on a comparison table of safety risk levels and preset risk thresholds, tiered safety risk warning information is generated.

[0033] For example, in a turning scenario on a wet highway surface, a distributed sensor network collected data showing a vehicle yaw rate of 0.3 rad / s, a sideslip angle of 4 degrees, a road friction coefficient of 0.6, and a steering wheel angle of 30 degrees. After processing this raw data through coupling relationship identification, the mutual information value between the yaw rate and the sideslip angle reached 0.85, indicating a high correlation between the two. The comprehensive safety performance score calculated using a three-layer index system was 75 points. After phase space mapping, the low-dimensional discriminant features obtained from this score were located in a certain region of the feature space. The distance relationship between this feature and the dynamic safety level boundary value showed that the current state was approaching the boundary of the "medium risk" region. Based on this, the system generated a yellow warning signal to remind the driver that the vehicle's dynamic stability had begun to decline and suggested reducing speed and avoiding sharp turns.

[0034] In one specific embodiment, the process of performing step S101 may specifically include the following steps:

[0035] Vehicle motion state data, including acceleration, yaw rate and sideslip angle, are collected through vehicle dynamic parameter sensors.

[0036] Environmental condition data, including rainfall, light intensity, and road surface friction coefficient, are collected through environmental monitoring sensors.

[0037] The driving behavior monitoring device collects driving operation data, including steering wheel angle, pedal pressure, and driver attention distribution.

[0038] Vehicle motion status data, environmental status data, and driving operation data are transmitted to the central processing unit via the on-board high-speed CAN bus network.

[0039] Real-time data calibration algorithms are used to compensate for sensor drift and correct system errors in the acquired data.

[0040] Based on the changing characteristics of different parameters, an adaptive sampling frequency adjustment mechanism is adopted to optimize the collection of vehicle motion state data, environmental state data, and driving operation data, thereby obtaining raw data on multidimensional safety influencing factors.

[0041] Specifically, vehicle motion data, including acceleration, yaw rate, and sideslip angle, are collected using vehicle dynamic parameter sensors. These sensors consist of multiple acceleration sensors, yaw rate sensors, and sideslip angle measurement devices. The acceleration sensors are installed near the vehicle's center of gravity and collect acceleration information in three axes, measured in m / s². 2 The data range is typically ±10g; the yaw rate sensor is installed at the center of the vehicle chassis to measure the vehicle's rotational speed around its vertical axis, in rad / s, with a typical measurement range of ±100° / s; the sideslip angle is indirectly calculated using optical or inertial measurement units combined with vehicle speed sensor data, representing the angle between the vehicle's actual direction of motion and its heading, in degrees, with an effective range typically ±15°. The raw signals collected by the sensors are initially filtered to generate standard-format data packets containing measurement values, timestamps, and sensor status information. Simultaneously, environmental monitoring sensors collect environmental status data, including rainfall, light intensity, and road friction coefficient. The environmental monitoring sensor network consists of rainfall sensors, light intensity sensors, and road friction coefficient detectors. Rain sensors are typically installed on the windshield area to quantify rainfall intensity by detecting the number and density of water droplets on the glass surface. The data is represented by discrete levels (0-5), where 0 indicates no rain and 5 indicates heavy rain. Light intensity sensors, located on the roof of the vehicle or near the rearview mirror, measure ambient illuminance in lux, ranging from 0 (complete darkness) to 100,000 lux (intense sunlight). Road surface friction coefficient detectors estimate road adhesion by analyzing tire-road contact characteristics or using specialized detectors; the value typically ranges from 0.1 (extremely slippery) to 1.0 (dry asphalt). The data collected by these sensors undergoes preliminary digitization to form an environmental status data package.

[0042] The driving behavior monitoring device collects driving operation data, including steering wheel angle, pedal pressure, and driver attention distribution. The steering wheel angle sensor, mounted on the steering column, measures the steering wheel's rotation angle, with a data range of ±720° and an accuracy typically of 0.1°. Pedal pressure sensors are installed in the mechanisms of the accelerator, brake, and clutch pedals (if applicable), measuring the pressure or displacement applied by the driver to the pedals; the data is typically normalized to 0-100% of the pedal travel. The driver attention monitoring system consists of an in-vehicle camera and an eye-tracking device, capturing the driver's facial expressions, eye movements, and head posture. Image processing algorithms analyze the attention distribution, outputting data including fixation point coordinates, fixation duration, and pupil diameter changes. These different types of data are transmitted to the central processing unit via the vehicle's high-speed CAN bus network. The CAN bus is a multi-master serial communication network protocol, operating at frequencies from 500kbps to 1Mbps, and employs differential signal transmission for strong anti-interference capabilities. During data transmission, each sensor node packages its measurement data into a standard CAN frame format. Each CAN frame contains a frame ID (indicating data priority and type), data length, data fields, and a checksum. Upon receiving the CAN frame, the central processing unit parses the different types of sensor data based on the frame ID and reassembles them into a complete data stream in chronological order. For example, yaw rate data might be transmitted as a CAN message with frame ID 0x220, sent every 10ms, while driver attention data might be transmitted as a message with frame ID 0x380, updated every 100ms.

[0043] Next, a real-time data calibration algorithm is used to compensate for sensor drift and correct system errors in the collected data. Sensor drift refers to the phenomenon where the sensor output deviates from the true value over time, mainly caused by factors such as temperature changes, vibration accumulation, and component aging. The real-time data calibration algorithm first identifies the drift baseline through zero-point detection under static conditions. For example, when the vehicle is stationary, the horizontal component of the acceleration sensor should be 0, and the yaw rate should be 0. The detected deviation value is recorded as the drift compensation value. Under dynamic conditions, the algorithm performs cross-verification through multi-sensor data fusion. For example, by comparing GPS speed data and wheel speed sensor data, the system error of the wheel speed sensor is identified. The correction process uses Kalman filtering or adaptive filtering techniques, combined with the historical performance characteristics of the sensors, to dynamically adjust the compensation coefficient, making the calibrated data closer to the true physical quantity. For example, when the yaw rate sensor detects an offset value of 0.02 rad / s in a stationary state, the system automatically subtracts this offset from subsequent measurements.

[0044] Based on the changing characteristics of different parameters, an adaptive sampling frequency adjustment mechanism is employed to optimize the collection of vehicle motion state data, environmental state data, and driving operation data, obtaining raw data on multi-dimensional safety influencing factors. The adaptive sampling frequency adjustment mechanism dynamically adjusts the sensor sampling frequency based on the parameter change rate and safety criticality. The algorithm first calculates the rate of change of the data, i.e., the difference between consecutive sampling points divided by the time interval. When the rate of change exceeds a preset threshold, the system automatically increases the sampling frequency of that parameter; when the parameter remains stable for a period of time, the system reduces the sampling frequency to save computational resources. Simultaneously, different parameters are assigned different base sampling frequencies and adjustment ranges according to their safety criticality. For example, during normal driving, the sampling frequency for yaw rate might be 20Hz, but when a sharp turn is detected, the sampling frequency automatically increases to 100Hz to capture rapidly changing vehicle dynamics; while ambient light intensity, as a relatively slow-changing parameter, might have a base sampling frequency of only 1Hz, increasing to only 5Hz even under rapidly changing light conditions. In this way, the system optimizes the efficiency and accuracy of data collection, forming a raw database of multi-dimensional safety influencing factors containing timestamps, numerical values, and quality markers.

[0045] In one specific embodiment, the process of performing step S102 may specifically include the following steps:

[0046] Outlier detection and smoothing are performed on the raw data of multidimensional safety influencing factors to obtain a preprocessed dataset.

[0047] For the factor pairs in the preprocessed dataset, the joint probability distribution is calculated using the kernel density estimation method to obtain the mutual information matrix between factors;

[0048] The preprocessed dataset is segmented based on a sliding time window, and the transfer entropy between factors is calculated through symbolic transformation to form a directional coupling strength matrix.

[0049] By fusing the mutual information value matrix and the directional coupling strength matrix, a multi-level factor coupling network topology is constructed.

[0050] Redundant connections in the topology of a multi-level factor-coupled network are pruned using a sparse representation learning method, while retaining key coupling paths.

[0051] The characteristics of the multi-level factor coupling network topology are analyzed from three scales: micro, meso, and macro, and a nonlinear factor coupling relationship network diagram is generated.

[0052] Specifically, outlier detection employs an improved Z-Score method. This involves calculating the mean and standard deviation within a moving window for each type of sensor data. A data point deviating from the mean by more than three times the standard deviation is marked as an outlier. In practice, for each sensor data sequence, the data is first arranged chronologically, and then a sliding window of 50 data points is set. Statistical features are calculated within this window. For detected outliers, a local interpolation replacement strategy is used, replacing the outlier with the weighted average of the normal data points before and after it. Data smoothing employs a Savitzky-Golay filter, a smoothing technique based on local polynomial fitting that effectively preserves the peak characteristics of the data while filtering out high-frequency noise. For slow-responding sensors such as environmental rainfall data, a Savitzky-Golay filter with a window length of 7 is used; for rapidly changing dynamic parameters such as yaw rate, a filter with a window length of 3 is used to preserve rapidly changing characteristics. Through these processes, spikes, jumps, and white noise in the original data are effectively removed, resulting in a continuously smooth preprocessed dataset.

[0053] For the factor pairs in the preprocessed dataset, the joint probability distribution is calculated using kernel density estimation to obtain the mutual information matrix between factors. Mutual information is an indicator in information theory used to measure the degree of interdependence between two random variables, representing the degree to which the uncertainty about one variable is reduced when the other is known. Calculating mutual information requires first estimating the marginal probability distributions and joint probability distributions of the variables. In this method, Gaussian kernel density estimation is used to achieve non-parametric probability density estimation. For any two security influencing factors X and Y, their data are first standardized to the [0,1] interval, and then the Gaussian kernel function is used to smooth the data points to estimate the joint probability density function. In the specific calculation process, for each pair of factors in the dataset, a two-dimensional planar grid is constructed, the data points are mapped onto the grid, and then the density value of each grid point is calculated using the kernel function. Based on the obtained probability density function, the mutual information value is calculated:

[0054]

[0055] Where MI(X,Y) represents the mutual information value between factors X and Y, p(x,y) is the joint probability density function of X and Y, and p(x) and p(y) are the marginal probability density functions of X and Y, respectively. The above calculation is repeated for all factor pairs to obtain an N×N mutual information value matrix, where N is the total number of safety influencing factors. The preprocessed dataset is segmented based on a sliding time window, and the transfer entropy between factors is calculated through symbolic transformation to form a directional coupling strength matrix. Transfer entropy is an asymmetric measure that can identify the directionality of information flow and helps reveal causal relationships between factors. The first step in calculating transfer entropy is to perform symbolic transformation on the data, that is, to convert continuous time series data into discrete symbol sequences. This method uses uniform binning technology to evenly divide the numerical range of each factor into 8 intervals, each interval corresponding to a symbol. For factor X, its time series {x1,x2,...,xT} is converted into a symbol sequence {s1,s2,...,sT}.

[0056] The formula for calculating the transfer entropy is:

[0057]

[0058] Among them TE Y→X x represents the transfer entropy from factor Y to factor X, which measures the degree of influence of factor Y on the future state of factor X; t+h This indicates the state of factor X at time t+h, where h is the prediction step size, usually set to 1. This represents the historical sequence of factor X in its k states prior to time t; This represents the historical sequence of factor Y in l states prior to time t; It is a joint probability distribution; and These are conditional probability distributions. In practical calculations, the historical lengths k and l are typically taken as 2-3 to balance computational complexity and information capture capability. The transfer entropy is calculated for all factor pairs to obtain the directional coupling strength matrix, where TE is an element of the matrix. ab This indicates the intensity of the information flow from factor b to factor a.

[0059] The mutual information matrix and the directional coupling strength matrix are fused to construct a multi-level factor-coupled network topology. The fusion process employs a weighted combination method, defining the overall coupling strength:

[0060] CE ab =α·MI ab +(1-α)·TE ab

[0061] Among them CE abThe comprehensive coupling strength between factor a and factor b is represented by α, which is a weighting coefficient ranging from [0,1] and adjusted according to the specific application scenario, typically set to 0.5. Based on the comprehensive coupling strength matrix, a multi-level network is constructed, where network nodes represent safety influencing factors, and the weights of connections between nodes are determined by CE_{ab}. To reflect the hierarchical nature of the factors, they are divided into three levels according to their functional characteristics: bottom-level sensor data nodes, middle-level state feature nodes, and high-level safety indicator nodes, forming a three-layer nested network structure.

[0062] This paper employs sparse representation learning to prune redundant connections in a multi-level factor-coupled network topology, preserving key coupling paths. A large number of weak connections in the network increase computational complexity and introduce noise interference. Sparse representation learning achieves network sparsity by minimizing reconstruction error while forcing most connection weights to approach zero. Specifically, a coupling strength threshold θ is first set, and for a comprehensive coupling strength CE... ab Connections less than θ are deleted. The threshold θ is determined adaptively, ensuring that approximately 20% of the total possible connections are retained. Then, the retained connections are ranked by importance, and L1 regularization is used to further reduce the weights of less important connections. In this way, the complex fully connected network is simplified into a sparse network containing key coupling paths, reducing computational complexity and improving network interpretability.

[0063] Finally, the characteristics of the multi-level factor coupling network topology are analyzed at three scales: micro, meso, and macro, generating a nonlinear factor coupling network graph. Micro-scale analysis focuses on the local structure of individual nodes and their directly connected nodes, calculating node degree, clustering coefficient, and local influence index. Meso-scale analysis focuses on the network's community structure, using the Louvain algorithm to identify tightly coupled factor groups and calculating the coupling strength within and between groups. Macro-scale analysis examines the global characteristics of the entire network, including indicators such as average path length, network diameter, and global efficiency. The results of these three scales are integrated into a single nonlinear factor coupling network graph, visually representing the coupling characteristics at different scales through node size, color, and line thickness.

[0064] For example, in the multi-factor coupling analysis of emergency braking scenarios, firstly, multi-dimensional raw data including vehicle speed, brake pedal pressure, tire slip ratio, and road friction coefficient are collected. Outlier detection is performed on this data, revealing several significant outliers in the tire slip ratio data, possibly caused by sensor jitter. These outliers are replaced with reasonable values ​​using local interpolation. After data smoothing, kernel density estimation is applied to calculate the mutual information values ​​between factor pairs. The results show that the mutual information value between brake pedal pressure and tire slip ratio is as high as 0.82, indicating a high correlation between the two; while the mutual information value between vehicle speed and ambient light intensity is only 0.08, showing a very low correlation. Further, the causal relationship between factors is determined through transfer entropy calculation. For example, the transfer entropy from brake pedal pressure to tire slip ratio is 0.75, while the reverse transfer entropy is only 0.12, clearly revealing the unidirectional influence of braking operation on tire slip. After fusing the mutual information matrix and the transfer entropy matrix, a multi-level network containing all safety factors is constructed, in which braking system-related factors form a clear, tightly connected sub-network. By setting a coupling strength threshold of 0.3, a large number of weak connections were removed, such as the weak correlation between vehicle speed and wiper speed. The final nonlinear factor coupling network diagram clearly shows the key factor chain in the emergency braking scenario: driver attention → brake pedal pressure → wheel pressure distribution → tire slip ratio → vehicle attitude. The factor coupling strength on this chain is much higher than that on other paths, indicating that this is the core influencing chain that should be focused on in safety analysis.

[0065] In one specific embodiment, the process of executing step S103 may specifically include the following steps:

[0066] Basic security indicators are extracted from the raw data of multidimensional security influencing factors to construct the underlying structure of the indicator system;

[0067] Based on the coupling path strength in the nonlinear factor coupling network diagram, key coupling factor combinations are identified, and mid-level coupling indices are generated.

[0068] Information entropy is calculated for the mid-level coupling indicators to obtain the indicator weight allocation coefficients.

[0069] By combining the mid-level coupling index with the index weight allocation coefficient, a top-level comprehensive security performance index is formed through fuzzy integral calculation.

[0070] Based on the characteristics of different driving scenarios, the top-level comprehensive safety performance index is calibrated in a scenario-based manner to obtain scenario-adaptive safety index values;

[0071] The safety index values ​​for scenario adaptability are standardized to form a comprehensive score for vehicle safety performance.

[0072] Specifically, basic safety indicators are extracted from the raw data of multidimensional safety influencing factors to construct the underlying structure of the indicator system. These basic safety indicators are numerical indicators directly calculated from raw sensor data, reflecting the fundamental safety characteristics of motor vehicles. During the extraction process, specialized processing methods are used for different types of raw data: for vehicle dynamic parameter data, braking distance indicators (theoretical braking distance calculated based on initial velocity and deceleration), lateral stability indicators (calculated based on the rate of change of yaw rate and sideslip angle), and longitudinal stability indicators (calculated based on acceleration changes and speed control deviations) are calculated; for environmental state data, road adhesion index (calculated based on friction coefficient and road surface type) and environmental visibility index (calculated based on a combination of light intensity and rainfall) are calculated; for driving behavior data, operation smoothness index (calculated based on the rate of change of steering wheel angle and pedal pressure) and driver attention index (calculated based on eye-tracking data) are calculated. These basic indicators are independent, directly reflecting the safety characteristics of motor vehicles in a single dimension, and together constitute the underlying structure of the indicator system. Based on the coupling path strength in the nonlinear factor coupling relationship network diagram, key coupling factor combinations are identified to generate mid-level coupling indicators. Mid-level coupling indices differ from bottom-level basic indices; they reflect the complex safety characteristics resulting from the interaction of multiple safety factors. The identification process first performs path strength analysis on the nonlinear factor coupling network diagram, extracting the factor connection paths with the highest coupling strength in the top 30%. Then, based on these critical paths, subsets of factors forming closed-loop or star-shaped structures are identified. These subsets typically represent tightly coupled functional units. For example, the "braking-road-tire" subset includes three factors: brake pedal pressure, road surface friction coefficient, and tire grip, all of which jointly affect the braking performance of a vehicle. For each identified factor subset, a corresponding coupling index calculation method is designed. Common mid-level coupling indices include: "Brake-Steering Coupling Stability Index" (reflecting the stability of a vehicle when simultaneously braking and steering), "Road-Tire-Suspension Coupling Adaptability Index" (reflecting the vehicle's suspension system's adaptability to different road conditions), and "Driver-Vehicle Response Coordination Index" (measuring the degree of matching between the driver's intention and the vehicle's actual response). These mid-level coupling indices are calculated by combining multiple basic indices, providing a more comprehensive reflection of the vehicle's safety performance under complex conditions. Information entropy is calculated for the mid-level coupling indicators to obtain the indicator weight allocation coefficients. Information entropy is an indicator that measures the uncertainty of data; the higher the entropy value, the greater the uncertainty. In determining indicator weights, using the information entropy principle means assigning lower weights to indicators with high information content (i.e., large dispersion) and higher weights to indicators with low information content (i.e., concentrated distribution). The calculation process first constructs a historical data frequency distribution table for each mid-level coupling indicator, dividing the indicator value range into several equally spaced intervals and statistically analyzing the sample frequency of each interval. Then, the information entropy of each indicator is calculated based on the frequency distribution.

[0073]

[0074] Where H(C) r ) represents the r-th mid-layer coupling index C r Information entropy, G is the number of intervals divided, f rg This represents the sample frequency of the r-th indicator in the g-th interval. Next, the difference coefficients for each indicator are calculated:

[0075]

[0076] Where D(C) r ) is the difference coefficient of the r-th mid-layer coupling index, and log(G) is the theoretical maximum entropy value. Finally, the normalized difference coefficients yield the weight allocation coefficients:

[0077]

[0078] Where W(C) r Let be the weight allocation coefficient of the r-th intermediate coupling index, and R be the total number of intermediate coupling indices. This method assigns lower weights to indices with higher dispersion and higher weights to indices with lower dispersion, reflecting the maximum information principle in information theory.

[0079] The top-level comprehensive security performance index is formed by combining the mid-level coupling index with the index weight allocation coefficient through fuzzy integral calculation. Fuzzy integral is a nonlinear integration method that can handle the interaction between indices, and is particularly suitable for handling situations where there is overlap or synergistic effect between indices. This method uses an improved Choquet fuzzy integral, and its calculation process is as follows: First, all mid-level coupling indices are standardized so that their values ​​are uniformly within the range of [0,1]:

[0080]

[0081] in These are the standardized r-th mid-layer coupling index values, Cr,min and C. r,max These are the historical minimum and maximum values ​​of the indicator. Then, the standardized indicators are sorted from largest to smallest:

[0082]

[0083] in This represents the index at the s-th position after sorting. Next, we calculate the fuzzy measure, which reflects the importance of the subset of indices:

[0084]

[0085] Where γ(A) s) is a subset A containing the top s indicators in the ranking. s The fuzzy measure, β, is the interaction parameter between indicators, determined based on the correlation analysis of the indicators, and typically ranges from [-1, 1]. Finally, based on the ranking and fuzzy measure, the Choquet fuzzy integral value is calculated as the top-level comprehensive security performance indicator:

[0086]

[0087] TSI is the top-level comprehensive security performance index, γ(A0) = 0. Choquet fuzzy integral can effectively capture the synergistic and conflicting effects between indices, and reflects the overall performance of complex systems better than simple weighted average.

[0088] The top-level comprehensive safety performance index is calibrated scenario-specifically based on the characteristics of different driving scenarios to obtain scenario-adaptive safety index values. The purpose of scenario-specific calibration is to make the safety performance assessment more closely reflect actual driving environments, as the importance and threshold standards of safety indicators differ in different scenarios. The calibration process first uses clustering methods to divide historical driving data into several typical scenarios, such as highway cruising, urban congestion, mountain road curves, and severe weather. For each scenario, a dedicated calibration model is constructed, and the calibration formula is:

[0089]

[0090] CSI q η is the scene adaptability safety index value under scene q. q It is the scene importance coefficient, which reflects the degree of danger of scene q. The larger the value, the higher the safety requirements in that scene. It is the scene environment response function, which depends on the current environment feature vector E. q This is used to adjust the sensitivity of safety indicators under different environmental conditions. For example, in wet and slippery road scenarios, the weight of indicators related to tire grip is automatically increased, while in high-speed cruising scenarios, the weight of indicators related to lane keeping is enhanced. Through scenario-based calibration, the safety assessment results become more environmentally adaptable and practically valuable.

[0091] Finally, the scenario-adaptive safety index values ​​are standardized to form a comprehensive vehicle safety performance score. The purpose of standardization is to transform safety index values ​​under various scenarios into a comprehensive score with a unified dimension that is easy to understand and compare. The standardization process first sets the scoring range to 0-100 points, and then establishes a non-linear mapping relationship:

[0092]

[0093] Where MCSS is the comprehensive safety performance score for motor vehicles, α is the slope parameter, controlling the steepness of the score change, and β...q This is the midpoint parameter of scenario q, representing the safety threshold of 50 points in that scenario. This sigmoid-type mapping provides higher resolution in the middle region while compressing extreme values, resulting in a more reasonable score distribution. The final comprehensive score directly reflects the safety performance level of the vehicle in the current driving scenario; a higher score indicates better safety performance.

[0094] In one specific embodiment, the process of executing step S104 may specifically include the following steps:

[0095] A time-series data matrix is ​​constructed based on the comprehensive safety performance score of motor vehicles, and the optimal time delay parameter is determined using the mutual information method.

[0096] The pseudo-nearest neighbor method is used to analyze time series data matrices to determine the optimal embedding dimension;

[0097] Based on the optimal time delay parameter and the optimal embedding dimension, the comprehensive score of motor vehicle safety performance is reconstructed into a high-dimensional phase space trajectory;

[0098] Calculate the nonlinear dynamic characteristic parameters of trajectories in high-dimensional phase space, including the maximum Lyapunov exponent, correlation dimension, and approximate entropy value;

[0099] Complex network features are extracted from high-dimensional phase space trajectories to obtain quantitative indicators of network topology features;

[0100] By combining nonlinear dynamic characteristic parameters and network topology characteristic quantification indicators, a low-dimensional discriminative feature set is obtained through nonlinear mapping transformation.

[0101] Specifically, a time-series data matrix is ​​constructed based on the comprehensive safety performance scores of motor vehicles, and the optimal time delay parameter is determined using the mutual information method. During the construction of the time-series data matrix, the continuously collected comprehensive safety performance scores of motor vehicles are arranged in chronological order to form a one-dimensional vector, where each element represents the comprehensive safety performance score at a given time point. The mutual information method, based on information theory principles, is used to determine the optimal delay value for time-series reconstruction by calculating the statistical dependence between the time series and its delayed version. In practice, for each candidate delay value, the mutual information value between the original sequence and the delayed sequence is calculated. First, the original data and delayed data are divided into several intervals, and the joint probability distribution and marginal probability distribution are statistically analyzed. Then, the mutual information value is calculated. As the delay increases, the mutual information value usually shows a trend of first decreasing and then stabilizing. The delay corresponding to the first local minimum of the mutual information value is selected as the optimal time delay parameter. This parameter represents the minimum time interval at which adjacent points in the time series begin to exhibit relative independence. Analyzing the time-series data matrix using the pseudo-nearest neighbor method to determine the optimal embedding dimension is another key step in phase space reconstruction. The core idea of ​​the false nearest neighbor method is that when the embedding dimension is insufficient, points that are far apart in a high-dimensional space may become neighbors in a low-dimensional projection, which is called a "false nearest neighbor". As the embedding dimension increases, the number of false nearest neighbors decreases. In the specific implementation, a series of candidate embedding dimensions are first set. For each dimension, a corresponding delay vector is constructed and the nearest neighbors of each point are identified. Then, it is checked whether these nearest neighbor points still maintain their nearest neighbor relationship after the dimension increases. If the ratio of the distance between two points in the higher-dimensional space to their distance in the current dimension exceeds a preset threshold, they are considered false nearest neighbors. The proportion of false nearest neighbors to the total number of points is calculated under different dimensions. When this proportion first drops below a certain threshold, the corresponding dimension is the optimal embedding dimension. This dimension is sufficient to unfold the phase space trajectory and avoid trajectory self-intersection.

[0102] Based on the optimal time delay parameter and optimal embedding dimension, the comprehensive score of vehicle safety performance is reconstructed into a high-dimensional phase space trajectory. The reconstruction is based on Takens' embedding theorem, which states that a phase space topologically equivalent to the original power system can be reconstructed using the delay coordinates of a univariate time series. During the reconstruction process, for each time point, a vector is constructed consisting of the current point and its subsequent delayed points. For example, if the optimal time delay is 5 and the optimal embedding dimension is 4, then the state vector at time point t is [MCSS(t), MCSS(t+5), MCSS(t+10), MCSS(t+15)]. Constructing state vectors for all possible t results in a series of points distributed in the high-dimensional space, forming the high-dimensional phase space trajectory. This trajectory preserves the dynamic characteristics of the original time series and reflects the evolution of vehicle safety performance over time.

[0103] Calculating the nonlinear dynamic characteristic parameters of high-dimensional phase space trajectories, including the maximum Lyapunov exponent, correlation dimension, and approximate entropy, is a crucial step in extracting key features from the trajectory. The maximum Lyapunov exponent measures the system's sensitivity to initial conditions, i.e., its degree of chaos. During calculation, two nearby points on the trajectory are selected, their separation over time is tracked, and the average separation rate is taken. A positive Lyapunov exponent indicates chaotic characteristics; a larger value indicates higher uncertainty. The correlation dimension characterizes the geometric complexity of the trajectory. During calculation, the correlation between trajectory point pairs at different distance thresholds is statistically analyzed, and the dimension value is determined through the power-law behavior of the correlation integral. A higher dimension indicates greater system freedom and more complex dynamic behavior. Approximate entropy measures the regularity and predictability of a time series, calculated by comparing the matching of patterns of different lengths. A higher entropy value indicates a more irregular sequence and lower predictability. These three parameters collectively describe the complex dynamic characteristics of vehicle safety performance changing over time. Extracting complex network features from high-dimensional phase space trajectories and obtaining quantitative indicators of network topology features is an important method for further characterizing trajectory properties. The core idea of ​​complex network feature extraction is to transform phase space trajectories into network structures and then analyze the network's topological characteristics. In practice, the phase space is first uniformly divided into multiple small regions, each corresponding to a node in the network. Then, network connections are established based on the movement of trajectory points between regions. If a phase space trajectory moves from region A to region B, a directed edge is created from node A to node B in the network, with the edge weight determined by the frequency of the transition. After constructing the network, topological characteristic indicators are calculated, including node degree distribution characteristics (such as average degree and degree distribution entropy), clustering coefficients (reflecting the local density of the network), centrality indicators (the importance of nodes in the network), and small-world properties (comparing the average path length of the network to that of a random network). These network characteristics can characterize the patterns and laws of security performance changes from a topological perspective.

[0104] Combining nonlinear dynamic characteristic parameters and network topology feature quantification indices, obtaining a low-dimensional discriminative feature set through nonlinear mapping transformation is a step in compressing high-dimensional features into a low-dimensional representation that facilitates classification. The main methods used for nonlinear mapping transformation are t-SNE and the improved Isomap algorithm. The t-SNE algorithm is particularly good at preserving the local structure of the data; its core is to construct conditional probability distributions between point pairs in both high-dimensional and low-dimensional spaces, and then minimize the KL divergence between them. The Isomap algorithm, on the other hand, reduces dimensionality by preserving the geodesic distances of the data on the manifold, thus better preserving the global geometric structure of the data. In practical applications, the advantages of both algorithms are often combined: first, Isomap is used to obtain the initial mapping, and then t-SNE is used for fine-tuning. Furthermore, to enhance the discriminative power of the features, Fisher's discriminant criterion and the maximum mutual information criterion are introduced for feature selection, eliminating redundant and noisy features. The final low-dimensional discriminative feature set typically has dimensions between 3 and 5, where each dimension is a nonlinear combination of the original high-dimensional features, effectively distinguishing different security state categories.

[0105] In one specific embodiment, the process of executing step S105 may specifically include the following steps:

[0106] Calculate the local density value and distance factor of each feature point in the low-dimensional discriminative feature set, and construct a density-distance decision graph;

[0107] Initial cluster centers were identified from the density-distance decision map using a density peak detection method.

[0108] Clustering is performed based on the distribution characteristics of the initial cluster centers to obtain initial security level clusters;

[0109] The effectiveness of the initial security level clusters is evaluated, and the optimal number of clusters is determined by calculating the silhouette coefficient.

[0110] A closed decision surface is constructed around the cluster boundary of the security level corresponding to the optimal number of clusters to form a static security level boundary.

[0111] Based on historical data analysis of the temporal change characteristics of security level clusters, dynamic boundary update rules are established to generate dynamic security level boundary values.

[0112] Specifically, in the vehicle safety performance classification method based on the coupling influence of multiple factors, adaptive density clustering and dynamic boundary determination are key steps. This process first calculates the local density value and distance factor of each feature point in the low-dimensional discriminant feature set, constructing a density-distance decision map. The local density value represents the degree of clustering of data points around a feature point, while the distance factor represents the minimum distance between that point and other high-density points. When calculating the local density value, for each feature point, the number of points whose distance to that point is less than the cutoff distance is counted, and then this number is divided by the total number of points to obtain the normalized local density. The cutoff distance is generally taken as the 2%-3% quantile of all point-to-point distances. The distance factor is the distance to the point with the smallest distance among all other points with a density greater than that point. If a point has the maximum local density, its distance factor is defined as the maximum distance between all points. Thus, each point has a pair of local density values ​​and distance factors. Plotting these two values ​​of all points on a two-dimensional plane forms the density-distance decision map. In this graph, points with both high local density and large distance factors are usually located in the upper right region of the graph, and these points are ideal candidate points for cluster centers.

[0113] Identifying initial cluster centers from the density-distance decision map using density peak detection is the first step in clustering. The core idea of ​​density peak detection is that cluster centers should have high local density and maintain a large distance from other high-density points. In practice, the decision value for each point is first calculated; this value is the product of local density and a distance factor. Then, all points are sorted from largest to smallest decision value, and the points with the largest decision values ​​are selected as initial cluster centers. The selection process also considers the minimum distance constraint between cluster centers to prevent excessive concentration of centers. In motor vehicle safety performance analysis, initial cluster centers typically correspond to typical safety state patterns, such as "high-stability driving state," "critically stable state," "slightly unstable state," and "severely unstable state." These center points are distributed in different regions of the feature space and can well represent different safety level categories.

[0114] Clustering based on the distribution characteristics of initial cluster centers to obtain initial safety level clusters is a crucial step in forming safety classifications. The partitioning process employs a density-based allocation strategy, where each non-center point is assigned to the cluster containing the cluster center with the strongest density correlation. Specifically, all non-center points are first sorted from highest to lowest local density. Then, for each non-center point, points with assigned cluster labels within its neighborhood (usually defined as within the cutoff distance) with a local density greater than that point are identified. This non-center point is then assigned to the cluster corresponding to the most frequently occurring cluster label. If no point with an assigned cluster label exists in its neighborhood, it is considered an outlier or noise point and may be temporarily left unassigned or assigned to a special "noise cluster." This top-down allocation method ensures that data points flow along the density gradient towards cluster centers, forming natural cluster boundaries. In motor vehicle safety performance classification, this process assigns low-dimensional discriminative feature points to different safety level clusters, with each cluster representing a safety state type.

[0115] Evaluating the effectiveness of initial safety level clusters and determining the optimal number of clusters through silhouette coefficient calculation are crucial steps in ensuring cluster quality. The silhouette coefficient is an internal metric for evaluating cluster quality, comprehensively considering both intra-cluster compactness and inter-cluster separation. For each data point, its silhouette value is calculated, reflecting the degree to which the point is correctly clustered. The calculation involves three steps: First, for point i, calculate its average distance a(i) to other points in the same cluster, representing the dissimilarity of point i within its own cluster; then, calculate the average distance b(i) of point i to all points in every other cluster, taking the minimum value, which represents the similarity of point i to its nearest non-cluster; finally, calculate the silhouette value s(i) of point i = (b(i) - a(i)) / max{a(i), b(i)}. The silhouette value ranges from -1 to 1, with a larger value indicating better clustering. The average silhouette value of all points is the silhouette coefficient of the cluster. By experimenting with different numbers of clusters (starting from 2 and increasing), the corresponding silhouette coefficients are calculated, and the number of clusters with the largest silhouette coefficient is selected as the optimal number of clusters. In motor vehicle safety performance classification, the optimal number of clusters is usually between 3 and 5, corresponding to different levels of safety status.

[0116] Constructing a closed decision surface around the cluster boundaries corresponding to the optimal number of clusters for each security level, forming a static security level boundary, is the foundation for achieving security state classification. A closed decision surface refers to the boundary surface that divides different security level clusters in the feature space, defining the dividing line between different security states. An improved Support Vector Data Description (SVDD) algorithm is used in the construction process, which generates a compact closed boundary for each cluster. For each security level cluster, all points in that cluster are first treated as positive samples, and points from other clusters as negative samples, then the SVDD model is trained. The core of SVDD is to find a hypersphere with the smallest radius, such that positive samples are located as close to the sphere as possible, and negative samples are located as far outside the sphere as possible. By introducing a kernel function, SVDD can handle non-spherical data distributions. The optimization objective is to minimize the volume of the hypersphere while maximizing the inclusion rate of positive samples. After solving the SVDD problem to obtain the support vectors and boundary parameters, these parameters are used to define a decision function that determines whether a new point is within a cluster. Combining the decision functions of each cluster forms the complete classification boundary, i.e., the static security level boundary.

[0117] Analyzing the temporal variation characteristics of safety level clusters based on historical data and establishing dynamic boundary update rules to generate dynamic safety level boundary values ​​is a key measure to adapt to dynamic changes in safety status. Static safety level boundaries cannot cope with the dynamic changes in vehicle safety status; therefore, a dynamic boundary update mechanism is needed. First, a large amount of historical data is collected, and the trajectory of safety status changes is recorded in a time series to analyze the evolution patterns of safety level clusters under different driving scenarios. Through analysis, key factors affecting safety level changes are identified, such as environmental conditions, vehicle speed changes, and driving operations. Then, boundary adjustment rules based on time series patterns are established, including expansion rules (contracting the boundary to provide early warning when an increasing trend in safety risk is detected) and contraction rules (appropriately widening the boundary when the safety status is stable). Simultaneously, an adaptive learning mechanism is introduced to continuously optimize the boundary based on new data. During the dynamic update process, a sliding time window technique is used to recalculate key parameters, such as local density thresholds and distance factor thresholds, within the most recent time window and adjust the boundary parameters accordingly. This dynamic boundary technique enables the system to adapt to changes in safety status under different driving conditions, providing more accurate safety classification.

[0118] For example, in this process, suppose we collect safety performance data for a certain vehicle model under various driving conditions. After processing through the aforementioned steps, we obtain a three-dimensional low-dimensional discriminative feature set containing 2000 feature points. First, we calculate the local density value of each feature point, choosing the 2.5% quantile of the point-to-point distance as the cutoff distance, which is approximately 0.15. For feature point A, we count 85 points with a distance less than 0.15, so its normalized local density is 85 / 2000 = 0.0425. Then, we calculate the distance factor and find that among the points with a local density greater than 0.0425 of point A, the closest is point B, with a distance of 0.58. Therefore, the distance factor of point A is 0.58. Similarly, we calculate the local density values ​​and distance factors of all points to construct a density-distance decision graph. Four distinct peak points were identified in the image, representing points with high local density values ​​and distance factors. These points are located in different regions of the feature space, corresponding to different safety state modes: Peak point 1 has a local density of 0.085 and a distance factor of 0.92, representing a "highly stable driving state"; Peak point 2 has a local density of 0.072 and a distance factor of 0.85, representing a "normal driving state"; Peak point 3 has a local density of 0.063 and a distance factor of 0.78, representing a "slightly unstable state"; and Peak point 4 has a local density of 0.055 and a distance factor of 0.81, representing a "significantly unstable state". Next, these four peak points were used as initial cluster centers, and all feature points were clustered according to a density gradient allocation strategy, resulting in four initial safety level clusters. The silhouette coefficient was then calculated for different numbers of clusters (2 to 6 clusters). The highest silhouette coefficient (0.68) was found with four clusters, confirming that four clusters were the optimal number. SVDD decision boundaries were then constructed around these four clusters to form static safety level boundaries. Finally, analysis of the temporal changes in safety status in historical data revealed that the transition from "normal state" to "slightly unstable state" often occurs faster on wet roads than on dry roads. Based on this, a corresponding expansion coefficient was set in the boundary dynamic update rules, enabling the system to identify potential unstable states earlier and provide advance warnings on wet roads. Through this dynamic boundary technology, the safety performance classification system can adapt to changes in safety status under different driving environments, providing more accurate graded warning information.

[0119] In one specific embodiment, the process of executing step S106 may specifically include the following steps:

[0120] The system receives dynamic parameters of motor vehicles, environmental conditions, and driving behavior data collected by a distributed sensor network through a real-time data acquisition interface, and constructs a real-time status data matrix.

[0121] A fast feature extraction algorithm is applied to the real-time state data matrix to obtain real-time feature vectors;

[0122] The real-time feature vectors are projected onto the feature space of the low-dimensional discriminant feature set through a phase space mapping transformation method to generate real-time low-dimensional discriminant points.

[0123] Calculate the positional relationship between the real-time low-dimensional discrimination point and the dynamic safety level boundary value to determine the current vehicle safety status category;

[0124] Based on the current classification of motor vehicle safety status and the dynamic changing trend of real-time low-dimensional discrimination points, a safety risk level is constructed.

[0125] Based on the comparison table of safety risk levels and preset risk thresholds, graded safety risk early warning information is generated.

[0126] Specifically, the vehicle safety performance classification method based on multi-factor coupling influence, after determining the safety level boundaries, needs to classify the real-time collected vehicle status and output warning information. First, it receives vehicle dynamic parameters, environmental conditions, and driving behavior data collected by a distributed sensor network through a real-time data acquisition interface to construct a real-time status data matrix. The real-time data acquisition interface acts as a bridge connecting the sensor network and the central processing unit, achieving data transmission via a high-speed CAN bus protocol. The acquired data is organized in a unified format, containing four basic fields: timestamp, sensor identifier, numerical value, and quality marker. After receiving the data, it is rearranged according to data type and time order to form a real-time status data matrix. Each row of this matrix corresponds to a time point, and each column corresponds to a sensor parameter. To handle the differences in sampling frequencies between different sensors, an interpolation alignment method is used to align all data to a unified time point. For example, when the yaw rate sensor sampling frequency is 100Hz, while the ambient temperature sensor sampling frequency is only 1Hz, linear interpolation is performed on the ambient temperature data to estimate the value at the intermediate time point, ensuring that each time point in the matrix has a complete parameter set. In addition, the real-time status data matrix also includes a sliding time window mechanism, which retains only the data for the most recent period (usually 10-30 seconds), ensuring both the real-time nature of the analysis and providing sufficient historical information for situational analysis.

[0127] Applying a fast feature extraction algorithm to the real-time state data matrix to obtain real-time feature vectors is a crucial step in data dimensionality reduction and feature representation. Fast feature extraction algorithms are feature computation methods optimized for real-time application scenarios, requiring both feature representation capabilities and computational efficiency. The algorithm first preprocesses the real-time state data matrix, including normalization, detrending, and smoothing. Then, it calculates time-domain features (such as mean, standard deviation, peak value, valley value, peak-valley difference, etc.), frequency-domain features (spectral characteristics obtained through Fast Fourier Transform), and statistical features (such as skewness, kurtosis, entropy, etc.). For motor vehicle safety performance evaluation, the focus is on extracting features related to dynamic stability, such as the fluctuation range of yaw rate, the rate of change of braking distance, and the spectral distribution of lateral acceleration. These raw features undergo preliminary dimensionality reduction using Principal Component Analysis (PCA), retaining principal components that explain more than 90% of the variance. During PCA, a pre-calculated feature vector matrix is ​​used to linearly transform the real-time features, avoiding recalculation of eigenvalue decomposition each time and improving computational efficiency. The final real-time feature vector typically has a dimension between 10 and 20, containing the main information from the original data.

[0128] Projecting real-time feature vectors onto the feature space of a low-dimensional discriminant feature set using a phase space mapping transformation method to generate real-time low-dimensional discriminant points serves as a bridge connecting the training model and real-time monitoring. The phase space mapping transformation method involves converting the extracted feature vectors into the same feature space as the training model, ensuring that real-time data and historical models are compared under the same reference frame. Specifically, the mapping parameters saved during the training phase, including the parameter matrix and reference point set of the t-SNE or Isomap algorithm, are first used to project the real-time feature vectors onto the low-dimensional feature space. For t-SNE mapping, a parameter inheritance strategy is adopted, using the embedded coordinates of the pre-trained model as initial values. Only a limited number of optimization iterations are performed on new data points, avoiding a complete recalculation of the embedded coordinates and improving real-time response capabilities. For Isomap mapping, a connection relationship is constructed between the real-time feature vectors and the k nearest neighbors in the training dataset, and the real-time features are projected onto the low-dimensional space based on geodesic distance calculations. In this way, low-dimensional discriminant points corresponding to the real-time state data are generated. The position of these points in the low-dimensional feature space intuitively reflects the current safety status characteristics of the motor vehicle.

[0129] Calculating the positional relationship between real-time low-dimensional discrimination points and dynamic safety level boundary values ​​to determine the current vehicle safety status category is the core step in classification decision-making. The positional relationship calculation is based on the distance metric from a point to a boundary, employing an improved nearest-point projection distance method. First, the distance between the real-time low-dimensional discrimination point and the boundary of each safety level cluster is calculated to find the nearest boundary and its direction. The distance calculation is based on the function value from the point to the SVDD decision boundary; a positive value is obtained inside the boundary, and a negative value is obtained outside the boundary, with the magnitude representing the relative distance to the boundary. Then, the safety level category to which the real-time discrimination point belongs is determined based on the distance value. If the discrimination point is located within the boundary of a certain safety level cluster, it belongs to that category; if it is located within the boundaries of multiple clusters, the cluster with the largest distance is selected as the category; if it is located outside all cluster boundaries, its most likely category is inferred based on the distance and direction to the nearest boundary, combined with historical trajectory information. For discrimination points near the boundary, a fuzzy classification concept is introduced, and its classification degree for each category is calculated for subsequent risk level assessment.

[0130] Constructing a safety risk level based on the current vehicle safety status classification and the dynamic changing trend of real-time low-dimensional discrimination points is a crucial step in risk assessment. The safety risk level considers not only the current status classification but also the changing trend, particularly the possibility of evolving into an unsafe state. Dynamic trend analysis is based on the motion trajectory of real-time low-dimensional discrimination points in the feature space, calculating the velocity vector (positional difference between adjacent time points) and acceleration vector (rate of change of the velocity vector) of the discrimination points. Then, the relative direction of these vectors to the safety level boundary is evaluated. If the velocity vector points to a more unsafe area and has a large amplitude, it indicates that the safety status is rapidly deteriorating, and the risk level should be increased accordingly. Furthermore, the distance between the real-time discrimination point and the safety level boundary is also considered; the smaller the distance, the higher the risk. A comprehensive risk score is calculated by integrating factors such as the discrimination point classification, boundary distance, motion speed, and motion direction. This score is then divided into different safety risk levels based on a set threshold, such as "Safe," "Caution," "Warning," "Danger," and "Emergency."

[0131] Generating tiered safety risk warning information based on a safety risk level and preset risk threshold lookup table is the final step in warning output. The preset risk threshold lookup table is a pre-defined mapping relationship between risk levels and warning content, based on historical data and expert knowledge. Different risk levels correspond to different warning methods, warning content, and warning urgency levels. When generating warning information, the preset lookup table is first consulted based on the current safety risk level to obtain the corresponding warning template. Then, considering the specific safety characteristics of the current situation, such as the type of unstable factor (e.g., skidding, fishtailing, insufficient braking), environmental conditions (e.g., slippery road surface, sharp curves), and driving operations (e.g., sharp turns, sudden braking), the warning template is parameterized and populated to form specific warning content. The warning content includes the risk level, risk description, possible consequences, and suggested countermeasures. Finally, an appropriate warning method is selected based on the warning urgency level, such as dashboard alerts, audible warnings, seat vibrations, or a combination of these warnings, ensuring that the driver can promptly notice the risk and take appropriate action.

[0132] For example, in a high-speed driving scenario, the real-time data acquisition interface receives data showing a yaw rate of 0.32 rad / s and a lateral acceleration of 4.2 m / s². 2 Data such as a steering wheel angle of 45 degrees and a road surface friction coefficient of 0.6 (for wet and slippery surfaces) were used to construct a real-time state data matrix. A fast feature extraction algorithm was applied to this matrix to calculate time-domain features such as the standard deviation of yaw rate (0.08 rad / s), frequency-domain features such as the dominant frequency of yaw rate (2.3 Hz), and statistical features such as the lateral acceleration skewness (1.2), forming a 15-dimensional real-time feature vector. This feature vector was projected onto a 3-dimensional feature space using a phase space mapping transformation method to generate a real-time low-dimensional discrimination point (0.68, -0.42, 0.15). The positional relationship between this discrimination point and the dynamic safety level boundary was calculated, revealing that it is located within the "slightly unstable state" cluster, relatively close to the boundary, and the distance to the boundary of the "significantly unstable state" cluster is decreasing. Analysis of the discrimination point's trajectory showed that its velocity vector (0.04, -0.03, 0.01) points towards the "significantly unstable state" cluster, indicating that the safety status is deteriorating. Based on the classification of "slightly unstable state," the proximity to the boundary, and the trend of moving towards a more unsafe area, a safety risk level of "Warning" is constructed. The system queries a preset risk threshold lookup table based on the "Warning" risk level, and combines this with the current slippery road surface and large steering angle to generate a warning message: "Warning: Vehicle lateral stability has decreased; there is a risk of skidding under the current slippery road conditions. It is recommended to slow down and avoid sharp steering maneuvers." Simultaneously, a yellow visual warning and a short audible alert are triggered to guide the driver to take appropriate measures to prevent the risk from escalating.

[0133] The above describes the vehicle safety performance classification method based on the multi-factor coupling effect in the embodiments of this application. The following describes the vehicle safety performance classification system based on the multi-factor coupling effect in the embodiments of this application. Please refer to [link / reference]. Figure 2 One embodiment of the motor vehicle safety performance classification system based on the coupling effect of multiple factors in this application includes:

[0134] The data acquisition module is used to collect dynamic parameters of motor vehicles, environmental conditions and driving behavior data through a distributed sensor network to obtain raw data on multi-dimensional safety influencing factors.

[0135] The calculation module is used to calculate the mutual information value and transfer entropy of the raw data of multidimensional security influencing factors, and generate a network diagram of nonlinear factor coupling relationship.

[0136] The module is used to construct a three-layer safety index system based on the nonlinear factor coupling relationship network diagram, and form a comprehensive score for motor vehicle safety performance.

[0137] The conversion module is used to convert the comprehensive safety performance score of motor vehicles through phase space mapping and extract key features to obtain a low-dimensional discriminative feature set.

[0138] The clustering module is used to apply an adaptive density clustering method to the low-dimensional discriminative feature set to determine the dynamic security level boundary value;

[0139] The classification module is used to classify the real-time collected vehicle status using the dynamic safety level boundary value and output graded safety risk warning information.

[0140] Through the collaborative efforts of the aforementioned components, a distributed sensor network is used to collect dynamic parameters of motor vehicles, environmental conditions, and driving behavior data, obtaining raw data on multidimensional safety influencing factors. This enables comprehensive and multi-faceted perception of the vehicle's safety status, providing a rich foundational data source for safety performance classification. By calculating mutual information values ​​and transfer entropy from the raw data of multidimensional safety influencing factors, a nonlinear factor coupling relationship network diagram is generated. This accurately captures and quantifies the complex nonlinear interactions between different safety factors, overcoming the limitation of traditional methods that ignore factor coupling. Based on the nonlinear factor coupling relationship network diagram, a three-layer safety index system is constructed, forming a comprehensive score for vehicle safety performance, establishing a system from raw data to final evaluation. The complete indicator chain of the evaluation makes the assessment results hierarchically interpretable. The comprehensive score of motor vehicle safety performance is transformed through phase space mapping and key features are extracted to obtain a low-dimensional discriminative feature set. Nonlinear dynamic analysis is applied to solve the problem of effectively representing high-dimensional data, improving the computational efficiency and accuracy of classification. Adaptive density clustering is applied to the low-dimensional discriminative feature set to determine dynamic safety level boundary values. Adaptive classification of safety states is achieved through artificial intelligence density clustering algorithms, overcoming the rigidity problem of traditional fixed threshold classification. The dynamic safety level boundary values ​​are used to classify the real-time collected motor vehicle states into multiple scenarios, outputting graded safety risk warning information, achieving accurate assessment and timely warning of safety risks. The artificial intelligence algorithm features in this scheme, especially mutual information and transfer entropy calculation, phase space mapping, and adaptive density clustering, make key contributions to the scheme: mutual information and transfer entropy algorithms can discover and quantify nonlinear correlation patterns from a large amount of raw data, breaking through the limitations of traditional linear analysis; phase space mapping and nonlinear dynamic feature extraction algorithms can capture the dynamic evolution law of motor vehicle safety states, enhancing the predictive ability of the system; the adaptive density clustering algorithm realizes the dynamic adjustment of safety state boundaries, improving the adaptability and accuracy of classification in different scenarios. The organic combination of these algorithms forms a complete intelligent analysis framework, enabling the motor vehicle safety performance classification method to have comprehensive capabilities of multi-dimensional perception, coupled analysis, dynamic adaptation and precise early warning, which significantly improves the comprehensiveness and accuracy of motor vehicle safety status assessment.

[0141] above Figure 2 The vehicle safety performance classification system based on multi-factor coupling influence in this embodiment of the invention will be described in detail from the perspective of modular functional entities. The vehicle safety performance classification device based on multi-factor coupling influence in this embodiment of the invention will be described in detail from the perspective of hardware processing.

[0142] Figure 3This is a schematic diagram of a vehicle safety performance classification device based on the coupling effect of multiple factors, provided by an embodiment of the present invention. The vehicle safety performance classification device 300 based on the coupling effect of multiple factors can vary significantly due to different configurations or performance characteristics. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors) and a memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and storage media 330 can be temporary or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the vehicle safety performance classification device 300 based on the coupling effect of multiple factors. Furthermore, the processor 310 may be configured to communicate with the storage media 330 and execute the series of instruction operations in the storage media 330 on the vehicle safety performance classification device 300 based on the coupling effect of multiple factors to implement the steps of the above-described vehicle safety performance classification method based on the coupling effect of multiple factors.

[0143] The vehicle safety performance classification device 300 based on the coupling effect of multiple factors may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The illustrated structure of the vehicle safety performance classification device based on the coupling effect of multiple factors does not constitute a limitation on the vehicle safety performance classification device based on the coupling effect of multiple factors provided by the present invention. It may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0144] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the motor vehicle safety performance classification method based on the coupling effect of multiple factors.

[0145] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0146] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a vehicle safety performance classification device based on multi-factor coupling effects (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0147] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for classifying the safety performance of a motor vehicle based on the coupling effect of multiple factors, characterized in that, The method includes: By collecting dynamic parameters of motor vehicles, environmental conditions, and driving behavior data through a distributed sensor network, raw data on multidimensional safety influencing factors can be obtained. The process of calculating mutual information values ​​and transfer entropy on the original data of the multidimensional security influencing factors to generate a nonlinear factor coupling network graph includes: performing outlier detection and smoothing on the original data of the multidimensional security influencing factors to obtain a preprocessed dataset; calculating the joint probability distribution of factor pairs in the preprocessed dataset using kernel density estimation to obtain a mutual information value matrix between factors; segmenting the preprocessed dataset based on a sliding time window and calculating the transfer entropy between factors through symbolic transformation to form a directional coupling strength matrix; fusing the mutual information value matrix and the directional coupling strength matrix to construct a multi-level factor coupling network topology; pruning redundant connections in the multi-level factor coupling network topology using a sparse representation learning method to retain key coupling paths; and analyzing the characteristics of the multi-level factor coupling network topology at three scales: micro, meso, and macro, to generate the nonlinear factor coupling network graph. A three-layer safety index system is constructed based on the nonlinear factor coupling relationship network diagram to form a comprehensive score for motor vehicle safety performance. The comprehensive safety performance score of the motor vehicle is transformed by phase space mapping and key features are extracted to obtain a low-dimensional discriminative feature set; An adaptive density clustering method is applied to the low-dimensional discriminative feature set to determine the dynamic security level boundary value; The dynamic safety level boundary value is used to classify the real-time collected vehicle status into multiple scenarios and output graded safety risk warning information.

2. The method for classifying the safety performance of a motor vehicle based on the coupling effect of multiple factors according to claim 1, characterized in that, The process involves collecting dynamic parameters of motor vehicles, environmental conditions, and driving behavior data through a distributed sensor network to obtain raw data on multidimensional safety influencing factors, including: Vehicle motion state data, including acceleration, yaw rate and sideslip angle, are collected through vehicle dynamic parameter sensors. Environmental condition data, including rainfall, light intensity, and road surface friction coefficient, are collected through environmental monitoring sensors. The driving behavior monitoring device collects driving operation data, including steering wheel angle, pedal pressure, and driver attention distribution. The vehicle motion status data, the environmental status data, and the driving operation data are transmitted to the central processing unit via the vehicle-mounted high-speed CAN bus network. Real-time data calibration algorithms are used to compensate for sensor drift and correct system errors in the acquired data. Based on the changing characteristics of different parameters, an adaptive sampling frequency adjustment mechanism is used to optimize the collection of vehicle motion state data, environmental state data, and driving operation data to obtain the original data of the multidimensional safety influencing factors.

3. The method for classifying the safety performance of a motor vehicle based on the coupling effect of multiple factors according to claim 1, characterized in that, The three-layer safety index system constructed based on the nonlinear factor coupling network diagram forms a comprehensive score for motor vehicle safety performance, including: Basic security indicators are extracted from the raw data of the multidimensional security influencing factors to construct the underlying structure of the indicator system; Based on the coupling path strength in the nonlinear factor coupling network diagram, key coupling factor combinations are identified, and mid-level coupling indices are generated. Information entropy is calculated for the aforementioned mid-level coupling index to obtain the index weight allocation coefficient; The mid-level coupling index is combined with the index weight allocation coefficient, and a top-level comprehensive security performance index is formed through fuzzy integral calculation. The top-level comprehensive safety performance index is calibrated according to the characteristics of different driving scenarios to obtain scenario-adaptive safety index values; The scenario-adaptive safety index values ​​are standardized to form the comprehensive safety performance score of the motor vehicle.

4. The method for classifying the safety performance of a motor vehicle based on the coupling effect of multiple factors according to claim 1, characterized in that, The process of transforming the comprehensive safety performance score of the motor vehicle through phase space mapping and extracting key features to obtain a low-dimensional discriminative feature set includes: A time-series data matrix is ​​constructed from the comprehensive safety performance score of the motor vehicle, and the optimal time delay parameter is determined by the mutual information method. The time series data matrix is ​​analyzed using the pseudo-nearest neighbor method to determine the optimal embedding dimension; Based on the optimal time delay parameter and the optimal embedding dimension, the comprehensive score of motor vehicle safety performance is reconstructed into a high-dimensional phase space trajectory; Calculate the nonlinear dynamic characteristic parameters of the high-dimensional phase space trajectory, including the maximum Lyapunov exponent, correlation dimension, and approximate entropy value; Complex network features are extracted from the high-dimensional phase space trajectory to obtain quantitative indicators of network topology features; By combining the nonlinear dynamic characteristic parameters and the network topology characteristic quantification index, the low-dimensional discriminative feature set is obtained through nonlinear mapping transformation.

5. The method for classifying motor vehicle safety performance based on the coupling effect of multiple factors according to claim 1, characterized in that, The step of applying an adaptive density clustering method to the low-dimensional discriminative feature set to determine the dynamic security level boundary value includes: Calculate the local density value and distance factor of each feature point in the low-dimensional discriminative feature set, and construct a density-distance decision graph; Initial cluster centers are identified from the density-distance decision map using a density peak detection method. Based on the distribution characteristics of the initial cluster centers, clustering is performed to obtain initial security level clusters; The effectiveness of the initial security level clusters is evaluated, and the optimal number of clusters is determined by calculating the silhouette coefficient. A closed decision surface is constructed around the security level cluster boundary corresponding to the optimal number of clusters to form a static security level boundary. Based on historical data analysis of the temporal change characteristics of the security level cluster, a boundary dynamic update rule is established to generate the dynamic security level boundary value.

6. The method for classifying motor vehicle safety performance based on the coupling effect of multiple factors according to claim 1, characterized in that, The process of using the dynamic safety level boundary values ​​to classify the real-time collected vehicle status into multiple scenarios and outputting graded safety risk warning information includes: The system receives vehicle dynamic parameters, environmental conditions, and driving behavior data collected by the distributed sensor network through a real-time data acquisition interface, and constructs a real-time status data matrix. A fast feature extraction algorithm is applied to the real-time state data matrix to obtain real-time feature vectors; The real-time feature vector is projected onto the feature space where the low-dimensional discriminant feature set is located through the phase space mapping transformation method to generate real-time low-dimensional discriminant points. Calculate the positional relationship between the real-time low-dimensional discrimination point and the dynamic safety level boundary value to determine the current vehicle safety status category; Based on the current vehicle safety status category and the dynamic change trend of the real-time low-dimensional discrimination point, a safety risk level is constructed. The graded safety risk warning information is generated based on the comparison table between the safety risk level and the preset risk threshold.

7. A vehicle safety performance classification system based on the coupling effect of multiple factors, characterized in that, For implementing the motor vehicle safety performance classification method based on the multi-factor coupling effect as described in any one of claims 1-6, the motor vehicle safety performance classification system based on the multi-factor coupling effect includes: The data acquisition module is used to collect dynamic parameters of motor vehicles, environmental conditions and driving behavior data through a distributed sensor network to obtain raw data on multi-dimensional safety influencing factors. The calculation module is used to calculate the mutual information value and transfer entropy of the original data of the multidimensional security influencing factors, and generate a nonlinear factor coupling relationship network graph. This includes: performing outlier detection and smoothing on the original data of the multidimensional security influencing factors to obtain a preprocessed dataset; calculating the joint probability distribution of factor pairs in the preprocessed dataset using kernel density estimation to obtain a mutual information value matrix between factors; segmenting the preprocessed dataset based on a sliding time window, calculating the transfer entropy between factors through symbolic transformation to form a directional coupling strength matrix; fusing the mutual information value matrix and the directional coupling strength matrix to construct a multi-level factor coupling network topology; pruning redundant connections in the multi-level factor coupling network topology using a sparse representation learning method, retaining key coupling paths; and analyzing the characteristics of the multi-level factor coupling network topology from micro, meso, and macro scales to generate the nonlinear factor coupling relationship network graph. The construction module is used to construct a three-layer safety index system based on the nonlinear factor coupling relationship network diagram to form a comprehensive score for motor vehicle safety performance; The conversion module is used to convert the comprehensive safety performance score of the motor vehicle through phase space mapping and extract key features to obtain a low-dimensional discriminative feature set; The clustering module is used to apply an adaptive density clustering method to the low-dimensional discriminative feature set to determine the dynamic security level boundary value; The classification module is used to classify the real-time collected vehicle status using the dynamic safety level boundary value and output graded safety risk warning information.

8. A vehicle safety performance classification device based on the coupling effect of multiple factors, characterized in that, It includes a memory and a processor, the memory storing a computer program that can run on the processor, and the processor executing the computer program to implement the motor vehicle safety performance classification method based on the multi-factor coupling influence as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it causes the processor to execute the motor vehicle safety performance classification method based on the multi-factor coupling effect as described in any one of claims 1 to 6.

Images