Driving behavior risk assessment method and system

Abstract

The invention discloses a driving behavior risk assessment method and system, and relates to the technical field of intelligent traffic and driving safety, and the method comprises the steps: collecting the operation parameter data of a vehicle, carrying out the feature extraction of the operation parameter data, extracting feature parameters to form a feature vector, and calculating the Mahalanobis distance score of the feature vector; according to the Mahalanobis distance score of the feature vector, carrying out risk grade division by adopting an improved hierarchical fuzzy inference system to obtain a risk grade score; and establishing a driver portrait model according to the risk level score, and when the risk level score exceeds a personalized risk threshold, outputting a driving risk behavior improvement suggestion to complete driving behavior risk assessment. According to the invention, a refined grading basis is provided for individual risk monitoring of the driver, dynamic monitoring and real-time intervention of driver behaviors are realized, traffic accidents can be reduced, and driving safety is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation and driving safety technology, and in particular to a driving behavior risk assessment method and system. Background Art

[0002] In recent years, with the rapid development of vehicle-mounted sensor technology and wireless communication technology, vehicle operation parameter collection and driving behavior risk assessment have become important research areas in intelligent transportation systems. Existing technologies have gradually evolved from the initial single data collection and risk judgment based on fixed thresholds to multi-source data fusion and multi-level information processing methods. Early driving risk assessment methods that relied on simple trigger mechanisms often used preset thresholds to judge single indicators such as vehicle acceleration, brake pressure, and speed changes, which was difficult to fully reflect the complexity of driving behavior. Methods based on data mining and statistical analysis were gradually introduced, describing the dynamic state of the vehicle through multi-level feature extraction, feature vector construction, and statistical distance quantification (such as Euclidean distance, Mahalanobis distance, etc.), and then using intelligent algorithms such as fuzzy reasoning and neural networks to assess driving risks.

[0003] However, although current technologies have made significant progress in the accuracy and real-time performance of risk identification, they still have deficiencies in processing data diversity, feature dimension redundancy, and threshold adaptability, resulting in some evaluation results lacking sufficient robustness and pertinence. Existing solutions often ignore individual driver behavior differences and dynamic changes in driving scenarios, making it difficult to form targeted risk behavior improvement recommendations, thus limiting the effectiveness of risk assessment systems in practical applications. Summary of the invention

[0004] In view of the problems existing in the existing driving behavior risk assessment methods, the present invention is proposed. Therefore, the problem to be solved by the present invention is how to provide a driving behavior risk assessment method and system.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a driving behavior risk assessment method, which comprises: Collecting vehicle operation parameter data, performing feature extraction on the operation parameter data, extracting feature parameters to form feature vectors, and calculating Mahalanobis distance scores of the feature vectors; The calculation of the Mahalanobis distance score of the feature vector includes: constructing a feature vector through vectorization processing based on the extracted feature parameters, calculating the difference between each data block and the standard feature template based on the feature vector, and calculating the Mahalanobis distance score; According to the Mahalanobis distance score of the feature vector, the improved hierarchical fuzzy inference system is used to divide the risk level and obtain the risk level score; The risk level classification using the improved hierarchical fuzzy inference system includes: Preset the initial fuzzy set, construct the membership function, and map the Mahalanobis distance score to the membership; According to the membership degree corresponding to the Mahalanobis distance score, a preliminary local risk assessment is conducted to form a local risk score based on a single indicator fuzzy rule, and a multidimensional fuzzy rule is constructed in combination with the operating parameter data; The local risk scores obtained are cross-validated to form local evaluation indicators, multiple local risk scores are hierarchically integrated, and the comprehensive risk membership function is constructed using fuzzy weighted average and maximum membership principle; The comprehensive risk membership function is defuzzified using the centroid method, and the risk level score is finally output and stored in the data recording system for model optimization; A driver profile model is established based on the risk level score. When the risk level score exceeds the personalized risk threshold, driving risk behavior improvement suggestions are output to complete the driving behavior risk assessment.

[0006] As a preferred embodiment of the driving behavior risk assessment method described in the present invention, the operating parameter data includes engine speed, instantaneous fuel consumption, vehicle speed change rate, brake pressure value, steering angle value, and lateral acceleration value and longitudinal acceleration value obtained by the vehicle-mounted inertial measurement unit.

[0007] As a preferred solution of the driving behavior risk assessment method of the present invention, the feature extraction of the operating parameter data includes: Each sensor monitors the operating status of different subsystems of the vehicle and transmits it to the acquisition module through the vehicle data bus, which pre-processes the original signal to generate a digital signal corresponding to each parameter and attaches a time stamp; According to the preset time window, the collected operating parameter data is stored in segments, and all the data received in each time window is grouped into a data block; According to the operating parameter data, the data is feature extracted to obtain feature parameters.

[0008] As a preferred embodiment of the driving behavior risk assessment method of the present invention, the formula of the Mahalanobis distance score is: ; In the formula, is the Mahalanobis distance score, X is the feature vector, is the reference feature mean vector, is the covariance matrix of the reference data set, and T is the transposed symbol.

[0009] As a preferred embodiment of the driving behavior risk assessment method of the present invention, the output of driving risk behavior improvement suggestions includes: Based on the risk level score, the multi-source characteristic parameters of the driver's operating behavior are collected and multi-dimensionally integrated with the risk level score to build a driver portrait model. , expressed as: ; in, represents the multidimensional behavior characteristic parameter vector in the i-th time period, represents the risk level score of the i-th time period; Introducing time-aware tags and scene context annotation fields to describe the time series correlation of behavioral features and operating environment tags respectively, enhancing the driver profile model to support cross-cycle behavioral pattern extraction, scoring trend analysis, and multi-dimensional behavioral association retrieval; Based on the historical risk score sequence in the driver portrait model, a personalized risk response model is constructed, the risk score distribution curve is obtained through the variable structure kernel density estimation method, and the personalized risk threshold of the current driver is calculated. , the formula is: ; Among them, KDE is the kernel density estimation function, This is the tolerance factor that the system adjusts adaptively based on driving style; Indicates that the kernel density estimation distribution is Quantile; If the current period risk level score Personalized risk threshold When the above conditions are met, the subsequent risk behavior identification and suggestion generation process is triggered; The risk behavior identification and suggestion generation process specifically includes: combining historical score correlation analysis with score curve fitting residual feedback mechanism to select leading behavior indicators that change with risk level; When a risk state is triggered, a feature subset is extracted from the driver profile model as the input vector of the neural network model; The neural network structure is an integrated multi-branch attention network, including a shared input layer and task branches. Each branch corresponds to a type of risky behavior. After extracting the low-level behavior pattern through a shared weight feature extractor, the branch modules output the probability value of the risk category respectively. Parse the risk behavior category labels and their probability values ​​output by the neural network, take the category with the highest confidence as the dominant risk behavior, and match the operation cycle features in the current driver portrait as semantic filling parameters; Utilize predefined composable suggestion templates and combine them with parameterized filling strategies to dynamically generate specific risk behavior improvement suggestions; If multiple coexisting risk behaviors are identified, a recommendation weight ranking mechanism is used to output a priority recommendation sequence based on the risk behavior identification confidence and the behavior influencing factor weights.

[0010] In a second aspect, the present invention provides a driving behavior risk assessment system, which includes: an acquisition module, used to collect vehicle operating parameter data, perform feature extraction on the operating parameter data, extract feature parameters to form a feature vector, and calculate the Mahalanobis distance score of the feature vector; a scoring module, used to perform risk level classification based on the Mahalanobis distance score of the feature vector using an improved hierarchical fuzzy inference system to obtain a risk level score; an output module, used to establish a driver portrait model based on the risk level score, and when the risk level score exceeds the personalized risk threshold, output driving risk behavior improvement suggestions to complete the driving behavior risk assessment.

[0011] In a third aspect, the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the processor executes the computer program, the steps of the driving behavior risk assessment method are implemented.

[0012] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, the steps of a driving behavior risk assessment method are implemented.

[0013] The beneficial effects of the present invention are: significantly improving the system's sensitivity and early warning capabilities to abnormal conditions, reducing the risk of misjudgment due to data redundancy or local anomalies, improving the accuracy of risk assessment, and also providing a refined classification basis for individualized risk monitoring of drivers. Dynamic monitoring and real-time intervention of driver behavior can reduce traffic accidents and improve driving safety. DETAILED DESCRIPTION

[0014] In order to make the above-mentioned purposes, features and advantages of the present invention more understandable, the specific implementation methods of the present invention are described in detail below in conjunction with the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0015] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0016] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0017] The first embodiment of the present invention provides a driving behavior risk assessment method, comprising: S1: Collecting vehicle operation parameter data, extracting features from the operation parameter data, extracting feature parameters to form feature vectors, and calculating Mahalanobis distance scores of the feature vectors; Specifically, the operating parameter data is collected by the vehicle-mounted sensors, including engine speed, instantaneous fuel consumption, vehicle speed change rate, brake pressure value, steering angle value, and lateral acceleration value and longitudinal acceleration value obtained by the vehicle-mounted inertial measurement unit; it includes a vehicle operating parameter data acquisition unit and a data segmentation storage unit. The vehicle operating parameter data acquisition unit collects multi-dimensional data such as vehicle speed, engine speed, fuel consumption, mileage, etc. from the vehicle dynamic system. The data segmentation storage unit divides the operating parameter data into time periods according to a pre-defined set time window, and archives and saves it with a data block identifier. The data block identifier is closely associated with the set time window to ensure that the subsequent processing modules can perform continuous analysis according to the data block sequence.

[0018] Each sensor monitors the operating status of different subsystems of the vehicle and transmits it to the acquisition module through the vehicle data bus. The acquisition module pre-processes the original signal, such as filtering and analog-to-digital conversion, to generate digital signals corresponding to each parameter, and adds accurate timestamps to ensure data continuity and consistency in the time domain.

[0019] The collected operating parameter data is stored in segments according to the preset time window. The preset time window is used as the basic unit of data segmentation, and the collection module collects all the data received in each time window into a data block.

[0020] The data information in each data block is organized in a unified format, including the sampling values ​​of all the above-mentioned operating parameter data and their corresponding time stamps. Each data block forms a continuous time series record, which is convenient for subsequent feature extraction and risk assessment processing.

[0021] During the data segmentation process, the pre-set time window and the data acquisition frequency are matched to ensure that the data block contains sufficient sampling points to reflect the dynamic changes in the vehicle's operating status.

[0022] Each data block adopts a unified naming rule and data format when stored, so that subsequent processing modules can directly call, parse and analyze it, ensuring the continuity, timeliness and accuracy of data collection and segmented storage, and providing a stable and reliable data foundation.

[0023] Extract characteristic parameters including engine speed fluctuation coefficient, fuel consumption change curve slope, vehicle speed mean square deviation, brake pressure peak ratio, steering angle change rate, lateral acceleration standard deviation, and longitudinal acceleration mean; based on the operating parameter data, perform primary, secondary, and comprehensive feature extraction on the data.

[0024] In the primary feature extraction stage, basic statistics such as mean, variance, maximum value, minimum value, etc. are obtained from the operating parameter data; In the secondary feature extraction stage, derived parameters such as acceleration change rate and fuel economy index are derived based on the vehicle dynamics model; In the comprehensive feature extraction stage, the time-frequency analysis method is used to obtain the signal frequency domain features. After the extraction results of each layer are normalized and weighted, it is ensured that the feature parameters have the distinguishability and robustness, providing a basis for subsequent data vectorization.

[0025] Based on the characteristic parameters, a high-dimensional characteristic vector is constructed through vectorization processing. After normalization and weight adjustment, the characteristic vector can reflect the global characteristics of the information contained in each data block.

[0026] Based on the feature vector, a statistical method is used to calculate the difference between each data block and the standard feature template, and the Mahalanobis distance is selected as the metric to calculate the Mahalanobis distance score. The score quantifies the degree to which the data block deviates from the standard in statistical distribution, providing an objective quantitative basis for risk assessment.

[0027] In each preset time window, the values ​​recorded by each sensor are obtained from the operating parameter data, including engine speed, instantaneous fuel consumption, vehicle speed change rate, brake pressure value, steering angle value, lateral acceleration value and longitudinal acceleration value. First, in the first-level extraction, basic statistics, arithmetic mean and variance are calculated for each parameter record sequence, and the maximum and minimum values ​​are recorded to describe the distribution range of the data.

[0028] In the second-level extraction, the differential components or derived indicators of the data (such as the gradient information reflecting the acceleration and deceleration of the vehicle in the vehicle speed change rate) are calculated based on the basic statistics, and frequency domain analysis is applied to some parameters. For example, the discrete Fourier transform (Mahalanobis distance score FT) is used to extract the spectral features in the engine speed signal to form richer feature parameters.

[0029] All statistics and derived indicators obtained through feature extraction at each level are combined into high-dimensional feature vectors according to a predetermined order, and each feature parameter is standardized to ensure that each component is in the same dimension.

[0030] Based on the feature vector, the degree of deviation between it and the reference feature distribution is calculated. The expression for calculating the Mahalanobis distance score is: ; In the formula, is the Mahalanobis distance score, which is used to quantify the degree of deviation between the current feature vector and the reference feature distribution, reflecting the severity of data anomalies. X is a feature vector, a high-dimensional data vector obtained by multi-level feature extraction, and each component corresponds to a certain vehicle operating parameter feature. is the reference feature mean vector, which represents the mean of each feature parameter calculated from the reference data set (representing the normal or baseline state). It is the covariance matrix of the reference data set, which describes the variance and covariance relationship between the feature parameters and reflects the statistical distribution characteristics of the reference data. T is the transposition symbol.

[0031] According to the Mahalanobis distance score, the data blocks are input into the improved hierarchical fuzzy inference system, which is composed of fuzzification unit, rule base unit and defuzzification unit.

[0032] S2: Based on the Mahalanobis distance score of the feature vector, an improved hierarchical fuzzy inference system is used to classify the risk level and obtain the risk level score; Specifically, based on the preset initial fuzzy sets of low risk, medium risk and high risk, a dynamically adjusted membership function is constructed.

[0033] An improved trapezoidal membership function model is adopted and adaptive adjustment parameters are embedded, whose values ​​are automatically updated according to historical risk data and current environmental status.

[0034] In addition, in order to enhance the boundary smoothness, nonlinear S-type or Z-type transformation is introduced into some risk intervals, so that the Mahalanobis distance score values ​​are mapped into low, medium and high risk membership in different intervals. Each membership reflects the matching degree of the Mahalanobis distance score in the corresponding risk interval.

[0035] A multi-level local fuzzy rule system is constructed. At the first level, a preliminary local risk assessment module is constructed based on the membership degree corresponding to a single Mahalanobis distance score value to form a local risk score based on a single indicator fuzzy rule.

[0036] Combined with other vehicle operating parameters (such as speed change rate, brake pressure, steering angle, etc.) and its dynamic state, multidimensional fuzzy rules are formed in the first layer, and composite rule expressions are designed so that the local risk score not only depends on the Mahalanobis distance score, but also reflects the interaction effects among other parameters.

[0037] For example, a rule can be set: when the Mahalanobis distance score is in the low-risk area and the auxiliary parameters show a stable state, the local risk score tends to be low; and when the Mahalanobis distance score has a significant degree of membership in both the medium and high-risk areas, and the vehicle speed and braking data show abnormal fluctuations, the local risk score is automatically adjusted upward.

[0038] After cross-validation, the first-level local risk scores form a set of local evaluation indicators, which provide basic data for subsequent levels.

[0039] Hierarchical fusion and comprehensive risk membership function construction, hierarchical fusion of multiple local risk scores in the first layer, using fuzzy weighted average and maximum membership principles, while introducing nonlinear correction factors and interaction terms to construct a comprehensive risk membership function.

[0040] The local scores of the same risk category are aggregated, and then the scores of different categories are weighted and synthesized. The weight parameters are dynamically adjusted based on historical data and real-time feedback to ensure that each local risk is reasonably reflected in the comprehensive evaluation.

[0041] During the fusion process, additional auxiliary parameters (such as real-time driving behavior and environmental variables) are involved in the adjustment to ensure that the final comprehensive risk membership function can comprehensively express the global risk level of the vehicle's operating status.

[0042] Defuzzification and risk level score determination,The comprehensive risk membership function is defuzzified using the centroid method and converted into a clear continuous risk level score.

[0043] During the defuzzification process, the integral interval is adaptively adjusted according to the real-time risk changes to adapt to nonlinear characteristics and dynamic environment; The resulting risk level score is used as the final output for vehicle safety monitoring and driving behavior intervention, and is stored in the data recording system for subsequent model optimization.

[0044] S3: A driver profile model is established according to the risk level score. When the risk level score exceeds the personalized risk threshold, driving risk behavior improvement suggestions are output to complete the driving behavior risk assessment.

[0045] Specifically, based on the risk level score, multi-source feature parameters related to the driver's operating behavior are collected and multi-dimensionally integrated with the risk level score to construct a driver portrait model, which is recorded as: ; in, represents the multidimensional behavior characteristic parameter vector in the i-th time period, Indicates the risk level score for this time period.

[0046] In order to enhance the structural usability of the model, time-aware tags and scene context annotation fields are introduced to describe the time series correlation of behavioral characteristics and operating environment labels (such as "nighttime", "congested", "slippery", etc.), respectively, to enhance the driver portrait model to support cross-cycle behavior pattern extraction, scoring trend analysis and multi-dimensional behavior association retrieval.

[0047] Based on the historical risk score sequence in the driver portrait model, a personalized risk response model is constructed. The risk score distribution curve is obtained by the variable structure kernel density estimation method. Combined with the skewness and kurtosis of the score distribution in the sliding window, the personalized risk threshold of the current driver is adaptively calculated. The threshold calculation is expressed as follows: ; Among them, KDE is the kernel density estimation function, This is the tolerance factor that the system adjusts adaptively based on driving style; Indicates that the kernel density estimation distribution is Quantile means that under this distribution, The risk score of is below this value; If the current period risk level score Exceeding Personalized Risk Threshold When the risk behavior is identified, the subsequent risk behavior identification and suggestion generation process is triggered.

[0048] The risk behavior identification and suggestion generation process specifically includes: Combine historical score correlation analysis with score curve fitting residual feedback mechanism to select leading behavior indicators related to risk level changes; When a risk state is triggered, a feature subset with high correlation with the current score is extracted from the driver portrait model as an input vector of the neural network model.

[0049] The feature subset screening method combines historical score correlation analysis (Pearson correlation coefficient) with the score curve fitting residual feedback mechanism to select dominant behavioral indicators that are significantly correlated with changes in risk levels.

[0050] The neural network structure is an integrated multi-branch attention network, including a shared input layer and multiple task branches, each branch corresponding to a risk behavior type (such as sudden lane change, fatigue driving, operation delay, etc.). After extracting the low-level behavior pattern through the shared weight feature extractor, the branch module outputs the probability value for each risk category.

[0051] The risk behavior category labels and their probability values ​​output by the neural network are parsed, and the category with the highest confidence is taken as the dominant risk behavior. The operation cycle characteristics in the current portrait (such as frequency, duration, triggering situation, etc.) are further matched as semantic filling parameters.

[0052] Utilize predefined composable suggestion module templates combined with parameterized population strategies to dynamically generate specific risk behavior improvement suggestions.

[0053] If multiple coexisting risk behaviors are identified, a recommendation weight ranking mechanism is used to output a priority recommendation sequence based on the risk behavior identification confidence and the behavior influencing factor weights.

[0054] Furthermore, this embodiment also provides a driving behavior risk assessment system, including: The acquisition module is used to collect the operating parameter data of the vehicle, perform feature extraction on the operating parameter data, extract the feature parameters to form a feature vector, and calculate the Mahalanobis distance score of the feature vector; The scoring module is used to classify the risk level according to the Mahalanobis distance score of the feature vector using an improved hierarchical fuzzy inference system to obtain a risk level score; The output module is used to establish a driver portrait model according to the risk level score, and when the risk level score exceeds the personalized risk threshold, output driving risk behavior improvement suggestions to complete the driving behavior risk assessment.

[0055] This embodiment also provides a computer device, which is applicable to the case of a driving behavior risk assessment method, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement all or part of the steps of the method described in the embodiment of the present invention as proposed in the above embodiment.

[0056] This embodiment also provides a storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the method in any optional implementation of the above embodiment is executed. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read-Only Memory, referred to as EPROM), programmable read-only memory (Programmable Read-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, disk or optical disk.

[0057] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be found in the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0058] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A driving behavior risk assessment method, characterized in that: include: Collecting vehicle operation parameter data, performing feature extraction on the operation parameter data, extracting feature parameters to form feature vectors, and calculating Mahalanobis distance scores of the feature vectors; The calculation of the Mahalanobis distance score of the feature vector includes: constructing a feature vector through vectorization based on the extracted feature parameters, calculating the difference between each data block and the standard feature template based on the feature vector, and calculating the Mahalanobis distance score; According to the Mahalanobis distance score of the feature vector, the improved hierarchical fuzzy inference system is used to divide the risk level and obtain the risk level score; The risk level classification using the improved hierarchical fuzzy inference system includes: Preset the initial fuzzy set, construct the membership function, and map the Mahalanobis distance score to the membership; According to the membership degree corresponding to the Mahalanobis distance score, a preliminary local risk assessment is conducted to form a local risk score based on a single indicator fuzzy rule, and a multidimensional fuzzy rule is constructed in combination with the operating parameter data; The local risk scores obtained are cross-validated to form local evaluation indicators, multiple local risk scores are hierarchically integrated, and the comprehensive risk membership function is constructed using fuzzy weighted average and maximum membership principle; The comprehensive risk membership function is defuzzified using the centroid method, and the risk level score is finally output and stored in the data recording system for model optimization; A driver profile model is established based on the risk level score. When the risk level score exceeds the personalized risk threshold, driving risk behavior improvement suggestions are output to complete the driving behavior risk assessment.

2. The driving behavior risk assessment method according to claim 1, characterized in that: The operating parameter data include engine speed, instantaneous fuel consumption, vehicle speed change rate, brake pressure value, steering angle value, and lateral acceleration value and longitudinal acceleration value obtained by the vehicle-mounted inertial measurement unit.

3. The driving behavior risk assessment method according to claim 2, characterized in that: The feature extraction of the operating parameter data comprises: Each sensor monitors the operating status of different subsystems of the vehicle and transmits it to the acquisition module through the vehicle data bus, which pre-processes the original signal to generate a digital signal corresponding to each parameter and attaches a time stamp; According to the preset time window, the collected operating parameter data is stored in segments, and all the data received in each time window is grouped into a data block; According to the operating parameter data, the data is feature extracted to obtain feature parameters.

4. The driving behavior risk assessment method according to claim 3, characterized in that: The formula for the Mahalanobis distance score is: ; In the formula, is the Mahalanobis distance score, X is the feature vector, is the reference feature mean vector, is the covariance matrix of the reference data set, and T is the transposed symbol.

5. The driving behavior risk assessment method according to claim 4, characterized in that: The output driving risk behavior improvement suggestions include: Based on the risk level score, the multi-source characteristic parameters of the driver's operating behavior are collected and multi-dimensionally integrated with the risk level score to build a driver portrait model. , expressed as: ; in, represents the multidimensional behavior characteristic parameter vector in the i-th time period, represents the risk level score of the i-th time period; Introduce time-aware tags and scene context annotation fields to describe the time series correlation of behavioral features and the operating environment labels respectively; Based on the historical risk score sequence in the driver portrait model, a personalized risk response model is constructed, the risk score distribution curve is obtained through the variable structure kernel density estimation method, and the personalized risk threshold of the current driver is calculated. , the formula is: ; Among them, KDE is the kernel density estimation function, This is the tolerance factor that the system adjusts adaptively based on driving style; Indicates that the kernel density estimation distribution is Quantile; If the current period risk level score Personalized risk threshold When the above conditions are met, the subsequent risk behavior identification and suggestion generation process is triggered; The risk behavior identification and suggestion generation process specifically includes: Combine historical score correlation analysis with score curve fitting residual feedback mechanism to select leading behavior indicators related to risk level changes; When a risk state is triggered, a feature subset is extracted from the driver profile model as the input vector of the neural network model; The neural network structure is an integrated multi-branch attention network, including a shared input layer and task branches. Each branch corresponds to a type of risky behavior. After extracting the low-level behavior pattern through a shared weight feature extractor, the branch modules output the probability value of the risk category respectively. Parse the risk behavior category labels and their probability values ​​output by the neural network, take the category with the highest confidence as the dominant risk behavior, and match the operation cycle features in the current driver portrait as semantic filling parameters; Utilize predefined composable suggestion templates and combine them with parameterized filling strategies to dynamically generate specific risk behavior improvement suggestions; If multiple coexisting risk behaviors are identified, a recommendation weight ranking mechanism is used to output a priority recommendation sequence based on the risk behavior identification confidence and the behavior influencing factor weights.

6. A driving behavior risk assessment system, based on the driving behavior risk assessment method according to any one of claims 1 to 5, characterized in that: include, The acquisition module is used to collect the operating parameter data of the vehicle, perform feature extraction on the operating parameter data, extract the feature parameters to form a feature vector, and calculate the Mahalanobis distance score of the feature vector; The scoring module is used to classify the risk level according to the Mahalanobis distance score of the feature vector using an improved hierarchical fuzzy inference system to obtain a risk level score; The output module is used to establish a driver portrait model according to the risk level score, and output driving risk behavior improvement suggestions when the risk level score exceeds the personalized risk threshold.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the driving behavior risk assessment method as described in any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the driving behavior risk assessment method as described in any one of claims 1 to 5 are implemented.

Images