A safety education content pushing system and method based on driving behavior analysis

Abstract

This invention discloses a safety education content delivery system and method based on driving behavior analysis, relating to the field of traffic safety technology. The proposed solution includes a driving data collection module, a driving behavior analysis module, a safety education resource library module, a personalized delivery decision module, a content delivery execution module, and an effect feedback optimization module. This invention not only accurately extracts driving behavior characteristics and assesses risks through multi-dimensional data collection and advanced models, identifying driver behavioral shortcomings and potential hazards, but also dynamically matches driver risk levels and learning styles based on a tagged resource library and priority model to deliver personalized educational content, solving the problem of homogenization. Furthermore, it constructs a closed-loop feedback optimization mechanism, continuously optimizing models, materials, and strategies based on feedback data to improve educational effectiveness. It adapts to various driving scenarios and usage habits, exhibiting strong compatibility and practicality, effectively correcting non-standard driving behaviors, and reducing the accident rate.

Description

Technical Field

[0001] This invention relates to the field of traffic safety technology, and in particular to a system and method for delivering safety education content based on driving behavior analysis. Background Technology

[0002] With the continuous growth of motor vehicle ownership, the traffic accident rate remains high. Among them, drivers' non-standard driving behavior is one of the main causes of accidents. Existing driving safety education mostly adopts a uniform delivery mode, which lacks precise analysis of the individual driving behavior characteristics of drivers. This results in the education content not matching the actual needs of drivers, and the education effect is greatly reduced. Currently, some driving behavior analysis systems can only achieve basic identification and warning of dangerous behaviors, and have failed to build a complete behavior analysis model and a personalized education content push mechanism. At the same time, the existing safety education content push relies heavily on manual screening, lacks in-depth mining of driving behavior data, and cannot dynamically adjust the push content and priority according to the driver's risk level and behavioral shortcomings. In addition, some technical solutions do not consider personalized factors such as driver learning style and driving scenario, which further reduces the pertinence and effectiveness of safety education. Therefore, there is an urgent need for a safety education content delivery system and method that can accurately analyze driving behavior characteristics, dynamically match educational needs, and intelligently push personalized content to solve the problems of insufficient targeting and poor educational effect in existing technologies. Summary of the Invention

[0003] The present invention proposes a safety education content delivery system and method based on driving behavior analysis, which solves the above-mentioned shortcomings of the prior art.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: A safety education content delivery system based on driving behavior analysis includes a driving data collection module, a driving behavior analysis module, a safety education resource library module, a personalized delivery decision module, a content delivery execution module, and an effect feedback optimization module. The driving data acquisition module is used to collect real-time driving operation data of the driver, vehicle operating status data and driving environment data, and to preprocess the collected data; The driving behavior analysis module constructs a driving behavior feature extraction model and a driving risk assessment model based on the preprocessed driving data, and outputs driving behavior feature vectors and driving risk level assessment results. The safety education resource library module is used to store multiple types and dimensions of safety education materials, and to tagged each educational material to generate a set of identification tags; The personalized push decision module combines driving behavior feature vectors, driving risk level assessment results, and driver personalized information to construct a priority push model and match the educational materials to be pushed and the push order. The content push execution module is used to push the educational materials to be pushed to the driver's terminal according to the push order, and collect push effect feedback data. The content push execution module includes a push unit and a feedback collection unit. The push unit supports multi-terminal push, including vehicle terminal, mobile APP, and WeChat official account, and selects the optimal push channel according to the driver's online status and terminal preference; The feedback collection unit is used to collect feedback data such as the driver's browsing time, completion rate, interactive comments, and assessment results of the pushed content, and transmits the feedback data to the effect feedback optimization module. After in-depth analysis by the effect feedback optimization module, it is synchronized to the driving behavior analysis module and the personalized push decision module for the optimization of the models and strategies of each module. The effect feedback optimization module is used to receive feedback data transmitted by the content push execution module, build a feedback analysis model, and output model optimization parameters, material optimization suggestions and push strategy adjustment schemes, which are then synchronized to the driving behavior analysis module, safety education resource library module and personalized push decision module to achieve closed-loop optimization of the entire process.

[0005] Furthermore, the driving data acquisition module includes an on-board sensor unit, an environmental perception unit, and a data preprocessing unit; The vehicle-mounted sensor unit includes a steering wheel angle sensor, an accelerator pedal sensor, a brake pedal sensor, a vehicle speed sensor, and a speed sensor, which are used to collect driving operation data such as the driver's steering operation, acceleration operation, and braking operation, as well as the vehicle's driving speed and engine speed operating status data. The environmental perception unit includes a GPS positioning module, a camera, a radar sensor, and a weather sensor, used to collect driving environment data for driving scenarios (urban roads, highways, rural roads), road condition information (congestion, smooth traffic, construction), distances to surrounding vehicles, and weather conditions (rain, snow, fog, high temperature, low temperature). The data preprocessing unit is used to clean (remove outliers and missing values), standardize, and fuse the collected data to generate structured driving data samples.

[0006] Furthermore, the driving behavior analysis module includes a feature extraction unit and a risk assessment unit; The feature extraction unit uses an improved convolutional neural network (CNN) model, takes structured driving data samples as input, and extracts key features of driving behavior, including features of rapid acceleration, emergency braking, sharp turning, speeding, fatigued driving, and distracted driving, to generate a driving behavior feature vector. The risk assessment unit constructs a hierarchical risk assessment indicator system, including operational risk indicators, state risk indicators, and environmental risk indicators. It adopts a risk assessment model based on gradient boosting tree, and combines driving behavior feature vectors and driving environment data to output driving risk levels (low risk, medium risk, high risk) and risk shortcomings.

[0007] Furthermore, the safety education resource library module includes a material storage unit and a tag generation unit; The educational materials stored in the material storage unit include video, audio, text, PPT, and simulation exercise programs, covering basic driving standards, dangerous behavior correction, emergency response skills, scenario-based driving techniques, and interpretation of laws and regulations. The tag generation unit creates multi-dimensional identification tags for each educational material, including behavior type tags (rapid acceleration, sudden braking, etc.), risk level tags (low risk, medium risk, high risk), scene type tags (urban roads, highways, etc.), weather adaptation tags (rainy days, snowy days, etc.), and material format tags (videos, images, etc.), generating an identification tag set and establishing a tag index.

[0008] Furthermore, the personalized push decision module includes a user information management unit, a keyword extraction unit, a tag matching unit, and a priority sorting unit; The user information management unit is used to store the driver's real-name authentication information, driving experience, vehicle type, historical assessment results, and personalized information on learning style preferences (visual, auditory, and practical). The keyword extraction unit extracts and pushes keywords based on the driving risk level assessment results, risk shortcomings, and driving behavior feature vectors. The keywords include risk behavior keywords, scenario keywords, and demand keywords. The tag matching unit is used to calculate the index difference between the push keyword sequence string and the identification tag sequence string. When the index difference is within a preset threshold range, it is determined that there is a mapping relationship between the two, and candidate educational materials are matched. The priority sorting unit prioritizes candidate educational materials based on risk level weight, urgency of behavioral shortcomings, learning style suitability, and historical push results, generating a list of educational materials to be pushed and the push order.

[0009] Furthermore, the effect feedback optimization module includes a feedback data receiving unit, a feedback data analysis unit, an optimization parameter generation unit, and an optimization scheme push unit; The feedback data receiving unit is used to receive feedback data transmitted by the content push execution module, including basic behavioral data (browsing time, completion rate, number of clicks), interactive feedback data (comment content, likes / favorites), and assessment result data (answer accuracy rate, practical operation score), and to classify and organize the data. The feedback data analysis unit uses natural language processing (NLP) technology and statistical analysis methods to explore the correlation between feedback data and push effects and driving behavior improvement. The feedback data analysis unit constructs a multi-dimensional feedback analysis model, uses NLP technology to analyze the sentiment and demand in the comments, and uses statistical analysis methods to explore the correlation between feedback data and push effects and driving behavior improvement, and identifies shortcomings in push content, model evaluation bias and strategy adaptation issues. Based on the analysis results, the optimization parameter generation unit outputs adjustment parameters for the driving behavior feature extraction model, risk assessment model, and priority push model, including model weight coefficients, threshold settings, feature extraction dimensions, etc. The optimization scheme push unit outputs material optimization suggestions to the safety education resource library (such as adding materials for a certain type of scenario or optimizing the presentation of existing materials), and outputs push strategy adjustment schemes to the personalized push decision module (such as adjusting the weight of the priority evaluation function or optimizing the tag matching threshold), and synchronizes the above parameters and schemes to the corresponding modules to complete the optimization.

[0010] A method for pushing safety education content based on driving behavior analysis, applicable to any of the above-mentioned safety education content pushing systems based on driving behavior analysis, includes the following steps: S1. Driving data collection and preprocessing: The driving data collection module collects the driver's driving operation data, vehicle operating status data and driving environment data. After cleaning, standardization and data fusion, structured driving data samples are generated. S2. Driving Behavior Analysis and Risk Assessment: The driving behavior analysis module is based on structured driving data samples. It extracts key features of driving behavior through a feature extraction model, generates driving behavior feature vectors, and calculates driving risk levels and risk shortcomings through a risk assessment model. S3. Educational Material Tagging and Resource Management: The safety education resource library module tags various types of stored safety education materials, generates identification tag sets, and establishes an index. S4. Personalized Push Decision: The personalized push decision module combines driving behavior feature vectors, risk level assessment results and driver personalized information to extract push keywords, match candidate educational materials and prioritize them to obtain a list of educational materials to be pushed and the push order. S5. Content Push and Effect Feedback: The content push execution module pushes the educational materials to be pushed to the driver's terminal in the order of push, and collects push effect feedback data, which is then transmitted to the effect feedback optimization module. S6. Feedback Analysis and Full-Process Optimization: The effect feedback optimization module performs in-depth analysis of feedback data, generates model optimization parameters, material optimization suggestions, and push strategy adjustment plans, and synchronizes them to the corresponding modules to complete the optimization, forming a closed-loop mechanism of "data collection - analysis - push - feedback - optimization".

[0011] Furthermore, the training process of the driving behavior feature extraction model and the construction process of the driving risk assessment model in step S2 are as follows: The training process includes: S21. Construct a driving behavior sample dataset, including normal driving samples and abnormal driving samples (rapid acceleration, sudden braking, sharp turns, speeding, fatigued driving, distracted driving), and label each sample with the corresponding behavior type label; S22. Perform data augmentation on the sample dataset, including random cropping, noise addition, and data rotation, to increase the sample size; S23. Divide the enhanced sample dataset into a training set and a test set, input them into the improved CNN model for training, and adjust the model parameters (learning rate, number of iterations, convolution kernel size) until the model test accuracy reaches the preset threshold to obtain the trained feature extraction model. The construction process includes: S2a. Construct a hierarchical risk assessment indicator system. The specific risk assessment indicators are as follows: The primary indicator is driving risk; Secondary indicators include operational risk, condition risk, and environmental risk; The three-level indicators include the frequency of rapid acceleration, the frequency of sudden braking, the frequency of sharp turns, the percentage of time spent speeding, the duration of fatigued driving, the number of times of distracted driving, potential vehicle malfunctions, the complexity of road conditions, and the severity of weather conditions. S2b. Use the Analytic Hierarchy Process (AHP) to determine the weight coefficients of each indicator; S2c, a risk assessment model based on the XGBoost algorithm, takes driving behavior feature vectors and the weights of each indicator as input, and outputs the driving risk level (low risk, medium risk, high risk) and the risk weakness items (the three tertiary indicators with the highest weight). The core calculation logic of the risk assessment model is as follows: ; in, The driving risk score is calculated (the risk level is determined based on the score range: low risk, medium risk, high risk). The total number of three-level indicators in the hierarchical risk assessment indicator system (n=9, including the frequency of rapid acceleration, the frequency of rapid braking, etc.). The weight coefficient of the i-th tertiary indicator (determined by the Analytic Hierarchy Process (AHP), satisfying ∑i=1nωi=1); The quantitative score of the i-th tertiary indicator (calculated by combining the driving behavior feature vector with actual driving data, such as the standardized score of the percentage of speeding time and the duration of fatigue driving).

[0012] Furthermore, in step S4, the construction and decision-making process of the priority push model includes: S41. Extract push keywords: Determine core keywords based on risk level (high risk corresponds to dangerous behavior correction and emergency response; medium risk corresponds to standardized operation and skills improvement; low risk corresponds to knowledge popularization and habit formation), determine special keywords based on risk shortcomings, and determine scenario keywords based on driving scenarios and weather. S42. Tag Matching: Calculate the index value of the push keyword sequence. and the index value of the identification label sequence string ,when When δ is a preset threshold, ranging from 0.1 to 0.3, the two are determined to be a match, and candidate educational materials are obtained through screening. S43. Priority Ranking: Constructing a Priority Evaluation Function ,in, The risk level is weighted (1.0 for high risk, 0.7 for medium risk, and 0.4 for low risk). The urgency weight of behavioral shortcomings is determined by a value ranging from 0.6 to 1.0. Learning style fit (value range 0.5-1.0), Weighting of historical push performance (value range 0.3-1.0). These are the weighting coefficients, and The candidate educational materials are sorted in descending order based on the evaluation function calculation results to obtain the push order.

[0013] Furthermore, the feedback analysis and overall process optimization in step S6 includes the following sub-steps: S61. Data Reception and Processing: Receive feedback data transmitted by the content push execution module, classify it into basic behavior data, interactive feedback data, and assessment result data, remove invalid data (such as duplicate clicks caused by misoperation, meaningless comments), and generate a standardized feedback dataset. S62. Multi-dimensional in-depth analysis: NLP technology is used to analyze the comments and identify drivers' satisfaction with the pushed materials (such as "the video is too long" or "the content is very practical") and potential needs (such as "the need for materials related to night driving"). Through correlation analysis, the correlation between the completion rate of a certain type of material and the corresponding improvement rate of driving behavior is determined (such as drivers with a completion rate of ≥90% for emergency braking materials have an average reduction of 40% in the subsequent occurrence of emergency braking behavior). S63. Optimization solution generation: For problems found in the analysis, such as the low accuracy of the risk assessment model in identifying fatigued driving, output the model feature extraction dimension adjustment parameters (add eyelid closure feature input), such as the completion rate of a certain type of scene material (rural road) is less than 60% after being pushed, output material optimization suggestions (add rural road scene simulation exercise material). S64. Optimization Plan Execution: The generated optimization parameters and suggestions are synchronized to the driving behavior analysis module, the safety education resource library module, and the personalized push decision module. Each module adjusts model parameters, updates materials, and optimizes strategies based on the received information to ensure that subsequent pushes are more in line with the driver's needs.

[0014] Compared with existing technologies, the beneficial effects of this invention are: 1. This invention, through multi-dimensional driving data collection, combined with an improved CNN model and XGBoost risk assessment model, achieves accurate extraction of driving behavior characteristics and scientific assessment of driving risks. It can accurately identify the driver's behavioral shortcomings and potential risks, providing a reliable basis for personalized education delivery. 2. This invention constructs a tagged safety education resource library, and combined with a priority push model, realizes dynamic matching of educational materials with drivers' needs. It can push personalized educational content based on drivers' risk level, behavioral shortcomings, learning style and driving scenarios, solving the problems of homogeneous educational content and insufficient targeting in the existing technology. 3. This invention introduces a push effect feedback mechanism. By collecting feedback information such as drivers' browsing data and assessment results, the model parameters are dynamically optimized, continuously improving the accuracy of push notifications and the educational effect. This can effectively correct drivers' non-standard driving behavior and reduce the traffic accident rate. 4. This invention supports multi-terminal push, adapts to different driving scenarios and driver usage habits, has good compatibility and practicality, and can be widely used in driving safety education scenarios for various motor vehicles such as private cars, ride-hailing vehicles, freight vehicles, and passenger vehicles. 5. This invention constructs a closed-loop optimization mechanism for the entire process through an effect feedback optimization module. By deeply analyzing the push effect feedback data, it can accurately identify the problems existing in each link, dynamically optimize model parameters, educational materials and push strategies, achieve continuous improvement in the effect of safety education, and further strengthen the effect of correcting driving behavior. In summary, this invention not only accurately extracts driving behavior characteristics and assesses risks through multi-dimensional data collection and advanced models, identifying driver behavioral shortcomings and potential hazards, but also dynamically matches driver risk levels and learning styles based on a tagged resource library and priority model to deliver personalized educational content, solving the problem of homogenization. Furthermore, it constructs a closed-loop feedback optimization mechanism to continuously optimize models, materials, and strategies based on feedback data, improving educational effectiveness. It is adaptable to various driving scenarios and usage habits, exhibiting strong compatibility and practicality, effectively correcting non-standard driving behaviors, and reducing the accident rate. Attached Figure Description

[0015] Figure 1 This is a block diagram of the overall system modules of a safety education content delivery system and method based on driving behavior analysis proposed in this invention. Figure 2 This is a flowchart illustrating the core process and method steps of a safety education content delivery system and method based on driving behavior analysis proposed in this invention. Detailed Implementation

[0016] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0017] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0018] Example 1 Reference Figure 1 - Figure 2 A safety education content delivery system based on driving behavior analysis includes a driving data collection module, a driving behavior analysis module, a safety education resource library module, a personalized delivery decision module, a content delivery execution module, and an effect feedback optimization module. The driving data acquisition module includes an onboard sensor unit, an environmental perception unit, and a data preprocessing unit; The vehicle sensor unit includes a steering wheel angle sensor, accelerator pedal sensor, brake pedal sensor, vehicle speed sensor, and speed sensor, which are used to collect driving operation data such as the driver's steering operation, acceleration operation, and braking operation, as well as the vehicle's driving speed and engine speed operating status data. The environmental perception unit includes a GPS positioning module, a camera, a radar sensor, and a weather sensor, which are used to collect driving environment data for driving scenarios (urban roads, highways, rural roads), road condition information (congestion, smooth traffic, construction), distances to surrounding vehicles, and weather conditions (rain, snow, fog, high temperature, low temperature). The data preprocessing unit is used to clean (remove outliers and missing values), standardize, and fuse the collected data to generate structured driving data samples. The vehicle sensor unit in the driving data acquisition module collects data such as the driver's steering angle, accelerator pedal, brake pedal, and vehicle speed through a steering wheel angle sensor, accelerator pedal sensor, brake pedal sensor, and vehicle speed sensor. The environmental perception unit determines the driving scenario as urban roads through a GPS positioning module, identifies traffic congestion through a camera, and collects weather data as rainy through a weather sensor. The data preprocessing unit cleans the collected data, removes outliers caused by sensor malfunctions, and generates structured driving data samples after standardization.

[0019] In this invention, the driving behavior analysis module includes a feature extraction unit and a risk assessment unit; The feature extraction unit uses an improved convolutional neural network (CNN) model, taking structured driving data samples as input, to extract key features of driving behavior, including features of rapid acceleration, emergency braking, sharp turning, speeding, fatigued driving, and distracted driving, and to generate driving behavior feature vectors. The risk assessment unit constructs a hierarchical risk assessment indicator system, including operational risk indicators, state risk indicators, and environmental risk indicators. It adopts a risk assessment model based on gradient boosting tree, and combines driving behavior feature vectors and driving environment data to output driving risk levels (low risk, medium risk, high risk) and risk shortcomings. The feature extraction unit in the driving behavior analysis module inputs structured driving data samples into the trained improved CNN model to extract key features such as the frequency of sudden braking (5 times / hour) and the number of times of distracted driving (3 times / hour), generating a driving behavior feature vector. The risk assessment unit, based on a hierarchical risk assessment index system and the XGBoost model, combined with environmental data such as rainy days and congested urban roads, calculates the driving risk level as medium risk, with the risk shortcomings being sudden braking behavior, distracted driving, and insufficient driving skills in rainy weather.

[0020] In this invention, the safety education resource library module includes a material storage unit and a tag generation unit; The educational materials stored in the material storage unit include video, audio, text, PPT, and simulation exercise programs, covering basic driving regulations, dangerous behavior correction, emergency response skills, scenario-based driving techniques, and interpretation of laws and regulations. The tag generation unit creates multi-dimensional identification tags for each educational material, including behavior type tags (rapid acceleration, sudden braking, etc.), risk level tags (low risk, medium risk, high risk), scene type tags (urban roads, highways, etc.), weather adaptation tags (rainy days, snowy days, etc.), and material format tags (videos, images, text, etc.), generating an identification tag set and establishing a tag index; The material storage unit in the safety education resource library module stores educational materials such as "Operating Procedures for Emergency Braking in Rainy Weather" (video), "Dangers of Distracted Driving in Congested Urban Roads" (text and images), and "Complete Guide to Driving Techniques in Rainy Weather" (PPT). The tag generation unit adds tags to the above materials respectively. For example, the tags for "Operating Procedures for Emergency Braking in Rainy Weather" are "emergency braking, rainy weather, medium risk, urban roads, video".

[0021] In this invention, the personalized push decision module includes a user information management unit, a keyword extraction unit, a tag matching unit, and a priority sorting unit; The user information management unit is used to store the driver's real-name authentication information, driving experience, vehicle type, historical assessment results, and personalized information on learning style preferences (visual, auditory, and practical). The keyword extraction unit extracts and pushes keywords based on the driving risk level assessment results, risk shortcomings, and driving behavior feature vectors. The keywords include risk behavior keywords, scenario keywords, and demand keywords. The tag matching unit is used to calculate the index difference between the push keyword sequence string and the identification tag sequence string. When the index difference is within the preset threshold range, it is determined that there is a mapping relationship between the two and the candidate educational materials are matched. The priority sorting unit prioritizes candidate educational materials based on risk level weight, urgency of behavioral shortcomings, learning style compatibility, and historical push results, generating a list of educational materials to be pushed and the push order. The user information management unit in the personalized push decision module stores the driver's personalized information: 3 years of driving experience, visual learning style preference, and qualified historical assessment results; the keyword extraction unit extracts push keywords as "sudden braking, distracted driving, driving in the rain, urban roads, medium risk"; the tag matching unit calculates the index difference between keywords and material tags to match candidate educational materials; the priority sorting unit calculates based on the priority evaluation function to determine the push order as "Sudden Braking Operation Specifications in Rainy Weather" (video) → "Hazards of Distracted Driving on Congested Urban Roads" (text and images) → "Complete Guide to Driving Techniques in Rainy Weather" (PPT).

[0022] In this invention, the content push execution module is used to push the educational materials to be pushed to the driver's terminal according to the push order, and collect push effect feedback data. The content push execution module includes a push unit and a feedback collection unit. The push unit supports push notifications from multiple terminals, including in-vehicle terminals, mobile apps, and WeChat official accounts, and selects the optimal push channel based on the driver's online status and terminal preferences. The feedback collection unit is used to collect feedback data such as the driver's browsing time, completion rate, interactive comments, and assessment results of the pushed content, and transmits the feedback data to the effect feedback optimization module. After in-depth analysis by the effect feedback optimization module, it is synchronized to the driving behavior analysis module and the personalized push decision module for the optimization of the models and strategies of each module. The content delivery execution module pushes the aforementioned educational materials sequentially via the driver's mobile app. The feedback collection unit collects data showing that the driver's viewing time for the "Rainy Weather Emergency Braking Operation Specifications" was 8 minutes (effective viewing), the completion rate was 90% (effective completion), and the assessment score was 85 points (pass). This feedback data is then transmitted to the effect feedback optimization module. After analyzing the data, the effect feedback optimization module finds that the driver has a high affinity for video materials (completion rate ≥ 90%) and outputs optimization suggestions: prioritize pushing video materials in the future. Simultaneously, the assessment score data is synchronized to the driving behavior analysis module to adjust the weighting coefficient of emergency braking behavior in the risk assessment model. In this invention, the effect feedback optimization module includes a feedback data receiving unit, a feedback data analysis unit, an optimization parameter generation unit, and an optimization scheme push unit; The feedback data receiving unit is used to receive feedback data transmitted by the content push execution module, including basic behavioral data (browsing time, completion rate, number of clicks), interactive feedback data (comment content, likes / favorites), and assessment result data (answer accuracy rate, practical operation score), and to classify and organize the data. The feedback data analysis unit uses natural language processing (NLP) technology and statistical analysis methods to explore the correlation between feedback data and push notification effectiveness and driving behavior improvement. The feedback data analysis unit constructs a multi-dimensional feedback analysis model, uses NLP technology to analyze the sentiment and demand in the comments, and uses statistical analysis methods to explore the correlation between feedback data and push notification effectiveness and driving behavior improvement, and identifies shortcomings in push notification content, model evaluation bias and strategy adaptation issues. Based on the analysis results, the parameter generation unit optimizes and outputs adjustment parameters for the driving behavior feature extraction model, risk assessment model, and priority push model, including model weight coefficients, threshold settings, and feature extraction dimensions. The optimization solution push unit outputs material optimization suggestions for the safety education resource library (such as adding materials for a certain type of scenario or optimizing the presentation of existing materials), and outputs push strategy adjustment plans for the personalized push decision module (such as adjusting the weight of the priority evaluation function or optimizing the tag matching threshold), and synchronizes the above parameters and plans to the corresponding modules to complete the optimization.

[0023] Example 2 Reference Figure 2 A method for pushing safety education content based on driving behavior analysis includes the following steps: S1. Driving data collection and preprocessing: The driver's driving operation data (steering, acceleration, braking), vehicle operating status data (vehicle speed, engine speed) and driving environment data (sunny weather, smooth road conditions) are collected through vehicle sensors and environmental perception equipment during the driving process on the highway. After cleaning, standardization and data fusion, structured driving data samples are generated. S2. Driving Behavior Analysis and Risk Assessment: Driving behavior features were extracted using an improved CNN model, revealing speeding behavior (15% of the time spent at speeds ≥120km / h), and generating a driving behavior feature vector. The XGBoost risk assessment model was used to calculate the risk level, which was determined to be high risk, with the risk weaknesses being speeding and insufficient emergency response capabilities on highways. S3. Educational Material Tagging and Resource Management: Materials such as "Hazards and Penalties of Speeding on Highways" (video), "Emergency Response Procedures for Highway Traffic Accidents" (simulation exercise procedure), and "Safe Driving Standards on Highways" (audio) were selected from the safety education resource library and tagged with "speeding, highway, high risk, video", "emergency response, highway, high risk, simulation exercise", and "high-speed driving, high risk, audio" respectively. S4. Personalized Push Decision: Combining personalized information such as driver learning style preferences (practical type) and 5 years of driving experience, the push keywords are extracted as "speeding, highway, emergency response, high risk". Candidate educational materials are matched and the push order is determined by the priority evaluation function: "Emergency Response Procedure for Highway Traffic Accidents" (simulation exercise program) → "Hazards and Penalties of Speeding on Highways" (video) → "Safe Driving Standards on Highways" (audio). S5. Content Push and Effect Feedback: The educational materials to be pushed are pushed in sequence through the vehicle terminal. The driver's completion rate of the simulation exercise is 95%, the assessment score is 92 points (excellent), the video viewing time is 10 minutes, and the assessment score is 88 points (pass). The feedback data is then transmitted to the effect feedback optimization module. S6. Feedback Analysis and Full-Process Optimization: The effect feedback optimization module analysis found that the completion rate and assessment scores of practical materials (simulation exercises) were higher than those of audio materials, and driver comments indicated that "simulation exercises make it easier to master emergency skills." Therefore, the optimization plan was output: the weight of material format adaptability in the priority evaluation function of the personalized push decision module was adjusted to give more weight to practical materials; at the same time, suggestions were pushed to the safety education resource library module to add simulation exercise materials related to highway emergency response, thus completing the full-process optimization.

[0024] In this invention, the training process of the driving behavior feature extraction model and the construction process of the driving risk assessment model in step S2 are as follows: The training process includes: S21. Construct a driving behavior sample dataset, including normal driving samples and abnormal driving samples (rapid acceleration, sudden braking, sharp turns, speeding, fatigued driving, distracted driving), and label each sample with the corresponding behavior type label; S22. Perform data augmentation on the sample dataset, including random cropping, noise addition, and data rotation, to increase the sample size; S23. Divide the enhanced sample dataset into a training set and a test set, input them into the improved CNN model for training, and adjust the model parameters (learning rate, number of iterations, convolution kernel size) until the model test accuracy reaches the preset threshold to obtain the trained feature extraction model. The construction process includes: S2a. Construct a hierarchical risk assessment indicator system. The specific risk assessment indicators are as follows: The primary indicator is driving risk; Secondary indicators include operational risk, condition risk, and environmental risk; The three-level indicators include the frequency of rapid acceleration, the frequency of sudden braking, the frequency of sharp turns, the percentage of time spent speeding, the duration of fatigued driving, the number of times of distracted driving, potential vehicle malfunctions, the complexity of road conditions, and the severity of weather conditions. S2b. Use the Analytic Hierarchy Process (AHP) to determine the weight coefficients of each indicator; S2c, a risk assessment model is constructed based on the XGBoost algorithm. It takes driving behavior feature vectors and the weights of each indicator as input, and outputs the driving risk level (low risk, medium risk, high risk) and the risk weakness items (the three tertiary indicators with the highest weights). The core calculation logic of the risk assessment model is as follows: ; in, The driving risk score is calculated (the risk level is determined based on the score range: low risk, medium risk, high risk). The total number of three-level indicators in the hierarchical risk assessment indicator system (n=9, including the frequency of rapid acceleration, the frequency of rapid braking, etc.). The weight coefficient of the i-th tertiary indicator (determined by the Analytic Hierarchy Process (AHP), satisfying ∑i=1nωi=1); The quantitative score of the i-th tertiary indicator (calculated by combining the driving behavior feature vector with actual driving data, such as the standardized score of the percentage of speeding time and the duration of fatigue driving).

[0025] In this invention, step S4, the construction and decision-making process of the priority push model includes: S41. Extract push keywords: Determine core keywords based on risk level (high risk corresponds to dangerous behavior correction and emergency response; medium risk corresponds to standardized operation and skills improvement; low risk corresponds to knowledge popularization and habit formation), determine special keywords based on risk shortcomings, and determine scenario keywords based on driving scenarios and weather. S42. Tag Matching: Calculate the index value of the push keyword sequence. and the index value of the identification label sequence string ,when When δ is a preset threshold, ranging from 0.1 to 0.3, the two are determined to be a match, and candidate educational materials are obtained through screening. S43. Priority Ranking: Constructing a Priority Evaluation Function ,in, The risk level is weighted (1.0 for high risk, 0.7 for medium risk, and 0.4 for low risk). The urgency weight of behavioral shortcomings is determined by a value ranging from 0.6 to 1.0. Learning style fit (value range 0.5-1.0), Weighting of historical push performance (value range 0.3-1.0). These are the weighting coefficients, and The candidate educational materials are sorted in descending order based on the evaluation function calculation results to obtain the push order.

[0026] In this invention, step S6, feedback analysis and overall process optimization, includes the following sub-steps: S61. Data Reception and Processing: Receive feedback data transmitted by the content push execution module, classify it into basic behavior data, interactive feedback data, and assessment result data, remove invalid data (such as duplicate clicks caused by misoperation, meaningless comments), and generate a standardized feedback dataset. S62. Multi-dimensional in-depth analysis: NLP technology is used to analyze the comments and identify drivers' satisfaction with the pushed materials (such as "the video is too long" or "the content is very practical") and potential needs (such as "the need for materials related to night driving"). Through correlation analysis, the correlation between the completion rate of a certain type of material and the corresponding improvement rate of driving behavior is determined (such as drivers with a completion rate of ≥90% for emergency braking materials have an average reduction of 40% in the subsequent occurrence of emergency braking behavior). S63. Optimization solution generation: For problems found in the analysis, such as the low accuracy of the risk assessment model in identifying fatigued driving, output the model feature extraction dimension adjustment parameters (add eyelid closure feature input), such as the completion rate of a certain type of scene material (rural road) is less than 60% after being pushed, output material optimization suggestions (add rural road scene simulation exercise material). S64. Optimization Plan Execution: The generated optimization parameters and suggestions are synchronized to the driving behavior analysis module, the safety education resource library module, and the personalized push decision module. Each module adjusts model parameters, updates materials, and optimizes strategies based on the received information to ensure that subsequent pushes are more in line with the driver's needs.

[0027] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A safety education content push system based on driving behavior analysis, comprising a driving data collection module, a driving behavior analysis module, a safety education resource library module, a personalized push decision module, a content push execution module, and an effect feedback optimization module; The driving data acquisition module is used to collect real-time driving operation data of the driver, vehicle operating status data and driving environment data, and to preprocess the collected data; The driving behavior analysis module constructs a driving behavior feature extraction model and a driving risk assessment model based on the preprocessed driving data, and outputs driving behavior feature vectors and driving risk level assessment results. The safety education resource library module is used to store multiple types and dimensions of safety education materials, and to tagged each educational material to generate a set of identification tags; The personalized push decision module is used to combine driving behavior feature vectors, driving risk level assessment results and driver personalized information to build a priority push model and match the educational materials to be pushed and the push order. The content push execution module is used to push the educational materials to be pushed to the driver's terminal according to the push order, and collect push effect feedback data; The effect feedback optimization module is used to receive feedback data and perform in-depth analysis to generate model optimization parameters, material optimization suggestions and push strategy adjustment schemes, which are then synchronized to the corresponding modules to achieve full-process optimization.

2. The safety education content delivery system based on driving behavior analysis according to claim 1, characterized in that, The driving data acquisition module includes an on-board sensor unit, an environmental perception unit, and a data preprocessing unit. The vehicle-mounted sensor unit includes a steering wheel angle sensor, an accelerator pedal sensor, a brake pedal sensor, a vehicle speed sensor, and a speed sensor, which are used to collect driving operation data such as the driver's steering operation, acceleration operation, and braking operation, as well as the vehicle's driving speed and engine speed operating status data. The environmental perception unit includes a GPS positioning module, a camera, a radar sensor, and a weather sensor, used to collect driving environment data such as driving scenarios, road conditions, distances to surrounding vehicles, and weather conditions. The data preprocessing unit is used to clean, standardize, and fuse the collected data to generate structured driving data samples.

3. The safety education content delivery system based on driving behavior analysis according to claim 1, characterized in that, The driving behavior analysis module includes a feature extraction unit and a risk assessment unit; The feature extraction unit uses an improved convolutional neural network model, taking structured driving data samples as input, to extract key features of driving behavior, including features of rapid acceleration, emergency braking, sharp turning, speeding, fatigued driving, and distracted driving, and to generate a driving behavior feature vector. The risk assessment unit constructs a hierarchical risk assessment indicator system, including operational risk indicators, state risk indicators, and environmental risk indicators. It adopts a risk assessment model based on gradient boosting tree, and combines driving behavior feature vectors and driving environment data to output driving risk level and risk shortcomings.

4. A safety education content delivery system based on driving behavior analysis according to claim 1, characterized in that, The safety education resource library module includes a material storage unit and a tag generation unit; The educational materials stored in the material storage unit include video, audio, text, PPT, and simulation exercise programs, covering basic driving standards, dangerous behavior correction, emergency response skills, scenario-based driving techniques, and interpretation of laws and regulations. The tag generation unit creates multi-dimensional identification tags for each educational material, including behavior type tags, risk level tags, scene type tags, weather adaptation tags, and material format tags, generates an identification tag set, and establishes a tag index.

5. A safety education content delivery system based on driving behavior analysis according to claim 1, characterized in that, The personalized push decision module includes a user information management unit, a keyword extraction unit, a tag matching unit, and a priority sorting unit; The user information management unit is used to store the driver's real-name authentication information, driving experience, vehicle type, historical assessment results, and personalized information on learning style preferences; The keyword extraction unit extracts and pushes keywords based on the driving risk level assessment results, risk shortcomings, and driving behavior feature vectors. The tag matching unit is used to calculate the index difference between the push keyword sequence string and the identification tag sequence string; The priority sorting unit prioritizes candidate educational materials based on risk level weight, urgency of behavioral shortcomings, learning style suitability, and historical push results, generating a list of educational materials to be pushed and the push order.

6. A safety education content delivery system based on driving behavior analysis according to claim 1, characterized in that, The effect feedback optimization module includes a feedback data receiving unit, a feedback data analysis unit, an optimization parameter generation unit, and an optimization scheme push unit; The feedback data receiving unit is used to receive feedback data transmitted by the content push execution module and to classify and organize the data. The feedback data analysis unit uses natural language processing technology and statistical analysis methods to explore the correlation between feedback data and push effects and driving behavior improvement. Based on the analysis results, the optimization parameter generation unit outputs adjustment parameters for the driving behavior feature extraction model, risk assessment model, and priority push model. The optimization scheme push unit outputs material optimization suggestions for the safety education resource library and outputs push strategy adjustment schemes for the personalized push decision module.

7. A method for pushing safety education content based on driving behavior analysis, applicable to the safety education content pushing system based on driving behavior analysis as described in any one of claims 1-6, characterized in that, Includes the following steps: S1. Driving data collection and preprocessing: The driving data collection module collects the driver's driving operation data, vehicle operating status data and driving environment data. After cleaning, standardization and data fusion, structured driving data samples are generated. S2. Driving Behavior Analysis and Risk Assessment: The driving behavior analysis module is based on structured driving data samples. It extracts key features of driving behavior through a feature extraction model and generates a driving behavior feature vector. It also calculates the driving risk level and risk weaknesses through a risk assessment model. S3. Educational Material Tagging and Resource Management: The safety education resource library module tags various types of stored safety education materials, generates identification tag sets, and establishes an index. S4. Personalized Push Decision: The personalized push decision module combines driving behavior feature vectors, risk level assessment results and driver personalized information to extract push keywords, match candidate educational materials and prioritize them to obtain a list of educational materials to be pushed and the push order. S5. Content Push and Effect Feedback: The content push execution module pushes the educational materials to be pushed to the driver's terminal in the order of push, and collects push effect feedback data, which is then transmitted to the effect feedback optimization module. S6. Feedback Analysis and Full-Process Optimization: The effect feedback optimization module performs in-depth analysis of feedback data, generates optimization parameters and solutions, and synchronizes them to the corresponding modules to complete the full-process optimization.

8. The method for pushing safety education content based on driving behavior analysis according to claim 7, characterized in that, The training process of the driving behavior feature extraction model and the construction process of the driving risk assessment model in step S2 are as follows: The training process includes: S21. Construct a driving behavior sample dataset, including normal driving samples and abnormal driving samples, and label each sample with the corresponding behavior type label; S22. Perform data augmentation on the sample dataset, including random cropping, noise addition, and data rotation; S23. Divide the enhanced sample dataset into a training set and a test set, input them into the improved CNN model for training, until the model's test accuracy reaches a preset threshold. The construction process includes: S2a. Construct a hierarchical risk assessment indicator system; S2b. Use the analytic hierarchy process (AHP) to determine the weight coefficients of each indicator. S2c: A risk assessment model is constructed based on the XGBoost algorithm, which outputs the driving risk level and risk shortcomings.

9. A method for pushing safety education content based on driving behavior analysis according to claim 7, characterized in that, In step S4, the construction and decision-making process of the priority push model includes: S41. Extract push keywords: Determine core keywords based on risk level, determine specific keywords based on risk weaknesses, and determine scenario keywords based on driving scenarios and weather. S42. Tag Matching: Calculate the index value of the push keyword sequence string and the index value of the tag sequence string; S43. Priority Sorting: Construct a priority evaluation function, sort the candidate educational materials in descending order based on the evaluation function calculation results, and obtain the push order.

10. A method for pushing safety education content based on driving behavior analysis according to claim 7, characterized in that, The feedback analysis and overall process optimization in step S6 include the following sub-steps: S61. Receive and organize push notification feedback data to generate a standardized feedback dataset; S62. Employ natural language processing technology and statistical analysis methods to conduct multi-dimensional in-depth analysis of the feedback data; S63. Based on the analysis results, generate model optimization parameters, material optimization suggestions, and push strategy adjustment plans; S64. Synchronize the optimized parameters and solutions to the corresponding modules to complete the full-process optimization.

Images