Driving behavior big data driven driver safety awareness assessment and optimization method and device

Abstract

The invention relates to the field of traffic safety, and relates to a driving behavior big data-driven driver safety awareness assessment and optimization method and device. Driving behavior data is collected through multiple sources such as a vehicle-mounted sensor, after preprocessing and feature extraction, an evaluation model is constructed by utilizing a machine learning algorithm, safety awareness levels are divided according to scores output by the model, personalized suggestions are provided, and regular updating and optimization can be carried out. The device is composed of a data acquisition assembly and the like, can be integrated on a vehicle-mounted system and other platforms, realizes interaction with a driver, and provides real-time evaluation results and improvement suggestions. The driver safety consciousness is evaluated and optimized comprehensively, in real time and in a personalized mode, the method has the advantages of comprehensiveness, real-time performance, personalization, integratability and the like, the traffic safety level can be improved, and the method is suitable for the field of traffic safety management.

Description

Technical Field

[0001] The present invention relates to the field of traffic safety, and specifically relates to technical means driven by driving behavior big data to realize the evaluation and optimization of drivers' safety awareness. Specifically, it relates to a method and device for evaluating and optimizing drivers' safety awareness driven by driving behavior big data. Background Art

[0002] In modern society, with the rapid growth of the number of motor vehicles and the increasing complexity of road traffic networks, traffic accidents have become a severe global problem. The casualties and property losses caused by traffic accidents every year are countless. A large number of accident investigations and analyses have shown that as the direct participants in road traffic, drivers' safety awareness and driving behaviors play a dominant role in the occurrence of traffic accidents. For example, fatigue driving can make drivers inattentive and slow to react, easily leading to rear-end collisions, crashes and other accidents; speeding will shorten the driver's reaction time and increase the braking distance of the vehicle. Once encountering unexpected situations, it is difficult to take effective avoidance measures in time; while behaviors such as frequent illegal lane changes and failure to use turn signals as required not only disrupt traffic order but also greatly increase the risk of scratches and collisions between vehicles. These bad driving behaviors reflect the weakness of drivers' safety awareness, which has become an important inducement for frequent traffic accidents.

[0003] Traditional methods for evaluating drivers' safe driving mainly include manual observation and simple driving behavior analysis. Manual observation usually relies on traffic police or professional observers to visually inspect drivers' behaviors at specific roads or times. This method not only consumes a large amount of manpower and time, but also has a limited observation range, making it difficult to achieve comprehensive and continuous monitoring of drivers' behaviors. At the same time, the results of manual observation are greatly affected by the subjective factors of observers. Different observers may have different judgments on the same driving behavior, resulting in the lack of objectivity and accuracy of evaluation results. Simple driving behavior analysis often based on a limited number of indicators, such as whether speeding, whether running a red light, etc., cannot deeply explore the safety awareness level behind drivers' behaviors and cannot comprehensively reflect the behavior patterns and potential risks of drivers during the entire driving process. For example, only relying on the indicator of whether speeding cannot judge drivers' driving stability at normal driving speeds, their ability to anticipate road conditions, and their awareness of observing other traffic rules and other safety awareness-related factors, making it difficult to accurately evaluate drivers' safety awareness and unable to provide targeted and effective improvement suggestions for drivers.

[0004] With the rapid development of information technology, the application of big data and intelligent technologies in the transportation field is becoming increasingly extensive and in-depth. Big data technology can collect, store, and process massive amounts of traffic data in real time, including vehicle driving data, driver behavior data, road environment data, etc., providing rich data resources for traffic management and safety assessment. In terms of driver safety awareness assessment, big data technology can comprehensively record various driving behavior details of drivers, such as speed change curves, acceleration distributions, steering operation frequencies, etc. By deeply analyzing these massive amounts of data accumulated over a long period, hidden behavior patterns and safety risk factors can be discovered, thereby achieving a more comprehensive and accurate assessment of driver safety awareness. At the same time, the development of intelligent technologies such as machine learning and deep learning provides powerful tools for data analysis and model construction. These technologies can automatically learn and extract features from big data, build complex assessment models, simulate the complex relationship between driver behavior and safety awareness, and break through the limitations of traditional assessment methods. For example, using deep learning algorithms can more accurately identify the fatigue driving behavior of drivers. By analyzing multi-modal data such as the driver's eye state, head movement trajectory, and driving operation characteristics, it can comprehensively judge whether the driver is in a fatigue state, providing a basis for timely intervention. In addition, intelligent technologies can also realize the real-time feedback of assessment results and the generation of personalized suggestions, interacting with drivers through various channels such as in-vehicle systems, mobile applications, etc., helping drivers timely understand their safety awareness levels, improve driving behavior targeted, and enhance traffic safety levels.

[0005] In summary, considering the importance of driver factors in traffic accidents and the deficiencies of traditional assessment methods, it is of great practical significance and urgency to develop a more effective driver safety awareness assessment and optimization method using big data and intelligent technologies, which is also the key problem that this invention is committed to solving. Summary of the Invention

[0006] In view of the deficiencies of the currently related existing technologies, the present invention provides a method and device for assessing and optimizing driver safety awareness driven by driving behavior big data. By integrating multi-source driving behavior data, applying advanced data processing technologies and machine learning algorithms, a precise safety awareness assessment model is constructed to achieve a comprehensive, real-time, and personalized assessment of driver safety awareness. And based on the assessment results, targeted optimization suggestions are provided for drivers, and feedback is carried out through various channels such as in-vehicle systems, mobile applications, and online service platforms, helping drivers timely improve their driving behavior, enhance their safety awareness levels, thereby effectively reducing the incidence of traffic accidents, providing innovative technical means and decision-making support for traffic safety management, and promoting the intelligent development of the traffic safety field.

[0007] To achieve the above object, a first aspect of the present invention provides a method for evaluating and optimizing driver safety awareness driven by driving behavior big data, including the following steps:

[0008] Data collection step: Collect behavioral data of the driver during the driving process through multiple data sources;

[0009] Data preprocessing step: Process the collected data, including removing abnormal data, filling in missing data, and normalization processing;

[0010] Feature extraction step: Extract key indicators characterizing the driver's safety awareness and driving behavior features based on the preprocessed data, and perform correlation analysis on the extracted features to remove redundant features;

[0011] Model construction and training step: Select a machine learning algorithm to construct a driver safety awareness ability evaluation model, divide the data set after feature extraction into a training set, a validation set, and a test set, use the training set to train the model, monitor the model performance using the validation set during the training process, and determine the final model according to the performance indicators of the test set;

[0012] Safety awareness evaluation step: Input the real-time collected and preprocessed driving behavior data into the trained evaluation model, output the driver safety awareness evaluation score and divide the safety awareness level;

[0013] Personalized optimization suggestion generation step: Develop personalized optimization suggestions for the driver according to the evaluation results and driving behavior features;

[0014] Model update and optimization step: Regularly collect new data, re-perform data processing and model training, and adjust the evaluation criteria and optimization suggestions in combination with the actual situation.

[0015] Further, in the data collection step, the data sources include in-vehicle sensors, in-vehicle intelligent terminals, and mobile intelligent devices. The collected data at least covers vehicle driving speed, acceleration change, steering angle, braking force, driving route, driving time, vehicle status information, and driver-vehicle interaction information, and further ensures the continuity and real-time transmission of data collection, and stores the collected data in real time.

[0016] Further, in the data preprocessing step, data cleaning specifically includes removing abnormal data points and error data records caused by sensor failures and signal interference, and correcting logical errors in the data; data filling uses an interpolation algorithm, and selects a suitable interpolation method according to the distribution characteristics of the data; data normalization uses a standardization method to make data with different dimensions within the same order of magnitude range.

[0017] Further, in the feature extraction step, the key indicators include driving operation features, driving state features, and fatigue driving-related features. Among them, the driving operation features include the number of hard accelerations, the number of hard decelerations, the proportion of the number of sudden lane changes, the speed variation coefficient, the number of passive brakes, the number of times the turn signal is not turned on, the maximum value of the steering wheel rotation, the number of speeding violations, and fatigue driving behavior recognition; the driving state features at least include the degree of deviation of the average driving speed from the road speed limit, the speed fluctuation coefficient, the proportion of the duration of speeding, the proportion of the duration of driving at a low speed for a long time, and the frequency of abnormal driving behavior in specific dangerous sections; the fatigue driving-related features at least include the driving duration distribution, the proportion of eyelid closure time, the frequency of abnormal head movement, and the regular change of driving operation; the number of hard accelerations is obtained by calculating the number of times the acceleration exceeds the set threshold; the number of hard decelerations is obtained by calculating the number of times the acceleration is lower than the set threshold; the proportion of the number of sudden lane changes is used to identify and quantify sudden lane change events during vehicle driving by whether the lateral acceleration exceeds the preset sudden lane change acceleration threshold; the speed variation coefficient is the ratio of the standard deviation of the vehicle speed to the average value; the number of passive brakes is the number of times the vehicle system actively triggers the brakes when it determines that the driver is driving dangerously; the number of times the turn signal is not turned on counts the number of times the turn signal is not turned on for a specified subject; the maximum value of the steering wheel rotation calculates the maximum steering angle of the driver's steering wheel; the number of speeding violations counts the number of times the vehicle exceeds the speed limit range; the fatigue driving behavior recognition is based on behavior pattern recognition and combined with external sensor information for judgment.

[0018] Further, in the model construction and training step, the machine learning algorithms include support vector machine (SVM), random forest (RF), and neural network (NN);

[0019] When using support vector machine (SVM), the safety awareness ability of the driver is classified by finding the optimal hyperplane, and its formula is: f(x) = w T x + b, where w is the normal vector, x is the feature vector, and b is the bias term. SVM selects the optimal hyperplane by maximizing the margin to achieve the classification of different safety awareness levels;

[0020] When using random forest (RF), multiple decision trees are constructed and classified in combination with a voting mechanism. The prediction results of each tree are voted, and finally the safety awareness level of the driver is determined. The mathematical representation of the RF model is: where N is the number of decision trees, is the prediction result of each tree;

[0021] When using neural network (NN), the method of deep learning is used to extract deep features in driving behavior through a multi-layer perceptron (MLP) and output a safety awareness score. The mathematical expression of the NN model is y = f(W (L) *f(W (L-1)....f(W (1) *x + b (1) ) + b (L-1) ) + b (L) ), where W is the weight matrix, b is the bias, and f is the activation function.

[0022] Furthermore, in the model update and optimization step, new driving behavior data is regularly collected and added to the original dataset, and the data preprocessing, feature extraction, model construction, and training steps are re-executed to continuously optimize the performance of the evaluation model. At the same time, the safety awareness evaluation criteria and optimization recommendation content are dynamically adjusted according to the actual situation, where the actual situation includes the latest traffic accident cases and changes in traffic regulations.

[0023] Furthermore, it includes comprehensively analyzing the evaluation results of three algorithms: Support Vector Machine (SVM), Random Forest (RF), and Neural Network (NN), and using a weighted average and voting mechanism to determine the final evaluation result of safety awareness. The weighted average calculation formula is:

[0024] where w SVM , w RF , w NN are the weights of each model, determined by the accuracy and other performance metrics of the model.

[0025] The second aspect of the present invention provides a device for evaluating and optimizing driver safety awareness driven by driving behavior big data, characterized in that: this device is used to integrate any of the methods in claims 1 to 7 above into the device, and includes the following components:

[0026] A data acquisition component, used to implement the data acquisition step, and collect the behavior data of the driver during the driving process through in-vehicle sensors, in-vehicle intelligent terminals, and mobile intelligent devices;

[0027] A data preprocessing component, used to execute the data preprocessing step, and perform abnormal data removal, missing data filling, and normalization processing on the collected data;

[0028] A feature extraction component, used to complete the feature extraction step, extract key indicators representing the driver's safety awareness and driving behavior characteristics based on the preprocessed data, and perform correlation analysis on the extracted features to remove redundant features;

[0029] A model construction and training component, used to implement the model construction and training step, select a machine learning algorithm to construct a driver safety awareness ability evaluation model, divide the dataset after feature extraction into a training set, a validation set, and a test set, use the training set to train the model, monitor the model performance using the validation set during the training process, and determine the final model according to the performance metrics of the test set;

[0030] A safety awareness assessment component, which is used to carry out the safety awareness assessment step, input the driving behavior data collected and preprocessed in real time into the trained assessment model, output the driver safety awareness assessment score and divide the safety awareness level;

[0031] A personalized optimization suggestion generation component, which is used to implement the personalized optimization suggestion generation step, and formulate personalized optimization suggestions for the driver according to the assessment results and driving behavior characteristics;

[0032] A model update and optimization component, which is used to execute the model update and optimization step, regularly collect new data, reprocess the data and train the model, and adjust the assessment criteria and optimization suggestions according to the actual situation;

[0033] An interaction component, which is used to interact with the driver, enabling the driver to obtain the safety awareness analysis results and improvement suggestions in real time through the in-vehicle system, mobile application or online service platform, where:

[0034] An in-vehicle system interaction sub-component. When the device cooperates with the in-vehicle system, the in-vehicle system is directly connected to the vehicle's sensors and communication system, receives the data from the data collection component and displays the analysis results;

[0035] A mobile application interaction sub-component. When the device cooperates with the mobile application, the mobile application uploads the driving data collected by the data collection component to the cloud for analysis, and enables the driver to view the historical driving behavior reports and improvement suggestions at any time;

[0036] An online service platform interaction sub-component. When the device cooperates with the online service platform, the online service platform receives the data collected by the data collection component for global data analysis, behavior summary and generation of long-term improvement suggestions to help the driver improve safety awareness.

[0037] Furthermore, the in-vehicle system interaction sub-component includes a data connection interface, a display unit and an audio prompt unit. The data connection interface is used to connect to the vehicle's sensors and communication system. The display unit is used to display the safety awareness assessment results and driving behavior data. The audio prompt unit is used to output improvement suggestions in voice.

[0038] Furthermore, the mobile application interaction sub-component includes a data upload unit, a report viewing interface, and a push notification unit. The data upload unit is used to upload driving data to the cloud. The report viewing interface is used for the driver to view historical driving behavior reports. The push notification unit is used to push improvement suggestions to the driver. The online service platform interaction sub-component includes a big data storage and analysis unit, a behavior summary unit, and a long-term suggestion generation unit. The big data storage and analysis unit is used to store and analyze driving behavior data. The behavior summary unit is used to summarize the driver's behavior. The long-term suggestion generation unit is used to generate long-term improvement suggestions.

[0039] The present invention adopts the above technical solutions and has at least the following beneficial effects:

[0040] 1. Comprehensively and accurately evaluate the driver's safety awareness: Through multi-source data collection and multi-dimensional feature extraction, covering various aspects such as driving operations, driving states, and fatigue driving, it can comprehensively and deeply evaluate the driver's safety awareness. For example, not only focusing on basic driving parameters such as speed and acceleration, but also combining fatigue driving-related features such as the proportion of eyelid closure time and the abnormal frequency of head movement to more accurately judge the driver's safety awareness level, overcoming the defects of single and incomplete indicators in traditional evaluation methods.

[0041] 2. Real-time feedback helps immediately improve driving behavior: By using on-vehicle systems, etc., to obtain and analyze driving behavior data in real time, it can immediately output safety awareness evaluation results and optimization suggestions. When the driver exhibits dangerous driving behaviors such as sudden acceleration and sudden steering, the on-vehicle system can immediately give voice prompts for correction to help the driver adjust driving behavior in a timely manner, effectively reducing the accident risk. This real-time advantage cannot be achieved by traditional evaluation methods.

[0042] 3. Personalized suggestions promote targeted improvement of safety awareness: According to the behavior patterns and evaluation results of different drivers, customized optimization suggestions are provided. Recommend training courses for drivers with low safety awareness, remind drivers at high risk of fatigue driving to rest and provide relief methods, etc., enabling drivers to targetedly improve their own problems and enhance safety awareness and driving skills, which is difficult to achieve with traditional methods.

[0043] 4. Integratability adapts to various application scenarios and enhances scalability: The present invention can be integrated into on-vehicle systems, mobile applications, or online service platforms to meet the needs of different scenarios. The on-vehicle system is convenient for real-time monitoring and feedback. The mobile application is convenient for drivers to view historical data and suggestions at any time. The online service platform can conduct global data analysis and long-term planning. This integratability makes the technology easy to promote and expand, and can better integrate with existing traffic management systems and intelligent vehicle technologies to provide broader support for traffic safety management.

[0044] 5. Comprehensive model evaluation improves evaluation accuracy and reliability: Multiple machine learning algorithms (such as support vector machines, random forests, neural networks) are used to construct an evaluation model, and the results are integrated through weighted averaging and voting mechanisms, making full use of the advantages of each algorithm and reducing the limitations of a single algorithm. Different algorithms have different sensitivities to different types of data and behavioral patterns. Comprehensive evaluation can more accurately reflect the driver's safety awareness and provide a more reliable basis for subsequent optimization suggestions, with higher accuracy compared to single-model evaluation methods.

[0045] 6. Data-driven continuous optimization ensures long-term effectiveness: Regularly update data and models, and adjust evaluation criteria and suggestions in combination with the latest traffic accident cases and regulatory changes, so that the evaluation method can always adapt to changes in the traffic environment and the evolution of driver behavior. For example, when new traffic regulations change the speeding standard or the determination of fatigue driving, the system can be updated in a timely manner to ensure the long-term effectiveness of evaluation and optimization and maintain a positive role in preventing traffic accidents. Brief Description of the Drawings

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0047] Figure 1 It is a flowchart of the driver safety awareness evaluation and optimization method of the present invention. Detailed Embodiments

[0048] Here, the exemplary embodiments will be described in detail, and the examples are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. On the contrary, they are only examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0049] Embodiment 1

[0050] As Figure 1 shown, in the first aspect of this embodiment, a driver safety awareness evaluation and optimization method driven by driving behavior big data is provided, including the following steps:

[0051] Data collection step: Collect the behavior data of the driver during the driving process through multiple data sources;

[0052] Data preprocessing step: Process the collected data, including removing abnormal data, filling in missing data, and normalizing the data;

[0053] Feature extraction step: Based on the preprocessed data, extract key indicators characterizing the driver's safety awareness and driving behavior features, and conduct a correlation analysis on the extracted features to remove redundant features;

[0054] Model construction and training step: Select a machine learning algorithm to construct a driver safety awareness ability evaluation model. Divide the dataset after feature extraction into a training set, a validation set, and a test set. Use the training set to train the model, monitor the model performance using the validation set during the training process, and determine the final model according to the performance indicators of the test set;

[0055] Safety awareness evaluation step: Input the preprocessed driving behavior data collected in real time into the trained evaluation model, output the driver safety awareness evaluation score, and classify the safety awareness level;

[0056] Personalized optimization suggestion generation step: Develop personalized optimization suggestions for the driver according to the evaluation results and driving behavior features;

[0057] Model update and optimization step: Regularly collect new data, re - perform data processing and model training, and adjust the evaluation criteria and optimization suggestions in combination with the actual situation.

[0058] As a preferred implementation manner, in the data collection step of this embodiment, the data sources include in - vehicle sensors, in - vehicle intelligent terminals, and mobile intelligent devices. The collected data at least covers vehicle driving speed, acceleration change, steering angle, braking force, driving route, driving time, vehicle status information, and driver - vehicle interaction information, and further ensures the continuity and real - time transmission of data collection, and stores the collected data in real time.

[0059] As a preferred implementation manner, in the data preprocessing step of this embodiment, data cleaning specifically includes removing abnormal data points and error data records caused by sensor failures and signal interferences, and correcting logical errors in the data; data filling uses an interpolation algorithm, and selects an appropriate interpolation method according to the distribution characteristics of the data; data normalization uses a standardization method to make data with different dimensions within the same order of magnitude range.

[0060] As a preferred embodiment, in the feature extraction step of this embodiment, the key indicators include driving operation features, driving state features, and fatigue driving-related features. Among them, the driving operation features include the number of hard accelerations, the number of hard decelerations, the proportion of the number of sudden lane changes, the speed variation coefficient, the number of passive brakes, the number of times the turn signal is not turned on, the maximum value of the steering wheel rotation, the number of speeding violations, and fatigue driving behavior recognition; the driving state features at least include the degree of deviation of the average driving speed from the road speed limit, the speed fluctuation coefficient, the proportion of the duration of speeding, the proportion of the duration of driving at a low speed for a long time, and the frequency of abnormal driving behavior in specific dangerous sections; the fatigue driving-related features at least include the driving duration distribution, the proportion of eyelid closure time, the frequency of abnormal head movement, and the regular change of driving operation; the number of hard accelerations is obtained by calculating the number of times the acceleration exceeds the set threshold; the number of hard decelerations is obtained by calculating the number of times the acceleration is lower than the set threshold; the proportion of the number of sudden lane changes is used to identify and quantify sudden lane change events during vehicle driving by whether the lateral acceleration exceeds the preset sudden lane change acceleration threshold; the speed variation coefficient is the ratio of the standard deviation of the vehicle speed to the average value; the number of passive brakes is the number of times the vehicle system actively triggers the brakes when it determines that the driver is driving dangerously; the number of times the turn signal is not turned on counts the number of times the turn signal is not turned on in a specified subject; the maximum value of the steering wheel rotation calculates the maximum steering angle of the driver's steering wheel; the number of speeding violations counts the number of times the vehicle exceeds the speed limit range; the fatigue driving behavior recognition is based on behavior pattern recognition and combined with external sensor information for judgment.

[0061] As a preferred embodiment, in the model construction and training step of this embodiment, the machine learning algorithms include support vector machine (SVM), random forest (RF), and neural network (NN);

[0062] When using support vector machine (SVM), the safety awareness ability of the driver is classified by finding the optimal hyperplane, and its formula is: f(x) = w T x + b, where w is the normal vector, x is the feature vector, and b is the bias term. SVM selects the optimal hyperplane by maximizing the margin to achieve the classification of different safety awareness levels;

[0063] When using random forest (RF), multiple decision trees are constructed and classified by combining a voting mechanism. The prediction results of each tree are voted, and finally the safety awareness level of the driver is determined. The mathematical representation of the RF model is: where N is the number of decision trees, is the prediction result of each tree;

[0064] When using neural network (NN), a deep learning method is used to extract deep features in driving behavior through a multi-layer perceptron (MLP) and output a safety awareness score. The mathematical expression of the NN model is y = f(W(L) *f(W (L-1) ....f(W (1) *x + b (1) ) + b (L-1) ) + b (L) )), where W is the weight matrix, b is the bias, and f is the activation function.

[0065] As a preferred implementation, in the model update and optimization step of this embodiment, new driving behavior data is regularly collected and added to the original dataset, and the data preprocessing, feature extraction, model construction, and training steps are re-executed to continuously optimize the performance of the evaluation model. At the same time, the safety awareness evaluation criteria and the content of optimization suggestions are dynamically adjusted according to the actual situation, where the actual situation includes the latest traffic accident cases and changes in traffic regulations.

[0066] As a preferred implementation, this embodiment includes comprehensively analyzing the evaluation results of three algorithms: Support Vector Machine (SVM), Random Forest (RF), and Neural Network (NN), and using weighted average and voting mechanisms to determine the final evaluation result of safety awareness. The weighted average calculation formula is:

[0067] where w SVM 、w RF 、w NN are the weights of each model, which are determined by the accuracy and other performance indicators of the model.

[0068] Embodiment 2

[0069] The second aspect of the present invention provides a driver safety awareness evaluation and optimization device driven by driving behavior big data. This device is used to integrate any of the above methods into the device and includes the following components:

[0070] A data collection component for implementing the data collection step, which collects the behavior data of the driver during the driving process through in-vehicle sensors, in-vehicle intelligent terminals, and mobile intelligent devices;

[0071] A data preprocessing component for executing the data preprocessing step, which performs abnormal data removal, missing data filling, and normalization processing on the collected data;

[0072] A feature extraction component for completing the feature extraction step, which extracts key indicators representing the driver's safety awareness and driving behavior characteristics based on the preprocessed data, and performs correlation analysis on the extracted features to remove redundant features;

[0073] The model construction and training component is used to implement the model construction and training steps. It selects a machine learning algorithm to construct a driver safety awareness ability evaluation model, divides the dataset after feature extraction into a training set, a validation set, and a test set, trains the model using the training set, monitors the model performance using the validation set during the training process, and determines the final model based on the test set performance metrics;

[0074] The safety awareness evaluation component is used to carry out the safety awareness evaluation step. It inputs the preprocessed driving behavior data collected in real time into the trained evaluation model, outputs the driver safety awareness evaluation score, and divides the safety awareness level;

[0075] The personalized optimization suggestion generation component is used to implement the personalized optimization suggestion generation step. It formulates personalized optimization suggestions for the driver based on the evaluation results and driving behavior characteristics;

[0076] The model update and optimization component is used to execute the model update and optimization steps. It regularly collects new data, re-performs data processing and model training, and adjusts the evaluation criteria and optimization suggestions according to the actual situation;

[0077] The interaction component is used to interact with the driver, enabling the driver to obtain the safety awareness analysis results and improvement suggestions in real time through the in-vehicle system, mobile application, or online service platform. Among them:

[0078] The in-vehicle system interaction sub-component, when the device cooperates with the in-vehicle system, the in-vehicle system is directly connected to the vehicle's sensors and communication system, receives the data from the data acquisition component, and displays the analysis results;

[0079] The mobile application interaction sub-component, when the device cooperates with the mobile application, the mobile application uploads the driving data collected by the data acquisition component to the cloud for analysis, and enables the driver to view the historical driving behavior reports and improvement suggestions at any time;

[0080] The online service platform interaction sub-component, when the device cooperates with the online service platform, the online service platform receives the data collected by the data acquisition component for global data analysis, behavior summary, and generation of long-term improvement suggestions to help the driver improve safety awareness.

[0081] As a preferred implementation manner, in this embodiment, the in-vehicle system interaction sub-component includes a data connection interface, a display unit, and an audio prompt unit. The data connection interface is used to connect to the vehicle's sensors and communication system, the display unit is used to display the safety awareness evaluation results and driving behavior data, and the audio prompt unit is used to output improvement suggestions in voice.

[0082] As a preferred embodiment, in this embodiment, the mobile application interaction sub-component includes a data upload unit, a report viewing interface, and a push notification unit. The data upload unit is used to upload driving data to the cloud. The report viewing interface is used for the driver to view historical driving behavior reports. The push notification unit is used to push improvement suggestions to the driver. The online service platform interaction sub-component includes a big data storage and analysis unit, a behavior summary unit, and a long-term suggestion generation unit. The big data storage and analysis unit is used to store and analyze driving behavior data. The behavior summary unit is used to summarize the driver's behavior. The long-term suggestion generation unit is used to generate long-term improvement suggestions.

[0083] Through the analysis method based on big data, the present invention can evaluate the safety awareness ability of drivers in real time and comprehensively, and provide personalized improvement suggestions, thereby effectively improving driving skills and safety awareness levels. Compared with the prior art, the present invention has the following advantages:

[0084] Comprehensiveness: Through the collection of multiple data sources and the extraction of multi-dimensional features, the safety awareness ability of drivers can be comprehensively evaluated.

[0085] Real-time: Provide real-time data analysis and feedback to help drivers immediately improve their driving behavior.

[0086] Personalization: Provide customized safe driving suggestions according to the behavior characteristics of different drivers.

[0087] Integratability: This method can be integrated into existing vehicle-mounted systems, mobile applications, and cloud service platforms, and has good scalability.

[0088] By integrating multiple algorithm models and adopting a weighted sum and voting mechanism, the present invention can more accurately evaluate the safety awareness ability of drivers, provide personalized safe driving suggestions, effectively reduce the incidence of traffic accidents, and improve the safety of drivers.

[0089] Embodiment III

[0090] A specific implementation scheme is provided in this embodiment as follows:

[0091] Scene setting and data preparation:

[0092] Select an area containing different road conditions and driving scenarios, such as the central business district of the city, highways, the outskirts of the city, etc., and organize 100 drivers to participate in the test. Install high-precision vehicle-mounted sensors on each test vehicle (such as the acceleration sensor with an accuracy of ±0.1m / s 2, the gyroscope accuracy is ±0.5° / s, the GPS positioning system positioning accuracy is within 5 meters, etc.), in-vehicle intelligent terminals (with powerful data processing and transmission capabilities), and a specially developed mobile application installed on the driver's smartphone (compatible with the vehicle system and stable in data transmission).

[0093] Detailed records and annotations are made on the road speed limits, road conditions information (such as the locations and characteristics of curves, steep slopes, construction sections, etc.) in the test area, as auxiliary data for subsequent analysis. At the same time, data on recent traffic accident cases in this area are collected for reference in model updating and evaluation standard adjustment.

[0094] Data collection and preliminary processing:

[0095] The driver drives in the test area for one week according to normal driving habits, during which the data collection system works continuously. The collected data includes vehicle driving speed (accurate to 0.1 km / h), acceleration change (recording the acceleration value every 0.1 second), steering angle (accurate to 0.1°), braking force (represented by the measured value of the brake pressure sensor), driving route (recording the trajectory through the GPS positioning system), driving time (accurate to seconds), vehicle status information (such as the status of light usage, door opening and closing status, etc.), and driver-vehicle interaction information (such as the frequency of manual operation buttons, the usage of voice commands, etc.).

[0096] After data collection is completed, data preprocessing is carried out first. In the data cleaning process, abnormal data points (about 3% of the total data volume) and incorrect data records (such as data with a negative speed value that does not conform to physical laws) generated due to short-term sensor failures (such as the acceleration sensor outputting an abnormally high value at a certain moment) or signal interference (such as the GPS signal being lost briefly in an area with strong electromagnetic interference) are identified and removed through algorithms. At the same time, logical errors in the data (such as unreasonable jumps in the driving route) are corrected. For a small amount of missing data (about 1% of the total data volume) caused by reasons such as equipment restart, the cubic spline interpolation algorithm is used for filling. Finally, the data is normalized using the standardization method, and data with different dimensions such as speed, acceleration, and steering angle are converted to the interval [-1, 1] for subsequent calculations and model input.

[0097] Feature extraction and analysis:

[0098] Based on the preprocessed data, features related to driving behavior are extracted. Calculate the number of hard accelerations (the number of times the acceleration exceeds 2 m / s 2 ), the number of hard decelerations (the number of times the acceleration is lower than -2 m / s 2 ), and the proportion of the number of sudden lane changes (the lateral acceleration exceeds 0.5 m / s 2It is identified as a sudden lane change event, and the proportion of it in the total number of lane changes is calculated), the speed coefficient of variation (the ratio of the standard deviation of vehicle speed to the average value), the number of passive brakes (the number of times the vehicle system actively triggers the brakes when it determines that the driver is driving dangerously), the number of times the turn signal is not turned on (the number of times the turn signal is not turned on during operations such as lane changes and turns is counted), the maximum value of the steering wheel rotation (calculate the maximum steering angle of the driver's steering wheel, accurate to 1°), the number of speeding times (the number of times the vehicle exceeds the speed limit range, and the speed limit value is determined according to road types and regional regulations) and the identification of fatigued driving behavior (monitor the proportion of the driver's eyelid closing time through the in-vehicle camera. When the closing time exceeds 1 second and occurs more than 3 times per minute, it is identified as a tendency of fatigued driving behavior, and a comprehensive judgment is made by combining the abnormal frequency of head movements, such as frequent nodding and shaking of the head).

[0099] Perform a correlation analysis on the extracted features and find that there is a certain positive correlation between the number of sudden accelerations and the number of speeding times (the correlation coefficient is about 0.6), and there is also a strong correlation between the maximum value of the steering wheel rotation and the proportion of sudden lane change times (the correlation coefficient is about 0.7). According to the analysis results, redundant features are removed, and relatively independent features such as the number of sudden accelerations, the speed coefficient of variation, the number of times the turn signal is not turned on, and the identification of fatigued driving behavior, which have a greater impact on the assessment of safety awareness, are retained. The feature data is sorted into a data set, where 70% is used as the training set, 20% is used as the validation set, and 10% is used as the test set.

[0100] Model construction, training and evaluation:

[0101] Three machine learning algorithms, support vector machine (SVM), random forest (RF) and neural network (NN), are selected to construct an evaluation model for the driver's safety awareness ability.

[0102] For the SVM model, set the kernel function to the radial basis function (RBF), find the optimal hyperplane through the grid search method, and adjust the hyperparameters C (penalty parameter, value range: [0.1, 100]) and γ (kernel function parameter, value range: [0.01, 10]) to select the optimal hyperplane by maximizing the margin to achieve the classification of different safety awareness levels. During its training process, the loss function uses the hinge loss function, and the optimization algorithm is the sequential minimal optimization (SMO) algorithm.

[0103] When constructing the random forest model, set the number of decision trees to 100, and the maximum depth of each tree to 10. When constructing each decision tree, randomly select the number of features as the square root of the total number of features. Through multiple sampling with replacement of the training set data, multiple decision trees are constructed, and each tree predicts the driver's safety awareness level. Finally, the voting mechanism is combined to determine the final prediction result.

[0104] In the neural network model, a three-layer structure is adopted. The number of nodes in the input layer is determined to be 4 according to the number of features, the number of nodes in the hidden layer is 8, and the number of nodes in the output layer is 1 (representing the safety awareness score). The ReLU function is selected as the activation function, the loss function is the mean square error (MSE) function, the optimization algorithm uses the stochastic gradient descent (SGD) algorithm, and the learning rate is initially set to 0.01 and gradually decays as the number of training rounds increases. During the training process, the weight matrix and bias are continuously adjusted through the backpropagation algorithm to gradually reduce the loss of the model on the training set.

[0105] During the training process, the validation set is used to monitor the model performance. For the SVM model, observe its accuracy (target value greater than 90%) and F1 value (target value greater than 0.85) on the validation set; for the random forest model, pay attention to the accuracy (target value greater than 92%) and recall rate (target value greater than 0.88); for the neural network model, focus on the mean square error (target value less than 0.05). When the performance metrics of the model on the validation set do not improve for 5 consecutive training rounds, stop the training. Finally, determine the final model according to the performance metrics on the test set. After testing, the accuracy of the SVM model on the test set is 91.5%, the accuracy of the random forest model is 93.2%, and the mean square error of the neural network model is 0.042.

[0106] Safety awareness assessment and personalized recommendations:

[0107] Input the preprocessed driving behavior data collected in real time into the trained evaluation model to output the driver safety awareness evaluation score (ranging from 0 to 100 points). For example, the evaluation score of driver A is 65 points, and the evaluation score of driver B is 45 points, etc. Divide the driver's safety awareness level into different levels according to the set thresholds (0 - 30 points is the low safety awareness level, 31 - 70 points is the medium safety awareness level, 71 - 100 points is the high safety awareness level).

[0108] Provide personalized optimization suggestions for drivers based on the evaluation results and driving behavior characteristics. For drivers with a relatively low level of safety awareness (such as Driver B), through in-vehicle system voice prompts and display on the screen, recommend them to participate in safety driving intensive training courses organized by offline driving schools, push safety driving knowledge video tutorials in the in-vehicle system, and at the same time remind them to pay attention to observing traffic rules and avoid dangerous driving behaviors such as sudden acceleration, sudden deceleration, and frequent lane changes. For drivers at risk of fatigue driving (judged by fatigue driving behavior recognition indicators), such as Driver A showing certain signs of fatigue during a long-distance drive, the in-vehicle system promptly gives a voice reminder for them to rest in the service area ahead for 20 minutes, play refreshing music, and at the same time push simple exercise methods for relieving fatigue (such as neck stretching, eye massage, etc.) on the mobile application. For drivers with poor driving operation habits (such as a relatively high number of times without using turn signals), after each operation without using turn signals, the in-vehicle system immediately gives a voice prompt for the correct operation method, provides a detailed specification for the use of turn signals on the mobile application, and reminds them to maintain good driving habits through push notifications.

[0109] Model Update and Optimization:

[0110] After a period of use (such as one month), collect new driving behavior data (about 5,000 pieces) and add it to the original dataset. Re-execute the data preprocessing, feature extraction, model construction, and training steps. During the model construction process, according to the characteristics of the latest collected data distribution and model performance requirements, fine-tune the hyperparameters of the SVM model (such as adjusting the C value to 5 and the γ value to 0.1), increase the number of decision trees in the random forest model to 120, adjust the number of hidden layer nodes of the neural network model to 10, and optimize the learning rate decay strategy. At the same time, in combination with traffic accident cases that occurred during the period (such as an accident caused by speeding on a certain section) and changes in traffic regulations (such as the speed limit adjustment for a specific section), dynamically adjust the safety awareness evaluation criteria (such as appropriately increasing the weight of speeding-related features) and the content of optimization suggestions (such as strengthening speed reminders and suggestions for drivers on a specific speed limit adjustment section). After the update and optimization, the performance of the model on the new test set is improved, such as the accuracy rate is increased by 3 - 5 percentage points, and the mean square error is reduced by 0.01 - 0.02.

[0111] Through this embodiment, the effectiveness and feasibility of the present invention in actual application scenarios are further verified, demonstrating the complete process from data collection to final model optimization, as well as how to provide practical and effective safety awareness evaluation and personalized optimization suggestions for drivers, which has a positive significance for traffic safety management.

[0112] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for evaluating and optimizing a driver's safety awareness driven by driving behavior big data, characterized in that: Including the following steps: Data collection step: Collect behavioral data of the driver during the driving process through multiple data sources; Data preprocessing step: Process the collected data, including removing abnormal data, filling in missing data, and normalization; Feature extraction step: Extract key indicators characterizing the driver's safety awareness and driving behavior characteristics based on the preprocessed data, and perform correlation analysis on the extracted features to remove redundant features; Model construction and training step: Select a machine learning algorithm to construct an evaluation model for the driver's safety awareness ability, divide the data set after feature extraction into a training set, a validation set, and a test set, use the training set to train the model, monitor the model performance using the validation set during the training process, and determine the final model according to the performance indicators of the test set; Safety awareness evaluation step: Input the real-time collected and preprocessed driving behavior data into the trained evaluation model, output the driver's safety awareness evaluation score, and classify the safety awareness level; Personalized optimization suggestion generation step: Develop personalized optimization suggestions for the driver based on the evaluation results and driving behavior characteristics; Model update and optimization step: Regularly collect new data, re-perform data processing and model training, and adjust the evaluation criteria and optimization suggestions in combination with the actual situation.

2. The method according to claim 1, characterized in that: In the data collection step, the data sources include in-vehicle sensors, in-vehicle intelligent terminals, and mobile intelligent devices. The collected data at least covers vehicle driving speed, acceleration change, steering angle, braking force, driving route, driving time, vehicle status information, and driver-vehicle interaction information, and further ensures the continuity and real-time transmission of data collection, and stores the collected data in real time.

3. The method according to claim 1, characterized in that: In the data preprocessing step, data cleaning specifically includes removing abnormal data points and error data records caused by sensor failures and signal interference, and correcting logical errors in the data; data filling uses an interpolation algorithm, and selects a suitable interpolation method according to the distribution characteristics of the data; data normalization uses a standardization method to make data with different dimensions within the same order of magnitude range.

4. The method according to claim 1, wherein: In the feature extraction step, the key indicators include driving operation characteristics, driving state characteristics, and fatigue driving-related characteristics. Among them, driving operation characteristics include the number of hard accelerations, the number of hard decelerations, the proportion of the number of hard lane changes, the speed variation coefficient, the number of passive brakes, the number of times without turning on the turn signal, the maximum value of the steering wheel rotation, the number of speeding violations, and fatigue driving behavior recognition; driving state characteristics at least include the degree of deviation of the average driving speed from the road speed limit, the speed fluctuation coefficient, the proportion of the duration of speeding, the proportion of the duration of driving at a low speed for a long time, and the frequency of abnormal driving behavior in specific dangerous sections; The features related to fatigue driving at least include driving duration distribution, proportion of eyelid closing time, abnormal head movement frequency, and regular changes in driving operations; the number of hard accelerations is obtained by calculating the number of times the acceleration exceeds the set threshold; the number of hard decelerations is obtained by calculating the number of times the acceleration is lower than the set threshold; the proportion of the number of sudden lane changes is used to identify and quantify sudden lane change events during vehicle driving by whether the lateral acceleration exceeds the preset sudden lane change acceleration threshold; the coefficient of variation of speed is the ratio of the standard deviation to the average value of the vehicle speed; The number of passive brakes is the number of times the vehicle system actively triggers the brakes when it determines that the driver is driving dangerously; The number of times of not turning on the turn signal counts the number of times of not turning on the turn signal in the specified subject; the maximum value of the steering wheel rotation calculates the maximum steering angle of the driver's steering wheel; the number of speeding counts the number of times the vehicle exceeds the speed limit range; the identification of fatigue driving behavior is based on behavior pattern recognition and combined with external sensor information for judgment.

5. The method according to claim 1, wherein: In the model construction and training step, the machine learning algorithms include Support Vector Machine (SVM), Random Forest (RF), and Neural Network (NN); When using a Support Vector Machine (SVM), the safety awareness ability of the driver is classified by finding the optimal hyperplane, and its formula is: f(x) = w T x + b, where w is the normal vector, x is the feature vector, and b is the bias term. The SVM selects the optimal hyperplane by maximizing the margin to achieve the classification of different safety awareness levels; When using the Random Forest (RF), classification is performed by constructing multiple decision trees and combining a voting mechanism. The prediction results of each tree are voted on to finally determine the driver's safety awareness level. The mathematical representation of the RF model is as follows: where N is the number of decision trees, is the prediction result of each tree; When using a neural network (NN), a deep learning method is adopted to extract deep features in driving behavior through a multi-layer perceptron (MLP), and a safety awareness score is output. The mathematical expression of the NN model is y = f(W (L) *f(W (L-1) ....f(W (1) *x + b (1) ) + b (L-1) ) + b (L) ), where W is the weight matrix, b is the bias, and f is the activation function.

6. The method according to claim 1, wherein: In the model update and optimization step, new driving behavior data is regularly collected and added to the original dataset, and the data preprocessing, feature extraction, model construction and training steps are re-executed to continuously optimize the performance of the evaluation model. At the same time, the safety awareness evaluation criteria and the content of optimization suggestions are dynamically adjusted according to the actual situation, and the actual situation includes the latest traffic accident cases and changes in traffic regulations.

7. The method according to claim 1, characterized in that, It includes comprehensively analyzing the evaluation results of three algorithms, namely support vector machine (SVM), random forest (RF), and neural network (NN), and using weighted average and voting mechanism to determine the final evaluation result of security awareness. The weighted average calculation formula is: Among them, w SVM , w RF , w NN are the weights of each model, which are determined by the accuracy of the model and other performance indicators.

8. A driver safety awareness evaluation and optimization device driven by driving behavior big data, characterized in that: This device is used to integrate any of the methods in claims 1 to 7 above into the device, and includes the following components: The data acquisition component is used to implement the data acquisition step, and collects the behavior data of the driver during driving through in-vehicle sensors, in-vehicle intelligent terminals, and mobile intelligent devices; The data preprocessing component is used to execute the data preprocessing step, and performs abnormal data removal, missing data filling, and normalization processing on the collected data; The feature extraction component is used to complete the feature extraction step, extracts key indicators representing the driver's safety awareness and driving behavior characteristics based on the preprocessed data, and performs correlation analysis on the extracted features to remove redundant features; The model construction and training component is used to implement the model construction and training step, selects a machine learning algorithm to construct an evaluation model for the driver's safety awareness ability, divides the dataset after feature extraction into a training set, a validation set, and a test set, uses the training set to train the model, monitors the model performance using the validation set during the training process, and determines the final model according to the performance indicators of the test set; The safety awareness evaluation component is used to carry out the safety awareness evaluation step, inputs the real-time collected and preprocessed driving behavior data into the trained evaluation model, outputs the driver's safety awareness evaluation score and divides the safety awareness level; The personalized optimization suggestion generation component is used to implement the personalized optimization suggestion generation step, and formulates personalized optimization suggestions for the driver according to the evaluation results and driving behavior characteristics; A model update and optimization component, which is used to perform model update and optimization steps, regularly collect new data, re - process data and train the model, and adjust evaluation criteria and optimization suggestions according to the actual situation; An interaction component, which is used to interact with the driver, enabling the driver to obtain safety awareness analysis results and improvement suggestions in real - time through in - vehicle systems, mobile applications or online service platforms, where: An in - vehicle system interaction sub - component. When the device cooperates with the in - vehicle system, the in - vehicle system is directly connected to the vehicle's sensors and communication systems, receives data from the data acquisition component and displays the analysis results; A mobile application interaction sub - component. When the device cooperates with the mobile application, the mobile application uploads the driving data collected by the data acquisition component to the cloud for analysis, and enables the driver to view historical driving behavior reports and improvement suggestions at any time; An online service platform interaction sub - component. When the device cooperates with the online service platform, the online service platform receives the data collected by the data acquisition component for global data analysis, behavior summary and generation of long - term improvement suggestions to help the driver improve safety awareness.

9. The driver safety awareness evaluation and optimization device driven by driving behavior big data according to claim 8, characterized in that The in - vehicle system interaction sub - component includes a data connection interface, a display unit and an audio prompt unit. The data connection interface is used to connect to the vehicle's sensors and communication systems. The display unit is used to display safety awareness evaluation results and driving behavior data. The audio prompt unit is used to output improvement suggestions in voice.

10. The driver safety awareness evaluation and optimization device driven by driving behavior big data according to claim 8, characterized in that, The mobile application interaction sub - component includes a data upload unit, a report viewing interface and a push notification unit. The data upload unit is used to upload driving data to the cloud. The report viewing interface is used for the driver to view historical driving behavior reports. The push notification unit is used to push improvement suggestions to the driver; The online service platform interaction sub - component includes a big data storage and analysis unit, a behavior summary unit and a long - term suggestion generation unit. The big data storage and analysis unit is used to store and analyze driving behavior data. The behavior summary unit is used to summarize the driver's behavior. The long - term suggestion generation unit is used to generate long - term improvement suggestions.

Images