Tunnel traffic flow and accident prediction model

Abstract

The invention provides a tunnel traffic flow and accident prediction model, and relates to the technical field of tunnel traffic. The tunnel traffic flow and accident prediction model comprises a multi-sensor data acquisition module, a traffic data source, a weather data source, a social media data source, a road network data source and a data integration platform. The multi-sensor data acquisition module is used for acquiring real-time monitoring data of the radar, the laser radar, the camera and the sound sensor, the traffic data source is used for acquiring historical traffic flow data, vehicle speed data, accident records and traffic event reports, and the weather data source is used for acquiring weather forecast and real-time weather data. The method has the advantages of multi-modal data fusion, real-time data updating, high accuracy, strong robustness, multi-task learning, edge calculation support and data security protection, and is higher in accuracy.

Description

Technical Field

[0001] The invention relates to the technical field of tunnel traffic, and in particular to a tunnel traffic flow and accident prediction model. Background Art

[0002] Existing tunnel traffic flow and accident prediction technologies mainly rely on traditional machine learning methods and simple deep learning models. These technologies are implemented through the following steps: First, data collection is carried out, including the use of sensor devices such as radars and cameras to collect information such as vehicle location and speed, obtaining historical traffic flow and accident records from traffic management departments, and obtaining weather forecasts and real-time weather data through the Meteorological Bureau or third-party weather services. Next, data preprocessing is carried out, including data cleaning for missing values ​​and outliers, and feature extraction such as time features, weather features, and vehicle features, and converting data into a format that can be processed by the model. Then, model construction and training are carried out, using traditional time series analysis methods such as ARIMA and SARI MA for traffic flow prediction, using traditional machine learning methods such as logical regression, decision trees, and support vector machines (SVM) for accident prediction, and using single-layer or shallow neural networks (such as MLP) or simple LSTM models for deep learning prediction. Next, we will evaluate the model, using MSE, RMSE, MAE and other indicators to evaluate the accuracy of traffic flow prediction, using accuracy, recall, F1 score and other indicators to evaluate the performance of accident prediction, and evaluate the performance of the model through cross-validation and test sets. Finally, we will deploy the model, integrate the model into the existing traffic management system, interact with data through the API interface, and develop a simple visualization interface to display the prediction results and real-time traffic conditions.

[0003] Existing tunnel traffic flow and accident prediction technologies have the following shortcomings: limited data processing capabilities, mainly including insufficient support for multimodal data. Existing technologies mainly process a single type of data (such as traffic flow data), and it is difficult to effectively integrate multimodal data (such as images, text, numerical values, etc.), as well as slow data updates, relying on historical data and manual updates, and unable to process and reflect current traffic conditions in real time. Limited model prediction capabilities, mainly manifested in low model complexity, using traditional machine learning methods and simple deep learning models, it is difficult to capture complex traffic patterns and long-term and short-term dependencies, and poor robustness. Under different traffic conditions and scenarios (such as holidays, bad weather, etc.), the prediction performance of the model may drop significantly. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a tunnel traffic flow and accident prediction model, which solves the problems of single data and poor robustness.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a tunnel traffic flow and accident prediction model, including a multi-sensor data acquisition module, a traffic data source, a weather data source, a social media data source, a road network data source, and a data integration platform;

[0006] The multi-sensor data acquisition module is used to obtain real-time monitoring data from radar, lidar, camera, and sound sensors. The traffic data source is used to obtain historical traffic flow data, vehicle speed data, accident records, and traffic incident reports. The weather data source is used to obtain weather forecasts and real-time weather data. The social media data source is used to obtain real-time traffic reports and user feedback from Twitter and Weibo platforms. The road network data source is used to obtain topological structure data of tunnels and surrounding roads. The data integration platform is used to obtain and integrate data from multiple data sources in real time, and to uniformly process and fuse different types of data through multimodal data processing technology.

[0007] Preferably, the prediction model also includes a data preprocessing module, and the data preprocessing module includes a data cleaning unit and a feature engineering unit.

[0008] Preferably, the prediction model also includes a data conversion unit for normalizing the numerical features to the same range and standardizing them so that their mean is 0 and their standard deviation is 1, and performing word segmentation and word embedding processing on the social media text data to convert them into a vector form that can be input into the model.

[0009] Preferably, the prediction model also includes a model construction module, and the model construction module includes a deep learning model, a graph neural network model, and a reinforcement learning model.

[0010] Preferably, the prediction model also includes a model training module, and the model training module includes a data partitioning unit and a parameter tuning unit.

[0011] Preferably, the prediction model also includes a model evaluation module, and the model evaluation module includes a traffic flow prediction evaluation unit, an accident prediction evaluation unit, and an uncertainty estimation unit.

[0012] Preferably, the prediction model also includes a real-time event detection module, and the real-time event detection module includes an anomaly detection unit and an event classification unit.

[0013] Preferably, the prediction model further includes a personalized prediction module, and the personalized prediction module includes a user behavior analysis unit, a dynamic adjustment unit, and a personalized recommendation unit.

[0014] Preferably, the prediction model also includes a traffic management system, and the model is integrated into the existing traffic management system to realize real-time application of the model.

[0015] The present invention provides a tunnel traffic flow and accident prediction model. It has the following beneficial effects:

[0016] The present invention provides a tunnel traffic flow and accident prediction model. Compared with the prior art, the present technology includes multimodal data fusion, which uses multimodal data processing technology to effectively fuse different types of data, improve the richness and accuracy of data, and real-time data update. Through edge computing and distributed architecture, real-time data collection and processing are realized to ensure that the model can timely reflect the current traffic conditions. LSTM, GRU and Transformer deep learning models are used to capture the long-term and short-term dependencies between traffic flow and accidents, improve the accuracy and robustness of prediction, use graph neural networks to model the road network structure, consider the correlation and influencing factors between roads, further improve the prediction accuracy, and use multi-task learning methods to simultaneously predict traffic flow and accident probability, improve the overall performance and generalization ability of the model, perform preliminary data processing and prediction on edge devices through edge computing, reduce data transmission delay, and improve response speed, use Isolation Forest and Autoencoder technologies to perform real-time anomaly detection and event classification, and timely discover and handle emergencies. DETAILED DESCRIPTION

[0017] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0018] An embodiment of the present invention provides a tunnel traffic flow and accident prediction model, including a multi-sensor data acquisition module, a traffic data source, a weather data source, a social media data source, a road network data source, and a data integration platform;

[0019] The multi-sensor data acquisition module is used to obtain real-time monitoring data from radar, lidar, camera, and sound sensors. The traffic data source is used to obtain historical traffic flow data, vehicle speed data, accident records, and traffic incident reports. The weather data source is used to obtain weather forecasts and real-time weather data, such as temperature, humidity, rainfall, and wind speed. The social media data source is used to obtain real-time traffic reports and user feedback from Twitter and Weibo platforms. The road network data source is used to obtain topological structure data of tunnels and surrounding roads, such as road length, curvature, slope, traffic light location, and surveillance camera location. The data integration platform is used to obtain and integrate data from multiple data sources in real time, and to uniformly process and fuse different types of data through multimodal data processing technology.

[0020] The prediction model also includes a data preprocessing module, which includes a data cleaning unit and a feature engineering unit.

[0021] The data cleaning unit is used to process missing values ​​and outliers. Missing value processing includes interpolation and mean filling methods. Outlier detection includes statistical methods and the machine learning algorithm Isolation Forest. The feature engineering unit is used to extract time features, weather features, vehicle features, road network features and event features. Time features include hours, days of the week, and holidays. Weather features include temperature, humidity, rainfall, wind speed, vehicle features include vehicle type, speed, driving direction, etc. Road network features include the topological structure of tunnels and surrounding roads, traffic light locations, and surveillance camera locations. Event features include historical and real-time traffic event locations, types, and processing times.

[0022] The prediction model also includes a data conversion unit, which is used to normalize the numerical features to the same range (such as [0,1]) and standardize them so that their mean is 0 and their standard deviation is 1, and to perform word segmentation and word embedding on the social media text data to convert them into a vector form that can be input into the model.

[0023] The prediction model also includes a model building module, which includes a deep learning model, a graph neural network model, and a reinforcement learning model.

[0024] The deep learning models are LSTM / GRU model and Transformer model, which are used to process time series data and capture the long-term and short-term dependencies between traffic flow and accidents. The graph neural network models are GCN and GAT, which are used to process the topological structure information of tunnels and surrounding roads and extract the features of nodes and edges. The reinforcement learning models are DQN and PPO, which are used to dynamically optimize traffic management and scheduling strategies and make the best decisions based on real-time traffic conditions.

[0025] The prediction model also includes a model training module, which includes a data partitioning unit and a parameter tuning unit.

[0026] The data partitioning unit divides the data into a training set (accounting for 70% of the total data), a validation set (accounting for 15% of the total data) and a test set (accounting for 15% of the total data). The parameter tuning unit selects the best model parameters through K-fold cross-validation and hyperparameter optimization methods (such as grid search, random search or Bayesian optimization).

[0027] The prediction model also includes a model evaluation module, which includes a traffic flow prediction evaluation unit, an accident prediction evaluation unit, and an uncertainty estimation unit.

[0028] The traffic flow prediction evaluation unit uses mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE) indicators for evaluation. The accident prediction evaluation unit uses accuracy, recall, and F1 score indicators for evaluation. The uncertainty estimation unit uses the Bayesian method to estimate the uncertainty of the prediction results and calculate the confidence interval.

[0029] The prediction model also includes a real-time event detection module, which includes an anomaly detection unit and an event classification unit.

[0030] The anomaly detection unit uses Isolation Forest and Autoencoder to detect abnormal events in traffic, such as vehicle failure, accidents, and congestion. The event classification unit uses Convolutional Neural Network (CNN) and Support Vector Machine (SVM) to classify the detected events, such as accident type and severity.

[0031] The prediction model also includes a personalized prediction module, which includes a user behavior analysis unit, a dynamic adjustment unit, and a personalized recommendation unit.

[0032] The user behavior analysis unit analyzes the user's driving behavior and historical data, extracts the user's driving habits, common travel time and other characteristics; the dynamic adjustment unit dynamically adjusts the prediction model according to user feedback and real-time data to improve the accuracy and practicality of the prediction; the personalized recommendation unit provides personalized travel suggestions and warning information based on the user's driving habits and real-time traffic conditions.

[0033] The prediction model also includes traffic management system, which integrates the model into the existing traffic management system to realize the real-time application of the model.

[0034] Edge computing devices are used for preliminary data processing and prediction, reducing data transmission delays and improving response speed. The cloud computing platform is used for model training and deployment, providing powerful computing and storage resources. The data security and privacy protection module includes a data encryption unit, an access control unit, a federated learning unit and an anonymization processing unit to ensure data security and privacy protection.

[0035] Specifically: Sensor data includes radar, lidar, camera, and sound sensor, which are used to monitor vehicle position, speed, driving direction, and sound in the tunnel in real time. Traffic data includes historical traffic flow data, vehicle speed data, accident records, and traffic incident reports. Weather data includes weather forecasts and real-time weather data, such as temperature, humidity, rainfall, and wind speed. Social media data includes real-time traffic reports and user feedback from platforms such as Twitter and Weibo. Road network data includes the topological structure of the tunnel and surrounding roads, the location of traffic lights, and the location of surveillance cameras.

[0036] Through the data integration platform, data can be acquired and integrated from multiple data sources in real time. Multimodal data processing technology can be used to uniformly process and fuse different types of data (such as images, text, and numerical values). Interpolation and mean filling methods can be used to handle missing values. Outliers can be detected and processed through statistical methods or machine learning algorithms (such as Isolation Forest).

[0037] Time features include extracting hour, day of the week, and holiday time features. Weather features include extracting temperature, humidity, rainfall, and wind speed weather features. Vehicle features include extracting vehicle type, speed, and driving direction features. Road network features include extracting topological structure features of tunnels and surrounding roads, such as road length, curvature, and slope. Event features include extracting historical and real-time traffic event features, such as accident location, accident type, and processing time.

[0038] Data conversion is to normalize numerical features to the same range (such as [0,1]). The numerical features are standardized so that their mean is 0 and their standard deviation is 1. Then the social media text data is processed by word segmentation and word embedding to be converted into a vector form that can be input into the model.

[0039] By processing time series data, the long-term and short-term dependencies between traffic flow and accidents are captured. The self-attention mechanism is used to process complex multimodal time series data to improve prediction accuracy. Combined with traditional time series analysis methods, the robustness of the model is enhanced.

[0040] The topological structure of the tunnel and surrounding roads is modeled as a graph, where each node represents a camera, sensor or traffic light, and each edge represents a road connection. The features of nodes and edges are extracted through graph neural networks (such as GCN and GAT), taking into account the correlation and influencing factors between roads. Convolution operations are performed on the graph to capture the structural information of the traffic network.

[0041] Use reinforcement learning DQN and PPO to dynamically optimize traffic management and scheduling strategies, and make the best decisions based on real-time traffic conditions. Use reinforcement learning models to continuously learn and adapt to new traffic patterns, and optimize traffic signal control and vehicle scheduling in tunnels.

[0042] In model training, 70% of the total data is used to train the model. 15% of the total data is used to tune the model parameters. 15% of the total data is used to evaluate the final performance of the model. K-fold cross validation is then used to select the best model parameters. The optimal hyperparameters are found through grid search, random search, or Bayesian optimization methods.

[0043] Traffic flow prediction uses mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE) indicators. Accuracy, recall, and F1 score indicators are used for accident prediction. Uncertainty estimation uses the Bayesian method to estimate the uncertainty of the prediction results and calculate the confidence interval.

[0044] Use traffic simulation software (such as SUMO and VISSIM) to verify the model and simulate the prediction effect under different traffic conditions. Test the performance of the model in different scenarios (such as holidays, bad weather, special events, etc.) to ensure the robustness of the model.

[0045] Isolation Forest is used to detect abnormal events in traffic, such as vehicle failures, accidents, and congestion. Autoencoders are used to detect abnormal patterns to improve the robustness and accuracy of detection. Convolutional neural networks (CNNs) are used to classify detected events, such as accident type and severity. Support vector machines (SVMs) are used as a classification algorithm to supplement CNN classification results and improve classification accuracy.

[0046] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A tunnel traffic flow and accident prediction model, characterized by: Including multi-sensor data acquisition module, traffic data source, weather data source, social media data source, road network data source, and data integration platform; The multi-sensor data acquisition module is used to obtain real-time monitoring data from radar, lidar, camera, and sound sensors. The traffic data source is used to obtain historical traffic flow data, vehicle speed data, accident records, and traffic incident reports. The weather data source is used to obtain weather forecasts and real-time weather data. The social media data source is used to obtain real-time traffic reports and user feedback from Twitter and Weibo platforms. The road network data source is used to obtain topological structure data of tunnels and surrounding roads. The data integration platform is used to obtain and integrate data from multiple data sources in real time, and to uniformly process and fuse different types of data through multimodal data processing technology.

2. A tunnel traffic flow and accident prediction model according to claim 1, characterized in that: The prediction model also includes a data preprocessing module, which includes a data cleaning unit and a feature engineering unit.

3. The tunnel traffic flow and accident prediction model according to claim 1, characterized in that: The prediction model also includes a data conversion unit for normalizing the numerical features to the same range and standardizing them so that their mean is 0 and their standard deviation is 1, and performing word segmentation and word embedding processing on the social media text data to convert them into a vector form that can be input into the model.

4. The tunnel traffic flow and accident prediction model according to claim 1, characterized in that: The prediction model also includes a model construction module, which includes a deep learning model, a graph neural network model, and a reinforcement learning model.

5. The tunnel traffic flow and accident prediction model according to claim 1, characterized in that: The prediction model also includes a model training module, and the model training module includes a data partitioning unit and a parameter tuning unit.

6. The tunnel traffic flow and accident prediction model according to claim 1, characterized in that: The prediction model also includes a model evaluation module, which includes a traffic flow prediction evaluation unit, an accident prediction evaluation unit, and an uncertainty estimation unit.

7. The tunnel traffic flow and accident prediction model according to claim 1, characterized in that: The prediction model also includes a real-time event detection module, which includes an anomaly detection unit and an event classification unit.

8. The tunnel traffic flow and accident prediction model according to claim 1, characterized in that: The prediction model also includes a personalized prediction module, which includes a user behavior analysis unit, a dynamic adjustment unit, and a personalized recommendation unit.

9. The tunnel traffic flow and accident prediction model according to claim 1, characterized in that: The prediction model also includes a traffic management system, which integrates the model into the existing traffic management system to achieve real-time application of the model.