Research On Dynamic Traffic State Prediction And Global Path Planning Based On Traffic Big Data

The role of traffic intelligence in alleviating traffic pressure and improving traffic convenience has been continuously strengthened,making it a hot topic in the field of intelligent transportation.At present,under the premise of continuous integration of high technology such as geographic information system(GIS),global positioning system(GPS),big data technology and computer technology,intelligent transportation systems can carry out more reliable and effective vehicle path planning,thereby reducing traffic jams.Probability of occurrence or alleviation of road congestion,reduce the occurrence of traffic accidents,and contribute to the realization of smart cities.As an important research direction in intelligent transportation systems,vehicle path planning has been paid attention to by domestic and foreign academic circles for many years and many research results have been obtained.Vehicle path planning based on predicted information can provide users with reliable traffic driving advice.Its research content mainly includes traffic information prediction,road network model construction and matching and path planning algorithms.This paper builds traffic information prediction model based on the real-time traffic information of Auto Navi Map,and then designs path planning algorithm based on this model.The main work is as follows:1.Obtain traffic big data and data preprocessing.Through the interactive interface provided by web crawlers and Auto Navi Maps,traffic information for a specific time period of 73 working days on some roads in Changsha area can be obtained.After obtaining the data,analyze the characteristics of the data,mainly including the temporal and spatial characteristics of traffic information.Data processing is performed according to the data characteristics;the nodes in the non-target area are eliminated,and the nodes with the same road information change over time in the same road section are represented as edges with weights to make the data more compact.2.Based on Le Net,two-layer RNN,GRU and LSTM models containing composite activation functions to build and train traffic prediction models,the short-term prediction accuracy of various models are compared when short-term traffic information and single-time traffic information are used as prediction conditions.Before training the CNN model,the characteristics and labels of the data are set for different prediction tasks;for the time series models of RNN,GRU and LSTM,the small batches of data are obtained from this arrays through adjacent sampling for training;use short-term traffic data and individual time traffic data as the forecast input respectively to analyze the forecast results of each model;when the forecast experiment is carried out with the traffic information at a single time as the forecast input,the data is increased through normal distribution to expand the dimension of the forecast input.3.The performance of three global path planning algorithms is compared,and an optimal energy consumption path planning method based on ant colony algorithm is proposed.The topology network required for path planning is constructed according to road network model theory;the performance of Dijkstra algorithm,A* algorithm and ant colony algorithm is compared when planning the shortest path in different road network scales;based on ant colony algorithm and the grid method,combined with the kinematics principle of the vehicle when driving,an optimal energy consumption path planning method is proposed.4.A path planning method based on traffic prediction is proposed.According to vehicle driving theory,kinematics theory and energy theory,combined with the characteristics of prediction data,the calculation method of path weights in changing road conditions,including driving time weights and vehicle energy consumption weights,is derived;based on the obtained weight calculation method combined with the Dijkstra algorithm,the path planning algorithm based on a dynamic environment is designed;a reverse-order RNN model is proposed to increase the prediction input of the CNN model;10 sets of actual traffic data are took as experimental samples,obtaining short-term traffic information through different methods,combining the improved Dijkstra algorithm for path planning,and use the results of path planning based on actual information as the reference group to analyze the pros and cons of the results obtained by each experimental group.5.For the obtained traffic data set,the state of the data nodes on the same road direction of the same road is always the same;when short-term traffic information is used as the prediction input,the CNN-based model has the best prediction effect,and when the traffic information at a single time is used as the prediction input,the RNN-based model has the best prediction effect;as the scale of the road network increases,the efficiency of the Dijkstra algorithm,A* algorithm,and ant colony algorithm decreases,and the A* algorithm has the highest efficiency;simulation experiments show that the path planning results of the improved Dijkstra algorithm combined with the RNN prediction model have the smallest error and the highest coincidence rate with the path planning results based on actual information.