Research On Traffic State Identification And Prediction Method Based On Floating Car Data

With the increasing number of motor vehicles,urban traffic problems are becoming increasingly serious.Traffic congestion,environmental pollution and traffic accidents are becoming serious obstacle to urban development.Accurate,real-time and reliable traffic state identification and prediction method is the key technology of intelligent transportation system,which is helpful to improve the efficiency of road network operation and is significant to traffic management control and public safety.The floating car technique is gradually gaining popularity in the research field of traffic forecasting due to its low difficulty in acquiring data,wide road network coverage and rich spatial and temporal information.This thesis aims to improve the performance of traffic state identification and prediction.In the background of data-driven,using deep learning with clustering and classification methods,the traffic state identification and prediction based on floating car data are deeply studied.The main contents and results are as follows:(1)ATT-CNN-LSTM model is constructed to predict short-term traffic flow.On the basis of in-depth analysis of the characteristics of floating car data,a long short-term memory network LSTM model is constructed for short-term traffic flow prediction,and then the LSTM baseline model is improved with attention mechanism and convolutional neural network CNN.Experiments show that the proposed ATT-CNN-LSTM model can effectively improve the short-term traffic flow prediction performance.(2)FCM model is constructed to identify the traffic state.The three characteristic indexes are selected,namely,average speed,travel time index and speed fluctuation.The k-means,FCM and DBSCAN algorithms are respectively used to cluster the historical data of floating cars.The contour coefficient is used as the evaluation index.The experimental results show that the FCM clustering effect is the best and can accurately divide the traffic state.(3)LSTM-FCM-KNN model is constructed to predict the traffic state.The FCM model was used to obtain the traffic state labels of the historical data,and then the KNN classification model FCM was used to obtain the data for training and analysis.Finally,the trained KNN model is utilized to predict the traffic status of the future short-time traffic flow prediction data obtained from the ATT-CNN-LSTM model,and the validity of the proposed traffic status prediction model is verified using the measured floating car data.