Research On Vehicle Trajectory Prediction Algorithm Based On Deep Learning

With the increasing demand of modern urban residents for traffic,traffic congestion and traffic accidents caused by the increasing pressure of traffic environment can not be ignored.The increase in the number of motor vehicles at forks and special sections has laid hidden dangers for the occurrence of road safety accidents.Through the information collection technology,the massive real-time traffic data collected are fully analyzed,and the travel habits,trajectory behavior patterns and social route information of residents are mined,so as to meet the needs of urban residents and social traffic information.Due to the value of trajectory data and the convenience of collection,vehicle trajectory prediction has become the main research basis of current transportation system.According to the data provided by trajectory prediction,quantitative and scientific analysis can be made in the short-term and dynamic changes.The driver can decide to change the travel route or travel time in advance,so as to alleviate the traffic jam,which is of great significance to the development of the city.First of all,this paper investigates the current research status of trajectory prediction,excavates the trajectory data of traffic websites,and through a series of data preprocessing such as data cleaning and feature normalization,combines the respective advantages of Kalman filter algorithm and fixed interval smoothing method,makes the improved Kalman filter smoothing algorithm process the trajectory sequence data,and improves the quality of trajectory data,And detailed description of the data processing process to show the effect.The first mock exam is deep learning model and machine learning model.Convolution neural network model,long and short time memory model,linear regression model,support vector regression model and Lasso regression model are established respectively.The consistency of each single model parameter is maintained,and the prediction results are compared with the same evaluation indexes.CNN model and LSTM model with better effect are selected.Finally,the first mock exam is used to try to determine the optimal proportion between the single models,and the CNN is the final one.LSTM combined forecasting model is applied to the system.Compared with the single prediction model,the error of the combined prediction model is significantly reduced,in which the mean square error is reduced by 6.52%,and the prediction effect is good,which provides scientific and reasonable suggestions for urban transportation system planning and scheduling.