Font Size: a A A

Research On Lane-changing Trajectory Model Based On Deep Learning

Posted on:2020-02-13Degree:MasterType:Thesis
Country:ChinaCandidate:S Y LiuFull Text:PDF
GTID:2392330602459453Subject:Transportation planning and management
Abstract/Summary:PDF Full Text Request
In recent years,the explosive growth of the number of motor vehicles has brought convenience to residents’travel,but at the same time,it inevitably caused a series of traffic problems,such as traffic congestion,traffic accidents,environmental pollution and so on.As one of the main driving behaviors,lane-changing behavior greatly affects road traffic safety and traffic flow characteristics.Moreover,lane-changing behavior is also one of the key issues in the field of assisted driving.In order to promote the development of automatic driving technology,and to better solve urban traffic problems,the lane-changing behavior is deeply analyzed by high-precision measured data,and a lane-changing trajectory model based on deep learning was proposed in this paper.The research contents are as follows:Firstly,the related theory of lane-changing behavior,the basic theory of deep learning and the acquisition and processing method of measured data of driving behavior were elaborated by classification and progressive method.Based on the analysis of errors in data,a scheme of adjusting acceleration,speed and position data was put forward,and the adjusted data was closer to actual driving.Based on the reconstructed vehicle trajectory data and the definition of lane-changing process,the data of subject vehicle lane-changing process and the position and speed data of the associated vehicles which had influence on lane-changing behavior had been extracted.According to the data of lane-changing process,the microscopic characteristics of lane-changing behavior had been analyzed from two aspects of lane-changing duration and speed.The lane-changing vehicles were connected with the associated vehicles,and the interaction between vehicles in lane-changing process had been described from speed and distance.In view of the characteristics of time series and deep learning in data feature extraction,a lane-changing trajectory model based on the Gated Recurrent Unit(GRU)was constructed to predict lane-changing trajectory;and the lane-changing trajectory model of the traditional BP neural network and the Long-Short-Term Memory(LSTM)neural network had been analyzed as a comparative experiments.The results showed that the lane-changing trajectory model based on the Gated Recurrent Unit(GRU)had fewer training times,faster training speed,and the predicted trajectory was more consistent with the actual trajectory.It shows that the GRU-based lane-changing trajectory model was more suitable than the BP-based or the LSTM-based neural network.
Keywords/Search Tags:Lane-changing behavior, Trajectory data, Deep learning, the Gated Recurrent Unit, Lane-changing trajectory model
PDF Full Text Request
Related items