Study On Short-term Traffic Flow Prediction Of Cities Based On LSTM And STCL3d Networks

The contradiction between the limited transportation resources of major cities in China and the increasing traffic demand is increasing due to the accelerating urbanization process,and intelligent transportation system plays an increasingly important role to solve this problem.How to accurately predict the traffic flow in the future,rational and effectively use and allocate existing transportation resources,and reduce traffic congestion are always important tasks of the intelligent transportation system.This paper extracts the traffic flow information from the taxi trajectory data and uses the deep learning method to realize the traffic flow prediction function.The main work is as follows:(1)According to the noise in the taxi trajectory data,the original data has been cleaned based on three aspects which are speed abnormality,orbital number abnormality and dwell time abnormality,and the map matching method has been used to extract road traffic flow parameters.The trajectory data of taxi has been mined,and the taxi operational situation and residents' travel situation in Shanghai have been extracted.The OD distribution rule of taxi operation has been analyzed visually by light map and heat map.(2)Aiming at the existing road traffic flow parameter prediction methods have low prediction accuracy and the spatial connection of the traffic flow parameters between the road segments is neglected,the traffic flow parameters of the current road segment and the upstream road segment have been input into the LSTM(Long Short-Term Memory)network model to achieve prediction of traffic flow parameters of multiple road sections(3)Putting the in-area and out-area flows of multiple consecutive blocks into the end-to-end model of STCL3d(Spatial-Temporal Convolutional LSTM Conv3d),which uses CNN to learn the spatial relationship between different areas,LSTM to learn the time relationship of traffic flow,Conv3 d to realize multi-time-space multi-region traffic flow parameter prediction.(4)The prototype system of traffic flow prediction has been designed,and the system is illustrated from the aspects of overall design,functional module design and visual interface.The prototype system makes it more convenient and intuitive for users to query traffic flow.The research shows that road speed needs to be smoothed and standardized before it is input into the network model,and the memorized LSTM model with BN layer can well fit the trend of road speed and predict road traffic flow parameters.STCL3 d model can express the spatial-temporal correlation relation of regional traffic flow,and realize the prediction function of regional traffic flow.In this paper,road traffic flow and regional traffic flow are modeled respectively.The processing of data,the transformation of data structure and the establishment of network model are of certain reference value for the study of other traffic flow prediction problems and timing data prediction problems.