Research On Short-term Prediction Method Of Traffic Flow Based On Spatial-Temporal Characteristics

The real-time accurate prediction of short-term traffic flow in road network helps the intelligent transportation system to better analyze the traffic conditions of the road network,which is significant to the road network traffic planning and traffic optimization control that emphasizes real-time.In order to effectively solve the short-term prediction of traffic flow,this paper studies the short-term prediction of traffic flow in a single road segment and the short-term prediction of traffic flow in a road network under the premise of analyzing the time and space characteristics of traffic flow.And the intelligent technologies such as machine learning technologies are used to predict.For the short-term forecast of traffic flow in a single road segment,this study proposes a hybrid model Res-LSTM.The proposed model decomposes the traffic flow average velocity sequence into three distinct components(trend,periodic,and residual).The long-and short-term memory network(LSTM)and residual neural network(ResNet)methods are used to process and model the trend and residual parts,respectively.Finally,Res-LSTM assigns different weights to different branches through training,and aggregates the periodic part of the average speed sequence of the traffic flow and the predicted output of the residual neural network and the long-and short-term memory network with the fully connected layer.This paper verifies the effectiveness and reliability of Res-LSTM in traffic flow velocity prediction by using the actual traffic data of several road segments in Qingyang District of Chengdu.In view of the short-term prediction of traffic flow in the road network,this study proposes a hybrid model SpAE-LSTM.The model considers the spatial relationship between traffic flows of road segments in the road network and their own temporal changes.It consists of a sparse self-encoder and a storage unit-based long-and short-term memory network(LSTM).The model extracts the spatial features between the road segments through the sparse self-encoder and combines the temporal characteristics of the traffic flow evolution captured by the LSTM to predict.Finally,the paper selects the actual traffic data of two different types of road networks to verify the performance of the SpAE-LSTM model and proves the accuracy of the model for short-term prediction of road network traffic flow.In summary,this study validates the effectiveness of the proposed two hybrid model schemes,Res-LSTM and SpAE-LSTM,for short-term traffic flow prediction.At the same time,it also explains that for the prediction problem of complex urban traffic flow,the hybrid model that combines the advantages of different methods is an effective and reliable solution.