Research On Forecasting Method Of Short-term Traffic Flow Using Spatio-temporal Information

At present,the number and types of vehicles in China are increasing dramatically,causing more and more traffic jams.Traffic jams not only caused environmental pollution,but also caused the waste of some public resources.Therefore,how to effectively reduce traffic pressure,improve the capacity of highways,and then ensure the convenience of residents to travel,has become an urgent problem to be solved in China’s transportation industry.Short-term traffic flow prediction as an important part of the Intelligent Transportation System(ITS)can effectively alleviate traffic congestion and improve transportation efficiency.Aiming at the problem of short-term traffic flow prediction,this paper studies the traffic flow prediction model using spatio-temporal information and deep learning method.The research work is as follows:Aiming at the traffic flow prediction of a single road segment,a prediction model based on GA-BP neural network is studied.Because the BP neural network is easy to fall into a local minimum during the training process and the convergence rate is slow,a GA-BP short-term traffic flow prediction model using a genetic algorithm to optimize the BP neural network is studied.The global optimization of the GA algorithm is used to find the optimal initial weight and threshold of the BP neural network,which improves the shortcomings of the BP neural network randomly selecting the initial value.It is verified that the GA-BP model has better prediction performance than the traditional BP model on the two traffic data sets of TRDL and PEMS.A bidirectional LSTM(Bi-LSTM)model considering the forward and reverse order of the traffic flow sequence is studied and proposed.In order to make full use of the time information of the traffic flow,the long-term and short-term memory network(LSTM)is used in the good prediction performance of processing time series problems,combining forward LSTM and reverse LSTM,while considering the forward and reverse time dependence of the data,according to the selection Model structure to predict traffic flow data.The experimental results show that when the number of hidden layers of the Bi-LSTM model is 3,the number of hidden nodes of each layer is 50,and the sliding time window size is 6,the prediction effect is the best;the Bi-LSTM model considering the bidirectional order of the traffic flow sequence is simpler The predictive performance of the directed LSTM network and other machine learning algorithms is better,which reflects the advantages of the Bi-LSTM model using the time characteristics of traffic flow data and considering forward and reverse directionality at the same time.For the prediction of traffic flow at multiple stations,a KNN-CNN-Bi LSTM model using space-time information is proposed.In order to fully consider and utilize the spatio-temporal information of the upstream and downstream traffic flows at the monitoring points to be tested,first use the KNN algorithm to sort the detection data at the upstream and downstream of the road monitoring points according to the Euclidean distance from the station to be measured,and select the combination with the best prediction performance Way,organize the input data into the form of a two-dimensional matrix,draw on the application of convolutional neural networks in the field of image recognition,use CNN to extract the spatial relationship between the data of different monitoring points,use Bi-LSTM to extract the time features,and finally use the full Connect the layer to denormalize the output of the final predicted traffic flow.The experimental results show that the KNN algorithm can effectively optimize the data set after selecting monitoring points,and the prediction accuracy is better and the time is shorter,and the KNN-CNN-Bi LSTM model that uses spatio-temporal information is better than the BiLSTM model that only uses time information and other The prediction effect of mainstream prediction models is better.