Research Of Short-term Traffic Flow Forecasting Based On Deep Learning

As urban traffic problems become more serious,intelligent transportation systems are becoming more and more important.It can effectively alleviate traffic congestion,improve the efficiency of road network,and provide information services for people’s traffic.The two core technologies of intelligent transportation system are traffic control and traffic guidance.Short-term traffic flow prediction can provide them with data foundation and is the core content of intelligent transportation system.This paper mainly studies short-term traffic flow prediction.The accuracy of short-term traffic flow prediction is affected by two factors:one is the accuracy of data collection;the other is the choice of prediction model.In terms of data collection,due to the high cost and low precision of the traditional use of fixed equipment to collect traffic flow data,this paper proposes a method based on map matching and artificial marker detection points.This method uses the vehicle GPS data as the data source,and through the operations of data preprocessing,map matching,manual marking detection points,and traffic flow extraction,the road traffic flow data with any position of any road as the detection point can be obtained.This method has the advantages of low cost,high precision,and intelligence.Short-term traffic flow prediction can be divided into traffic congestion prediction and traffic flow prediction.Therefore,in terms of model design,because most current models cannot fully consider the time and space characteristics of traffic flow,this paper introduces convolutional neural networks into traffic congestion prediction.In the congestion prediction,the traffic flow data is converted into the congestion level data,and then the convolutional neural network is used for feature extraction and classification prediction,and the congestion level of a certain road segment in the future time period is obtained,which proves that the convolutional neural network is effective in extracting traffic flow characteristics.In order to provide more basic data to the intelligent transportation system,traffic flow prediction is required.Because deep learning has better feature learning ability and distributed expression ability,it can better describe the internal changes of data,so deep learning is applied to traffic forecasting.This paper proposes a model that uses deep convolutional neural networks to extract traffic flow data features and input the extracted feature vectors into a support vector regression model to achieve traffic flow prediction.This model also considers time and space information and historical data,and road influence factors.In order to verify the accuracy of the two prediction models presented in this paper,experiments were carried out based on the data provided by Drip Trips.Experiments show that the prediction accuracy of the congestion prediction model proposed in this paper can reach 91.79%,which has a good prediction effect.Comparing the traffic flow prediction model proposed in this paper with the autoregressive integral moving average model and the support vector regression model,it is found that the model has better prediction results,that is,the degree of fitting is higher and the performance index is better.