Study On Short-Term Traffic Flow Velocity Prediction Based On LSTM-SVR Mixed Model

With the development of today’s society and the continuous progress of science and technology,there are now a variety of transportation modes to facilitate our travel,along with China’s economic development and people’s consumption level,the number of private cars is also increasing.However,with the continuous increase of the number of cars,the pressure on urban traffic is also increasing various problems also appear,such as traffic congestion,air pollution,traffic accidents,etc.In order to solve many problems in the field of traffic,the intelligent traffic system emerges with the efforts of relevant experts and scholars.Intelligent transportation system has achieved satisfactory results from reducing road traffic pressure,reducing road congestion rate and improving people’s travel efficiency.As an important premise of intelligent transportation,traffic flow prediction has been widely used in recent years.At present,the factors to be considered in the short-term traffic flow velocity prediction are as follows:first,how to find a suitable data set to study and analyze the traffic velocity;second,how to effectively preprocess the data to facilitate our prediction;third,how to choose the appropriate prediction model to predict the traffic flow speed.Short-time traffic flow prediction plays a very important role in the application of traffic control.Short-term traffic flow is characterized by nonlinearity,wide time span,uncertainty and instability.Therefore,on the basis of analyzing the disadvantages of single prediction model,this thesis proposes a combined model algorithm based on LSTM neural network and SVR by using the Q-traffic data set which is newly developed by baidu.LSTM model is a kind of recursive neural network model,which can learn the rules of long-term dependence more effectively than the general neural network model.Therefore,LSTM neural network model is more suitable for time series data such as traffic flow velocity,and SVR is a common algorithm in the field of traffic flow prediction.First,we built a composite model and made predictive analysis of traffic flow data.Through our simulation experiment,we can see that our composite model has significantly improved the accuracy compared with a single model.Based on the combination of model,We modified for the LSTM neural network of com-bination model above in the fourth chapter of this thesis,use Seq2Seq attention mechanism model to rebuild LSTM neural network,and use the new information extraction algorithm to extract the values from the dataset of new information source code into the neural network model,combining with the traffic speed data as input of the model.Through experiments,we can see that the modified new combination model algorithm performs better in a larger data set,which is further improved on accuracy compared with the combination model proposed to chapter 3 of this thesis.Then set up a control experiment on the speed prediction of the traffic network,the results show that the combined model algorithm proposed in the third chapter is better than the new fusion model in the aspect of traffic speed prediction,indicate that our combined model algorithm based on LSTM and SVR is better than the modified model algorithm proposed in chapter 4 in the generalization of traffic flow prediction while the confusion model we have put forward in the fourth chapter is more suitable for prediction of traffic flow speed in a single section.