Research On Short-Term Traffic Flow Prediction Based On Deep Learning

With the rapid development of national urbanization,the number of motor vehicles in the city is increasing day by day,followed by traffic congestion,traffic noise,environmental pollution and other problems are also troubling people.The emergence of intelligent transportation systems offers new solutions to these traffic problems,as an important part of the intelligent traffic subsystem,the short-term traffic flow prediction can provide the traffic management department with reliable traffic data in the future.Traffic control platform can use the predicted traffic flow data to carry out traffic diversion and path planning.It can be seen that the accuracy of the prediction data directly affects the play of the role of the entire intelligent transportation system.Therefore,it is great academic and practical significance to study an accurate and stable method for the prediction of short-term traffic flow on urban roads.In recent years,the improvement of short term traffic flow prediction method has been gradually deepened with the development of deep learning theory.In this paper,a prediction model based on deep learning is constructed based on the advantages of Long Short Term Memory(LSTM)network in time series data processing,aiming at the problems of structural lag and complex computation of some existing traditional traffic flow prediction models.Aiming at the problem that some prediction models only face a single road segment and the input data of the model is not fully processed.In this paper,the actual road network structure is selected,and the original traffic flow sequence is decomposed and reconstructed by heuristic threshold denoising algorithm to achieve the purpose of denoising.By calculating the correlation coefficient of traffic flow data of each section of the road network,the compression matrix of traffic flow data of the road network is constructed.Data de-noising and correlation analysis of road network data minimize the interference of data to the model.At the same time,It make the prediction be considered at the road network level.The actual traffic flow data is used to verify the model.Firstly,the LSTM-1 and LSTM-2 models are trained with original data and de-noising processing data respectively.The effectiveness of the wavelet threshold de-noising and LSTM model combined with prediction method is proved by comparing performance evaluation indexes.Then,the prediction model proposed in this paper is compared with the other four models by setting different prediction time steps.According to the simulation results,the average prediction accuracy of the LSTM-2 model proposed in this paper is 94.62%.When the prediction time step was increased from 10 min to 20 min,the increase and decrease difference of root mean square error of LSTM-2 model was only 2.31%.Among several comparison models,the prediction accuracy and stability of LSTM-2 model were the best,which indicated that theprediction method proposed in this paper was efficient and reliable.On this basis,the differences of traffic flow prediction models in weekend and non-weekend periods are considered.Finally,based on the traffic flow predicted in this paper,the congestion discrimination system is defined.According to the congestion discrimination system,the congestion discrimination of the selected research section is carried out and the congestion discrimination result table is given.It has realized the transformation of research results from theory to practical application.