Research On Short-Term Traffic Flow Prediction Method Based On Characteristic Analysis

With the rapid development of economy,the increasing level of urbanization and the continuous growth of car ownership,the traffic congestion,traffic safety and environmental pollution problems are becoming more and more serious.As one of the most effective ways to solve urban traffic problems,the Intelligent Traffic System(ITS)has received worldwide attention.Traffic control and induction,as the core application of ITS,play an important role in alleviating traffic congestion,improving traffic safety and reducing environmental pollution.In order to realize traffic control and induction,accurate traffic flow prediction is needed.Scholars at home and abroad have done a lot of research in the field of traffic flow prediction and proposed many pioneering and innovative traffic flow prediction methods.However,since most research focus on the optimization of models and algorithms,considering less about the characteristics of traffic flow data,it is difficult to make breakthrough in accuracy and stability.Therefore,based on the analysis of traffic flow data,this paper proposes several prediction models that adapt to the characteristics of traffic flow,realizing accurate traffic flow prediction and laying foundation for efficient traffic control and induction.First,by summarizing the current research status,the characteristics of traffic flow on urban expressway are analyzed.According to the spatio-temporal correlation of the traffic flow,a 3D shape function spatio-temporal interpolation method is proposed to recover the missing traffic data and reduce the subsequent prediction error caused by the abnormal data.According to the chaotic characteristics of traffic flow,the Phase Space Reconstruction(PSR)method is introduced to construct the input dataset of the prediction model and avoid the subjective error caused by selection of the input dataset.Second,based on the chaotic characteristics of traffic flow,the Echo State Network(ESN),which is suitable for chaotic time series prediction,is introduced.Combining the PSR and ESN,the traffic flow prediction model is established.The experimental results show that the proposed model outperforms traditional traffic prediction models.Besides,in order to solve the hyper parameter selection problem of the model,the Mind Evolution Algorithm(MEA)is introduced.By employing convergence and alienation operations,MEA can search the optimal combination of reservoir pool size and parameters of the ESN.The experimental results show that MEA has fast convergence speed and can avoid being trapped in the local optimum,and the MEA-ESN model has higher prediction accuracy.Finally,in order to make breakthrough in predicting the traffic flow with large fluctuation,a traffic flow prediction model based on decomposition and reconstruction is established.From the perspective of data structure,the Empirical Mode Decomposition(EMD)method with adaptive decomposition ability is innovatively applied to traffic flow prediction.EMD can convert mixed and non-stationary time series into several stable decomposition time series,and thus enhancing the predictability of the traffic data.Experimental results show that the proposed decomposition and reconstruction method can improve the accuracy of traffic flow prediction dramatically.At the same time,in order to solve the limitation cause by the single input,a multivariate decomposition and reconstruction method is proposed.According to the correlation of the traffic flow parameters,the model converts the speed and occupancy data into theoretical flow data,decomposing and reconstructing the observed and theoretical flow at the same time,constructing input dataset containing both of them.The experimental results prove that multivariate decomposition and reconstruction method can further improve the accuracy of traffic flow prediction.There are 56 figures,20 tables and 92 references.