The Short-Term Traffic Flow Prediction Based On Combined Model

Traffic congestion is one of the main problems faced in the governance process of all big cities in the 21 st century.Many cities are deploying various technologies to improve current traffic congestion.A useful way to try to alleviate traffic congestion is to perform short-term and it is also a key part of the(ITS)of intelligent transportation system.The accuracy of shortterm traffic flow prediction determines the quality of ITS in terms of traffic control and traffic guidance functions.This paper conducts research on the prediction of short-term traffic flow.(1)Detailed description of the development status of PSO,and based on this,a phased variation dynamic particle swarm optimization algorithm(SPSO)is proposed.In the early stage of the algorithm operation,some particles with too low fitness are mutated by updating their position to the average value of the particle position with the highest current fitness value;in the later stage of the algorithm operation,the number of evolutionary stagnation steps is used as the trigger condition(a preset Threshold,if it exceeds,perform the operation),random perturbation of individual extreme value and global extreme value.After the targeted mutation operation in the above phases,the diversity of the particle swarm can always be controlled within a reasonable range,thereby effectively improving the global convergence ability.(2)In the traffic flow prediction,the existing models will have large random fluctuations when forecasting,and the prediction error will exceed the safety limit,which will bring serious consequences to traffic flow guidance and traffic travel decision-making.The paper proposes a method based on EMD and SPSO.And GRU neural network traffic flow prediction model(EMD-SPSO-GRU).In this model,the experimental data is decomposed by the EMD algorithm to obtain a limited number of eigenmode functions(IMF);then the initial weights and thresholds of the GRU neural network are optimized using the strong global optimization capability of SPSO,and the optimized GRU is used The model learns the short-term time series law of each IMF component and makes predictions;finally,the predicted value of each IMF component is added to obtain the final predicted value.(3)The performance of SPSO and EMD-SPSO-GRU network models were compared and verified through experiments.For the MHPSO algorithm,mainly through comparison with PSO and QPSO,The optimization performance of this algorithm is verified mainly by three aspects:algorithm extensibility,population diversity and benchmark function optimization results,for the EMD-SPSO-GRU model,public traffic is used The stream data set is used for prediction,and the prediction performance is verified by comparison with the prediction results of BP,RNN,GRU and EMD-GRU models.