Research On Traffic Short-term Passenger Flow Forecast Based On Improved Neural Network

With the development of the economy,more and more cars are used in our country,leading to serious traffic congestion problems in city.Public transportation can effectively alleviate the problem of traffic congestion,while rail transit,an important part of the modern urban transportation system,plays an irreplaceable role.Therefore,the research on the short-term passenger flow prediction of urban rail transit helps to ensure the efficient and safe travel of residents.This thesis first filters and organizes the data set.Then,this thesis analyzes the correlation of the passenger flow data,and deletes the data with low overall correlation coefficient to improve the quality of the data sample.Due to the non-stationary and non-linear characteristics of the short-term passenger flow of urban rail transit,this thesis proposes to use a neural network model to predict the short-term passenger flow combined with historical data.Considering the shortcomings of Back Propagation(BP)neural networks,this thesis further proposes two methods to improve the prediction accuracy of BP neural network models.One is from the perspective of the structural parameters of the neural network,using Genetic Algorithms(GA)optimize the threshold and weight combination of the neural network.The other is from the data point of view,using a Modified Ensemble Empirical Model Decomposition(MEEMD)algorithm to decompose and reorganize time series data optimize operation.The two methods were combined with the BP neural network to construct two combined prediction models,namely the GA-BP model and the MEEMD-BP model.The MEEMD algorithm decomposes the data to obtain components with different correlation coefficients from the original data.According to the different recombination methods of these components,the MEEMD-BP model is determined as MEEMD-BP1,MEEMD-BP2 and MEEMD-BP3.In this paper,the prediction results of BP neural network model and the two types of combination models are compared and analyzed by using three evaluation indexes,namely root mean square error(RMSE),mean absolute error(MAE)and goodness of fit R~2.The experimental results show that the prediction accuracy of all combined prediction models is greatly improved compared to the prediction accuracy of the single BP neural network model.Compared with the BP neural network model,the RMSE of the GA-BP model is reduced by 18.76%,the MAE is reduced by 19.51%,and the R~2 is increased by 12.22%.Among the three MEEMD-BP models,the MEEMD-BP2 model has the best prediction results.Compared with the single BP neural network model,the RMSE of MEEMD-BP2 model is reduced by 34.86%,the MAE is reduced by 32.60%,and the goodness of fit R~2 is increased by 20.15%.The results show that combined prediction model has a good effect in improving the accuracy of short-term passenger flow prediction,while the MEEMD-BP combined model using the MEEMD algorithm to improve data quality has a greater improvement on BP neural network model.