Prediction For Short-term Passenger Flow Of Subway Based On Improved PSO Optimized LSTM Neural Network

As a new,efficient and green means of public transport,the subway can effectively relieve traffic pressure and solve the contradiction between supply and demand in large and medium-sized cities to a large extent.Short-term passenger flow prediction is an important decision basis for subway dispatching.Accurate subway short-term passenger flow prediction can provide a guarantee for optimal operation of the subway.Based on the data of the subway swiping card,this paper analyzes the distribution characteristics of passenger flow;sums up the methods of passenger flow prediction,improves the particle swarm optimization algorithm(PSO),and constructs a subway short-term passenger flow prediction model based on an improved particle swarm optimization(IPSO)optimized Long Short-term Memory(LSTM)neural network,the optimal historical position of particles is taken as the optimal parameters of LSTM neural network(the number of iterations,the learning rate and the number of neurons in the hidden layers),the optimized parameters are used to build LSTM neural network and the passenger flow in the subway station is predicted,which greatly improves the operation and management of the subway.The specific research contents are as follows:The parameters of LSTM neural network are optimized by ipso algorithm.The optimized parameters are used to build LSTM neural network and the passenger flow in the metro station is predicted,(1)Data processing and analysis of subway passenger flow.Firstly,swiping card data of the subway card is processed to get the number of passengers entering the station in a given time interval to construct the passenger flow data of time series.Secondly,the empirical mode decomposition method is used to decompose the data of passengers on working days and non-working days to reduce the interference of data noise.Finally,this paper analyzes the relativity of week day and non-week day passenger flow,current passenger flow and historical passenger flow,and the passenger flow distribution under different time granularities.(2)Taking into account the problem that the standard particle swarm optimization algorithm cannot sufficiently distinguish between the global search and the local search-making it is easy to fall into the local extremum-the particle swarm optimization algorithm is improved.Firstly,the velocity updating formula and position updating formula are improved,and the inertia weight and time factor are changed from a fixed value to an adaptive value that changes dynamically with the number of iterations,so as to improve the global searching ability of the particle swarm optimization algorithm.Secondly,the adaptive mutation function is introduced to make it possible for the particle swarm optimization algorithm to have the ability of jumping out of local range.Finally,the Sphere function,Rastrigin function and Schwefel 2.22 function are selected to test the improved particle swarm optimization algorithm,which proves that the improved particle swarm optimization algorithm has higher searching ability.(3)Analysis and verification of prediction examples.Taking the passenger flow prediction of Shanghai subway station as an example,the incoming passenger flow of Lujiazui station for the following day is predicted.The improved particle swarm optimization algorithm is used to optimize the learning rate,the number of iterations and the number of neurons in the hidden layers of the LSTM neural network,and the optimized parameters are used to construct an IPSO-LSTM model to predict shortterm subway passenger flow,construct LSTM,PSO-LSTM and Multilayer Perceptron(MLP)three subway short-term passenger flow prediction models,compare the prediction accuracy through mean absolute percentage error,root mean square error,mean absolute error and coefficient of determination.The results show that the prediction errors of the IPSO-LSTM model are all smaller than those of other models,which proves that the IPSO-LSTM model constructed in this paper has a better prediction effect.