Research On Short Traffic Flow Forecast Based On IGA-LSTM

The transportation industry is the main artery of economic development.With the rapid economic development,people are becoming more and more dependent on various means of transportation.The continuous increase of various vehicles has brought tremendous pressure to road traffic,traffic accidents and other problems have also increased.And people’s demand for a more complete Intelligent Transport System(ITS)has also become stronger.Traffic flow forecasting,as the research focus of ITS,is also receiving more and more attention.How to improve the prediction accuracy and speed of short-term traffic flow,and provide timely and accurate advice to the traffic management department,is the research significance of this paper.Therefore,in this paper,a short-term traffic flow prediction model is proposed based on the long and short-term memory neural network(LSTM)optimized by the improved genetic algorithm.The main tasks completed in this paper are as follows:(1)First,this paper do some research and summarize on the relevant research on shortterm traffic flow at China and abroad.The study found that short-term traffic flow data is a kind of time series data with strong nonlinearity and randomness.LSTM is effective in processing time series data.Therefore,this paper selects LSTM as the basic model for short-term traffic flow forecasting research.(2)An improved genetic algorithm(IGA)is proposed.Aiming at the problem of premature convergence and poor local convergence of traditional genetic algorithm(GA),it is proposed in the coding process of genetic operation a encoding method based on integer coding.In order to enhance the optimization ability of the algorithm,an error-based fitness function is introduced in the selection operation.To speed up the convergence rate of the population and prevent local convergence,a new crossover operation and mutation operation are adopted.(3)Propose a short-term traffic flow prediction model based on IGA optimized LSTM.Generally,it is necessary to manually set the model parameters of the LSTM.The setting of these parameters will greatly affect the model’s fitting ability,training speed and final prediction effect.Aiming at the inherent defects of neural network manual tuning,such as time-consuming and inaccurate tuning,this paper uses IGA to optimize the initial parameters of LSTM and establishes an IGA-LATM model.(4)The data collected by various sensors or devices often have different problems or defects.In order to ensure the prediction effect,data preprocessing is carried out.The nearest neighbor substitution method and regression model method are used for the missing data,and the Holt-Winters’ seasonal method is used for the outlier data.Then simulated experiments were carried out on the established IGA-LSTM model.First,IGA-LSTM and GA-LSTM and LSTM prediction comparison experiments proved the optimization effect of IGA on LSTM.Then there is a comparative test between IGA-LSTM and GA-BP model and PSO-WNN,which proves that the IGA-LSTM model has excellent comprehensive performance in traffic flow prediction.