Short-Term Traffic Flow Prediction Based On Multiple Model Combination

With the development of The Times and the gradual improvement of people’s living standards,more and more people tend to own their own scooters in addition to choosing public transportation when traveling.However,it is difficult to reach the balance between the increasing traffic flow and the limited road resources,which makes the road traffic face a severe test.The traditional traffic flow forecasting technology and method can not provide accurate results for the increasingly complex traffic situation forecasting,which promotes the development of intelligent transportation system.In order to avoid various traffic problems such as road congestion,it is extremely important to master the initiative of traffic control.The purpose is to predict the future traffic flow of the road in advance.In this study,an LGSS model based on ensemble learning and combination of multiple algorithms is designed for short-time traffic flow prediction.Through comparison of experimental results,it is found that other models of this model have higher prediction accuracy.Part of the data from the Performance Measurement System(PEMS)was used in the experiment.Firstly,the data were divided into training set and test set according to the ratio of 4:1,and the data were screened and preprocessed.Secondly,the experiments in this paper,the basic model to determine for how long memory neural network(LSTM),door control unit(GRU helped)circulation,stacked(SAEs)three kinds of traditional models,since the encoder to training and prediction of data,and compare three predicted values and real values,using five kinds of assessment indexes of three kinds of model prediction accuracy analysis.Finally,the three models are combined with the simple cycle unit(SRU)model through the method of ensemble learning to form a new model of short-term traffic flow prediction--LGSS prediction model.The model is used to analyze the original data,and the point-by-point product method is used in the calculation process.This method can carry out parallel operation,and it can also prevent the gradient from disappearing.By comparing the predicted results of several iterations,it is confirmed that this model is more beneficial to the training of gradient algorithm of deep network.The experimental results show that: from the analysis of multiple evaluation indexes,the prediction effect of the combined model of LGSS designed in this study is greatly improved compared with the traditional single model.At the same time,the SRU can be used to carry out parallel operation.When the model iteration 4 times has the highest prediction accuracy,the time of each execution is shortened to 0.7491 s,which reduces the time consumption by 30%compared with the traditional single model,which fully verifies the superiority of this model.