Research And Prediction On The Traffic Based On Multi-Model Stacking

The soaring number of cars brought from urbanization has caused serious social problems such as traffic congestion.The congestion problem of highway toll stations is an urgent problem to be solved,because the traffic volume gathered at the toll stations during peak or peak roads often exceeds the carrying capacity of the transportation system and toll stations.This problem can be solved based on traffic flow to develop a response plan to ease traffic,but this needs the ability to accurately predict future traffic flows.However,traffic flow patterns vary with different random factors,such as weather conditions,holidays,time of day,and so on.Predicting future traffic flows is a well-known problem.This dissertation focuses on predicting short-term traffic flow through multimodel fusion.First,the temporal and spatial characteristics of traffic flow are analyzed and several classic methods for predicting traffic flow are introduced in this dissertation.Next,the time series model that will be used is also explained.Then the integrated learning and decision trees are illustrated,including random forest and GBDT,and explains its principle and steps,and also proposes some improvement methods.Finally,the simulation prediction,after introducing the concept of evaluation criteria and model fusion and using the mean absolute error(MAE)as the evaluation criteria,three models for prediction are used in this dissertation.The first model is a time series model,but the prediction of a time series model is prone to extreme values.The second and third are random forests and GBDT.If model fusion is not applied,the MAE of each individually trained model is high,and in the visualization it can be seen that each model has its own shortcomings.The first step of model fusion is to take the average of the random forest and GBDT results as the output of the decision tree.The second step is to linearly weight the output of the decision tree and the results of the time series model to obtain the final result.After calculating the MAE and comparing with the previous model,the final result has the highest precision,which verifies the effectiveness of the proposed method.