Reasearch And Implemention Of Traffic Flow Forecasting Model Based On Machine Learning

With the rapid development of the economics and the continuous improvement of the level of urbanization,people's quality of life has been improved,however,it has also brought serious traffic congestion.How to make rapid and accurate forecasting of the traffic flow with the historic traffic flow on the city section is a major research topic in the field of intelligent traffic.The traditional methods of dealing with traffic flow prediction can be divided into mathematical model-based methods(such as Kalman filter model,time series model,etc.)and mathematical model prediction methods(such as neural network model Parameter regression model,etc.).However,traditional methods have shown some limitations in dealing with the increasing complexity of traffic flow data,which is mainly manifested as follows:(1)Lots of algorithms show the limitations in coping with nonlinear problems(2)The non-stationary characteristics of the traffic flow greatly affect the accuracy of the model(3)the challenge of efficiency for a large number of samples.In recent years,with the popularity of data mining,machine learning and other data-oriented technology,traffic flow prediction is increasingly combined with above algorithms,which leads to a significant improvement in prediction accuracy.The Pems data of the California Department of Transportation are used as experimental data in this paper.According to the non-stationary characteristics of traffic flow,this paper proposes a two stage ordered clustering model which combines with DBSCAN algorithm and the optimal segmentation algorithm,to achieve clustering for ordered sample with less cost in the absence of prior knowledge.Based on the ordered clustering model,this paper proposes time-segmented support vector machine model.Based on the goodness of fit as an index,this model is proved that it can achieve the ideal regression accuracy.The paper also proposes a historic data weighting model for generating traffic flow sequence.The model utilize time-segmented support vector machine model to generate reference value,which will be weighted with historic data.Through iterating the above process,traffic flow sequence is generated and is proved to achieve high accuracy compared with real sequence.Finally,label propagation algorithm is introduced to divided all the sampling time in the experimental data into three kinds of modes of transportation(rising point,drop point,stationary point),according to the characteristic of each sampling time.On the basis of the classification results,random forest model is introduced to recognize corresponding traffic modes for real-time traffic flow sequence.This model has achieved the desired results in the main performance indicators.