Research On Road Traffic State Recognition And Prediction Based On Machine Learning

Nowadays,traffic congestion has always been a major “stubborn disease” in the development of large and medium cities both at home and abroad.Excessive congestion of urban roads not only reduces commuting efficiency,but also causes an immeasurable negative impact on social and economic development,residents’ living quality,and regional ecological environment.It seriously restricts the healthy and sustainable development of urban traffic.The Intelligent Transportation System(ITS)provides a feasible path for the governance of urban traffic congestion.Among them,accurate and real-time traffic state recognition and prediction is the prerequisite for ITS to realize effective traffic control and guidance,so as to achieve the purpose of alleviating urban traffic congestion.This paper starts from solving the actual traffic congestion problem,takes traffic state recognition and prediction as the research object,uses GPS trajectory data as the research foundation,summarizes relevant research at home and abroad.Then,based on the use of machine learning algorithms,this paper establishes a traffic state recognition and prediction model to improve the effect of traffic state recognition and prediction.The main content includes the following points:(1)It preprocesses the GPS trajectory data and proposes solutions to various problems that may occur in the trajectory data.The geometric map matching algorithm was used to match the processed data with the help of the FME platform.According to the specific calculation method of traffic flow parameters,the average instantaneous speed and average travel speed of the specified section were extracted based on GPS data.(2)Combined with the characteristics of traffic state classification,a traffic state recognition model based on Fuzzy C-means clustering(FCM)and K-nearest Neighbor(KNN)algorithm was established by taking the average travel speed and speed variation coefficient as characteristic parameters.Firstly,the FCM model is used to divide the historical traffic data into four states: congested,slow,smooth and unimpeded,then the model classifies each sample into one of the four categories.On this basis,the KNN model is used to carry out traffic state recognition on the new input data samples with Euclidean distance as the measurement standard.The experimental results show that the accuracy of the KNN model is up to 99.0%.(3)Based on the analysis of traffic flow characteristics and traffic prediction methods,a traffic flow parameter prediction model based on Discrete Wavelet Transform(DWT)combined with Gated Circulation Unit(GRU)and Back Propagation Neural Network(BPNN)was constructed,and the prediction results were put into the KNN model to realize indirect traffic state prediction.The accuracy of the combined model is verified by predicting the three road sections with large differences in traffic patterns and comparing them with ARMA,LSTM,GRU,SAEs,and other models;further,the model is used to predict the traffic state of the remaining 80 roads in the road network,which again verifies the robustness and portability of the model.Finally,the above prediction results are put into the KNN model to predict the traffic state of the road section,and the prediction accuracy is as high as 90%,which verifies the feasibility of using the indirect prediction method to predict the traffic state.