City Road Travel Time Prediction Based On Map Matching And Machine Learning

Intelligent Transportation System(ITS)is one of the effective ways to alleviate the pressure on urban road traffic.It has gradually become an important link in the development of smart cities.Travel time prediction is one of the key information for ITS research.It can not only provide pedestrians with a variety of A variety of travel options can also provide a certain theoretical basis for traffic management and traffic guidance.At the same time,it can also reflect the effect of urban road network accessibility and connectivity from the side.In view of the challenges and difficulties faced by the current travel time prediction of urban road networks,based on the collected floating car data,this paper does a corresponding research on the travel time prediction of urban road vehicles.The main research contents are as follows:(1)Analyze the temporal and spatial distribution characteristics of the floating car GPS basic data set,add expanded data such as congestion,weather,and temperature on the basis of the original basic data set,and initially form an expanded data set.Aiming at the problem of offset data and null values in the data set,box plots and Lagrangian interpolation are used to identify outliers and missing values in the data set,which improves the quality of the GPS basic data set and extended data set.(2)Hidden Markov model based on trajectory segmentation to match the collected GPS point data with the actual road network.In the matching process,the longest common sub-sequence method is used to determine the distance,length and direction of the GPS sub-trajectory and the road segment sequence,and then determine the degree of similarity between the two,so as to match the GPS trajectory data with the electronic map of Beijing.Based on the matched GPS points and the actual road network,the vehicle trajectory data is mined to obtain relevant characteristic data such as the number of turns,the number of crossings,road traffic,space occupancy,road density,etc.,thereby obtaining map-based data The matched extended feature data set.(3)In the study of travel time prediction in city road,a fusion model based on light GBM and XGBoost algorithms is proposed.The two models were used to screen the features of the models respectively,and the feature data set was divided into key features and common features by sorting the importance of each feature in the two models.According to the predictive power of the two models(the deviation between the model’s predicted value and the true value),the reciprocal error method is used to assign weights to the optimal single-model prediction solutions of the Light GBM and XGBoost models to achieve linear weighted fusion of the models.(4)Use the excavated and expanded road travel time data set to test and analyze the established fusion model.Divide all the features and key features in the feature dataset into two cases as the input of the light GBM and XGBoos models.The experimental results show that when the full features are used as input,the prediction effect of the model is not optimal,and only the key features are used as the model.When input,the MAPE of the fusion model is 3.1%,and the optimization effect is more significant.