Research On Urban Traffic State Prediction Based On Spatiotemporal Characteristics

As a worldwide problem,urban traffic congestion is a key factor limiting the development of intelligent transportation.The continuous development of the economy has improved the level of urbanization and motorization,and the number of cars has increased year by year,making the limited road capacity unable to carry the rising traffic flow,causing traffic congestion problems.Traffic state prediction is an important part of the intelligent transportation system.Accurate traffic state prediction can assist route planning and guide vehicle scheduling,and is one of the important means to alleviate traffic congestion.Based on the high-dimensional and sparse characteristics of traffic data,this paper combines the matrix factorization algorithm and ensemble learning algorithm to achieve accurate prediction of traffic status.This paper mainly completes the following work:(1)Analysis of the spatiotemporal characteristics of the traffic state.In the time dimension,the time distribution of traffic congestion is analyzed,and the analysis is carried out in detail from the perspectives of the same day in different weeks and different days in the same week;in the spatial dimension,this paper divides the spatial relationship into four categories according to the number of upstream and downstream sections.They are one-in-one-out,one-in-multiple-out,multiple-in-one-out,and multiple-in-multipleout,and typical road sections are selected for each spatial structure for analysis and description.(2)Construction of traffic state prediction model.According to the matrix decomposition algorithm,the traffic state matrix is decomposed from the two dimensions of time and space.The time dimension uses the periodic factor method to predict the time hidden factor in the future,and the spatial dimension uses the Louvain community partition algorithm and the K-means clustering algorithm to obtain the spatio connect.Due to the implicit feedback problem of matrix decomposition,the integrated learning algorithm Catboost is integrated to obtain the final traffic state prediction result.(3)Based on the abstract road network data in Xi’an,the traffic state prediction experiment is carried out.In this paper,the weighted F1 Score is used as the model evaluation index.The model test results show that the weighted F1 Score is 0.89,and the model effect is good.This paper selects the morning peak hours,evening peak hours,low peak hours and typical congested road sections to analyze the results,and proves that the model performs well in each time period and each road section.Finally,this paper selects the local road network in the experimental data for empirical research.It uses the model proposed in this paper to predict the traffic status of key road sections in the road network,and proposes a reasonable traffic diversion strategy and the best driving route suggestion,which alleviates local traffic congestion and helps the development of smart transportation.