Short-term Urban Traffic Flow Forecasting

With the risen of the economic strength,the process of urbanization in China is gradually accelerating.And the number of large cities and even megalopolises has risen rapidly,which has put forward many requirements for urban traffic management.Under this background,the research and application of urban traffic intelligence have been effectively developed,and the Intelligent Transport System(ITS)has also been widely used.As one of the basic data in urban traffic,the accurate forecast of urban traffic flow has a significant meaning for ITS.Therefore,the research topic of this paper is the short-term urban traffic flow forecasting.In order to improve the accuracy of traffic flow forecasting,researchers have proposed various methods,which are mainly based on time series analysis algorithms.In these methods,traffic flow forecasting is regarded as a time series analysis problem,and the historical traffic flow data is used to discover the trend of traffic flow.However,these methods still have some drawbacks when applied to complex urban road networks.This is mainly because the urban traffic flow is event-sensitive,especially in the case of some traffic accidents or large-scale activities,the trend of traffic flow will deviate from the common pattern in historical data.To make accurate forecast under these conditions,we need to analyze and model the relationship between the traffic flow on upstream and downstream roads.In this paper,we define the traffic transition between upstream and downstream roads as transfer flow and record the ratio of this transfer flow to the traffic flow on upstream road as the transfer ratio,so as to model the relationship between the traffic flow on upstream and downstream roads.Thus,we can quantify the impact on traffic flow at the time of the incident and use the traffic flow on upstream road to forecast the trend of traffic flow on downstream road.Therefore,we propose a new Combination of Transfer Ratio and Road Similarity(CoTRRS)model in this paper.In order to obtain a priori knowledge of the transfer ratio,we first use the Continuous Bag-of-Words(CBOW)model to extract road similarity between upstream and downstream roads from the vehicle trajectory data.Then,we use a back propagation neural network(BP neural network)as the basic structure of our model to forecast the traffic flow,in which the transfer ratio and road similarity between the upstream and downstream roads are used as the input.Finally,to verify the performance of the CoTRRS model,we have performed a large number of experiments on a real urban traffic dataset.And the empirical study reveals that the CoTRRS model outperforms the baselines.