Research On Short-Term Traffic Flow Prediction And Route Recommendation

As the important element of city progress, traffic is not only the essentialchannel in hominine distribution and physicaldistribution, but also the ligament of connecting city. Among the research of traffic problems, short-term traffic flow prediction and route recommendation should be closely linked to our daily life. This paper will study and illustrate these two hot traffic issues.Based on the traffic flow data, Short-term traffic flow prediction is to predict the traffic flow about next delta t moment. Traditionally, people forecast the traffic flow by using autoregressive model (AR), moving average model (MA), historical average model (HA) and autoregressive moving average model (ARIMA). These linear prediction models generally use the least squares (LS) to estimate model’s parameters and the calculation is not complicated. However, the model itself failed to reflect the uncertainty and the nonlinearity of traffic flow and it is also unable to overcome the influence of random disturbance. With the prediction interval becomes shorter, the prediction accuracy of these models becomes very poor. In recent years lots of literatures have been concentrating on neural network and truly improved the accuracy of traffic flow prediction but it loses power in complex urban environment. Aim at the deficiency of the existing methods, we put forward a plane moving average algorithm. This new approach assembles information from relevant traffic time series and has the following advantages:(1) it integrates both individual and similar flow patterns in making prediction, (2) the training data set does not need to be large, (3) it has more generalization capabilities in predicting unpredictable and much complex urban traffic flow than previously used methods.In respect of route recommendation, path optimization strategy has always been a hot problem in the study on traffic administration and control system and differs in different application Fields. The commonly used path optimization strategy contains A-star algorithm, Dijkstra algorithm and Behrman-Ford algorithm, etc. These algorithms focus on independent travel time vector rather than the correlation between adjacent roads, so the rationality of its final recommendation is questionable.According to these problems, we present a path bundling model. This model is innovative and uses the vehicle traffic characteristics and different travel time among the adjacent roads to compute the rational recommended routes.To assess our approaches, we have performed extensive experiments on a real data set, and the results give evidence of its superiority and rationality over existing methods and have practicalsignificance.