A new approach is proposed by integrating time series model with nonparametric approach to predict short-term congested freeway traffic flows. The key difference between the paper and the previous researches is that the proposed model can dynamically adapt to different traffic scenarios with a new introduced variable z(sub t) to represent the interactions of exogenous unquantifiable or non-easily quantifiable impact factors to the traffic flow pattern. Two variants, a single-step prediction model and multi-step model of the proposed approach are studied and compared. The application results of these two models to a real world traffic network showed that the multi-step model is better than the single-step model. Furthermore, the built-in B+ tree structure of this approach improves the computation and searching speed, making the future real time application possible. For the covering abstract see ITRD E134653.
Abstract