| Summary: | Short-term prediction of dynamic turning movement proportions at intersections is very important for intelligent transportation systems, but it is impossible to detect turning flows directly through current traffic surveillance devices. Existing prediction models have proved to be rather accurate in general, but not precise enough during every time interval, and can only obtain the one-step prediction. This paper first presents a Bayesian combined model to forecast the entering and exiting flows at intersections, by integrating a nonlinear regression, a moving average, and an autoregressive model. Based on the forecasted traffic flows, this paper further develops an accurate backpropagation neural network model and an efficient Kalman filtering model to predict the dynamic turning movement proportions. Using Bayesian method with both historical information and currently prediction results for error adjustment, this paper finally integrates both the above two prediction models and proposes a Bi-Bayesian combined framework to achieve both one-step and two-step predictions. A case study is implemented based on practical survey data, which are collected at an intersection in Beijing city, including both historical and current data. The reported prediction results indicate that the Bi-Bayesian combined model is rather accurate and stable for on-line applications.
|
|---|
