A Dynamic Combination Forecast Model for Analysis Transport Volume Time Series

A dynamic combined forecasting model for transport freight volume time series prediction is established. The time-varying combined weights are computed with the Bayesian posterior probability based on each local predictor´s performance. This method´s forecast performance is reliable, because it tracks the real-time prediction precision of the combined models and adjusts their credit values (weights) according to their past predictive error. The forgetting factor is proposed as a threshold in order to avoid the singular forecasting model´s performance change so intensely over different time intervals as to cause unimaginable effect to the latter online weights computation. In error evaluation system, the performance of the proposed dynamic combination forecast model outperforms the singular predictor used respectively as well as some conventional combination forecasting methods.