Abstract:

To address the limited accuracy of single-model forecasting and challenges faced by combined models in handling abnormal data fluctuations，this study proposes a novel forecasting method integrating mutation point correction into a pruned exact linear time （PELT）-grey prediction model（GM）-seasonal autoregressive integrated moving average （SARIMA） combined model. This method initially employs the PELT algorithm to detect fluctuations in freight turnover data and identify change points. The Grey GM（1，1） model is then used to correct anomalies at these change points，enabling the dataset to better meet the stationarity and randomness requirements for the SARIMA model. Finally，based on the optimized dataset，the SARIMA model is used to perform predictions on the refined data. Using freight turnover data from Beijing as a case study，comparative analysis of different hybrid models reveals that the proposed model exhibits superior performance than other combined models，with significant reductions in mean squared error and mean absolute error and a coefficient of determination close to 1. The PELT-GM-SARIMA model is structurally simple and can better adapt to time-series data with missing values or frequent anomalies，resulting in more accurate predictions. This study presents a more effective approach for traffic predictions in highway transportation planning and investment decision making.

Key words: transportation economics, PELT-GM-SARIMA model, freight turnover, change-point correction, transportation planning