Gross Ton-Kilometer forecasting models for freight trains of Northern-Central Indian railways

Indian Railways (IR) earns about 70% of its revenue from freight traffic. The goods loading is highly erratic in nature which creates the problem of inaccurate forecasting of the electric power required. Thus, IR requires precise forecasting models to deal with this problem. In this paper, Gross Ton KiloMeter (GTKM) earned by freight traffic is forecasted using traditional statistical models, i.e., exponential smoothing, moving averages, weighted moving averages, trend corrected exponential smoothing (Holt's method) and Holt-Winter's (HW) approach, with proper experimentation. Real time GTKM data of a common type of goods train, viz., BCN, BOXN, etc is considered for experimentation. The data is observed to be highly nonlinear in nature. Proposed HW model provides impressive forecasting results with a satisfactory figure of MAPE as 8.63%.