Short-Term Load Forecasting for Indian Railways Using Machine Learning Models

Railway load forecasting is vital for the efficient operation and management of railway systems, but it poses significant challenges due to its scale, diversity, and complexity. While many machine learning models exist for load forecasting, there is a critical need for an accurate model capable of handling the inherent uncertainty in railway loads. To address this, we propose an accurate model that divides the dataset into three subsets with varying time gaps to capture the dataset error effectively. Each subset is trained using three different techniques Prophet, Triple Exponential Smoothing, and SARIMA. The outputs from these techniques are then fed into a dataset reformation to ensure forecasted values are in equal intervals. Finally, the reformed data is given as input to a Linear Regression model to refine the forecast further and reduce forecasting errors. The proposed model demonstrates superior performance compared to existing techniques, achieving reductions in MAPE and RMSE for each state. The MAPE obtained using the proposed model for UP(ISTS), Delhi, Punjab, UP(STU), and Maharashtra were 5.83, 15.72, 9.50, 3.91, and 3.62, respectively, and the RMSE in MW were 9.93, 3.21, 4.87, 16.82, and 17.97, respectively. The Proposed model is adept at handling uncertain load patterns and dataset errors, making it a fitting solution for accurate short-term load forecasting in Indian Railways.