MODELING OF THE SOLAR POWER OUTPUT FORECAST SYSTEM FOR HYDERABAD RAILWAY STATION (INDIA) USING TRANSFER LEARNING AND BAYESIAN REGRESSION

Abstract

With the rapid adoption of photovoltaic systems on the railways and their integration into the electricity grid, it has become a necessity to accurately forecast the photovoltaic output at their intended site of use for effective energy management to mitigate the instability of the grid caused by the intermittency of solar power. Inadequate data for the newly installed solar power plants is a severe bottleneck in forecasting and energy management. The research aims to overcome this by modelling a solar power forecast system using transfer learning from a pretrained Bayesian regression model capable of forecasting solar irradiance at the same location. The weather data from the data access viewer of NASA power website was used in the study. The predicted results were compared with the actual solar power output data from the datalogger of the Hyderabad, India Railway Station solar power plant. The resultant root-mean-square-error obtained is 96.50, and the mean absolute percentage error is 11.23%. The proposed method showed superior results compared to XGBoost and linear regression.