Day-ahead Solar Power Generation Forecasting using LSTM and Random Forest Methods for North Eastern Region of India

A tremendous increase in the renewable energy capacity addition worldwide has been observed in the recent times. India is also aiming to make renewables, the major source of power to meet its ever-increasing power demand. At present, while the total solar installed capacity of the country is around 58 GW with a daily average generation of around 200 MU, the North Eastern Region (NER) of the country has only 204 MW of solar power installed capacity as in the year 2022. Assam is one of the leading states in NER for installing solar photovoltaic (PV) independent solar parks connected to high voltage systems with an installed capacity of 199 MW. There is a huge potential in NER for adding more solar power capacity in the future. In this paper, Block-wise (15-minutes) Day ahead Solar Power generation is forecasted for Rowta PV plant in Assam using Long Short Term Memory (LSTM) and Random Forest (RF) methods for a week considering the available weather data of NER. The objective of this paper is to provide a reliable method of forecasting for System Operators and Generating utilities that can ease the process of day ahead solar power forecasting. In terms of overall forecasting error, LSTM emerges as the better forecasting technique compared to Random Forest method. The approach is validated through SCADA data from Rowta PV plant. The performance of LTSM method is further compared with NARX with exogenous input method, another popular ANN method in the field of forecasting. Forecast results by these two methods for a day have also been presented in this paper.