FORECASTING SOLAR ELECTRICITY GENERATION POTENTIAL ON 18 MAJOR RESERVOIR SURFACES IN SRI LANKA USING LONG SHORT-TERM MEMORY (LSTM) MODEL

Abstract

Sri Lanka’s energy sector remains heavily reliant on fossil fuels, with coal and oil generating about 50% of total electricity as of 2023. Despite significant solar irradiance potential, solar energy contributes just 5% to the national grid. A major constraint to large-scale solar expansion is land-use conflict with agriculture, urban development, and ecological conservation. Floating Solar Photovoltaic (FPV) systems, deployed on reservoir surfaces, offer a sustainable alternative by utilizing underutilized water bodies without competing for land. This study focuses on forecasting the monthly solar energy potential across 18 major reservoirs in Sri Lanka, considering a 1 m² panel area for each site, using a Long Short-Term Memory (LSTM) model. The results of this study will help assess the feasibility of deploying FPV systems. Solar energy generation potential was derived by adjusting measured irradiance for environmental factors such as cloud cover, precipitation, atmospheric pressure, and surface reflectivity, combined with panel efficiency based on a 1 m² panel area. A separate LSTM model was trained for each site using a 12-month input sequence to predict the subsequent month’s solar energy output. Data were standardized and split into training (80%) and testing (20%) subsets. Each model incorporated dropout and early stopping to mitigate overfitting, and performance was evaluated using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). Historical data revealed strong seasonality, with average daily outputs ranging from 1–4 kWh/m², and peaks exceeding 5 kWh/m² in certain months. February to April were generally the most productive months. RMSE ranged from 0.64 to 0.89, with MAPE values below 0.57, indicating relatively accurate forecasts. Site-specific variability influenced model performance. Future studies could explore integrating climate change projections and reservoir-specific operational data to improve long-term forecasting accuracy and support strategic planning for FPV system deployment.