Comparative Analysis of LSTM-Based PV Power Forecasting Models with Climate-Adaptive Feature Selection in Abuja, Nigeria

Abstract

In this research, we analyse how Long Short-Term Memory (LSTM) models can predict photovoltaic (PV) power output, in Abuja, Nigeria by selecting specific climate features and model configurations. The rising energy needs due to population growth and urbanisation emphasise the importance of sustainable energy sources. This study aims to improve the accuracy of PV power forecasts for integrating power into the current electrical grid and enhancing energy management strategies. By analysing data from the ERA5 dataset that includes various climatic features, we rigorously trained and assessed the LSTM models. Our results indicate that specific window sizes and combinations of features notably enhance forecasting accuracy with a window size of 6 and a mix of meteorological and solar radiation features showing the performance metrics (MAE, RMSE, R²). The study also underscores the significance of autocorrelation and cross-correlation analyses in optimizing model setups. Our findings suggest that LSTM models can accurately predict PV power output offering insights for maximizing energy usage in urban areas with similar climates. This research contributes to efforts aimed at reducing reliance on fossil fuels and promoting sustainable energy solutions. Future endeavours will explore integrating real-time data and incorporating additional climatic features to further refine forecasting models.