Forecasting Global Horizontal Irradiance to Evaluate Battery Capacity under PV Uncertainty in a Rural Area of Bangladesh

In today's competitive energy market, renewable energy (RE) sources have the disadvantage of difficulty in precise forecasting and dispatching their output. To mitigate this uncertainty, this study presents a comparative analysis between forecasting models for Global Horizontal Irradiance (GHI) which is directly related to solar power generation and a battery capacity management system from the perspective of Bangladesh's rural area. The proposed system consists of a forecasting model for GHI prediction to estimate the photovoltaic (PV) electricity generation and a battery management system to store and distribute the generated power. The forecasted GHI can be used to estimate the PV electricity generation which can later be used to estimate the capacity of the battery energy storage system. A deep neural network model Long Short-Term Memory (LSTM) and an Artificial Neural Network (ANN) model were implemented to forecast the GHI. On a dataset built with the data for any rural area of Bangladesh, this is the first application of deep learning-based models to predict the GHI to the best of the authors' knowledge. The results show that the ANN model that was implemented has a 15.6% better RMSE and 20.86% MAE score on the test data than the LSTM model and has better accuracy. The estimation of the power generation was used to model a small-scale battery management system consisting of PV integration, distribution system, and loads. The battery management system is designed such that the power can be supplied to the main grid and also the abundant power can be stored for use later. The battery management system is designed in the MATLAB/Simulink environment.