Forecasting Solar Irradiation on Solar Tubes Using the LSTM Method and Exponential Smoothing

Abstract

Sunlight is an alternative energy source that can be used as a substitute for fossil fuels. Renewable energy potential has not been widely utilized, especially in Indonesia. Utilization of sunlight, one of which is done indoors to save electricity and the source is not limited. This study aims to predict solar irradiance to determine the value of sunlight intensity in an area as the main source of the utilization of renewable electrical energy through the solar tube system with the LSTM method. This low-cost system offers a renewable way and considers the potential for solar radiation as an energy-efficient alternative based on the intensity of light captured by the solar tube. This research uses two methods. The LSTM method is a recurrent neural network forecasting technique that can study deeply and extract temporal relationships in data because of its large architecture. The exponential smoothing method is part of the time series forecasting technique and is used when the dataset has no cyclic variance and trend. Data collection was carried out in sunny conditions because it represents a stable condition in sunlight. The results obtained from the two methods are evaluated with RMSE and MAE values to choose the optimal approach. Due to lower RMSE and MAE values in this comparison, LSTM performs better than Multiple Repeat and Exponential Smoothing in terms of performance.