Solar Power Prediction in IoT Devices using Environmental and Location Factors

Abstract

Energy-harvesting IoT nodes need to conserve their energy to remain operating without interrupting. By predicting input power supply, IoT nodes could appropriately schedule or adjust data transmission interval to match available energy for lasting operations. In this work, we explore the effectiveness of using environmental and location factors, including light intensity, temperature, humidity, facing directions of a solar panel, as well as historical input power data to help predicting the solar input power of IoT nodes. Various time series and machine learning models including EWMA, WCMA, SARIMAX, and LSTM are fitted, tuned, and compared to determine significant factors and best-performing model. Our results reveal that the facing direction has a significant impact on the input power generated and model hyperparameters. Among the models investigated, SARIMAX yields the lowest prediction errors around 11% - 26%.