Blended Temperature Forecasting Model for Thailand Using Multiple Data Sources

Abstract

With the escalation in the cost of electricity, there has been a noticeable inclination towards the installation of solar photovoltaic (PV) systems in multiple regions across Thailand. The increase in PV installations has led to an electricity demand that fluctuates depending on the prevailing weather conditions, creating challenges in managing and regulating electricity demand. In order to support electricity regulators in managing the fluctuations, it is crucial to implement a solar power forecasting system for individual households. One of the critical variables in forecasting solar power generation, besides solar irradiance, is temperature. This study introduces a temperature prediction system for every geographic location in Thailand at a 10x magnification level, which provided an hourly temperature for each location in the country. The proposed model integrated input data from three open-source platforms, namely Meteostat, Weatherapi, and IBM Weather. Utilizing the capabilities of each input source, the deep learning model was employed. The system, powered by the proposed model, achieved a Mean Squared Error (MSE) of 1.17 °C when compared to the actual data acquired from the Meteorological Department of Thailand.