”ANALISIS MODEL RECURRENT NEURAL NETWORK UNTUK MEMPREDIKSI FLUKS RADIASI MATAHARI DI NUSA TENGGARA TIMUR DAN DKI JAKARTA

• 主要作者: Delvin, Rachel
• 格式: Final Project
• 語言: Indonesia
• 主題: Astronomi
• 在線閱讀: https://digilib.itb.ac.id/gdl/view/60914
• 機構: Institut Teknologi Bandung

Indonesia has a high potential in utilizing solar irradiance as its source of electrical energy by installing photovoltaic cell (PV). The operation of solar powered electrical system is hard since solar irradiance can't be controlled hence the prediction of it is necessary. This research proposed the use of Recurrent Neural Network (RNN), an Artificial Neural Network which created specifically to handle sequential data, to complete the task of predicting one hour-ahead solar irradiance. The goal of this research is to analyze two types of RNN models, Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRU) and find models with the best performance at predicting solar irradiance in two location, DKI Jakarta dan East Nusa Tenggara. DKI Jakarta has the potential of producing solar powered electricity using PV as much as 33 TWh/year and East Nusa Tenggara as much as 1,025 TWh/year. Models learn from 9 input parameters which include meteorological data, geographic, and time information through regression. Dataset used in this research contains data from 25 location points from both provinces ranging from 2015-2020 obtained from National Solar Radiation Database (NSRDB). The best RNN model built from this research for the DKI Jakarta test dataset is GRU with RMSE, MAE, dan R2 respectively equal to 138.51 W/m2, 95.36 W/m2, and 0.79. On the other hand, the best RNN model for the East Nusa Tenggara test dataset is LSTM with RMSE, MAE, dan R2 respectively equal to 130.88 W/m2, 86.65 W/m2, and 0.84. The computational time for each of the models cost less than 50 minutes.