ELECTRICITY PRODUCTION PREDICTION OF MICRO-HYDRO POWER PLANT (MHPP) USING MULTILAYER LSTM, LSTM SEQUENCE TO SEQUENCE, AND LINEAR REGRESSION MODELS

• Main Author: Ading Wasana, Sandhi
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/87578
• Institution: Institut Teknologi Bandung

In the energy transition era, a transformation is needed to integrate clean electricity generation derived from renewable energy sources (RES). However, the development of electricity generation based on RES faces challenges, one of which is the fluctuating electricity production caused by various factors. Therefore, electricity production prediction is crucial to estimate the amount of electricity that can be generated and integrated into the main system. This research aims to develop an electricity production prediction model using various methods. The dataset utilized includes electricity production data from the past five years and external data such as rainfall. During the pre-processing stage, production data was processed to reduce null values and noise in the dataset using Wavelet Transform (WT). Subsequently, the data was processed using three methods: Linear Regression, LSTM, and Multilayer LSTM, to identify the best model. Evaluation metrics used include RMSE, R², and MAE. The study results indicate that combining historical production datasets with rainfall data, processed using WT and the Multilayer LSTM method, yielded metric scores with RMSE of 0.178, R² of 0.973, and MAE of 0.079. These scores are significantly better than models without rainfall data and other methods such as LSTM Seq2Seq and Linear Regression. The primary contribution of this research is the demonstration that combining historical production datasets and rainfall data can improve the accuracy of prediction models. Additionally, noise reduction through WT has been proven to significantly enhance accuracy, especially in the Multilayer LSTM model. The accuracy improvement is evident in the two case study datasets: Lubuk Gadang and Sangir Hulu power plants. These findings are expected to serve as a reference for developing electricity production prediction models based on RES to support the integration of clean energy into the main system.