OPTIMIZATION OF ELECTRICITY LOAD FORECASTING USING DEEP LEARNING FOR POWER SYSTEM OPERATION PLANNING AT PT PLN (PERSERO) UP3B PAPUA

• Main Author: Putra Prasatya, Dwi
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/86674
• Institution: Institut Teknologi Bandung

To meet the increasing electricity demand in Indonesia, PT PLN (Persero) has implemented key measures based on the Minister of Energy and Mineral Resources Regulation No. 20 of 2020 concerning the Electric Power System Grid Code. One of the Grid Codes is the Scheduled and Dispatch Code (SDC), which ensures the readiness and efficiency of the system. Accurate load forecasting, which is a critical component of the SDC, remains a challenge due to the limitations of traditional methods that rely on growth percentage from previous periods. This study utilizes deep learning models, namely Multi-Layer Perceptron (MLP) and Long Short-Term Memory (LSTM), to improve forecasting accuracy for the Jayapura and Sorong systems. SCADA data from 2020-2024 is used, with hyperparameter tuning through grid search. Model evaluation shows that LSTM outperforms MLP, where in the Jayapura system, LSTM achieves MAE 1.0519, RMSE 1.6528, and MAPE 0.0135, slightly better than MLP with MAE 1.0597, RMSE 1.6804, and MAPE 0.0136. In the Sorong system, LSTM also performs better with MAE 1.3782, RMSE 1.8657, and MAPE 0.0333, compared to MLP with MAE 1.6192, RMSE 2.1661, and MAPE 0.0390. These results indicate that LSTM provides more accurate predictions for load planning. LSTM also reduces forecasting errors compared to manual methods, with error reduction of 70.6% in Jayapura and 53.5% in Sorong. These findings highlight the potential of LSTM in enhancing accuracy and operational planning efficiency for a more reliable and economical power system.