MEDIUM-TERM POWER LOAD FORECASTING FOR PLN P2B JAVA BALI USING ARTIFICIAL NEURAL NETWORK AND SARIMAX MODEL

• Main Author: Hardono Hutama - NIM: 13514031 , Andri
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/25555
• Institution: Institut Teknologi Bandung

The problem studied in this Final Assignment is daily maximum load forecasting for PLN P2B Java Bali by using historical maximum, minimum, and average load. This Final Assignment intends to find appropriate models for prediction and develop supporting application based on those models. Besides historical data, day of the week, type of holiday, and separation based on area are used to support the prediction. From techniques which are commonly used for prediction, artificial neural network (ANN) and SARIMAX model is chosen. An experiment with both techniques is conducted to find appropriate configurations so that the resulting models can model power load well. Those configurations will be evaluated and compared with each other and with the PLN performance target. Both models are unable to reach MAPE of 2% which is targeted by PLN. Based on train data, the result of SARIMAX which have MAPE of 2.4% is more accurate compared to the result of ANN which have MAPE of 2.7%. Evaluation in terms of model building speed and standard deviation of errors show that SARIMAX model performs better. After evaluation is conducted, an application which can build similar models for prediction is developed. This application plays a role to ease prediction making. The ease in prediction making is ensured by testing the functional requirements of the application. It is concluded that ANN and SARIMAX model can be used to predict power load even though the MAPE of both proposed models do not reach the targeted MAPE. An application is also successfully developed to facilitate prediction making so that it is easier. <br />