Analysis of the Best Neural Network Configuration for Predicting Household Customer Kwh Sales in Banda Aceh City

Abstract

Energy consumption (kWh) is critical to the operation of electrical systems. Predictive modeling optimizes energy usage, increasing power system efficiency. This study created an artificial neural network (ANN) architecture to estimate energy consumption (kWh) for home users in Banda Aceh. The ANN topology consisted of 5 input layers, 5-25 hidden layers, and one output layer. This study used two scenarios: first, the ANN topology was trained using the logsig activation function, and then the tansig activation function was used for training. Based on training simulations, the ANN architecture with 5 input layers, 5 hidden layers, and 1 output layer has the lowest Mean Squared Error (MSE) of 0.00035. The next phase involved testing this ANN topology. The next stage is to analyze the ANN architecture with 5 input layers, 5 hidden layers, and 1 output layer using the testing technique. Based on the testing technique, the ANN architecture with 5 input layers, 5 hidden layers, and 1 output layer had a MAPE value of 3.34%.