Electrical Load Forecasting Study Using Artificial Neural Network Method for Minimizing Blackout

Abstract

Bantul Regency is one of the areas where the demand for electricity consumption in every year increases gradually, this is because Bantul is a developing regency with a number of natural attractions. So that every year will experience the growth of development followed by an increase in the need for electrical energy. With this issue, it is necessary to forecast the number of consumers and the electricity demand so that the providers of electrical energy PT. PLN (Persero) can provide as needed and hopefully can minimize blackout. In doing forecasting, the method used is Artificial Neural Network (ANN) that is run with Backpropagation. The advantage of this method are the convenience in formulating the experience and knowledge of forecasters, and very flexible in changing the rules of forecasters. The Levenberg-Marquardt training algorithm, the Graident Descent Variable Learning Rate and Quasi Newton are used. The most accurate results are seen by looking at the smallest average error percentage rate generated on all three training algorithms. So it is possible to do things related to the prediction and forecasting. The results of training with trainlm, traingdx and trainbfg show that the resulting error is small enough that is 0.194%, 0.15%, and 0.14%. From the results of the training shows that Artificial Neural Network (ANN) is good to be applied into prediction or forecasting.