Short-term load forecasting via artificial neural network

Abstract

Load forecasting is a one of important element in Operation and Planning Division to predict behavior of load in future. There are several ways to forecast the load demand which can be categorized into two which are classical approach and artificial intelligence. Artificial intelligence has shown ability to solve with this nonlinear characteristics and do not require any complex mathematical formulation. One of technique in artificial intelligence is Neural Network. This project focused on the neural network as a tool to project the electricity demand in the future. Before forecasting, first neural network must have to be designed by determine input, hidden layer and output numbers. The effectiveness of the neural network as a load forecasting is simulated using MATLAB Neural Network toolbox. In the neural network, back propagation technique is used as a learning algorithm and compares the final value with actual data by using Mean Absolute Percentage Error (MAPE) as percentage error between the output and actual value. Load forecasting using neural network has been tested using three models. Each of these models has different architecture in order to test the performance of neural network. The result shows that the load forecasting can be done by using neural network and MAPE of the models had achieved below than 10%.