Prediction of electricity sales using neural based inverse distance weighting method

Abstract

Prediction of electricity sales becomes important for State Electricity Company of Indonesia (PLN) in order to estimate the Statement of Profit and Loss in next year. To obtain good predictive results may require many variables and data availability. There are many available methods that do not require so many variables to get predicted results with a good approximation. The aim of this study was to predict electricity sales by using an interpolation method called IDW (Inverse Distance Weighting). Several data samples are mapped into Cartesian coordinates. The data samples used are power connected to the household (X-axis), to industry (Y-axis), and electricity sales (Z value). Firstly, the sampled data clustered by using SOM algorithm. The Z value in each cluster is predicted by using the IDW method. The prediction results of IDW method are then optimized using ANN-BP (Artificial Neural Network Back Propagation). The trained net structure is then used to predict the electricity sale in next year.