SHOR T-TERM LOAD FORECASTING FOR INDUSTRIAL CUSTOMERS USING FASART AND FASBACK NEURO-FUZZY SYSTEMS

This paper studies the problem of Short-Term Load Forecasting (STLF) for industrial customers. Since they have a large impact on power consumption and a par- ticular load demand, an accurate forecast is specially impor- tant. For this task we study the application of two neuro- fuzzy systems, FasArt and FasBack, in addition to other techniques such as Multilayer Perceptron (MLP) with the backpropagation (BP) learning algorithm, as well as stan- dard statistical Autoregressive Integrated Moving Average (ARIMA) processes. The experimental study is performed using real data provided bya majorSpanish company. While the most accurate predictions are achieved similarly with FasBack and MLP, the former features easy knowledge ex- traction and on-line learning capabilities that make FasBack a better choice.