A Novel Electric Load Demand Forecaster Using Taguchi’s Method and Artificial Neural Network

The use of Artificial Neural Network (ANN) for electric load forecasting has been proposed in many studies. Among these studies, the daily peak load or total load with weather consideration was mostly predicted in order to dispatch high-quality electricity or assess electric load efficiently for power utilities. However, load demand forecasting from the standpoint of consumers is seldom discussed. With the global market competition, enterprises invest in instruments to cut down on large electricity bills of operating costs. A formal study shows that the regular ANN training model was inadequate to deal with volatile load patterns, especially in Very Short-Term Electric Demand Forecasting (VSTEDF). In this paper, we present Taguchi’s and rolling training methods of ANN for VSTEDF. By using this proposed rolling training model, the electric load demand is predicted precisely every 2 minutes. The forecasting error is smaller than 3%. Compared with the conventional ANN model and Grey model, the proposed Taguchi-ANN-based predictor has better accuracy in the application of VSTEDF. The improved Taguchi-ANN-based electricity demand forecaster in conjunction with the PC-based electricity demand-control system is a cost-effective and efficient means to manage the usage of electricity.

Keywords: Artificial neural networks (ANN), electric load forecasting.