Short-term load forecasting: Multi-level wavelet neural networks with holiday corrections

Short-term load forecasting plays a central role in reliable system operation by Independent System Operators and in making prudent bid decisions by market participants. Accurate forecasting is difficult in view of the complicated effects on load by various factors. In addition, it is difficult to forecast holidays as well as the days before and the days after in view of their particular load patterns and very limited data. In this paper, a multi-level wavelet neural network method is developed to forecast tomorrow's load. To effectively forecast the load for holidays as well as the days before and the days after, a correction coefficient scheme with holiday grouping is developed. Numerical results for a simple example and for Midwest-ISO's load demonstrate the effectiveness of multi-level wavelet neural networks, correction coefficients, and holiday grouping.