Highly Accurate Short-term Electric Load Forecasting in Consideration of Equalization of Learning Data

点击翻译按钮获取中文摘要

This project deals with the preconditioned intelligent systems and the equalization of learning data. Under competitive and deregulated power systems, short-term load forecasting plays a key role to provide input information with generation scheduling. To compete with other players in power markets, minimizing the maximum error of load forecasting is of main concern. The erroneous results bring about keeping extra power generation reserve in their own company or purchasing more expensive electricity from other companies. As result, power system operators are interested in the reduction of the errors. The preconditioned intelligent system proposed by the author is one of good solutions. By classifying learning data into some clusters, an intelligent system is constructed at each cluster. The method is more effective in terms of model accuracy and computational time. However, it has a drawback that each duster has different performance that comes from underlearning due to the available data. In this study, a method for equalizing the number of learning data is proposed to alleviate underlearning. According to the Kohonen network of artificial neural network, a set of similar data is constructed to reconstruct learning data. In addition, several methods for clustering and the application of the preconditioned intelligent system to fault location in power systems are investigated