Forecasting electricity load with optimized local learning models

Electricity load forecasting plays an important role in the strategy management of electricity power system. Learning methods such as artificial neural networks, and more recently, support vector regression machines (SVR) have been introduced to this field. In practices we often expect a fast forecasting, while standard algorithms based on the whole data set are time consuming. To this end, in this paper we introduce local learning strategy considering only a subset of candidates in the neighborhood of the test point and present a general form for it. Concretely, we consider the combination of KNN and εε-SVR for its powerful generalization ability and simplicity. As for model optimizing, Pattern Search method is used for model selection and multiple kernels are developed to improve the performance. Intensive experiments on a real world electricity load forecasting have been carried out and the results show that our methods can improve the performance at a reduced computation cost. Consequently local learning strategy provides a promising alternative for fast electricity load forecasting.