Combination approaches for short term load forecasting

Short term load forecasting for day ahead operations is an important task of an electric distribution company. Forecasting errors directly impact the economics of the distribution company in a market scenario. Many categories of methods like, expert system, artificial neural network and time series analysis, have been developed for short term load forecasting. We compare and contrast these methods on a utility data set. It is seen that no method can be said to be consistently better or worse than the other. Therefore, this paper explores the idea of development of a combination forecast from the three individual forecasts. The combination forecast is shown to have better expected performance than any one of the individual forecasts. Different methods of combining forecasts like, in proportion to probability of success, combining with weights calculated from variance minimization, on the basis of eigenvector of covariance matrix and median forecast are considered. Results of an urban electric distribution company´s data is used to demonstrate the efficacy of the approaches.