A deterministic annealing approach for short term load forecasting

This paper presents a new algorithm for short term load forecasting based on the deterministic annealing technique. The forecasting function is a piecewise model whose complexity is limited by the Shannon entropy of the partition. The advantages of this method include: flexibility, less model complexity, high forecasting accuracy and global optimal property. The new approach can be used to alleviate the restriction for the partitions and the tendency to be trapped in local minimum of the existing methods such as multilayer neural networks, radial basis function networks, classification and regression trees, multiple adaptive regression splines and fuzzy logic method. The practical numerical tests for short term load forecasting problems show that the prediction accuracy achieved by deterministic annealing method is significantly higher than that obtained by multilayer neural networks and radial basis function networks.