Hybrid neural network model for short-term load forecasting

Short-term load forecasting has always been the essential part of reliable and economic operation in power systems. In this paper, a hybrid neural network model combining rough set theory, principal component analysis, dynamic clustering analysis and ant colony optimization algorithm is presented. First, rough set theory is used to eliminate redundant influential factors that don't exert tremendous effect on power load. Next, principal component analysis is employed to minimize the correlations in the selected factors. Then, using dynamic clustering analysis, the historical load data are divided into several groups. According to the similarity between hourly load to be forecasted and classified categories calculated by grey relational analysis, typical samples are selected and corresponding neural network model for hourly load, training by ant colony optimization algorithm, is established. Finally, the forecasting results using actual load of Chekiang province in China proves that the proposed model is satisfactory.