| Short-term load forecasting has been essential for economic and reliable power system operation.Accurate STLF plays a critical role in generation scheduling,reliability analysis,and system operation and optimization.With the deregulation of power system,STLF is more important than ever before since it is used to make decisions on energy generation and purchase。Artificial neural network is a valid tool that can adequately model the complex behavior of the electric load.Ensemble methods are effective to improve the performance of individual predictors.The task of building an effective ensemble model can be broken into two subtasks: developing diverse and accurate member predictors,and then combining them to form a composite predictor.This paper presents a nonlinear ensemble of partially connected neural networks for short-term load forecasting.Partially connected neural networks are chosen as individual predictors due to their good generalization capability.A group-based chaos Genetic Algorithm is developed to generate diverse and effective neural networks,utilizing the population information of GA.A novel pruning method is employed to develop partially connected neural networks in a deterministic way.To further enhance prediction accuracy,an ANN-based nonlinear ensemble of partially connected neural network predictors is developed.The proposed nonlinear ensemble neural network is evaluated on a PJM market dataset and an ISO New England dataset with promising results of 1.76% and 1.29% error respectively,demonstrating its capability as a promising predictor.To capture the variability and uncertainty of electric load,this paper proposes a TRUST-TECH based Versatile interval forecasting method.Versatile distribution function,optimized by TRUST-TECH method,is used to represent the error distribution of electric load forecasting.According to the point forecasting result and its corresponding Versatile distribution,we can get reasonable interval at any confidence degree. |
PDF Full Text Request