Multi-View Neural Network Ensemble for Short and Mid-Term Load Forecasting

Accurate load forecasting is essential to the operation and planning of power systems electricity markets. In this paper, an ensemble radial basis function neural networks (RBFNNs) proposed which trained by minimizing localized generalization error for short-term mid-term forecasting. Exogenous features extracted from series (with long memory multi-resolution wavelet transform) in various timescales are used train RBFNNs. Multiple RBFNNs fused as model with high capability using a weighted fusion method based on model. Experimental results three practical datasets show that compared other methods, reduces mean absolute percentage (MAPE), squared (MSE), (MAE) at least 0.12%, 8.46 (MW) 2 , 0.83 MW (i.e., predict daily peak next month), respectively, MAPE, MSE 0.19%, 2009.69 0.30%, 3697.18 half-hour-ahead day-ahead forecasting, respectively.