A hybrid short-term load forecasting with a new input selection framework

Abstract

This paper proposes a hybrid STLF (short-term load forecasting) framework with a new input selection method. BNN (Bayesian neural network) is used to forecast the load. A combination of the correlation analysis and ℓ2-norm selects the appropriate inputs to the individual BNNs. The correlation analysis calculates the correlation coefficients between the training inputs and output. The Euclidean distance with respect to a desired correlation coefficient is then calculated using the ℓ2-norm. The input sub-series with the minimum Euclidean norm is selected as the most correlated input and decomposed by a wavelet transform to provide the detailed load characteristics for BNN training. The sub-series whose Euclidean norms are closest to the minimum norm are further selected as the inputs for the individual BNNs. A weighted sum of the BNN outputs is used to forecast the load for a particular day. New England load data are used to evaluate the performance of the proposed input selection method. A comparison of the proposed STLF with the existing state-of-the-art forecasting techniques shows a significant improvement in the forecast accuracy.