Short-term Load Forecasting Using Deep Belief Network with Empirical Mode Decomposition and Local Predictor

This paper proposes a novel load forecast algorithm framework, which combines the Deep Belief Network(DBN) with Empirical Mode Decomposition(EMD) and Local Predictor(LP). Empirical Mode Decomposition(EMD) is applied to decompose the load curve for denoising. DBNs are graphical models which learn to extract a deep hierarchical representation of the training data. LP is used to select samples based on Euclid distance. These selected samples compose the neighbor set which are used to fine tuned the network. Therefore, Deep Belief Network combined with Empirical Mode Decomposition and Local Predictor (EMD-DBNLP) can be applied to forecast the load and provide more accurate and reliable results. The data used for this experiment is system loads from the AEMO on NSW. The performance of EMD-DBNLP is compared with other algorithms, including Least Squares Support Vector Machines(LS-SVMs), Back Propagation-Artificial Neural Network (BP-ANN), traditional Deep Belief Network(DBN) and Deep Belief Network with Empirical Mode Decomposition(EMD-DBN) respectively. The forecast results demonstrate that the proposed algorithm achieves a better performance than others.