Application of Pruned Bilinear Recurrent Neural Network to load prediction

Prediction of electric load by using Pruned Bilinear Recurrent Neural Network (PBRNN) is proposed and presented in this paper. The PBRNN was developed to alleviate the computational cost associated with the Bilinear Recurrent Neural Network by using a pruning procedure. Since electric loads have a time-series characteristic, a prediction scheme based on the PBRNN can be an optimal candidate for the electric load prediction problem. Experiments are conducted on a load data set from the North-American Electric Utility (NAEU). Results show that the Pruned BRNN-based prediction scheme outperforms the conventional Multi- Layer Perceptron Type Neural Network (MLPNN) in terms of the Mean Absolute Percentage Error (MAPE).