A novel fuzzy-based ensemble model for load forecasting using hybrid deep neural networks

Abstract

A novel, hybrid structure for week-ahead load forecasting is presented. It is the energy market evolution that compels its participants to require load predictions whose accuracy cannot be provided by traditional means. The proposed implementation combines attributes from ensemble forecasting, artificial neural networks and deep learning architectures. The proposed model initially clusters the input data using a novel fuzzy clustering method for creating an ensemble prediction. For each cluster created, a new regression approach is applied to model locally the load forecasting problem. Following a two-stage approach, initially, a radial basis function neural network (RBFNN) is trained using three-fold cross-validation and the hidden layers of the best three RBFNNs are used to transform the input data to a four dimensional dataset. Then, a convolutional neural network (CNN) is deployed receiving as input the latter dataset. Thus, a neural network is formed consisting of a radial basis function (RBF), a convolutional, a pooling and two fully-connected layers. Both RBFNNs and CNNs are trained with the Adam optimization algorithm within the Tensorflow deep learning framework. The proposed model is designed to predict the hourly load for the next seven days and its effectiveness is evaluated in two different case studies; namely the Hellenic interconnected power system and the isolated power system of Crete. Both case-studies exhibit the superior performance of the proposed model when compared to state-of-the-art and traditional load forecasting schemes.