Short-term wind speed forecasting approach using Ensemble Empirical Mode Decomposition and Deep Boltzmann Machine

In the recent past, significant growth in renewable generation and integration with grid have resulted in diversified experiences for planning and operation of modern electric power systems. Electrical power system planners and operators have to work with technical issues of photovoltaic and wind resources integration into the grid to provide clean, reliable, safe, and affordable energy for people around the globe and also to minimize the use of fossil fuels. Wind energy is a fairly dependable source of renewable energy for generating electricity in spite of its highly non-linear and chaotic nature. But the prediction of such data demands highly non-linear temporal features. A new robust hybrid deep learning strategy (HDLS) is developed for enhanced prediction accuracy by preprocessing the raw input. The most effective signal decomposition technique, ensemble empirical mode decomposition (EEMD) is used for preprocessing. This technique decomposes the input into finite intrinsic mode functions and a residue after which training input matrices are established. In the next step, each Deep Boltzmann Machine (DBM) model is constructed by stacking four restricted Boltzmann machines (RBM). The training input matrices formed by each of the extracted intrinsic mode functions and a residue are applied to each DBM. Then the summation of all the predicted results is evaluated to attain the final result of time-series. For adequate performance assessment, hybrid deep learning strategy is developed for analyzing wind farms in Telangana and Tamilnadu. Finally, the proposed deep learning strategy is found to give more accurate results in comparison with existing approaches. (C) 2019 Elsevier Ltd. All rights reserved.