Using a Hybrid Model based on Gated Recurrent Unit NeuralNetworks, Wavelet Transformation, Random Forest, andParticle Swarm optimization algorithm to predict Wind Speed

Wind energy is a form of renewable energy harnessed from the kinetic energy of the wind. It is a clean source of power that produces no greenhouse gas emissions. This has attracted significant attention from electricity industry administrators and investigators. This heightened interest has prompted numerous researchers to devote their time and expertise to forecast wind patterns and overcome challenges in wind power generation. In this study, a novel approach is presented, integrating gated recurrent unit (GRU) neural networks and discrete wavelet transform (DWT) as a method for decomposing signals to predict wind speeds. Additionally, the proposed methodology incorporates a feature selection process wherein random forest (RF) assesses the level of interdependence between input and output features, while the particle swarm optimization algorithm (PSO) determines optimal dependency values for effective feature selection. The effectiveness of the proposed model was evaluated using a dataset comprising ۸,۷۶۰ wind speed data points. The results were assessed using error metrics such as MAPE, MSE, MAE. The outcomes exhibited acceptable accuracy, demonstrating the efficiency of the proposed model.

Keywords:

Data decomposition , Gated Recurrent Unit neural networks , Random Forest.