Hybrid machine learning neural network architectural models for forecasting of wind speed in renewable energy applications

Abstract

Wind energy is a non-linear entity due to its relationship with wind speed and based on the stochastic nature of the natural wind source. When an error occurs during the forecasting of the wind speed, this results in an unexpected error factor in the generation of power from wind energy sources. Thus, it is highly important to predict the wind speed in a most possible accurate manner so as to avoid the presence of errors in wind energy generation. Basically, the development of neural network architectural models have paved a way for better prediction accuracy since it learns and get adapted with respect to the past experiences. Due to which, the applicability of neural network architectures shall enhance wind speed prediction to a higher rate and this may permit the grid sectors to operate in an effective manner to meet the demands of the end user customers. Considering the factors, this thesis has modelled novel techniques that are scalable, effective and plausible models and the summary of findings with significant contributions are presented here. Firstly, a new wavelet based ELMAN JORDAN neural network is modelled with its architectural design for forecasting the wind speed for the considered locations of the wind farm. Basically, ELMAN JORDAN neural network design is a recurrent neural network model and the concept of wavelet functions are introduced into this network for better convergence and to attain minimal square error. The newly modelled neural network is tuned for its weights and bias parameters employing the proposed hybrid particle swarm optimization (PSO) algorithm and ant lion optimization (ALO) algorithm. The advantages of both particle swarm optimization algorithm and ant lion optimization algorithm are hybridized effectively to construct this iv iv hybrid PSO-ALO algorithm. Numerical simulations are carried out employing the proposed hybrid optimization based neural network models for the considered wind farm datasets and the attained results prove the efficacy and effectiveness of the pr