Neural Network Based Reinforcement Learning for Maximum Power Extraction of Wind Energy

The main challenge faced with wind-based energy harnessing is the fluctuating and varying nature of the occurrence of wind and capturing the maximum available power from it. Maximum power can be extracted if the wind speed is forecasted and maximum power point of operation of the wind energy conversion system is tracked. The main prediction techniques used in forecasting are physical, spatial correlation, statistical, and artificial intelligence. The maximum power tracking algorithms available for wind turbines include the perturb and observation method, power signal feedback method and optimal tip speed ratio. The problem incurred with all commonly used maximum power point tracking techniques used are the lack of experiential learning of the control algorithm so as to operate with maximum efficiency with all wind conditions. This paper discusses the use of artificial neural network based reinforced learning algorithm to predict the wind speed and operate the wind energy conversion system at maximum power point. The controller is also equipped with self-learning capability such that it re-evaluates the parameters when the old data has outdated and revise the algorithm with its experiential learning algorithm. The control technique is validated by simulating a 5 MW Permanent magnet synchronous generator based wind energy conversion system in the MATLAB. The performance of the system is compared with the maximum power point tracking algorithm using perturb and observation method and the results are discussed.