Intelligent maximum power extraction control for wind energy conversion systems based on online Q-learning with function approximation

This paper proposes an intelligent maximum power point tracking (MPPT) algorithm for variable-speed wind energy conversion systems (WECSs) based on an online Q-learning algorithm. Instead of using the conventional Q-learning that uses a lookup table to store the action values for the discretized states, artificial neural networks (ANNs) are used as function approximators to output the action values by using the electrical power and rotor speed of the generator as inputs. This eliminates the need for a large storage memory. The proposed method learns the optimal speed control strategy of the WECS by updating the connecting weights of the ANNs, which has a lower computational cost than the conventional Q-learning method. Moreover, the knowledge of wind turbine characteristics or wind speed measurement is not required in the proposed method. The proposed method is validated by simulations for a WECS equipped with a doubly-fed induction generator (DFIG) and experimental results for an emulated WECS equipped with a permanent-magnet synchronous generator (PMSG).