Neural Network Based Controller for an AC Microgrid Connected to a Utility Grid

Microgrid architecture includes distributed energy resource (DER) such as wind power (WP), solar power (SP), and battery banks (BB) for storage, etc, where an appropriate power electronic interface for each DER is integrated and controlled in order to achieve flexible power sharing. Grid disturbances are the most known challenges of microgrid control, which may lead to destabilize the DER behavior, and can produce a discontinuity in power generation. In this paper, neural network based controllers for an alternative current (AC) microgrid composed of WP, SP, BB, and load demand (LD) connected to utility grid are studied. A neural sliding mode linearization (N-SM-L) controller controls the powers injected to utility grid through the power electronic converter controller of each DER. By using a recurrent high order neural networks (RHONN) identifier, wich is trained with an extended kalman filter (EKF) online, an adequate model is obtained under different grid conditions, which helps to reject the disturbances, ensuring stability, and improving robustness. The proposed microgrid architecture is first simulated using simpower toolbox of Matlab and then it is real-time simulated in an Opal-RT (OP5600) simulator.