Virtual inertia regulation based high-order observer for weak microgrids using deep sequential learning

The high penetration of sustainable energy resources (SERs) into the hybrid Micro-grid system (HMGS) is equipped with AC/DC/DC power converters to connect such generation units to the bus. Since these power electronic interfaces do not have rotational mass, the HMGS's inertia will be reduced accordingly. This study introduces the concept of a virtual frequency controller (VFC) in the energy storage systems (ESSs) employed in a low-inertia HMGS under high penetration of SERs. This paper introduces a new approach to adaptive VFC for low-inertia HMGSs which includes three major innovations: (1) implementation of a high-order extended disturbance observer (HOEDO) for real-time estimation and compensation of unknown dynamics and SER-induced disturbances; (2) applying the deep learning deep sequential action-value learning (SAVL) algorithm to adjust the parameters of the controller-observer in an unsupervised fashion to adapt to system variability and uncertainties without a priori models; and (3) the inclusion of virtual rotor dynamics emulation during primary and secondary control loops to enhance responsiveness akin to synchronous generators. By maximizing a cumulative reward function, the agents of deep SAVL are trained in such a way that the controller is adaptively designed according to SERs/load variations. By deploying the training capability of deep neural networks, the suggested adaptive VFC technique improves the system's inertia and damping features under high-penetration of SERs. Moreover, comparative responses of the suggested adaptive VFC controller with conventional virtual inertia control (CVIC) controlled by model predictive control (MPC), fractional-order proportional-integral-derivative (FOPID) control, the conventional virtual inertia and virtual damping control-based PI controller, and CVIC-based two-layer multiple MPC (TLMMPC) methods infer that the suggested scheme not only exhibits more suppressed fluctuations but also demonstrates a faster response to the dynamic variations.