MPC-based Realtime Power System Control with DNN-based Prediction/Sensitivity-Estimation (2106.02794v1)

Abstract: This paper presents a model predictive control (MPC)-based online real-time adaptive control scheme for emergency voltage control in power systems. Despite tremendous success in various applications, real-time implementation of MPC for control in power systems has not been successful due to its online computational burden for large-sized systems that takes more time than available between the two control decisions. This long-standing problem is addressed here by developing a novel MPC-based adaptive control framework which (i) adapts the nominal offline computed control, by successive control corrections, at each control decision point using the latest measurements, (ii) utilizes data-driven approach for prediction of voltage trajectory and its sensitivity with respect to control using trained deep neural networks (DNNs). In addition, a realistic coordination scheme among control inputs of static var compensators (SVC), load-shedding (LS), and load tap-changers (LTC) is presented with a goal of maintaining bus voltages within a predefined permissible range, where the delayed effect of LTC action is also incorporated in a novel way. The performance of the proposed scheme is validated for IEEE 9-bus as well as 39-bus systems, with $\pm 20\%$ variations in nominal loading conditions. We also show that the proposed new scheme speeds up the online computation by a factor of 20 bringing it down to under one-tenth the control interval, making the MPC-based power system control practically feasible.