Committee: 

Dr. Saeedifard, Advisor 

Dr. Graber, Chair

Dr. Grijalva

Abstract: The objective of the proposed research is to model and design the control of multiport autonomous reconfigurable solar power plant (MARS) which integrates the utility-scale solar photovoltaic (PV) power plants and energy storage system (ESS) into transmission ac grid and HVdc links. The MARS system incorporates advanced hierarchical control methods to provide primary frequency response, reactive power support, and transient stability to the ac grid. This proposal constructs a high-fidelity switched system model of the isolated MARS; this model is developed using numerical stiffness-based hybrid discretization and hysteresis relaxation to reduce the simulation time. In addition, an interpolating method is introduced to the PV submodules to help increase the minimum simulation timestep and maintain high accuracy in the simulation results. The MARS system suffers from unbalanced capacitor voltages under certain operating conditions. The updated energy balancing criteria (EBC) are proposed to reduce the capacitor voltage disparity and maintain stability under arbitrary operating conditions. In addition, two types of machine learning based EBC criteria are developed and compared. The artificial neural network (ANN) outperforms the random forest (RF) algorithm and is implemented in the hierarchical control of MARS to disable/enable the EBC depending on the MARS input power dispatch command. The developed algorithms and control mechanisms are validated with PSCAD/EMTDC simulations as well as the designed hardware in the loop evaluation platform.