PhD student Temitope Amuda with Advisor A. Domínguez-García

We consider the problem of voltage regulation in power distribution networks with inverter-based resources (IBRs) whose reactive power output can be controlled. The problem is formulated as a stochastic optimization program, which is solved online using a modified version of the projected stochastic gradient descent (PSGD) algorithm. The proposed PSGD-based algorithm leverages the sensitivities of changes in bus voltage magnitudes to changes in the reactive power setpoints of the IBRs. We propose a method for learning such sensitivities online using a recursive least squares estimator (rLSE). To ensure the proper operation of the rLSE, the sequence of incremental changes in IBR reactive power setpoints must remain persistently exciting. This requirement is guaranteed by design through a mechanism that is built into the controller. We demonstrate that the sensitivity learning method is amenable for distributed implementation, which is advantageous from a reliability point of view. We showcase the effectiveness of the framework through numerical simulations using the IEEE 123-bus distribution test feeder in Figure 1. This research is supported by the U.S. DOE.

Figure 1: Numerical simulation results of the modified IEEE 123-bus network, visualizing the buses’ voltage magnitudes trajectories for some of the buses after load perturbations at the buses. The reference voltage magnitudes are depicted in black dashed lines, the voltage magnitudes trajectories without control action are depicted in blue lines, and the voltage magnitudes trajectories with control action are depicted in red lines.