This talk will be divided into two parts. The first one deals with the problem of controlling the total power provided by a collection of Distributed Energy Resources (DERs). We assume the model of the underlying power network that the DERs are connected to is partially unknown; thus, it needs to be estimated prior to solving the aforementioned control problem. We formulate this estimation problem as a box-constrained quadratic program and solve it online using the projected gradient descent algorithm. To resolve the potential issue of collinearity in the measurements, we introduce random perturbations in the DER power injections during the estimation process.
The second part of the talk deals with the problem of controlling a set of voltage regulation devices in a power distribution network so as to mitigate the impact of variability in uncontrolled power injections. We formulate this problem as a Markov decision process and propose a data and computationally efficient batch reinforcement learning algorithm to solve it. To circumvent the “curse of dimensionality” resulting from the large state and action spaces, we propose a sequential learning algorithm to learn an action-value function for each voltage regulation device, based on which the optimal control action can be directly determined.