Controlling Smart Inverters Using Proxies: A Chance-Constrained DNN-Based Approach

Coordinating inverters at scale under uncertainty is the desideratum for integrating renewables in distribution grids. Unless load demands and solar generation are telemetered frequently, controlling given approximate grid conditions or proxies thereof becomes a key specification. Although deep neural networks (DNNs) can learn optimal inverter schedules, guaranteeing feasibility largely elusive. Rather than training DNNs to imitate already computed power flow (OPF) solutions, this work integrates DNN-based policies into OPF. The proposed trained through two OPF alternatives that confine voltage deviations on average as convex restriction of chance constraints. be driven by partial, noisy, proxy descriptors current conditions. This important when has solved an unobservable feeder. DNN weights via back-propagation upon differentiating AC equations. An alternative gradient-free variant also put forth, which requires only solver avoids computing gradients. Such practically relevant calculating gradients cumbersome prone errors. Numerical tests compare control schemes with setpoints terms optimality feasibility.