DNN-based policies for stochastic AC OPF

A prominent challenge to the safe and optimal operation of modern power grid arises due growing uncertainties in loads renewables. Stochastic flow (SOPF) formulations provide a mechanism handle these by computing dispatch decisions control policies that maintain feasibility under uncertainty. Most SOPF consider simple such as affine are mathematically resemble many used current practice. Motivated efficacy machine learning (ML) algorithms potential benefits general for cost constraint enforcement, we put forth deep neural network (DNN)-based policy predicts generator real time response The weights DNN learnt using stochastic primal–dual updates solve without need prior generation training labels can explicitly account constraints SOPF. advantages over simpler their enforcing safety limits producing near solutions demonstrated context chance constrained formulation on number test cases.