Learning Regionally Decentralized AC Optimal Power Flows with ADMM

One potential future for the next generation of smart grids is use decentralized optimization algorithms and secured communications coordinating renewable (e.g., wind/solar), dispatchable devices coal/gas/nuclear generations), demand response, battery & storage facilities, topology optimization. The Alternating Direction Method Multipliers (ADMM) has been widely used in community to address such problems and, particular, AC Optimal Power Flow (AC-OPF). This paper studies how machine learning may help speeding up convergence ADMM solving AC-OPF. It proposes a novel machine-learning approach, namely ML-ADMM, where each agent uses deep learn consensus parameters on coupling branches. also explores idea only from runs that exhibit high-quality properties, filtering mechanisms select these runs. Experimental results test cases based French system demonstrate approach significantly.