A Distributed Two-Stage Framework for the Co-Scheduling of Power, Heat and Hydrogen Networks with Heterogeneous Frequency Supports

This paper proposes a distributed two-stage frame-work for the co-scheduling of power, heat and hydrogen networks with primary and enhanced frequency supports. To address the bilinear terms inherent in the Nadir frequency constraint, a distributed two-stage framework in the form of a fully-distributed computational pipeline is proposed. In the first stage, FedOnce, a distributed learning architecture with one-shot communication, is proposed to coordinate the energy sectors to efficiently predict the total frequency reserves. These predictions serve as warm starts to generate accurate linearizations of the bilinear Nadir terms, which are then embedded into the second-stage convex optimization of power, heat and hydrogen networks through the decentralized alternating-direction method of multipliers with pre-keyed encryption. The underlying distributed computational pipeline helps resolve the dilemma between computational effi-ciency and information privacy, through both parallel learning and advanced encryption. Case studies demonstrate its significant improvements in computational efficiency (by 2 orders of mag-nitude), approximation gaps (by 4.2%) and frequency response characteristics.