Distribution network reconfiguration based on the "switch pair" concept and discrete SAC algorithm

This paper proposes an active distribution network reconfiguration method based on the "switch pair" concept and a discrete SAC algorithm. First, the method models photovoltaics (PV) and on-load tap changers (OLTC). Then, it clarifies the radial network constraint that must be satisfied when tie/sectionalizing switches (TS) operate. Next, the Gumbel-Softmax technique is introduced as a "softening" approach, representing discrete distributions as continuous probability distributions, thereby adapting the traditional SAC algorithm for discrete space models. A masking mechanism is also incorporated to filter out invalid joint actions of TS and OLTC. Finally, case studies demonstrate the superiority of the proposed method. The "switch pair" concept effectively reduces the action space of the model, enhancing the exploration and convergence capabilities of the algorithm.