Skip to main navigation Skip to search Skip to main content

Hybrid DRL-SOCP Based Two-Stage Co-Optimization of GFM Siting and Multi-Resource Routing for Cyber-Physical-Transportation Resilience

  • School of Electrical Engineering

Research output: Contribution to journal › Article › peer-review

Abstract

Extreme sequential hazards trigger severe spatiotemporal (ST) cascading failures across interdependent Cyber-Physical-Transportation (CPT) networks. Overcoming the limitations of isolated power grid analyses, this paper proposes a hybrid Deep Reinforcement Learning (DRL) and Second-Order Cone Programming (SOCP) two-stage proactive-reactive co-optimization framework, in which the reactive stage adopts a per-step bi-level structure that couples a DRL routing policy (upper level) with an SOCP physical-feasibility engine (lower level). First, a physics-informed diffusion model generates ST cascading hazards, guiding proactive tri-network hardening and Grid-Forming Energy Storage System (GFM-ESS) siting. Second, a physics-constrained proximal policy optimization with heterogeneous action subspaces (PC-PPO-HAS) agent dynamically routes excavators, Unmanned Aerial Vehicles (UAVs), and repair crews to resolve cross-layer bottlenecks like road blockages and communication blind spots. These high-dimensional routing actions are strictly constrained by a lower-level SOCP engine that embeds a provably tight second-order cone AC branch-flow model and the GFM droop response to guarantee steady-state physical feasibility and voltage/frequency support in the post-fragmentation islands. Case studies show that the proposed tri-network framework maintains high resilience to aftershocks (75.06% time-averaged functionality in the most severe scenario) and, by offloading routing complexity to offline training, generates a full 24-hour recovery schedule in 3.39 s, meeting real-time dispatch requirements.

Original languageEnglish
JournalIEEE Transactions on Smart Grid
DOIs
StateAccepted/In press - 2026
Externally publishedYes

Keywords

  • CPT systems
  • distribution network resilience
  • DRL
  • GFM inverter
  • sequential cascading hazards

Fingerprint

Dive into the research topics of 'Hybrid DRL-SOCP Based Two-Stage Co-Optimization of GFM Siting and Multi-Resource Routing for Cyber-Physical-Transportation Resilience'. Together they form a unique fingerprint.

Cite this