A Master–Slave Deep Learning Framework for Real-Time Transient Stability-Constrained Optimal Power Flow

Abstract

Transient stability-constrained optimal power flow (TSC-OPF) is a robust approach for ensuring the dynamic security of power systems in economic dispatch. However, its adoption has been hindered by the computational complexity of optimization incorporating transient stability constraints (TSC). In this paper, the TSC-OPF optimization problem is converted into a machine learning task, and we propose a master-slave deep learning framework as an end-to-end data-driven approach to achieve real-time TSC-OPF. This framework is constructed by a master model dedicated to TSC-OPF learning and multiple slave models responsible for validating TSCs against individual contingencies. An integrated training algorithm is also proposed to adaptively coordinate the training of these models, gaining awareness of TSC compliance while pursuing economical dispatch solutions. The proposed framework and training algorithm have been validated on New England and Australian power systems, confirming the capability and scalability of the data-driven system to deliver real-time TSC-OPF solutions within a millisecond. An ablation study demonstrates our approach's strong TSC compliance and excellent fidelity of dispatch solutions. Implementing TSC-OPF in real-time can significantly improve power system operability, particularly in environments with increased variations from energy resources and reduced system strength.