MMoE-based Lyapunov neural network for transient stability analysis of power systems with renewable energy

With the expansion of power system scale and the increasing penetration of renewable energy units, fast and accurate transient stability analysis has become indispensable for ensuring system security and stability. This paper proposes an intelligent transient stability assessment method based on Lyapunov’s direct method. First, a candidate Lyapunov function change rate approximation method using finite differences is presented, which integrates the neural Lyapunov approach with limited observations of complex power systems to expand applicability. Second, a Lyapunov network construction strategy based on the multi-gate mixture-of-experts (MMoE) architecture is developed to resolve the gradient conflict between physical consistency and data fitting. Third, a multi-task learning framework is designed to adaptively balance the constraints of Lyapunov properties and the guidance of empirical trajectories. Finally, a stability assessment index based on the Lyapunov network and QuickHull algorithm is proposed to enable transient stability evaluation and ROA estimation. Case studies on an improved single-machine infinite-bus (SMIB) system and the CSEE-RAS system demonstrate that the proposed method effectively constructs a neural network satisfying Lyapunov function properties, achieving fast and accurate transient stability assessment for complex power systems.