Prediction of Maximum Frequency Deviation in Power Systems Based on Self-Attention Mechanism

With the rapid integration of renewable energy, the inertia and capacity levels of power systems have significantly declined, posing greater challenges to frequency stability. Existing methods often rely on centralized or simplified models, neglecting the heterogeneity of inertia distribution and transient dynamics, and thus fail to accurately identify critical risk nodes. To address this, this paper proposes a novel approach that integrates the selfattention mechanism with Accumulated Local Effects (ALE) analysis, enabling high-precision prediction of the maximum frequency deviation while enhancing model interpretability. Simulation results demonstrate that the proposed method effectively reveals the impacts of disturbance power, load status, and generator inertia on system dynamic responses, providing a unified framework and theoretical support for security assessment and control strategy optimization in modern power systems.