Deep-Learning-Directed Preventive Dynamic Security Control via Coordinated Demand Response

TL;DR

The paper tackles the challenge of preventing rotor-angle instability from rare three-phase short-circuit faults in bulk power systems. It introduces an end-to-end deep learning framework that combines a convolutional neural network with an attention mechanism to predict out-of-step conditions from time-domain simulation data and drive preventive Coordinated Demand Response (CDR). Key contributions include a task-agnostic resilience framework based on time-domain simulations, a CNN-Att architecture that captures spatiotemporal fault patterns, and a transfer-learning-based weighted majority voting scheme (TL-WMV) to activate DR. Results on the IEEE 39-bus test system show improved early OOS prediction and a reduced probability of collapse under preventive DR, highlighting practical potential for reducing blackout risk without additional hardware.

Abstract

Unlike common faults, three-phase short-circuit faults in power systems pose significant challenges. These faults can lead to out-of-step (OOS) conditions and jeopardize the system's dynamic security. The rapid dynamics of these faults often exceed the time of protection actions, thus limiting the effectiveness of corrective schemes. This paper proposes an end-to-end deep-learning-based mechanism, namely, a convolutional neural network with an attention mechanism, to predict OOS conditions early and enhance the system's fault resilience. The results of the study demonstrate the effectiveness of the proposed algorithm in terms of early prediction and robustness against such faults in various operating conditions.

Figures (3)

• Figure 1: The proposed DLD-DSC via CDR.
• Figure 2: The blurred colormap representation of the tabular data.
• Figure 3: The comparison of CNN-Att accuracy in rotor angle instability identification.