Generative Adversarial Networks for Real-time Stability of Inverter-based Systems

TL;DR

The paper tackles real-time stability assessment for islanded, droop-controlled distribution networks, where online domain-of-stability estimation is computationally prohibitive. It proposes conditional Generative Adversarial Networks (cGANs) to learn the stability hyperspace offline and deliver configuration-specific stability regions online with substantial speedups. Across single- and multi-configuration scenarios, the approach achieves high accuracy in reproducing stability regions and demonstrates scalability to multiple network topologies. The work shows that cGANs can provide non-conservative, real-time guidance for supervisory control to tune droop gains and improve power sharing while accommodating network reconfiguration and renewables variability.

Abstract

In islanded systems with droop-controlled sources, the droop coefficients need to be tuned in real-time using supervisory control to maintain asymptotic stability. In contrast to offline tuning methods, online domain-of-stability estimation yields non-conservative droop gains in real-time, ensuring good power sharing performance as the operating point varies. The challenge in the conventional online domain-of-stability estimation process is its unscalability and high computational complexity. In this paper, an efficient alternative using conditional Generative Adversarial Networks (cGANs) is described. We demonstrate that the notion of power system stability can be learned by such deep neural networks, and that they can offer a scalable alternative to conventional domain-of-stability estimation methods in islanded distribution systems. The implementation of cGANs-based stability assessment is described for an LV distribution test case and its advantages demonstrated.

Figures (5)

• Figure 1: Schematic of 5-node ring-main distribution system.
• Figure 2: Training procession for simple GANs with learning rate $8\times10^{-6}$.
• Figure 3: Stability region obtained from 20000 samples from GANs is plotted in red. The theoretical stability region is shown in blue. The points identified by the GANs not in the actual stability region are shown in green.
• Figure 4: Training procession for cGANs with learning rate $8\times10^{-6}$.
• Figure 5: Stability region (shown in red) for Inverter-1 identified using cGANs for 4 system configurations at epoch 1907. Corresponding theoretical regions are indicated in blue, obtained from the traditional method. Green circles denote erroneously projected points of stability by cGANs method.