Deep Ensemble With Proliferation of PV Energy for Bidirectional Evaluation of Voltage Stability Margin

This paper proposes an algorithm for the bidirectional evaluation of voltage stability margin (VSM) with a large photovoltaic (PV) power penetration. To address the variability caused by loads and PV for bidirectional VSM, a deep ensemble model with a simultaneous point and probabilistic predictions is developed, which takes advantage of advanced metering infrastructure and weather data. The deep ensemble model consists of offline- and online-trained ensemble, which uses optimization blend and k-means, respectively. The proposed approach can also consider topology changes and environmental impacts (e.g., solar irradiance and temperature), with a major bearing on load/PV models and network parameters. The validity of the proposed approach is verified through its applications to a modified IEEE 33-bus system and IEEE 123-node test feeder, followed by VSM sensitivity analyses (Sen-ELM) and tolerance analyses with the consideration of input errors. The numerical results demonstrate that the proposed VSM approach has a strong generalization capability considering the proliferation of PV energy.