Machine Learning Enabled Steady-State Security Predictor as Deployed for Distribution Feeder Reconfiguration

Abstract

The distribution network is reconfigured by modifying the topology arrangement of the network feeders. Because the voltage stability of the distribution networks can differ within a range following network reconfiguration, the calculation of steady-state voltage stability plays a significant role in real-time feeder reconfiguration. Examining the state of security and estimating it for the next operational configuration is crucial for making real-time decisions. Online security evaluation needs minimal complexity and computing time. Standard methods of assessing the distribution network’s steady-state voltage stability can be insufficient for online and real-time environments due to their high complexity and long computing period. This study proposes a machine learning (ML) approach for classifying configuration states and adopts the decision tree technique to interpret the online applications in the feeder reconfiguration. For the classification, the single line equivalent L-index voltage stability and switching configurations of the feeders are employed as training information for the ML models. A modified IEEE 14-bus and 30-bus test systems verifies the feasibility of the suggested solution. Once trained, the proposed system provides a quick and accurate classification for unknown configurations of the specific security state in 0.3 seconds.