Proactive Topology Optimization and Service Restoration for Improved Distribution System Outage Management

It is estimated that close to 90% of outages in the electric power grid originate in the distribution system. Although the use of advanced metering infrastructure has increased situational awareness in the distribution system, current approaches to outage management are often reactive and do not fully leverage insights from outage prediction models for the service restoration process. Hence, this work aims to provide a holistic strategy for combining outage prediction and service restoration in the outage management process. First, a detailed analysis of an outage dataset is conducted in order to gain insights into the frequency and duration of outages in a distribution system. Two machine learning techniques, random forest and gradient boosting, are used to rank outage features, including outage causes, climate descriptions, and failed equipment, to determine which outage features have the greatest impact on average outage duration in a distribution system. Following that, this dissertation proposes a proactive topology optimization and service restoration framework that leverages forecasts from outage prediction models to mitigate the impacts of predicted outages in the distribution system. The proposed framework is formulated as a mixed integer linear programming (MILP) problem with the objectives of minimizing the load lost prior to the outage and maximizing the restorable load when the outage occurs at the predicted locations. The MILP model was simulated using the Python Optimization Modeling Objects (Pyomo) package, an open-source tool, and solved using the CPLEX solver. Using modified versions of the IEEE 13-node and 123-node test feeders, the framework considers three optimization cases: single outage, multiple outage, and weighted multiple outage. In addition, a sensitivity analysis based on the weighted multiple outage case is presented in order to determine the optimal topology to operate in given a range of probabilities for the possible outage locations in the distribution system. Furthermore, the MILP model used in this work is validated by comparing its power flow results with those obtained from OpenDSS, an open-source simulation tool for electric power distribution systems. The results show that the MILP model provides a reasonable approximation of the nonlinear power flow model. Overall, this dissertation provides a method for improving situational awareness within the distribution system. Using the proposed approach, distribution system operators can determine what topology to operate in ahead of predicted outages, thereby reducing the loads left out of service.