Predictive Analytics and Optimization for Improved Electric Power Network Reliability and Operation

Abstract

Advances in sensor technology, data storage and signal processing enable methods to indirectly monitor many complex engineering systems. The aim of this dissertation is to present novel statistical and optimization methods that exploit real time sensor information to derive predictive failure risk assessments and decision models to enhance reliability and profitability in electric power systems. We first focus on developing and solving large-scale optimization models to compute sensor-driven optimal operational and maintenance decisions for a fleet of power plants. Operational decisions relate to the well-known unit commitment problem, which identifies dispatch and commitment profiles that satisfy demand requirements, yet are optimized against real-time degradation levels of each power plant. Maintenance decisions focus on deriving optimal fleet level condition-based maintenance schedules that exploit potential economic and stochastic dependence existing among the individual power plants. The decisions are performed while adhering to constrains, such as generation and ramping limits of the power plants, capacities of transmission lines, network reliability, etc. We also consider the interaction between the operational load on the power plants, and their corresponding rate of degradation. This interaction is particularly important since it significantly affects the remaining life of the power plants and the optimal maintenance decisions. Finally, we extend this framework to opportunistic maintenance of windfarms that considers the significant cost reductions arising from grouping the turbine maintenances together. The effectiveness of our approach is illustrated in extensive experiments.