Abstract
Climate change places an immense stress on existing vulnerabilities within our current energy infrastructure via two primary challenges. First, operational management and control of smart city microgrids is becoming more challenging due to the intermittent behavior from high levels of renewable energy penetration. Second, given the increased frequency of extreme climate events and the uncertainties associated with complex energy systems, ensuring resilience against system anomalies and climate catastrophes is becoming progressively more difficult. Harvesting energy in a sustainable manner to reduce both cost and emissions while sustaining the power grid’s stability, is an active area of research to which we aim to contribute.
In this thesis, we propose a novel framework for near real-time dispatch in power system planning. Our proposed framework is based on the dynamic data driven applications systems (DDDAS) paradigm and integrates modern machine learning techniques to map scenarios derived from our proposed resource-aware scenario selection (RSS) algorithm to their corresponding solutions from a mixed integer linear program (MILP). We investigate MG operation under our proposed framework at the substation level using the IEEE-18, IEEE-30, and IEEE-33 test networks. To the best of our knowledge, our approach is the first throughout the literature to: i) develop an RSS algorithm that determines the most probable operating scenarios energy system is likely to experience for flexible disaster planning, ii) integrate power flow physics, DSM practices, power quality into an alternating current (AC) unit commitment (UC) problem, and iii) implement machine learning (ML) for deploying the solutions from existing MILP dispatch models in real-time, and iv) sustain the ML model’s predictive performance in a nonstationary environment to mitigate concept drift. The proposed framework shows promising results in optimizing the design and operation plan of an MG.