SE PhD Final Oral Defense of Enes Bilgin

TITLE: Distributed Load Participation In Energy And Reserve Co-Optimizing PowerMarketsABSTRACT: In recent years, the rate of integrating renewable energy resources into the US power system is increasing rapidly. Although renewables are proving economic energy supply options, their volatility and intermittency can result in prohibitively-high-cost for securing reserves that are needed to guarantee instantaneous energy balance and power system stability. Regulation Service (RS) reserves, a critical type of bi-directional Capacity Reserves, are today provided by expensive and environmentally unfriendly centralized fossil fuel generators. This dissertation investigates low-cost, demand-side provision of RS reserves. This is a challenging undertaking since loads must first promise reserves in the Hour Ahead Markets, and then be capable of responding to the dynamic ISO signals by adjusting their consumption effectively and efficiently. To this end, we investigate a decision support framework that enables Smart Neighborhood Operators (SNOs) to become demand side RS reserve providers.The dissertation focuses on:(I) Finding optimal/near optimal SNO control policies to modulate aggregate electricity consumption in response to the dynamic ISO signal. To accomplish this, we (i) describe the Dynamic Programming (DP) problem of responding optimally to ISO signals, and propose and implement an Approximate Policy Iteration and a Modified Actor Critic approach to solve for the optimal policy, (ii) solve the problem for a population of duty cycle appliances with realistic thermodynamics using Reinforcement Learning, (iii) propose a smart thermostat design and drive it by an adaptive control policy, an approach that performs excellently for systems whose dynamics and dynamically changing consumer preferences are not known or observed beyond total power consumption.(II) Developing a decision support framework that assists SNOs to bid for reserve optimally in a competitive Hour Ahead market, based on (i) statistical analysis employing tracking error and expected utility variance characterizations derived under (I) to estimate probabilistic constraints on the maximal RS reserves that can be offered to the market, (ii) use of the analytic properties of the optimal policies as well as numerical results derived under (I) to propose and calibrate a describing function of expected policy implementation costs, and (iii) select reserves that maximizes Hour Ahead clearing price based revenues minus SNO’s RS reserve provision cost.COMMITTEE: Advisor: Michael Caramanis, SE/ME; Ioannis Paschalidis, SE/ECE; Christos Cassandras, SE/ECE; Pirooz Vakili, SE/ME; Chair: Hua Wang, SE/ME

When 12:00 pm to 2:00 pm on Friday, April 11, 2014
Location 15 Saint Mary's Street, Rm 105