Kwon, Soongeol (2017-04). Demand-Side Management for Energy-efficient Data Center Operations with Renewable Energy and Demand Response. Doctoral Dissertation.
Thesis
In recent years, we have noticed tremendous increase of energy consumption and carbon pollution in the industrial sector, and many energy-intensive industries are striving to reduce energy cost and to have a positive impact on the environment. In this context, this dissertation is motivated by opportunities to reduce energy cost from demand-side perspective. Specifically, industries have an opportunity to reduce energy consumption by improving energy-efficiency in their system operations. By improving utilization of their resources, they can reduce waste of energy, and thus, they are able to prevent paying unnecessary energy cost. In addition, because of today's high penetration of renewable generation (e.g. wind or solar), many industries consider renewable energy as a promising solution to reduce energy cost and carbon pollution, and they have tried to utilize renewable energy to meet their power demand by installing on-site generation facilities (e.g. PV panels on roof top) or making a contract with renewable generation farms. Moreover, it is becoming common for energy markets to allow industries to directly purchase electricity from them while providing the industries with day-ahead and real-time electricity price information. In this situation, industries have an opportunity to adjust purchase and consumption of energy in response to time-varying electricity price and intermittent renewable generation to reduce their energy procurement cost, which are called demand response. Considering these opportunities, it is anticipated that the industrial sector can save a significant amount of energy cost, however, time-varying behavior, uncertainty and stochasticity in system operations, power demand, renewable energy, and electricity price make it challenging to determine optimal operational decision. Motivated by the aforementioned opportunities as well as challenges, this dissertation focuses on developing decision-making methodologies tailored for demand-side energy system operations to improve energy-efficiency based on energy-aware system operations and reduce energy procurement cost by utilizing renewable energy and demand response in an integrated fashion to optimally reduce energy cost. For practical application, this dissertation considers real-world practices in data centers including their operations management and power procurement for the following research tasks: (i) develop a server provisioning algorithm that dynamically adapts server operations in response to heterogeneous and time-varying workloads to reduce energy consumption while providing performance guarantees based on time-stability; (ii) propose stochastic optimization models for optimal energy procurement to determine purchase and consumption of energy based on day-ahead and real-time energy market operations considering utilization of renewable energy based on demand response; (iii) suggest a decision-making model that integrate the proposed server provisioning algorithm with energy procurement to achieve energy-efficiency in data center operations to reduce both energy consumption and energy cost against variability and uncertainty. In terms of methodologies, this study uses operations research techniques including deterministic and stochastic models, such as, queueing analysis, mixed-integer program, Markov decision process, two-stage stochastic program, and probabilistic constrained program. In conclusion, this dissertation claims that renewable energy, demand response, and energy storage are worth to be considered for data center operations to reduce energy consumption and procurement cost. Although variability and uncertainty in system operations, renewable generation, and electricity price make it challenging to determine optimal operational decisions, numerical results show that the proposed optimization problems can be efficiently solved by the developed algorithm. The proposed decision-making methodologies can also be extended to other in
