OPTIMAL CONTROL OF DATACENTER ENERGY STORAGE FOR DEMAND RESPONSE
Open Access
- Author:
- Mamun, Abdullah-Al
- Graduate Program:
- Mechanical Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- February 21, 2017
- Committee Members:
- Hosam Fathy, Dissertation Advisor/Co-Advisor
Hosam Fathy, Committee Chair/Co-Chair
Christopher D. Rahn, Committee Member
Horacio Perez-Blanco, Committee Member
Anand Sivasubramaniam, Outside Member - Keywords:
- Battery
Lithium-ion
Stochastic control
Genetic algorithm
Optimal control
Particle filter
Physics-based modeling
datacenter
energy management - Abstract:
- This dissertation examines the problem of optimizing the use of electrochemical energy storage devices for demand response in datacenters. Demand response refers to the adjustment of total datacenter electricity demand in response to changing electricity prices. The research in this dissertation is motivated by three critical challenges in datacenter demand response with batteries. First, using existing energy storage in the emergency uninterruptible power supply (UPS) system for demand response might impact the availability of emergency power. Second, if a separate energy Lithium-ion energy storage is used for demand response, using the battery to meet demand response goals without a health-conscious control policy might accelerate battery degradation and reduce expected battery lifetime. Third, proper knowledge of battery aging behavior to construct a health-conscious control policy may not be available. In a broad sense, this dissertation is motivated by the datacenter industry’s growing electricity demand, and the associated costs and carbon emissions. In 2013, US datacenters overall used 91 billion kWh of electricity, at an estimated cost of $6.7b. By the year 2020, this consumption is projected to increase up to 140 billion kWh/year, which is equivalent to the output of 50 power plants with nearly 150 million metric ton of carbon emission. The current literature relies on empirical battery models in its examination of the datacenter demand response problem. These models are quite limited in their ability to capture the fundamental physical phenomena affecting battery behavior. Therefore, the ability of energy storage systems to handle demand response loads needs to be studied from the electrochemical point of view. In addition to using a suitable battery model, a control scheme is also necessary to optimally utilize batteries for demand response. Optimal battery utilization, in this case, means using the batteries to minimize the electricity cost as much as possible with minimum battery degradation. If not optimized for health, a demand response control policy might cause premature failure and shorter end of life (EOL) of batteries and offset the economic benefit of demand response. This research intends to study how amortized capital and operating expenses can be reduced by utilizing energy storage devices from the perspective of electrochemical battery model-based control. This research also focuses on developing a collective learning algorithm to learn battery degradation behavior over time in a distributed datacenter setting to improve health-conscious demand response. The above ideas are presented in this dissertation with the following studies: The first study uses a one-dimensional, physics-based model of a valve-regulated lead-acid (VRLA) battery to examine the degree to which battery energy availability during power outages is affected when datacenter UPS systems are used for demand response. The second study builds on the idea of using a separate Lithium-ion battery pack for demand response assuming the availability of a separate energy storage system for emergency power. This study focuses on developing an optimal control strategy for demand response to (i) maximize the dollar savings attainable through peak shaving, while (ii) minimizing battery degradation. The dissertation solves this multi-objective optimization problem using a second-order model of battery charge dynamics, coupled with a physics-based model of battery aging via solid electrolyte interphase (SEI) growth. The third study presents a stochastic control framework to handle the inherent uncertainty in datacenter power demand for health-conscious demand response using lithium-ion battery packs. The optimal control problem is formulated as a stochastic dynamic programming (SDP) problem where uncertain power demand is modeled as a first-order Markov chain. The fourth study examines the degree to which a large-scale datacenter employing distributed lithium-ion batteries for demand response can learn the aging and degradation dynamics of the underlying batteries by measuring their input/output current/voltage data. Altogether these studies form a foundation for improving current practices of energy storage modeling and control for demand response in terms of electricity cost, and battery aging.