Date

2015

Document Type

Dissertation

Degree

Doctor of Philosophy

Department

Industrial Engineering

First Adviser

Snyder, Lawrence V.

Other advisers/committee members

Defourny, Boris; Kishore, Shalinee

Abstract

Today, one main challenge that the energy industry faces is the ability to increase energy efficiency. This requires effort from two entities within the energy system – the rule/policy maker from the upper level who creates standards to properly regulate energy usage or energy-trading processes. Another entity is the rule follower, who reacts to the rules or polices wisely to maximize its own benefits taking into consideration economic and quality-of-life issues. This thesis studies three problems to help the rule follower increase its benefit in the electricity market. In the first problem, a power consumer aims to dispatch its power usage over time given different electricity price rates, appliance characteristics, and power limit constraints. We propose an approximate dynamic programming (ADP) algorithm, which works well for small instances. We also propose several scheduling policies for very fast solutions of large scale problems. We show that sorting the requested appliances according to their operating urgency improves the cost. This result allows the power consumer to dispatch the power usage properly and quickly. In the second problem, a demand charge cost is incurred according to the peak load for some large power consumers. A battery system is introduced for peak shaving. Under load uncertainty, we develop a stochastic DP model to solve the problem optimally as well as a Sample Average Approximation (SAA) algorithm for real time implementation. We also introduce several Naive Algorithms for comparison with the DP and SAA algorithm. Finally, we introduce a real time SAA algorithm and test its performance on a data set consisting of 365 days. This algorithm is very effective in terms of generating real time power dispatch plans. In the third problem, we look at the solar power trading problem between a PV farm and the grid operator who imposes complex constraints on the power profile. We propose a Mixed Integer Programming model assuming perfect knowledge of the PV output. Then we relax some of the constraints, and develop a dynamic programming model for the stochastic load problem. We show that a threshold structure battery inventory solution exists for the relaxed problem. We also propose a dynamic programming model with respect to the key constraints from the grid operator.

Share

COinS