Mathematical Optimization Techniques For Managing Selective Catalytic Reduction For Coal-fired Power Plants

Abstract

Coal is one of the most important energy sources in the U.S. However, it is also one of the biggest air polluters where the major emissions generated from coal combustion are oxides of nitrogen (NOx). NOx leads to ozone formation and make people more susceptible to respiratory illness. The US Environmental Protection Agency (EPA) has steadily tightened the regulation for NOx emissions that can be discharged into the atmosphere. Many techniques and technologies can all assist with NOx removal. However, to meet upcoming EPA mandates, more aggressive technique such as Selective Catalytic Reduction (SCR) is highly recommended. SCR is an emissions control technique that primarily reduces harmful emissions of NOx. To maintain SCR performance, catalyst layers may be added or replaced to improve NOx reduction efficiency. To make these changes, power plants must be temporarily shut down, and SCR maintenance during scheduled power plant outages can be very expensive. Consequently, developing fleet-wide SCR management plans that are both efficient at reducing NOx and limiting operating costs would be extremely de-sirable. In this dissertation, we propose an SCR management framework that finds an optimal SCR management plan for the fleet number of plants that minimizes NOx emissions or total operating costs using mathematical optimization techniques. In the first part of this dissertation, we propose an SCR schedule generation and optimization algorithm (SGO) to solve the fleet SCR management problem. SCR schedule generation enumerates the set of possible outage schedules by recursion. An optimal set of these generated schedules are then selected by a 0-1 large scale integer program. The main approaches for SGO are recursion, branch-and-bound, and Pareto efficient frontiers. Although SGO is very effective and can yield a good result within a reasonable amount of time, the problem size can get larger and the computational time can increase exponentially. In the second part of this dissertation, we address this limitation by replacing SGO with a multi-commodity network flow problem (MCFP). We first formulated the MCFP as a relaxed problem to solve the fleet SCR management problem without a constraint on average daily NOx. Edges are generated instead of schedules to represent the flow of all SCR catalyst layers for the fleet. The MCFP relaxed problem is solved by a 0-1 integer program. We then address the average daily NOx constraint limitation by introducing MCFP with schedule elimination constraints (MCFPwSEC). The MCFPwSEC algorithm uses a single cut per iteration to incorporate an average daily NOx constraint into the model. We then reduce the computational time further with the introduction of a multi-cut MCFPwSEC. Multi-cut MCFPwSEC similarly eliminate infeasible solutions per iteration based on a heuristic algorithm. Then, we further explore additional ways to reduce the computational time further with discussions on a reactor potential (RP) constraint. Finally, we discuss future extensions of this research.