|
Clean Combustion Power
Neural networks optimize boiler operation and reduce emissions
What if a power plant worked like the human brain, learning to adapt and change behavior based on inputs from throughout its system? That is precisely the goal of a neural network, a relatively new approach to boiler optimization that extracts real-time data from a Distributed Control System (DCS) to adjust plant performance. The result is more efficient operations and reduced emissions — once thought to be mutually exclusive objectives.
In today’s energy market, utilities are under pressure to produce low-cost electric power while meeting increasingly stringent environmental regulations. In issuing the "NOx SIP Call," the U.S. Environmental Protection Agency has set new requirements for reducing emissions of nitrogen oxides (NOx) to an average of 0.15 pounds per million Btu in the eastern half of the United States. With a recently revised compliance deadline of May 31, 2004, utilities are looking for low-cost ways to reduce NOx emissions without sacrificing performance.
Exploring Viable Options
Traditional solutions involve a trial-and-error combination of NOx reduction strategies, such as switching to Powder River Basin coal. The western coal’s low-nitrogen content produces less NOx, but its high moisture and low Btu content may require a plant retrofit or result in a performance degradation. Other options include low NOx burners, overfire air (OFA) and post-combustion controls.
"The option that is talked about the most is selective catalytic reduction (SCR)," said Jeff Arroyo, fuels and combustion project manager for Black & Veatch’s Energy Services Group. "It’s a post-combustion technology that is very expensive, but it gets compliance in one fell swoop. The other options have to be done in combination to have a chance of achieving the same results. But together, they’re all less expensive than a SCR."
Recent advances in computer hardware and software technology have made boiler optimization a viable alternative to one or more of the options that require large capital outlays. Based on neural network computing, which uses real-time data to learn from plant operation experience, boiler optimization can reduce emissions while improving plant performance. Neural network models differ from traditional engineering and statistical computing models because they accurately make predictions for complex, time-varying, non-linear relationships.
Excerpt from "Clean Combustion Power," Solutions Magazine, Vol. 21, No. 3, 2001, pp. 12–13. Copyright 2001 Black & Veatch, Kansas City, Mo. |
