Intelligent Systems Report € March € 1996 € Volume 13 € No. 3

Neural nets clean up utility plants

Ohio Edison (Akron, Ohio) is utilizing an online, real-time, neural network-based closed loop supervisory control system. The system, NeuSight from Pegasus Technologies Corp. (Painesville, Ohio), is optimizing the combustion process in a coal-fired utility boiler, reducing nitrogen oxide (NOx) emissions and loss on ignition (LOI), while improving unit heat rate. Ohio Edison, Ohio Department of Development, the U.S. Department of Energy, the Environmental Protection Agency, and Pegasus Technologies sponsored the program under a National Industrial Competitiveness Through Energy, Environment and Economics grant.

Nitrogen in the coal accounts for as much as 80% of total NOx from coal combustion. Currently, the only available methods for reducing NOx emissions from coal-fired boilers are either installing catalytic conversion equipment for stack gas treatment or retrofitting low-NOx burners in the flue-gas stream. Although each technology reduces NOx emissions by 30% to 75%, each costs more than $5 million to install in a typical plant.

Combustion control is an emerging alternative for reducing emissions, particularly neural network-based control systems (see ISR, August 1995). Pegasus Technologies, for instance, has developed a neural network system for improving the thermal efficiency of large, coal-fired utility plants. For less than $250,000, Pegasus can install its NeuSight system to control NOx emissions.

The neural network core of the product is CAD/CHEM, a generalized process optimizer developed by AI Ware Inc. (Cleveland, Ohio). The software is able to model nonlinear physical processes.

NeuSight, which operates on a UNIX-based workstation, communicates with a digital control system (DCS) through a computer interface unit. Using a nonlinear model, the system performs many "what if" simulations to determine the combustion set-points required to reduce NOx emissions. The model periodically updates itself as the neural network learns from new data. Reacting to fuel fluctuations, soot buildup and equipment performance, the system generates new set-points and downloads them to the DCS automatically.

Optimizing a boiler for both thermal efficiency and emissions reduction requires that dynamic set-point targets be generated in real-time to balance trade-offs between NOx emission control, heat rate improvement and unburned carbon. Pegasus uses the neural network to identify 50 to 100 set-point parameters that include plant output, emissions requirements, ambient operating conditions, equipment age and maintenance. The NeuSight computer model adjusts the plant control system for on-line response to changes in the combustion process.

According to Ohio Edison, implementation of the NeuSight system at its New Castle Station reduced NOx emissions while improving the unit heat rate. Limited data suggests a dramatic reduction (30%) in the amount of unburned carbon in the ash.

Click here to return to Table of Contents for the ISR March issue