Optimization follows DCS retrofit
Following a DCS retrofit at Entergy Corp.'s Independence station in Arkansas, engineers there felt more was needed, so an optimization system was added. It combines a neural network approach with model predictive control (Figure 2) —the former to minimize NOx emissions, the latter to minimize heat rate while providing tighter control of peak steam temperatures during dispatch. An expert-system-based sootblower package completed the optimization system.
2. Minimizing NOx and heat rate. Model predictive control and neural networks were integrated to create an optimization package for Entergy Corp.'s Independence Station in Arkansas. Source: Invensys Systems Inc.
According to a paper authored by Entergy's Steven Coker and engineers from Invensys Systems Inc. and presented at the 48th Annual Joint ISA/EPRI Power Industry Symposium last June, the optimization system paid for itself in fuel savings within months. NOx emissions were reduced by 20% to 25% over and above the DCS retrofit levels.
The complexity of the twin-furnace-design, 800-MW units fired by Powder River Basin coal made it difficult even for the new DCS to minimize NOx emissions within the envelope of other performance constraints. Specifically, the three objectives of the program were to:
- Drive the unit to minimize NOx emissions continuously, even during dispatch.
- Improve unit heat rate.
- Trim peak reheat and superheat temperatures during dispatch. Peak temperatures were suspected of contributing to tube failures, forced outages, and unreliability. The sootblower expert system avoids steam temperature impacts and potential tube erosion and also contributes to lower NOx emissions and heat rate improvement.
Invensys Systems' Consulting Engineer Don Labbe notes in the paper that the tangentially fired twin-furnace design offered many opportunities to reduce NOx. as there are 152 air dampers, eight coal mills, and 64 primary air ports in the furnace. The development of control models was based on those parameters found to influence NOx emissions. The neural network technology used here reportedly allows faster training time, which aids in continuously updating the models.
Model predictive control, notes Labbe, is well suited for situations in which controlled variables and manipulated variables are related by thermodynamic, chemical, or control relationships. It provides rapid response and stable control within constraint limits. For example, final steam temperature is related to superheat spray, with a particular gain based on steam properties and a time response based on metal mass and steam flow.
The issue with the sootblowers was this: The system was effective at responding to gas path pressure drop and large accumulations of slag, but usually at the sacrifice of energy distribution (specifically, lower superheat steam temperatures, with attendant impact on unit heat rate). The expert system now "optimizes" among the various objectives. Operators now receive advisories and support data that help them select which sootblowers to activate and when.
Email
Comments
Print
Reuse
