nox reduction and boiler optimization - neuco, inc gen_conf_2015_sherco_u2_boiler...x reduction and...
TRANSCRIPT
Sherco Unit 2NOx Reduction and Boiler Optimization
2015 Power-Gen Conference
Jason Grimm (Xcel Energy)
Steve Piche (NeuCo)
Outline
Background
Equipment
Project Objectives
Parallel Projects
Combustion Optimization
Project Overview
Novel Neural Network Training Algorithm
Results
Conclusions
1Proprietary and Confidential
Sherco Unit 2 Commissioned in the late 1970s
730 MWg 8 corner tangentially fired CE Boiler
Seven 1003 RP Mills, six required for full load
Wet scrubbers with electrostic precipitators
Single, shared stack with Unit 1
Wind-Following
Continuously cycle to follow demand. Turn down to about 40% of load capacity
Typically burn blend of 70% Black Thunder or Jacob’s Ranch and 30% Absaloka fuels
Proprietary and Confidential 3
Unit 2 Existing NOx Control
Mid 1990s OEM Designed LNCSFS III Low NOx system
Four levels of SOFA ports in 8 Corners
Closed Couple Over-Fired Air
Low NOx Burners
Opposed blade dampers and individual ABB damper positioners
Common stack CEMS
1 NOx and 1 CO analyzer per economizer outlet duct
Proprietary and Confidential 4
Equipment Challenges
Unit 2 has older low-NOx furnace design and therefore cannot perform as well as Unit 1
Unit 1&2 common stack with common CEMS
Poor coal fineness and distribution
Limited NOx, CO, and O2 instrumentation
Unit 1 overhaul spring of 2015 Unit 2 solo operation
Project Goals
Goal: Reduce and maintain NOx emissions at least from a shared stack below 0.15 lb/MMBtu over a 30 day rolling averege starting January, 2015. At start of the project, goal was not being met.
Penalty: If the NOx is not maintained below the 0.15 lb/MMBtu average, the units must be derated to achieve the goal.
Shared Stack: Emission are measured in a shared stack from two 730 MW T-fired units with no SCRs.
CO Constraint: CO emissions must be maintained below 400 ppm.
Proprietary and Confidential 6
Optimization Strategy
1. Establish optimal baseline boiler curves
2. Improve fuel fineness and fuel distribution
3. Optimize boiler performance through optimization software and operational strategies
Proprietary and Confidential 7
Boiler Tuning with SmartBurn March 3 – 14, 2014
Determine baseline performance
Adjusted SOFA curves – Reduce CO and NOx,, balance O2
SOFA Yaw adjustments – Reduce CO, balance O2
Adjusted and “locked” SOFA tilts at + 20° - Max NOx Reduction
September 2 – 5, 2014
Adjusted windbox/furnace dP curve – maintain AA damper pos. ~25%
Adjusted to final SOFA damper curves
Adjusted O2 Curve and fuel-air damper curves
February 23 – 27, 2015 (Unit 2 Solo Operation)
Major Yaw Adjustments – Reduced CO from ~1000ppm ~100ppm at 2.5% O2 without Optimizer!
Full Load NOx ~0.155 lb/MMBtu with worst mill configuration!
8
Dynamic Classifier Retrofit
All 7 mills retrofitted with Loesche dynamic classifiers
Goals
>75% through 200 mesh
>99.9% through 50 mesh
Maintain current throughput
Installation Schedule
August, 2014 – March, 2015
Project included online coal flow measurement system
Courtesy of Loesche Energy
Systems
Proprietary and Confidential 9
Provides real-time closed-loop optimization of fuel and air
biases in the boiler as frequently as every minute.
Using:
– Model Based Optimization
– Models built using data from the unit (dynamic and neural net models)
To Improve:
– NOx
– CO
– Heat rate
– Steam temps
– Opacity
CombustionOpt
Proprietary and Confidential 12
System Integration
Proprietary and Confidential 13
Courtesy of CombustionOpt
V3.0 User Manual
Each server reads plant data
and sends optimized
outputs to the DCS via PI
NeuCo has remote
connections via VPN to each
server
Model Based Optimization
Proprietary and Confidential 14
Manipulated
Variables
(MVs)
Controlled & Disturbance
Variables
(CVs & DVs)
Optimizer Model
Model Based Optimizer
Setpoints &
Constraints
SP
DCS / Plant
Output and Gradient
Model Options:
- Dynamic models use linear dynamic models (Model Predictive Control)
- Steady State models use neural networks (Neural Network Optimization)
Available Manipulated Variables
Model Predictive Control (Optimize every minute)
O2 Bias
Windbox to Furnace Differential Pressure Bias
Neural Network Optimizer (Optimized every 10 minutes)
Aux Air Damper Biases (16)
Feeder Biases (7)
Over-fired Air Biases (12)
Number of available biases is 35.
Proprietary and Confidential 17
Optimizer must observe boundary and rate of
change constraints on the manipulated variables.
Next Generation Neural Network Model
Confidence can be used in both training the models (input selection) and for optimization.
Design of experiments for collecting data can be based upon model confidence.
Sherco Unit 2 Updated to V3 software in September, 2014
18Proprietary and Confidential
Neural
Network
Model
Aux Air Biases (t+1)
Feeder Biases (t+1)
SOFA Biases (t+1)
NOx(t+1) – NOx(t) and
Neural
Network
Model
+-
+
Aux Air Biases (t)
Feeder Biases (t)
SOFA Bias (t)
Next Generation: Predict a change in NOx and the confidence in that change
due to changes in manipulated variables.
Variance(NOx(t+1) – NOx(t))
Sherco 2 CombOpt V3 UpgradeSeptember, 2014
More Routine DOEs
Daily DOEs
Keep models up to date
Models now use all variables, not just furnace specific
Extended bias limits on 22 and 23 OFA dampers
Completed DOE allowing 22 OFA dampers to go 5% 100% open
Showed 22 elevation OFA dampers have a large effect in controlling CO and NOx.
Currently allow OFA 22 elevation dampers to go 5% 50% at full loads
23 elevation OFA Dampers now have a limit of +/- 20%
45 Day Test Period Results are collected from a 45 day test period (Nov. 16, 2014 to Dec. 31,
2014).
Optimizer was on 95% of the time during the test period.
Proprietary and Confidential 21
Optimizer Off Optimizer On
45 Day Test Period Results
Proprietary and Confidential 23
Bottom Line: 10% NOx Reduction
Note: The data shown is average boiler outlet duct analyzer values, not the
reportable CEMS data
Conclusion & Lesson’s Learned
Boiler Optimization is a very dynamic effort requiring many pieces of equipment to be running in optimal condition including
Pulverizers - coal fineness and distribution
Furnace equipment - burners, SOFAs, tilt systems, air dampers, etc
Boiler controls - draft control, base boiler curves established
As the optimizer “champion”, it is essential to have adequate knowledge of YOUR boiler
What makes it tick, how far can you push it…
Conclusions & Lesson’s Learned
It is imperative to keep an open line of communication with the optimizer service engineering team in order to share ideas and make progress
It is vital to have a good working and trusting relationship with I&C and operators
The work on Sherco Unit 2 shows that when the above equipment is optimized and relationships are in place, NeuCo’s 3rd generation boiler optimizer can reduce it’s NOx an additional 10%
Detailed Results
Proprietary and Confidential 29
Optimizer Off Optimizer On
- Results collected over a 45 day period from Nov 16, 2014 to Dec 31, 2014.
- Optimizer was on 95% of the time during the test period.
Gross MW 669.04
NOx Ave 0.183
NOx East 0.188
NOx West 0.178
CO Average 221
CO East 278
CO West 165
O2 Average 3.16
O2 East 3.03
O2 West 3.28
Gross MW 676.44
NOx Ave 0.166
NOx East 0.175
NOx West 0.157
CO Average 258
CO East 341
CO West 176
O2 Average 2.9
O2 East 2.79
O2 West 3.01