Neural networks prove effective at NOx reduction

19 May 2000

The availability of low cost computer hardware and software is opening up increasing possibilities for the use of artificial intelligence concepts, notably neural networks, in power plant control applications, delivering lower costs, greater efficiencies and reduced emissions. Brad J. Radl, Pegasus Technologies, Mentor, OH, USA

Many power plants have been retrofitted with distributed control systems (DCS) that provide a data collection repository of operating parameters recorded in real time. The availability of these data coupled with the flexibility of the DCS to implement sophisticated control strategies, enable artificial intelligence (AI) technologies like neural networks to identify optimal set points for key process parameters, thereby optimising performance and emission control.

One example of a neural network system is the NeuSIGHT combustion optimisation system, developed by Pegasus Technologies, a subsidiary of KFx Inc. A recent survey conducted by Resource Data International of Pegasus’ NeuSIGHT installations has shown that AI solutions can help reduce NOx emissions and improve heat rate. In addition to these benefits, utilities noted that the solution enabled either deferral or elimination of capital expenditures on other NOx control technologies, such as low NOx burners, SNCR and SCR – and reduction of the variable SCR and SNCR maintenance and consumables costs.

Pegasus’ NeuSIGHT combustion optimisation software is installed at more than 25 sites, most operating in closed loop supervisory control. It uses the leading neural network technology from Computer Associates as its modelling engine, leveraging this technology by coupling it with coal-plant implementation expertise.

The benefits can be illustrated using three recent case studies.

The Ameren Labadie station

The Ameren Labadie plant in Missouri uses Powder River Basin coal. It has a 600 MW tangentially fired boiler with a single furnace that has six firing levels each with fuel and auxiliary air compartments surrounding the burner pipe. At each firing level, there is a concentric auxiliary air compartment and an offset auxiliary air compartment. These provide for biasing of the fireball towards the furnace walls or into the traditional firing circle.

Because the Labadie unit is equipped with low NOx concentric firing system (LNCFS) Level 3 NOx control technology, it has two levels of close coupled overfire air (CCOFA) compartments and five levels of separated overfire air (SOFA) compartments. The main burner assembly, which includes the air and auxiliary air nozzles, tilts upward and downward. The SOFA tilts independently from the main burner tilts.

The NeuSIGHT optimisation package uses neural network technology to analyse the current operating input data and compute biases and set points (outputs) for key parameters that control boiler combustion and, hence, boiler performance. The biases that are computed reflect current boiler operation.

The NeuSIGHT system incorporates “on-line training” to update its predictive capabilities as boiler operation, the performance of auxiliaries and fuel properties change.

The system is implemented on a closed loop basis, ie outputs from the optimisation package are sent directly into the control system. NeuSIGHT provides optimisation information to the control system over the entire range of boiler operation, from full load to one-third load. Under boiler upset conditions NeuSIGHT can be bypassed manually by the operators, in much the same way that other operating controls can be overridden. Provisions are also made for operators to put individual sets of controllable variables into and out of NeuSIGHT control. This capability has helped increase operator acceptance and in the long run is helping to achieve performance improvement goals.

The system is set up to optimise boiler operation for the reduction of NOx and the improvement of heat rate. NOx information is obtained from the continuous emissions monitoring equipment. Heat rate is computed directly by NeuSIGHT in a separate but simultaneously running real-time program. The operating staff can input to NeuSIGHT the relative weights between these parameters to control boiler operation to achieve optimisation goals that change due to total system requirements.

NOx reductions achieved. The Labadie unit uses 24 controllables to meet target values for the desired outputs. To date at Labadie, NOx reductions of over 30 per cent have been achieved. This is a further reduction from the baseline values that were already lowered significantly through the installation of LNCFS Level 3 hardware and the use of Powder River Basin coal.

NOx reduction has been the major concern at this station but work is underway to evaluate the impact that NeuSIGHT has on the heat rate and the furnace exit gas temperature.

Lambton units 3 and 4

The goal of the Fossil Business Unit (FBU) of Ontario Power Generation (formerly Ontario Hydro) is to improve its annual heat rate by 2 per cent by the end of 2000 and to reduce NOx emission rate levels by 10 per cent from the 1996 levels. This requires that each station or unit sustain an average of 2 per cent improvement above effects such as higher capacity factor and ambient conditions.

Ontario Power Generation’s Lambton Generating System is 24 km south of Sarnia, Ontario on the east bank of the St Clair River. Electricity is produced by four 510 MW coal-fired units and is delivered to the Ontario Power grid via the 230 kV output system.

The boiler of each generating unit was manufactured by Combustion Engineering with double furnace, radiant superheater and reheater, single drum, controlled circulation and is tangential-fired with 48 burners. There are six pulverisers in each generating unit with two primary air fans, two forced draft fans, two induced draft fans and one precipitator. A wet-scrubbed FGD system was added to both units 3 and 4. These units were also retrofitted with the Bailey INFI-90 DCS control system for the overall unit control including the FGD systems. Low NOx burners and separate overfire air ports were installed at unit 4 to significantly reduce NOx emissions. A total of 162 and 175 key station process parameters were selected as inputs to the NeuSIGHT neural network model for units 3 and 4, respectively. Both average corrected NOx and corrected heat rate were identified as the targets for optimisation. A total of 26 and 38 controllable parameters were selected for optimisation purposes for units 3 and 4, respectively. The main controllable parameters for unit 3 are as follows:

auxiliary air dampers (7 levels);

excess O2;

mill outlet temperatures (6 mills);

and primary air dampers (6 mills).

Since low NOx burners and SOFA ports are installed at unit 4, SOFA dampers and burner tilts were also included as the controllable parameters.

NOx reductions achieved. Performance tests completed for unit 3 showed a NOx reduction of 15 - 25 per cent. A 0.5 per cent improvement in heat rate was also achieved concurrently with the reduction in NOx , depending on the operating conditions. Due to the fact that unit 4 had low NOx burners with separate overfire air ports, the baseline NOx level is nominally 60 per cent of that for unit 3. The reduction in NOx at unit 4 resulting from the neural network system was 10-15 per cent.

With the success of the demonstration programme, Ontario Hydro decided to extend installation of the neural network technology to the remaining units at Lambton.

Parish 8

W A Parish unit 8 is a base-loaded 600 MW tangentially fired Combustion Engineering boiler firing Powder River Basin coal. The unit has a single furnace with six mills, each with a pair of auxiliary air dampers. One of the six mills is always out of service for maintenance activities. The boiler has an extra auxiliary damper located above the top mill, which is used for overfire air. The control parameters used by the neural network model to optimise the boiler operation are the feeder rates, overfire air, excess oxygen and auxiliary air.

The original bid specification for the Parish 8 neural network system called for an advisory control system because the unit is not equipped with a modern DCS. The data from the boiler instrumentation are collected by a Honeywell DAS. When Reliant prepared the specification, it was generally believed (and accepted) that closed-loop control could not be achieved without a DCS. However, Reliant and Pegasus were able to provide a low-cost closed-loop control solution, which has now been successfully implemented.

In this system, the DAS is used to gather most of the data for the NeuSIGHT control system. Other data which are not captured by the Honeywell system (eg, excess oxygen and overfire air set points) are collected via an Allen-Bradley PLC. The optimised set points are output from the NeuSIGHT model to the boiler control system through the same PLC.

The significance of the successful closed-loop application of NeuSIGHT to a boiler without a DCS is that it greatly increases the number of units in the industry for which neural network optimisation is now possible. Boilers with less sophisticated control systems, and for which the economics of an upgrade to a modern DCS are not favourable, will now be able to reap the benefits of the latest in closed-loop control technology.

A series of tests was conducted on the unit in which the key operating parameters were varied to assess their impact on NOx and heat rate. The engineering information from this testing was then used to develop the neural network control model. This core data was then saved, and combined with new data so that the model can be automatically retuned as the boiler process drifts, and as maintenance conditions change.

NOx reductions achieved. The NeuSIGHT control system on Parish 8 completed a two-week acceptance test in June 1999. Results of the testing showed that the NOx emissions were reduced by 15 per cent. A quantitative analysis of the impact on heat rate had not been performed at the time of writing, but the consensus of the plant personnel was that the heat rate had decreased. An unusual operating constraint on the boiler is that it has a CO emissions permit limit of 0.059 lb/MMBtu, which translates to about 50 ppm. Therefore, the control system must carefully consider the CO emissions while determining the optimum set points for minimising NOx. The baseline CO emissions for the boiler are well below the permit limit, and the results of the testing confirmed that the NeuSIGHT control system also maintained CO emissions below the permitted emissions limit.

International expansion

While Pegasus enjoys a leading position in the US boiler optimisation market, it is also very interested in expanding into international markets. This effort have just begun, with an initial installation in Poland now underway. If this installation proves successful, several additional sales have been contracted with the same parties.

This first European installation has produced a number of inquiries from other generating companies in the same region.

Pegasus also participates in a partnership with DSS, a joint venture between Rolls Royce and Science Applications International Corporation (SAIC). This partnership is intended to strengthen Pegasus’ European marketing efforts.

In addition, Pegasus has recently submitted its first bid (as a subcontractor) for a NeuSIGHT installation in India.

Potential to save millions

Experience to date shows that by applying neural network based combustion optimisation software, utilities can reduce emissions, increase plant efficiency, and have the potential to save millions of dollars in capital and operating costs.

Linkedin Linkedin   
Privacy Policy
We have updated our privacy policy. In the latest update it explains what cookies are and how we use them on our site. To learn more about cookies and their benefits, please view our privacy policy. Please be aware that parts of this site will not function correctly if you disable cookies. By continuing to use this site, you consent to our use of cookies in accordance with our privacy policy unless you have disabled them.