| OCR Text |
Show American Flame Research Committee International Symposium September 30,1996 were compiled into a flow chart type process. The operators follow successive steps .in th~ flow chart until NOx emissions are reduced below the predefined limit. The steps descnbed In the flow chart instruct the operators to first reduce the number of mills in service (if possible), introduce air biases, reduce excess oxygen to the boiler, or begin periodic soot blowing. This process is an iterative method to bring NOx down to a target level by changing one control at a time and monitoring the response of the system. This process enabled them to achieve a level I emission rate of 0.49 lb/MMBTU and a level II rate of 0.42 lb/MMBTU. InEC 1M Installation at Kintigh Station The InEC1M system was installed at Kintigh station in early 1996 to begin data collection. There are over 350 sensor points which are transferred to the InEC 1M computer over a digital data link. In addition, there are three manual burner settings for each burner which are entered into the InEC1M system manually. InEC 1M AI Model Development A proprietary LMCS development process was utilized to generate the AI NOx model architecture. The resulting NOx model incorporated 182 signals out of over 450 possible signals. Some of the signals which are used by to model NOx are: • Excess 02 sensors distributed throughout the gas duct after the economizer • Feeder and mill data • Primary and secondary air flows • Burner settings • Steam temperatures The N Ox models have demonstrated an average error of approximately 5 percent. The tons/hr signal, which was utilized to develop the InEC 1M NOx models, was a CEM preprocessed 15 minute average of the instantaneous NOx reading. LMCS believes that we would have been able to obtain a lower average error if an instantaneous reading of NOx emission rate was available. Typical plots of actual NOx versus predicted NOx are shown for two days of operation in figures 4 and 5. There are several other models which are required by the InEC 1M system at Kintigh station to perform its control and optimization task effectively. A model of CO is required to ensure that the CO limit of 100 ppm is not exceeded. For a given NOx target and load condition, the CO limit may not be physically realizable. In this case, the InEC 1M system will alert the operator that the CO limit can not be met. A sample plot of the CO model output versus average excess 02, is shown in figure 6. This plot shows that the CO model has learned the characteristic knee shaped CO curve. There are also several other models used by the InEC TM system for performance optimization. One such model is a fly ash loss-on-ignition model which was developed using manual LOI samples. -5- |
