Prediction Of Nox Using Support Vector Machine For Gas Turbine Emission At Putrajaya Power Station

The monitoring of flue gas emission from power generation is mandatory in Malaysia where the conventional way of measuring the emission is using field instrumentation (in-situ or extraction type) commonly known as Continuous Monitoring Emission System (CEMS). This system requires high maintenance cost on the instrumentation, spare parts/kits and calibration gases. The purpose of this research is to study alternative way of measuring the emission gases using artificial intelligence prediction method called support vector machine. This research will concentrate on the prediction of NOx only, one of the flue gasses emitted by power generation plant. The research approach is by collecting the data from gas turbine parameters such as inlet guide vane position, compressor inlet temperature, gas control valves position, NOx reading from CEMS and many more. The data will be run for supervised learning and creating a model to predict NOx emission. The prediction models used for this study were radial basis function and polynomial. The predicted results were compared with the reading from the CEMS and the percentage average of accuracy recorded was 98.14% for radial basis function model and 96.95% using polynomial model. The result shows that the prediction method to predict NOx is reliable and accurate. The complete implementation of this software will enable the power generation to lower the maintenance cost since the maintenance of this system is only on the server/PC maintenance and yearly relative accuracy test audit (RATA) which is very minimum compare to CEMS maintenance on the field instruments, parts/kits and calibration gasses. Keyword: Gas Emission Prediction, NOx, Support Vector Machine (SVM), Continuous Emission Monitoring System (CEMS)