Study on Modeling and Prediction of NOx Emissions in Coal-Fired Boiler in a Standard 500MW Power Plant

Abstract

Due to strict restrictions on coal-fired power plants by environmental regulation authorities to limit the amount of oxides of nitrogen (NOx) emissions, NOx prediction and control practices in industries have become common. Therefore, it is important to develop techniques to address the underlying issues arising from the utilization of pulverized coal in utility boilers to assist the power plant designers and operators to minimize harmful emissions from the stack and run the operation cleanly and efficiently. During the coal combustion process, NOx are major pollutants. The indirect enhancement of the greenhouse effect, depletion of stratospheric ozone, photochemical smog and acid rain are some of the adverse effects of NOx. In coal-fired power plants, NOx emission should be measured accurately and reliably and reduced in order to ensure the compliance of plant emissions with stringent emission limits imposed by government environmental bodies, while keeping the operation economically optimized and secure. This study has been divided into two major parts, (1) offline modeling and prediction, (2) online modeling and prediction. Offline Modeling and Prediction: For precise and reliable prediction of NOx emissions from tangentially fired pulverized coal boiler using Least Squares Support Vector Machines (LSSVM), it is crucial to tune the hyper parameters of LSSVM. This work presents a novel Teaching-Learning-Self-Study-Optimization (TLSO) algorithm with its application on tuning of hyper-parameters of LSSVM to predict the NOx emissions from a tangentially fired pulverized coal boiler in a standard 500MW power plant in South Korea. The original teaching-learning-based optimization (TLBO) gives uniformly distributed and randomly selected weight to the amount of knowledge to a learner at each phase, i.e., teacher phase and learner phase. This uniformly distributed and randomly selected weight causes the algorithm to converge the average cost of learners in a moderate number of iterations. Li and his coworkers intensified the teacher and learner phase by introducing weight-parameters in order to improve the convergence speed in term of iterations in 2013 and called it ameliorated teaching-learning-based optimization (ATLBO). Whereas the criterion of a good evolutionary optimization algorithm is to be consistent in converging the cost of objective function. For this, it should include intensification for local search as well as diversification for global search in order to reduce the chances of trapping in local minimum. Some students naturally tend to study by themselves by the means of library and internet academic resources in order to enhance their knowledge. This phenomenon is termed as self-study and is introduced in proposed TLSO’s learner phase as a diversification factor (DF). Various other evolutionary algorithms such as ant colony optimization (ACO), particle swarm optimization (PSO), teaching-learning-based-optimization (TLBO), ameliorated teaching-learning-based-optimization (ATLBO) and two variants of TLSO are also developed and compared with proposed TLSO in term of consistency to converge to global minimum. Results reveal that the TLSO was found consistent not only for a higher number of functions among 20 benchmark functions but also for NOx prediction application. Results also show that the predicted NOx emissions through LSSVM tuned with TLSO keeps good track of measured NOx and maintains generalization accuracy of NOx emissions prediction. Online Modeling and Prediction: The deviation in process environment with time may change the values of some of the process variables. This drift in process variables may deteriorate the reliability of the prediction of the developed model. Also, the accurate and reliable real-time estimation of NOx emissions is indispensable for the implementation of successful prediction and control of NOx emission from a coal-fired power plant. To tackle with this issue, in the second part of the study, an online approach is proposed to utilize LSSVM and output bias in a novel real-time fashion. This work proposes a real-time update scheme using LSSVM to build a real-time version for real-time prediction of NOx emissions. Incorporation of LSSVM in the update scheme enhances its generalization ability for long-term predictions. The proposed real-time model based on LSSVM (LSSVM-scheme) is applied to NOx emission process data from a tangentially fired pulverized coal boiler in a standard 500MW power plant in South Korea to compare the prediction performance of NOx emission with real-time model based on partial least squares (PLS-scheme). Prediction results reveal that LSSVM-scheme predicts robustly for a long passage of time with higher accuracy in comparison with PLS-scheme.