Development of Nitrogen Oxides (NOx) Emission Prediction Model of Nam Phong Power Plant with Machine Learning

Abstract

Nitrogen Oxides (NOx) are harmful gases to human health and the environment. These emissions primarily result from fuel combustion in engines and industrial processes. To meet regulatory requirements, the Nam Phong Power Plant in Thailand has implemented Continuous Emission Monitoring Systems (CEMS) to measure and report NOx emissions to regulatory authorities. However, considering the high costs associated with installing and maintaining CEMS, as well as recent changes in Thai legislation allowing for predictive NOx measurement methods, it is worth exploring the use of Machine Learning as a reliable method for estimating NOx emissions accurately. In this study, a comprehensive comparison was conducted on six Machine Learning algorithms: Linear Regression, Decision Tree, Random Forest, XGBoost, K-Nearest Neighbors, and Backpropagation Multilayer Perceptron Neural Network. Among these models, Random Forest emerged as the top performer, exhibiting superior performance metrics, including the lowest MAE, MAPE, and the highest R² scores. These results underscore the potential accuracy and reliability of Random Forest in predicting NOx emissions. Furthermore, research on feature importance has revealed the significant influence of certain parameters on model accuracy. These parameters include steam injection flow, steam injection temperature, and ambient conditions. The influence of controllable factors, such as the temperature of steam injection, on NOx emissions is noteworthy. These findings not only hold promise for enhancing the precision of predictive models but also present opportunities to decrease NOx emission levels while maintaining plant efficiency.