Evaluating GRNN, Decision Tree, and Random Forest: A Gas Turbine Emission Prediction Comparative Study

The exhaust CO, CO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf>, O<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf>, SO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf>, and NO gas emissions of a gas turbine-powered compressor unit under specific operating conditions are studied. Due to the high costs of hardware, maintenance, and calibration, Predictive Emissions Monitoring Systems (PEMS) is a more potential alternative to traditional CEMS for monitoring gas turbine emissions. PEMS provides a cost-effective and precise solution to traditional hardware-based emissions monitoring by employing algorithms to predict emissions. Market PEMS models use empirical approach modeling with limited data. This research explores Decision Trees and Random Forest for a new model that can handle more data of multiple inputs and output and compares its performance to the GRNN modelling approach. This study analyzed one million data points on gas turbine emissions (collected from 2021 to 2023) and found Random Forest to be the most accurate prediction method, while Decision Tree offers a good balance for smaller datasets, and Generalized Regression Neural Network (GRNN) is best for simpler data.