Physics-Informed Predictive Emissions Monitoring for Industrial Gas Turbines: Digital Twin Modeling, Validation, and Deployment

Abstract

Operators are increasingly seeking cost-effective alternatives to Continuous Emissions Monitoring Systems (CEMS) to meet regulatory and ESG goals. This paper presents the development and validation of a Predictive Emissions Monitoring System (PEMS) for industrial gas turbines, offering a digital solution that predicts NOx and CO emissions using operational data. PEMS employs a hybrid modeling approach that combines machine learning with physics-based principles, ensuring that predictions remain grounded in combustion fundamentals. The models incorporate engine-specific parameters, including critical dimensions, fuel properties, and thermodynamic relationships and are trained and validated using statistically rigorous methods on hundreds of data points from factory and field tests. Results demonstrate prediction accuracy within ±2 ppm for NOx levels below 10 ppm and within ±20% for higher concentrations, consistent with EPA PS-16 guidance. Integrated health monitoring ensures model reliability by tracking key indicators such as combustor pressure drop, fuel valve positions, and key temperatures. PEMS is deployed via a digital platform, enabling real-time emissions monitoring, model management, and cloud/edge integration. This work demonstrates how digitalization and advanced, physics-informed modeling can provide a scalable, low-maintenance solution for emissions reporting and optimization, reducing the need for costly hardware-based monitoring systems.