Аңдатпа
Industrial emissions of nitrogen oxides (NOx) and carbon monoxide (CO) pose significant risks to human health and climate. Conventional monitoring techniques do not enable early prediction of regulatory exceedances. This work suggests a strategy for predicting CO and NOx emissions utilizing a gas turbine unit's digital twin architecture and machine learning algorithms. Four regression models - linear, polynomial (second degree), Ridge, and Lasso - were used and contrasted. The models were trained using historical data from an industrial facility (more than 35,000 records for the years 2020–2025), including key technological parameters such as ambient conditions, compressor discharge pressure (CDP), and turbine inlet temperature (TIT). The test results showed that the polynomial model was the most accurate: the coefficient of determination R2climbed to 0.78, and the mean squared error (MSE) decreased by 22%. Sensitivity analysis revealed that CO emissions are positively impacted by compressor discharge pressure (CDP) and negatively impacted by turbine inlet temperature (TIT). Unlike existing studies based on simulated or standalone models, this work demonstrates a closed-loop digital twin prototype validated on real industrial sensor data with integrated diagnostics and sensitivity analysis. The proposed approach improves the accuracy of environmental performance assessment and can be used in proactive environmental management systems, ESG reporting, and sustainable industrial operations.