Machine Learning-Based Prediction of the CO<sub>2</sub> Concentration in the Flue Gas and Carbon Emissions from a Waste Incineration Plant
Yifei Ma, Pinjing He, Fan Lü, Hua Zhang, Shengjun Yan, De-Biao Cao, Hongju Mao, Dan Yu Jiang
Abstract
Monitoring the CO 2 concentration in flue gas (CO 2 _G) is crucial to accurately calculate the direct carbon emissions associated with waste incineration. In this study, random forest (RF) and extreme gradient boosting (XGBoost) algorithms were used to predict CO 2 _G, using 21 operating variables from a municipal solid waste (MSW) incineration plant as input variables. The results showed a strong prediction performance for both the RF and XGBoost-based models with R 2 values of 0.932 and 0.903, respectively. A feature importance analysis identified key variables used for model retraining, resulting in R 2 values of 0.917 and 0.894, respectively. Based on the predicted and measured values of CO 2 _G and a balance calculation, the direct carbon emissions from waste incineration were determined. The emissions based on the predicted CO 2 _G value ranged from 283.38 to 348.39 kgCO 2 -eq/t, while the emission based on the measured value was 269.21 kgCO 2 -eq/t. To further validate the accuracy of the calculation results, the physical composition of MSW in the incineration plant was analyzed, resulting in a direct carbon emission estimate of 257.59 kgCO 2 -eq/t. These findings demonstrate the effective application of machine learning (ML)-based CO 2 _G predictions and overcome the labor-intensive and data-lagging aspects of carbon emission accounting in waste incineration.