Hybrid GRNN–SVR Model for Predicting CO Emissions in Cement Manufacturing

The goal of this study is to forecast carbon monoxide (CO) emissions from the MASS Cement Industry in Sulaymaniyah, Kurdistan–Iraq, based on daily operational data obtained from January 1 up to December 31, 2023. A total of five explanatory variables were included: the amounts of limestone (LS, tones per hour; t/h), clay (t/h), iron (t/h), sand (t/h), and heavy fuel oil (HFO, liters per hour; L/h) consumed. The response variable was CO emissions, measured in milligrams per normal cubic meter (mg/Nm³). The study included three models- Generalized Regression Neural Network (GRNN), support vector regression (SVR), and Hybrid GRNN-SVR (the combination of both methods). The results established strong relationships between the selected inputs and CO emissions. Of these raw materials, limestone, clay, and iron oxide are critical in clinker production and thus greatly affected emission levels through combustion and chemical reactions, whereas HFO consumption was directly associated with fuel efficiency and emission intensity. Among the tested models, the hybrid GRNN-SVR achieved the highest predictive accuracy with lower RMSE(0.9661) and MAE(0.6241) over individual GRNN and SVR models. It is concluded that integrating GRNN with SVR offers a more robust alternative for predicting and monitoring CO emissions from cement manufacturing processes. Future work should consider including additional environmental factors (e.g., temperature and humidity) and optimizing fuel consumption, particularly HFO, to further reduce CO emissions and enhance operational sustainability.