Prediction of CO 2 mole fraction via CO 2 CPU process using different machine learning algorithms

Abstract

Abstract There is a general industrial procedure called compression and refining unit to catch CO2 from the flue gases produced during oxyfuel combustion. This research discusses the application of decision trees, Adaboosting, random forests, machines that support vectors, and k-nearest neighbor classifiers and gradient boosting in predicting CO2’s mole fracion from flue gases of oxyfuel’s combustion emitted from the power plant. First of all, a training and test dataset was developed using the different variables. Then, a total of 491 simulations were performed and the mole fraction of CO2 was examined. The anticipated outcomes suggested that six machine learning algorithms that rank performance from excellent to poor, RF, GB, AB, DT, KNN, and SVM can be picked to forecast the mole fraction of CO2. Important features were detected by SHAP and the best algorithm was chosen by cross-validation. Results were shown that The RF algorithm enjoyed a great CO2 mole fraction ability to predict and displayed the very best ability for generalization and most reliable prediction precision among all four with an accuracy of 97%. After that LIME was used to explain the results of the RF algorithm. Out of the various variables studied, the pressure of the multistage compressor had the highest effect on the CO2 mole fraction. These results show that machine learning can be used as a reliable predictor of CO2 performance capture within the CPU process.