Machine learning-aided model for predicting oily sludge pyrolysis under various feedstock and operating conditions
Cheng Lu, Dixuan Li, Beidou Xi, Guangji Hu, Jianbing Li
Abstract
Oily sludge pyrolysis technology has the advantages of potential recovery of valuable resources and safe disposal of non-recoverable residues. However, experimentally determining the optimal pyrolysis operating conditions is time-consuming and expensive. In this study, a machine learning (ML) approach was developed to predict and optimize the oily sludge pyrolysis process. Among the six machine learning models, eXtreme Gradient Boosting (XGB) was found to have the best prediction results. A multi-task XGB model was then developed with oily sludge ultimate and proximate composition and pyrolysis operating conditions as the modeling inputs. The modeling results indicated that the sludge ash and hydrogen contents as well as the pyrolysis temperature are the most critical factors affecting pyrolysis process and its performance. The contribution of sludge ultimate composition to the pyrolysis performance is 42.5 %, followed by sludge proximate properties (35.8 %) and pyrolysis operating conditions (21.7 %). The multi-task XGB ML model achieved an average R 2 of 0.90 through model verification. The ML-aided modeling approach provides new insights for understanding and optimizing the oily sludge pyrolysis. • Machine learning models were developed to predict oily sludge pyrolysis products. • XGB showed optimal performance (test R 2 of 0.93–0.94) for single-/multi-task models. • Pyrolysis is most affected by sludge ash and hydrogen content and pyrolysis temperature. • The Multi-task XGB model can be used to optimize the pyrolysis oil and gas yield. • Model validation was conducted to verify the accuracy of multi-task XGB modeling.