Abstract:To rapidly predict changes in residual oil content after the pyrolysis of oily sludge and to guide the optimization of pyrolysis process parameters, this study collected a dataset comprising 228 samples and employed machine learning methods to predict the oil content in the oily sludge pyrolysis residues. Several factors were used as input variables, including final pyrolysis temperature, pyrolysis time, heating rate, nitrogen flow rate, initial oil content, water content, and residue content of the oily sludge. The oil content in the pyrolysis residues was used as the output variable. Methodologically, this study applied four advanced machine learning algorithms in an innovative manner: Gradient Boosting Decision Trees (GBDT), eXtreme Gradient Boosting (XGB), Support Vector Machines (SVM), and Random Forests (RF), to construct high-precision prediction models for the oil content in pyrolysis residues. These models were rigorously trained and cross-validated on a dataset comprising 228 samples to ensure their generalization ability and prediction accuracy. The results showed that the coefficients of determination (R2) for the GBDT, XGB, SVM, and RF models on the test set reached 0.8716, 0.8667, 0.8356, and 0.9171, respectively, providing initial validation of the effectiveness of these machine learning models in predicting oil content in oily sludge pyrolysis residues. To further improve the predictive performance of the models, this study introduced the Bayesian Optimization Algorithm (BOA) to fine-tune the hyperparameters of the models. After BOA optimization, the R2 values of the four models significantly increased to 0.9012, 0.9001, 0.8965, and 0.9204, respectively. Among them, the Bayesian-Optimized Random Forest (BO-RF) model exhibited the best predictive performance, demonstrating high consistency on the test set and extremely high accuracy in predicting the dynamic trends of oil content in oily sludge pyrolysis residues. Furthermore, through feature importance analysis, it was found that the final pyrolysis temperature, initial oil content in the sludge, and pyrolysis duration were the most critical factors influencing the oil content in the residues. In summary, by introducing advanced machine learning algorithms combined with a Bayesian optimization strategy, this study successfully constructed high-precision prediction models for the oil content in oily sludge pyrolysis residues. The BO-RF model, in particular, offers an effective and accurate approach for predicting oil content. This achievement contributes to enhancing the pyrolysis process of oily sludge, boosting resource utilization efficiency, and advancing sustainable waste treatment methods. It provides strong support for the pyrolysis treatment of oily sludge at both theoretical and practical levels, opening up new perspectives and approaches for environmental management and resource recovery research. Close-