A comparative data mining approach for the prediction of energy recovery potential from various municipal solid waste

Knowledge of higher heating value (HHV) is critically important in the techno-economic analysis and optimal performance of thermal waste-to-energy (WTE) systems. In this work, we proposed a rapid and cost-effective method to accurately estimate the high heating value for various municipal solid wastes using a novel machine learning technique and the ultimate analysis. To this end, we relied on a comprehensive experimental HHV dataset from a diverse range of geographical origins and eight various waste classifications (i.e., 30 paper, 12 textile, 12 rubber and leather, 29 MSW mixture, 34 plastic, 61 wood, 20 sewage sludge, and 53 other wastes). In particular, we developed nonlinear HHV equations for fast, accurate, straightforward applications (e.g., especially when limited elemental information is available) using robust gene expression programming (GEP) algorithm. Meanwhile, we analyzed the accuracy gains from the application of more complex and nonlinear models such as a novel support vector machine (RBAS-SVM) and a feed-forward neural network (FFNN). While coefficient of determination (R2) of the time-saving and straightforward GEP equations is 0.966, the SVM and FFNN models have 0.973 and 0.978 R2. Besides, root mean square error (RMSE) of GEP, SVM, and FFNN models are 1.57, 1.44, and 1.25, respectively. The results of statistical performances for the entire dataset showed that average absolute error (AAE) of the proposed models in a range of 0.87–1.12%. Finally, the presented models were compared to those developed by other authors regarding the comprehensiveness (dataset size and diversity), accuracy, and validity (statistical performances) for the entire dataset. •A rapid and cost-effective approach for accurate modeling of HHV is proposed.•252 experimental MSW data from a diverse geographical origins are utilized.•Our proposed HHV models outperform the accuracy of the previous studies.