The construction of a process to produce bio-oil from biomass pyrolysis, as well as optimizing and controlling its operation, requires accurate prediction of yield under varying process conditions and feedstock properties. The existing models often fail to capture the complex relationship between bio-oil yield and feedstock properties and operating parameters. This study applies three well-known machine learning (ML) classes, i.e., adaptive neuro-fuzzy inference systems (ANFIS), artificial neural networks, and least-squares support vector regression to predict the bio-oil yield. 419 sets of experimental measurements about the achievable bio-oil yield from 40 biomass types at a wide range of pyrolysis temperature, heating rate, residence time, and gas flow rate are used for training these intelligent models and monitoring the reliability of their simulation performance. The relevancy test approved that the gas flow rate and heating rate, with the Pearson correlation coefficients of 0.392 and − 0.202, have the highest impact on the bio-oil yield. The statistical accuracy monitoring of the ML models confirmed that the ANFIS model outperformed all alternatives, achieving the mean absolute error (MAE), root mean square error (RMSE), absolute average relative deviation (AARD), and correlation coefficient (R) of 2.18, 3.69, 6.45 %, and 0.95541, respectively. This outstanding simulation performance of the ANFIS model is related to its hybrid architecture that integrates interpretable fuzzy rules with artificial neural network adaptability. The applicability domain investigation identifies seven outliers and one out-of-leverage sample among the experimental databank.
扫码关注我们
求助内容:
应助结果提醒方式:
