Biochar production via pyrolysis of various organic waste has potential to reduce dependence on conventional energy sources and mitigate global warming potential. Existing models for predicting biochar yield and compositions are computationally-demanding, complex, and have low accuracy for extrapolative scenarios. Here, two data-driven machine learning models based on Multi-Layer Perceptron Neural Network and Artificial Neuro-Fuzzy Inference System are developed. The data-driven models predict biochar yield and compositions for a variety of input feedstock compositions and pyrolysis process conditions. Feature importance assessment of the input dataset revealed their competitive significance for predicting biochar yield and compositions. Overall, the predictive accuracy of the models was up to 12.7% better than the Random Forest and eXtreme Gradient Boosting machine learning algorithms reported in the literature. The models developed are valuable for environmental footprint assessment of biochar production and rapid system optimization.

通过各种有机废物的热解生产生物炭有可能减少对传统能源的依赖并减轻全球变暖的可能性。用于预测生物炭产量和成分的现有模型计算要求高、复杂，并且对于外推场景的准确性低。在这里，开发了两个基于多层感知器神经网络和人工神经模糊推理系统的数据驱动机器学习模型。数据驱动模型预测各种输入原料成分和热解过程条件的生物炭产量和成分。输入数据集的特征重要性评估揭示了它们在预测生物炭产量和成分方面的竞争意义。总体而言，模型的预测准确度高达 12。比文献中报道的 Random Forest 和 eXtreme Gradient Boosting 机器学习算法好 7%。开发的模型对于生物炭生产的环境足迹评估和快速系统优化很有价值。