【摘要】 文章利用从文献中收集的305组水热炭的基础特性数据，采用决策树、随机森林、梯度提升树3种机器学习算法建立水热炭基础特性的单任务和多任务预测模型，并利用SHAP法研究输入特征参数对水热炭基础特性影响的差异。结果表明：在3种机器学习算法中，梯度提升树模型在水热炭基础特性的多任务和单任务预测过程中，均体现出最高的准确性，测试集的平均相关系数分别为0.89和0.87，均方根误差分别为0.34和0.37；通过SHAP法对梯度提升树模型的输入特征参数进行评价，发现水热反应温度和原料中C元素含量是影响水热炭产率、高位热值和C元素含量的最主要参数。通过构建水热炭基础特性的预测模型，有利于优化水热炭制备工艺，降低实验成本，提高水热炭制备工艺的经济效益。更多还原

【Abstract】 In this work, 305 sets of data of hydrochar’s basic properties was collected from the references. Then, the single-task and multi-task prediction models of hydrochar’s basic properties(mass yield, higher heating value, and carbon content) were established based on three types of the machine learning algorithms(the decision tree, the random forest, and the gradient boosting decision tree). Results showed that among the three types of the machine learning algorithms, the gradient boosting decision tree model was the best algorithm, where the average determination coefficient values of the test set were 0.88 and 0.87, and the root mean square error values were0.34 and 0.37. The SHAP method was used to evaluate the input characteristic parameters during the modeling by using the gradient boosting decision tree. The dominant influence factors for the prediction of the mass yield, higher heating value, and carbon content of the hydrochar were the hydrothermal reaction temperature and the C content in raw biomass. The construction of the prediction model of the hydrochar’s basic properties was favorable to reduce the cost for the optimization of the hydrochar production conditions.更多还原