基于机器学习的粉煤灰活性分类预测  

Reactivity classification prediction of coal fly ash based on machine learning

作　　者：胡涛 武梦婷 胡巍 陈秋松[1] 齐冲冲 HU Tao;WU Mengting;HU Wei;CHEN Qiusong;QI Chongchong(School of Resources and Safety Engineering,Central South University,Changsha 410083,China)

机构地区：[1]中南大学资源与安全工程学院,湖南长沙410083

出　　处：《中南大学学报（自然科学版）》2023年第10期3829-3839,共11页Journal of Central South University:Science and Technology

基　　金：国家自然科学基金资助项目(52004330)。

摘　　要：针对粉煤灰活性快速分类问题,基于随机森林融合数据集分析、参数调整等方法,建立粉煤灰(CFA)活性智能分类模型。利用准确率、召回率、精确率和ROC曲线下面积SAUC这4种评估指标对模型进行评估。此外,使用特征重要性、部分依赖图(PDP)和机器学习解释(SHAP)模型3种方法来衡量特征物质的贡献。研究结果表明:模型的准确率为85.45%,召回率为97.56%,精确率为84.29%,SAUC为0.92。K_(2)O、Fe_(2)O_(3)、Na_(2)O和Al_(2)O_(3)对SHAP模型影响较大。随着K_(2)O、Na_(2)O等特征物质占比增加,粉煤灰呈现高活性的概率增加;而随着Al_(2)O_(3)等特征物质占比增加,粉煤灰呈现高活性的概率降低。所建立的模型可快速划分粉煤灰活性并判断其是否具有作为辅助胶凝材料的潜力。Aiming at the problem of rapid classification of coal fly ash reactivity,the coal fly ash(CFA)reactivity prediction model was established based on random forest fusion dataset analysis and parameter tuning method.The accuracy,recall,precision and area under curve(SAUC)were used to evaluate model performance.Furthermore,feature importance,partial dependence plot(PDP)and SHapley Additive exPlanations(SHAP)were used to measure the contribution of characteristic component.The results show that the accuracy of the model is 85.45%,the recall is 97.56%,the precision is 84.29%,and the SAUC is 0.92.The K_(2)O,Fe_(2)O_(3),Na_(2)O and Al_(2)O_(3)have strong influences on the SHAP model.The probability of coal fly ash showing high reactivity increases as the percentage of characteristic compound such as K_(2)O and Na_(2)O increases while the probability of coal fly ash showing high reactivity decreases as the percentage of characteristic compound such as Al_(2)O_(3)increases.The established model can quickly divide the coal fly ash reactivity and determine whether it has the potential as supplementary cementitious materials or not.

关 键 词：粉煤灰活性 随机森林 特征重要性 部分依赖图 SHAP 

分 类 号：X752[环境科学与工程—环境工程]