基于灰狼优化支持向量机回归与SHAP值的锡冶炼能耗预测  

Energy Consumption Prediction of Tin Smelting Based on Grey Wolf Optimized Support Vector Machine Regression and SHAP Values

作　　者：马朝君 彭巨擘 袁海滨 郑光发 么长慧 章夏冰 冯早 MA Chaojun;PENG Jobo;YUAN Haibin;ZHENG Guangfa;YAO Changhui;ZHANG Xiabing;FENG Zao(Research and Development Center,Yunnan Tin Industry Group(Holding)Co.,Ltd.,Kunming 650200,China;Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,China)

机构地区：[1]云南锡业集团(控股)有限责任公司研发中心,昆明650200 [2]昆明理工大学信息工程与自动化学院,昆明650500

出　　处：《有色金属（冶炼部分）》2024年第2期1-7,共7页Nonferrous Metals（Extractive Metallurgy)

基　　金：云南省科技厅基础研究合作项目(202101BC070001-023)。

摘　　要：锡冶炼过程综合能源消耗占整个锡生产过程90%,存在很大节能潜力。针对锡冶炼过程综合能耗机理模型难以建立、导致预测准确度不高的问题,提出灰狼优化的支持向量机回归(GWO-SVR)模型用于锡冶炼过程综合能耗的预测,并以某锡冶炼厂为例,将所提模型与SVR、RF(随机森林)、BP(反向传播神经网络)、LR(线性回归)模型进行比较。结果表明,GWO-SVR模型可获得最理想的预测结果,在预测精度上相比于其他机器学习算法有着巨大优势。此外,使用SHAP值从全局解释和单样本解释两个方面解释所建立的GWO-SVR模型,可视化特征对输出的贡献,增加了GWO-SVR的可解释性,并以此制定可靠的节能策略。The comprehensive energy consumption of tin smelting process accounts for 90%of the entire tin production process,which has great energy-saving potential.Address to the difficulty in establishing the comprehensive energy consumption mechanism model of tin smelting process and the low prediction accuracy,the Gray Wolf Optimization Support Vector Machine Regression(GWO-SVR)model was proposed to predict the comprehensive energy consumption of tin smelting process.Taking a tin smelter as an example,the proposed model was compared with the SVR,RF(Random Forest),BP(Back Propagation Neural Network)and LR(Linear Regression)models.The results show that the GWO-SVR model yields the most desirable prediction results,and has great advantages over other machine learning algorithms in terms of prediction accuracy.Furthermore,using SHAP values to explain the GWO-SVR model from both global interpretation and single-sample interpretation and visualize the contribution of features to the output increases the interpretability of GWO-SVR,and thus develops a reliable energy-saving strategy.

关 键 词：锡冶炼预测模型 模型可解释性 支持向量机回归 灰狼优化算法 

分 类 号：TF814[冶金工程—有色金属冶金]