文献类型：期刊文献

英文题名：Multi-objective interpretation machine learning framework for Synergistically optimizing energy consumption and surface quality in electrolytic copper foil production

作者：Yang, Zhengwu[1,3];Zhang, Xinwan[1,3];Hu, Huawei[1,3];Shi, Yaqi[1,3];Xu, Wenxuan[1,2,3];Huang, Huiting[1,3];Zhang, Wenzhuo[1,3];Fu, Dan[4];Lu, Zhihao[1,3];Jia, Daqing[1,3];Chang, Dingming[1,3];Zhang, Lehua[1,2,3,4]

机构：[1]East China Univ Sci & Technol, Natl Engn Res Ctr Ind Wastewater Detoxicat & Resou, Shanghai 200237, Peoples R China;[2]Shihezi Univ, Sch Chem & Chem Engn, Shihezi 832003, Xinjiang, Peoples R China;[3]East China Univ Sci & Technol, State Environm Protect Key Lab Environm Risk Asses, Shanghai 200237, Peoples R China;[4]Shanghai Engn Res Ctr Heavy Met Pollut Control & R, Shanghai 200031, Peoples R China

年份：2025

卷号：525

外文期刊名：JOURNAL OF CLEANER PRODUCTION

收录：;EI(收录号：20253719161879);WOS:【SCI-EXPANDED(收录号:WOS:001573374400001)】；

基金：This work was supported by the National Natural Science Foundation of China (NSFC) (22478119) , the Jixi graphite industry unveiling sci-entific and technological research project (JKJB2023H03) , and the Xinjiang Tianchi Talent Introduction Project.

语种：英文

外文关键词：Electrolytic copper foil; Non-dominated sorted genetic algorithm-II; SHapley additive exPlanations; Multi-objective optimization; Interpretability

摘要：The rapid growth of the electronic chip industry, fueled by the rise of artificial intelligence, is driving increasing demand for high-quality electrolytic copper foil (ECF). A major challenge for ECF production is achieving highquality output in a cost-effective and low-carbon manner. In this study, a Composite-Feature-Multi-ObjectiveInterpretation (CFMI) framework based on machine learning (ML) was developed to simultaneously optimize energy consumption and CF surface roughness during ECF production. The results show that the eXtreme Gradient Boosting (XGBoost) model outperforms the other seven ML algorithms in predicting both energy consumption (RMSE = 50.39 kWh.t(-1) , R-2 = 0.93) and CF surface roughness (RMSE = 0.30 mu m, R-2 = 0.73). SHapley Additive exPlanations (SHAP) analysis identifies current density, temperature, and sulfuric acid con-centration as primary factors influencing energy consumption, while deposition time, current density, and copper ion predominantly affect CF surface roughness. Moreover, SHAP-based feature clustering effectively reduces dimensionality while maintaining model accuracy, with XGBoost showing less than 1 parts per thousand R-2 degradation. Furthermore, the proposed multi-objective Non-dominated Sorting Genetic Algorithm-II (NSGA-II) outperforms the traditional Response Surface Methodology (RSM), achieving 59.63 kWh.t(-1) energy savings (equivalent to a 33.63 kg CO2.t(-1) reduction) at a target CF roughness of 0.60 mu m. The Pareto front further confirms the excellent dual-optimization capability of NSGA-II, which achieved a 10.15 % energy reduction while maintaining product quality. This study provides new insights for intelligent copper foil production and enables greener manufacturing paradigms.