Abstract: Ceramic manufacturing as a typical energy-intensive consumption manufacturing industry,its energy saving and consumption reduction has been one of the hot issues of concern.Enterprises can find entry points for energy saving and consumption reduction through energy efficiency prediction to reduce production energy consumption and improve production energy efficiency.By analyzing the energy consumption data in the production process,an energy efficiency prediction model was established,which could accurately predict the energy consumption of the production process and provide support for the optimization of energy efficiency to realize the green manufacturing and sustainable development of high-energy-consuming industries.To address the above objectives,an energy efficiency prediction method for ceramic manufacturing based on feature-weighted Stacking ensemble learning was proposed.By analyzing the prediction performance and relevance of different models,the linear regression,extremely randomized tree,extreme gradient boosting and K nearest neighbors were identified as the base learners.Then,the different base learners were feature-weighted according to the prediction accuracy.Finally,the predictions from the weighted base learner were integrated using a light gradient boosting machine algorithm as a meta-model for prediction.The proposed method was validated through the ceramic manufacturing dataset,and the results showed that the prediction accuracy of the feature-weighted Stacking ensemble learning model was significantly higher than that of the traditional Stacking ensemble learning prediction model and the single-base learner model,which proved the validity of the proposed method,and provided theoretical support for the realization of the green manufacturing and sustainable development.

Key words: ceramic manufacturing, energy efficiency forecasting, Stacking ensemble learning model, feature weighting

摘要： 陶瓷制造作为典型的高能耗制造行业,其节能降耗一直是备受关注的热点问题之一。企业可通过能效预测找到节能降耗的切入点,从而降低生产能耗和提高生产能效。通过分析生产流程中的能耗数据,建立能效预测模型,准确预测生产过程的能源消耗,并为能效优化提供支撑,以实现高能耗产业的绿色制造与可持续发展。针对上述目标,提出基于特征加权Stacking集成学习的陶瓷制造能效预测方法,首先,通过分析不同模型的预测性能和相关性,确定线性回归、极端随机树、极限梯度提升树和k-最近邻作为基学习器;然后,根据预测精度对不同基学习器进行特征加权;最后,将加权后基学习器的预测结果进行集成,使用轻量级梯度提升算法作为元模型进行预测。基于陶瓷制造数据集,对提出的方法进行验证,结果表明:特征加权Stacking集成学习模型的预测精度要显著高于传统Stacking集成学习预测模型和单一基学习器模型,证明了所提方法的有效性,为实现绿色制造与可持续发展提供理论支撑。

关键词: 陶瓷制造, 能效预测, Stacking集成学习模型, 特征加权