基于证据理论的钢球磨煤机料位不确定性量化研究

Uncertainty quantification of coal fill level in ball mill based on evidence theory

针对钢球磨煤机具有多变量、强耦合和非线性等特性使其面临监测精度低、难度大、性能不稳定等问题.首先设计了现场参数采集实验并利用灰熵关联理论对球磨机料位及其辅助变量间的相关性进行了分析.然后从结构参数、优化准则以及参数距离出发对传统证据理论进行改进,提出了具有鲁棒性和自适应性的新型证据k-NN(Robust Adaptive Evidence k-Nearest Neighbors,RAEk-NN)分类器,再以RAEk-NN分类器构建料位的证据回归多模型,并结合非线性偏最小二乘(Nonlinear Partial Least Squares,NPLS)和支持向量机(Support Vector Machine,SVM)建立了D-S融合法则料位不确定性量化组合模型.结果表明:所提出的组合模型能够实现更精确的料位预测结果,更适应于工况多变的复杂情况,可用于实际生产过程.

The ball mill is characterized by multiple variables,strong coupling and nonlinearity,which makes it face the problems of low accuracy,high difficulty and unstable performance in monitoring.To address this,first,the field parameter acquisition experiment is designed,and the correlation between coal fill level in ball mill and its auxiliary variables is analyzed by using the grey entropy correlation theory.Then,the traditional evidence theory is improved in structural parameters,optimization criteria and parameter distances,and a new Robust and Adaptive Evidence k-Nearest Neighbors(RAEk-NN)classifier with robustness and adaptivity is proposed.Next,the RAEk-NN is used as the classifier to construct the evidence regression multiple models of material level.Meanwhile,using the Nonlinear Partial Least Squares(NPLS)and Support Vector Machine(SVM),the combination rule of the Dempster-Shafer(D-S)evidence theory is adopted to establish the uncertainty quantification combined model.The results show that the proposed combined model can achieve more accurate prediction results of coal fill level,and it is more suitable for the complex situation with variable working conditions,thus can be used in the actual production process.