基于综合建模方法的铅锌烧结块成分预测

Prediction for Composition of PbZn Agglomerate Based on Integrated Modeling

针对铅锌烧结过程异常复杂的实际情况,提出了一种既可保证预测精度又满足配料计算对数据完备性要求的铅锌烧结块成分预测智能集成模型。该模型综合了机理与多种智能建模方法的优点,对于正常生产情况(即数据完备区),通过模糊分类/组合以及神经网络NN分段描述方法建立了成分预测的监督式分布神经网络模型;对于异常或不常用工况(即数据不完备区),通过专家经验规则修正部分假定或统计参数方式建立经验机理模型;采用串、并联形式将2种模型有机结合,并通过专家推理进行集成协调与更新修正,形成智能集成模型,实现成分可靠、准确的在线预测。在实际生产中运用该模型,烧结块铅、锌成分预测的相对误差分别为1.51%和0.41%。

Considering the complexity in Pb-Zn sintering process, an intelligent integrated modeling is presented to assure the composition prediction precision of Pb-Zn agglomerate and to meet the requirements of the data completeness by blending computation, which synthesizes the advantages of mechanism modeling and other intelligent methods. Firstly, under normal production conditions (or on the complete data section), a supervised distributed neural networks (SDNN) is established by fuzzy classification/combination and the description method of different neural networks (NN) on different subspace; then, under abnormal/unusual production conditions (or on the incomplete data section), an expertise and mechanism based (EM) model is built, in which some hypothetical or statistical parameters are modified by empirical rules; lastly, the EM model and the SDNN are integrated in series and parallel structure, and the integrated model is coordinated and updated by use of expert reasoning to realize the on-line prediction of composition. The intelligent integrated model is applied to industrial process, and the relative errors of prediction for lead and zinc are 1.51% and 0.41%, respectively.