工业过程数据具有高斯和非高斯混合分布的特点。独立因子分析(IFA)采用一维高斯混合模型拟合任意的因子分布,因此可以处理高斯和非高斯混合的问题。虽然在给定因子数的前提下变分IFA算法可以有效地缩短建模时间,但是独立因子数的选择仍然需要较长的计算时间。此外,若IFA的因子数选择不当,会造成部分因子的信息遗留在观察变量的残差中,导致GSPE监控指标的监控性能变差。为了解决IFA在实际应用中存在的问题,本文结合了IFA和FA方法。首先使用FA辅助IFA选取独立因子数,以进一步减小IFA建模时间;其次使用FA对IFA的残差进行再处理,以解决由于独立因子数选择不当造成的问题。最后将该方法应用于田纳西-伊斯曼(TE)过程和乙烯裂解炉过程的监控中,实验结果验证了该联合方法的有效性。
Many industrial process variables have the characteristics of non-Gaussian mixture distribution. Independent factor analysis (IFA ) algorithm utilizes one dimension Gaussian mixture model to approximate any factor distribution such that it can address the problem of the Gaussian and non-Gaussian mixture. Although the variational IFA with given factors can reduce the modeling-time, it still takes a lot of time to determine an optimal number. Especially, an inappropriate number may make theinformation of partial factors be remained in the residuals of the observed variables,whichwill result in the poor monitoring performance on the GSPE index. Aiming at the above problem in the application of IF A % this paper proposes a combined method of IFA and F A . Firstly, FA algorithm is used in determining theindependent factor number so as to reduce the modelling time of IFA . And then,FA is further utilized to re-process the residual of IFA so that the remained partial factors’ information in residual can be fully employed. Finally, the monitoring experiment in the Tennessee-Eastman (T E ) process and the ethylene cracking furnace verifies the validity of the proposed method.