Abstract:

Many industrial process variables have characteristics of high-dimension and not strictly obeying the Gaussian distribution. A method was proposed to solve these problems of the industrial process. The method was based on LTSA algorithm and combined index to improve monitoring performance. Firstly, the local tangent space alignment (LTSA) algorithm was used to get the sub-manifold of low dimension from the normal sample data to achieve dimensionality reduction. The two-step strategy was used to get the new statistical model. Then, the non-Gaussian statistical value and the Gaussian statistical value were constructed. Therefore, the new statistical value weighted by these two statistical values was intended to achieve monitoring of the process. Finally, the proposed method was applied to the Tennessee Eastman (TE) process and the ethylene cracking furnace to demonstrate its effectiveness.

Key words: algorithm, integration, process control, non-linearity, non-Gaussian, combined index, local tangent space alignment algorithm, Tennessee Eastman process

摘要：

很多实际工业过程数据都具有高维、非线性且不严格服从高斯分布等特点。为处理数据维数高且是高斯分布和非高斯分布的混合体等问题,实现高效的过程监控,提出了一种基于LTSA和联合指标的非高斯过程监控方法。首先采用局部切空间排列(LTSA)算法从正常样本数据中提取低维子流形以实现维数约减;然后基于非高斯-高斯两步策略建立统计模型并得到非高斯统计量和高斯统计量,再对其进行加权得到新的统计量以实现对过程的监控;最后将该方法应用于田纳西-伊斯曼标准测试平台和实际乙烯裂解炉的过程监控,说明了所提方法的有效性。

关键词: 算法, 集成, 系统工程, 非线性, 非高斯, 联合指标, 局部切空间排列算法, 田纳西-伊斯曼过程