Abstract:

A performance assessment method based on multi-space nonlinear iterative partial least squares(Ms-NIPLS) and Gaussian process regression(GPG) is proposed to solve the insufficient accuracy caused by the non-linear relationship between input and output data of chemical process. First, this method divides the similar historical datasets into different sets of performance grades, extracting the feature subspaces of training datasets by Ms-NIPLS method, and then using GPR to obtain a non-linear mapping structure between the feature subspaces and the performance grades labels to achieve off-line modeling. With the model obtained, current performance can be assessed online, and the transition performance coefficient is constructed to distinguish the steady-state performance grades and the transition performance states between the steady-state performance grades. Finally, the method in this paper is applied to the online performance assessment of ethylene cracking process to illustrate the effectiveness and accuracy of the performance assessment method proposed.

Key words: multi-space, nonlinear iterative partial least squares, Gaussian process regression, model, steady state, transition, performance assessment

摘要：

针对化工过程中因输入输出数据间非线性关系造成在线性能评估准确度不足的问题，提出一种基于多数据空间非线性迭代偏最小二乘和高斯过程回归（multi-space nonlinear iterative partial least squares and Gaussian process regression，Ms-NIPLS-GPR）的性能分级评估方法。首先将性能相近的过程历史数据划分成不同性能等级的集合，利用Ms-NIPLS方法提取不同性能等级训练数据的特征子空间，然后用GPR获得特征子空间与性能等级标签之间的非线性映射结构，建立输入数据与性能等级之间的离线模型。得到模型后，在线评估当前过程性能等级，同时通过构造过渡性能系数来区分稳态性能等级和稳态性能等级间的过渡性能状态。最后，将该方法应用到乙烯裂解过程在线性能评估中，说明该性能评估方法的有效性和准确性。

关键词: 多数据空间, 非线性迭代偏最小二乘, 高斯过程回归, 模型, 稳态, 过渡, 性能评估