Nonlinear process monitoring method based on multi-kernel principal component analysis model

Abstract

The invention discloses a nonlinear process monitoring method based on a multi-kernel principal component analysis model, and aims to establish and fuse KPCA models corresponding to a plurality of kernel functions, thereby avoiding the problem of kernel function selection, and implementing effective nonlinear process monitoring on the basis of the KPCA models. According to the method, firstly, allcommonly-used kernel function types are considered, and the kernel function selection problem is avoided. Therefore, the method is high in universality. Secondly, a plurality of different nonlinear process monitoring models are established by using a plurality of kernel functions respectively, so that the advantage of multi-model modeling is brought into full play. In other words, the fault detection effect of the method is not weaker than that of any process monitoring model using a single kernel function. By integrating the two advantages, the method provided by the invention overcomes thedefects of the traditional KPCA-based process monitoring method, and is a more preferable nonlinear process monitoring method.

Description

technical field [0001] The invention relates to an industrial process monitoring method, in particular to a nonlinear process monitoring method based on a multi-core principal component analysis model. Background technique [0002] The large-scale and high-efficiency production of modern industrial process objects put forward higher and higher requirements for real-time monitoring of process operation status. Timely detection of fault conditions during the operation of process objects is the fundamental way to ensure product quality. It can be said that the research on process monitoring technology with fault detection as the core task has been accompanied by the process of industrial development. Today, the data-driven process monitoring method is the most mainstream implementation technology, which is mainly due to the wide application of measuring instruments and computer technology. The measurable and storable industrial process sampling data has laid a solid foundation ...

AI Technical Summary

Problems solved by technology

In the absence of historical fault data, it is impossible to select a kernel function suitable for the monitored object from the perspective of improving the fault detection effect

Even if the sampling data of some fault conditions can be provided in some process object historical databases, the types of fault conditions are countless, and the KPCA model obtained by using these limited fault data to help select the kernel function may only be aimed at these faults type, and the detection of failures that raise unknown types may be ineffective

Therefore, the choice of kernel function in the KPCA-based nonlinear process monitoring method is a problem worthy of attention, but it has not been deeply studied so far.

Images