EKLPP (Ensemble Kernel Locality Preserving Projections) based fault detection method for continuous process

Abstract

The invention relates to an EKLPP based fault detection method for a continuous process, and belongs to a kernel parameter selection based fault detection method for the continuous process. Historical data in normal states serves as a training set of modeling data, a Gaussian kernel function carries out kernel transformation on the modeling data, and an EKLPP method is used to carry out modeling of a nonlinear continuous process and fault detection. A training data matrix is standardized. A KLPP method is used to convert the training data matrix into a kernel matrix to replace data to be detected, a Bayesian decision is used to convert detection results into the form of probabilities of faults, and whether data in the moment is normal is determined according to whether an integrated statistical quantity exceeds a control limit. If the integrated statistical quantity exceeds the control limit, the data in the moment has faults. If test shows that a system has faults, workers need to find out the condition and eliminate risks. The method of the invention can be used to solve the problem that suitable parameters of different faults are different.

Description

technical field [0001] The invention relates to a non-linear process fault detection method, in particular to a partial fault detection method dealing with different fault core parameter selections. Background technique [0002] As one of the important fields of industrial production, the chemical production process is increasingly showing nonlinear characteristics. The traditional global fault detection method can only retain the global information of the original data and destroy its local structure. How to effectively extract the faults in the production process? The local information of the original data to monitor the nonlinear process has become an important content of fault detection technology research. [0003] The kernel locality preserving projection method has obvious advantages in dealing with the problem of preserving the local structure of nonlinear data. However, people usually use the Gaussian kernel function when processing data with the traditional kernel...

AI Technical Summary

Problems solved by technology

However, the faults in the production process are usually not unique, and the kernel parameters selected from this are not necessarily applicable to all faults. For different faults, the applicable parameters will be significantly different.

Images