Nonlinear process fault detection method based on differential locality preserving projection (DLPP)

Abstract

The invention discloses a nonlinear process fault detection method based on differential locality preserving projection (DLPP) and relates to a nonlinear process fault detection method. Batch data acquired in a normal production process is taken as a training set of modeling data, three-dimensional data is expanded to be two-dimensional, a nearest neighbor of each sample in a data set is found, and differential operation is performed on the samples and their nearest neighbors; a locality preserving projection model is established, a projection matrix is found, and a control limit of a square prediction error (SPE) is calculated through kernel density estimation; for a batch sample of a new time point k, data is expanded to form a row of vectors; the data of the new time point is projected to a model of the DLPP, and an SPE statistical amount is calculated; according to whether the statistical amount exceeds the control limit of modeling, whether the data of the time point is normal is determined; if the statistical amount exceeds the control limit, there are faults in data samples of the time point; and otherwise, the data samples are normal. When a test indicates that the faults occur in a system, work personnel needs to examine conditions timely so as to eliminate dangers.

Description

technical field [0001] The invention relates to a nonlinear process fault detection method, in particular to a nonlinear process fault detection method based on differential local preservation projection. Background technique [0002] Nonlinear processes exist in real industrial processes and are widely used in the production of high-quality, high value-added products, such as: machinery, papermaking, metallurgy, food processing, etc., and are an important industrial production process. Therefore, fault detection in nonlinear processes has been receiving much attention. Nonlinear data structures often play an important role in fault detection. Therefore, this puts high demands on the performance of nonlinear process fault detection. [0003] The difference algorithm can effectively eliminate the nonlinear structure of the data, and at the same time maintain the internal structure of the data, optimize the fault detection of the nonlinear process, so as to ensure production...

AI Technical Summary

Problems solved by technology

These methods reduce the data from a high-dimensional space to a low-dimensional space by retaining the principal components, changing the data distribution, but the relationship between the data in the initial feature space and the corresponding data manifold cannot be maintained after this global transformation, so in the new Some information about the data in the initial feature space will definitely be lost in the space

How to ensure that local information is not lost, there is currently no effective method to find more intrinsic useful information hidden in high-dimensional observation sets

Images