A Fast Fault Detection Method Based on Random Projection and k-Nearest Neighbors

Abstract

The invention discloses a quick fault detection method based on random projection and a k-nearest neighbor method, and belongs to the technical field of monitoring and diagnosis of an industrial process. The random projection and the k-nearest neighbor method are combined; by use of the advantage of distance retention of the random projection and the high performance of the k-nearest neighbor method for processing non-Gaussian, nonlinear and multi-working-condition problems of data, the industrial process is monitored. Compared with other methods in the prior art, the method disclosed by the invention has the advantages that the calculation complexity can be reduced, furthermore, the detection performance of the k-nearest neighbor method in a dimension reduction sub space can be guaranteed, and quick and accurate detection can be realized.

Description

technical field [0001] The invention belongs to the field of industrial process monitoring and fault diagnosis, in particular to a fast fault detection method based on random projection and k-nearest neighbor. Background technique [0002] For process monitoring and fault diagnosis problems, most traditional methods use multivariable statistical process monitoring technology (Multivariable Statistical Process Monitoring, MSPM), in which principal component analysis (Principal Component Analysis, PCA) and partial least squares (Partial Least Squares, PLS) The representative methods have been successfully applied in industrial process monitoring. The traditional MSPM method assumes that the process data obeys Gaussian distribution, the relationship between variables is linear and the data comes from a single operating condition, but the actual measurement data is difficult to meet these assumptions, and often presents non-Gaussian, nonlinear and multi-working conditions and o...

AI Technical Summary

Problems solved by technology

[0004] Although, PC-kNN can solve the problem of high computational complexity

However, the distance between samples after PCA dimensionality reduction cannot be guaranteed, that is, the distance between samples in the original space cannot be guaranteed in the pivot subspace, which will inevitably affect the detection performance of kNN in the pivot subspace, because kNN uses the distance between samples for fault detection

That is to say, samples that are judged to be faulty in the original space (obviously deviate from the normal sample set) may be judged as normal samples in the pivot subspace, that is, false negatives; otherwise, the samples that are judged to be normal in the original space Samples (close to the normal sample set), may be judged as abnormal samples in the principal component subspace, that is, false positives

Images