A system and method for online update of ocsvm monitoring model

Abstract

The invention relates to an on-line updating system and method for an OCSVM monitoring model. The on-line updating system for the OCSVM monitoring model is characterized in that the system comprises a data collecting device and a monitoring model on-line updating module, a sample standardized module, a judging module, an out-bound sample updating module and an in-bound sample updating module are built in the monitoring model on-line updating module; the data collecting device digitalizes collected real-time process variables into real-time process data and then transmits the real-time process data to a sample standardization sub-model, and the sample standardization sub-model transmits the current OCSVM monitoring model and a novel standardized sample to a judgment sub module; the judgment sub module transmits the judged out-bound sample, the judged in-bound sample and the current OCSVM monitoring model respectively to the out-bound sample updating sub-module and the in-bound sample updating sub-module, and the OCSVM monitoring models output by the out-bound sample updating sub-module and the in-bound sample updating sub-module are both used as current monitoring models. The on-line updating system and method for the OCSVM monitoring model can be widely applied to the on-line update of non-linear process monitoring models in the practical industrial process.

Description

technical field [0001] The invention relates to an online updating system and method of a nonlinear model, in particular to an online updating system and method of an OCSVM (One-Class Support Vector Machine, single-class support vector machine) monitoring model. Background technique [0002] OCSVM is a support vector machine method that only uses a single class sample for modeling during training. This method projects the sample points into the feature space, and uses a tangent plane to separate the sample projection points from the origin of the feature space at the maximum interval, so that the area where most samples are located in the original space and the area without sample distribution are separated by a decision boundary. Realize the detection of abnormal sample points. Therefore, OCSVM can be used to solve the problem of complex production process online monitoring with sensor data having non-Gaussian distribution and nonlinear correlation characteristics. The di...

AI Technical Summary

Problems solved by technology

At present, the LASVM method can only give the approximate solution of the monitoring model during the online update process, and the calculation accuracy of the model is not high.

It can be seen that the existing OCSVM monitoring model online update method has shortcomings such as inaccurate monitoring effect or high computational complexity.

Images