Single-class support vector machine kernel parameter optimization method based on sample edge points and internal points

Abstract

The invention provides a single-class support vector machine kernel parameter optimization method based on sample edge points and internal points. According to the method, normalization processing ona target type data set sample is carried out; according to each sample point and the geometric relationship of a sample adjacent to the each sample point, edge points of the sample and internal pointsare selected; the nearest neighbor and the farthest neighbor of each edge point and the internal point in a specified sample set are searched; according to the spatial distribution condition of the sample, a standby set of kernel parameters is determined; through each parameter value of the kernel parameter standby set, the corresponding Gaussian kernel function is constructed; an appropriatenessindex of nuclear parameters is calculated, and a parameter value corresponding to the maximum value of the appropriateness index is taken as the optimal kernel parameter value. The method is advantaged in that automatic optimization of the single-class support vector machine nucleus parameters can be realized, training a single-class support vector machine is not needed in the optimization process, classification accuracy of the single-class support vector machine is made to be high through the optimized parameters, and the method has wide application prospects in the fault detection and newand abnormal point detection field.

Description

technical field [0001] The invention relates to a parameter optimization method, in particular to a single-class support vector machine kernel parameter optimization method based on internal points of sample edge points. Background technique [0002] Fault detection detects faults in the production process in time by monitoring various variables in the production process of the product to ensure product quality. In the actual production process, most of them are normal samples, and faulty samples usually mean economic losses, so they are difficult to obtain and the number is very rare. Moreover, these small number of fault samples are only from some faults, they are not representative and cannot cover all fault sample areas. Establishing a binary classification model with such a large difference in the number of normal samples and fault samples will lead to deviations in the model and cannot accurately detect faults. In response to this situation, the researchers proposed ...

AI Technical Summary

Problems solved by technology

Although this type of method has a small amount of calculation, it does not consider the geometric relationship between samples, resulting in poor performance of the optimized parameters.

Images