Optimized kernel principal component analysis fault monitoring method based on chaotic particle swarm

Abstract

The invention discloses an optimization kernel principal component analysis fault monitoring method based on a chaotic particle swarm. The method comprises the steps of obtaining an initial data matrix; wherein the initial data matrix comprises a normal data sample matrix and a fault data sample matrix; mapping the initial data space into an implicit feature space through nonlinear mapping, and performing nonlinear feature transformation in the implicit feature space; establishing a fault monitoring model by taking the initial data matrix as training data; acquiring test data; and inputting the test data into the fault monitoring model, and carrying out fault on-line monitoring on the test data. Kernel function parameters of kernel principal component analysis are optimized through a chaotic particle swarm optimization algorithm to find out the optimal nonlinear characteristics and accurately monitor nonlinear faults, so that the monitoring delay time is shortened, and the fault monitoring precision is improved.

Description

technical field [0001] The invention relates to the field of fault monitoring, in particular to a chaotic particle swarm-based optimization kernel principal component analysis fault monitoring method. Background technique [0002] In recent years, Kernel Principal Component Analysis (KPCA), as an advanced principal component analysis method, has been widely used to monitor the nonlinear characteristics of industrial processes. KPCA maps the original input space to a high-dimensional feature space through a nonlinear mapping function. When performing nonlinear feature transformation, the selection of the kernel function will seriously affect the monitoring performance of KPCA, so choosing an appropriate kernel function is the key. [0003] The determination of the KPCA kernel function currently mainly includes k-cross-validation error estimation and leave-one-out error estimation methods, but they are limited to the existing general kernel function and predetermined parameter...

AI Technical Summary

Problems solved by technology

Someone proposed to use the gradient descent method to solve the kernel function optimization problem, but this algorithm needs to calculate the partial derivative of the objective function to the optimized parameters. If the partial derivative of the objective function to a certain parameter does not exist or cannot be solved due to the complexity of the calculation, use The gradient descent method cannot achieve kernel function optimization

Someone also proposed to use the improved genetic algorithm to optimize the kernel function, but the genetic al...

Images