近几年，国内外学者把高斯过程方法应用到了各个领域。高斯过程是基于贝叶斯

In recent years, domestic and foreign scholars have applied the Gaussian process method to various fields. <br>The Gaussian process is a parameter-free kernel method based on the Bayesian learning theory framework62), which is used for supervised learning and has been successfully applied to regression and classification. Since the Gaussian process adopts the kernel function under the framework of statistical learning theory, the nuclear machine has strong nonlinear performance and can solve more complex problems without having to pay a huge computational cost. This also avoids the problem in the neural network to a certain extent. The "curse of dimensionality" problem 63. More importantly, the Gaussian process provides a paradigm for Bayesian learning.According to the training sample, it can be converted from the prior distribution to the posterior distribution, and can infer the hyperparameters of the kernel function: while the selection of hyperparameters by SvM is usually Only empirical methods or cross-validation methods can be used. Compared with ANN and SVM, the outstanding advantages of the Gaussian process are: it is easy to implement without sacrificing performance, can adaptively obtain hyperparameters during the model construction process, has a complete Bayesian formulation, and has a clear prediction output. Probabilistic interpretation of, and can directly realize multi-class classification, these advantages have been verified in Yixian 672. In recent years, domestic scholars have successively applied Gaussian processes in image classification, remote sensing image classification, geotechnical engineering, traffic flow forecasting, process optimization scheduling, soft sensor modeling and other fields, and the research has received unprecedented attention. However, the application literature of equipment fault diagnosis based on Gaussian process is still very few, especially the literature that will be applied to equipment performance evaluation has not been reported.

In recent years, scholars at home and abroad have applied gaussian process methods to various fields.<br>Gaussian process is based on Bayesian learning theory framework of non-parameter nuclear method 62), used for supervised learning, has been successfully applied to regression and classification. Because the Gaussian process adopts the nuclear function under the framework of statistical learning theory, the nuclear machine has strong nonlinear performance, which can solve more complex problems without paying a huge computational cost, which also avoids the "dimensional disaster" problem 63 in the neural network to a certain extent. What's more, the Gaussian process provides a paradigm for Bayesian learning, which can be converted from a priori distribution to a post-test distribution based on the training sample, and can be inferred from the nuclear function superparametrics: SvM's selection of superparametrics usually uses empirical or cross-validation methods. Compared with ANN and SVM, the outstanding advantages of Gaussian process are: easy to implement without sacrificing performance, adaptive acquisition of super parameters in the model construction process, full Bayesian formulaic representation, predictive output with clear probability explanation, and can directly implement multi-class classification, these advantages have been verified in the contribution 672. In recent years, domestic scholars have applied the Gaussian process to image classification, remote sensing image classification, geotechnical engineering, traffic flow prediction, process optimization scheduling, soft measurement modeling and other fields, the research has received unprecedented attention. However, there is still very little application literature for device fault diagnosis based on Gauss process, especially the literature that will be applied to equipment performance evaluation has not been reported.

In recent years, scholars at home and abroad have applied Gaussian process method to various fields.<br>Gaussian process is a nonparametric kernel method based on Bayesian learning theory, which is used for supervised learning and has been successfully applied to regression and classification. Because Gaussian process uses the kernel function under the framework of statistical learning theory, the kernel machine has strong nonlinear performance, and can solve more complex problems without huge computational cost, which also avoids the "dimension disaster" problem in neural network to a certain extent. More importantly, Gaussian process provides a paradigm for Bayesian learning. According to the training samples, it can transform from prior distribution to posterior distribution, and can infer the super parameters of kernel function. However, SVM can only choose the super parameters by empirical method or cross validation method. Compared with ANN and SVM, Gaussian process has the following advantages: it is easy to implement without sacrificing performance, can adaptively obtain super parameters in the process of model construction, has complete Bayesian formula representation, has clear probability interpretation of prediction output, and can directly realize multi class classification. These advantages have been verified in Yixian 672. In recent years, domestic scholars have applied Gaussian process to image classification, remote sensing image classification, geotechnical engineering, traffic flow prediction, process optimization scheduling, soft sensing modeling and other fields, and the research has received unprecedented attention. However, the application literature of equipment fault diagnosis based on Gaussian process is very few, especially the literature of equipment performance evaluation has not been reported.<br>