Performance degradation model precision verification method based on leave-one-out cross validation

Abstract

The invention discloses a performance degradation model precision verification method based on leave-one-out cross verification. The method comprises the following steps: 1) collecting data; 2) determining an alternative model form; 3) fitting to obtain an alternative model equation; 4) calculating a fitting error of the alternative model; 5) calculating the predicated error of the alternative model; 6) evaluating the model by means of the predicated error; and 7) determining an optimal model. According to the method, the advantages and disadvantages of the performance degradation model are comprehensively and completely evaluated from the two perspectives of the fitting error and the prediction error, and a new thought is provided for the aspects of model verification and selection in engineering research.

Description

[0001] 1. Technical field [0002] The invention provides a performance degradation model accuracy verification method based on leave-one-out cross-validation, which belongs to the technical field of reliability engineering. [0003] 2. Background technology [0004] The research on the performance degradation model accuracy verification method based on leave-one-out cross-validation has extensive practical significance. At present, statistical methods are mainly used in model selection. Statistical methods are mainly based on data, using mathematical models, and extracting data. Features, so as to summarize the laws and trends of the data, and provide a basis for subsequent prediction and analysis. However, in practical applications, there are mostly cases with abnormal data. Usually, people will select some candidate models at the beginning of the research, then compare the candidate models with the original data, and then specify some evaluation indicators according to the n...

AI Technical Summary

Problems solved by technology

When selecting a model, it seems that the process is relatively easy and the principle is relatively simple, but there are many practical problems in general engineering research. If the sample size is large, we can directly use all the data to calculate the prediction Error, in this case, it is not difficult to calculate the prediction error, and at the same time, it has no meaning for life prediction

In most cases of practical application, there are often not enough samples, and we can only estimate the prediction error by some method.

Images