Material performance prediction method based on dynamic selection training set

Abstract

The invention provides a material performance prediction method based on a dynamically selected training set, which comprises the following steps: selecting a specific training set for each group of test data to train a model and predict the model, and selecting the training set by combining Euclidean distance and Gaussian distribution. The method is better than using all original data as a training set and using data of the same cluster as the test data as the training set, the model trained by the former predicts the test data, and the average absolute error is 34.92% and 24.85% lower than the average absolute error of the former predicts the test data and the average absolute error of the latter predicts the test data.

Description

technical field [0001] The invention relates to the field of material performance prediction, in particular to a material performance prediction method based on dynamically selected training sets. Background technique [0002] For the performance prediction of materials, the existing technologies are mainly divided into two categories: one is to predict the specific properties of specific materials through experiments, which is time-consuming and laborious; the second is to use machine learning methods to predict material properties. The training set trains the model, and then uses the trained model to predict the new test data. According to the different methods of selecting the training set, it can be divided into two categories, one is to select all the original data as the training set, and the other is to select The data of the same cluster is used as the training set, but the prediction accuracy of the two methods has a large error. Through analysis, it is found that t...

AI Technical Summary

Problems solved by technology

[0002] For the performance prediction of materials, the existing technologies are mainly divided into two categories: one is to predict the specific properties of specific materials through experiments, which is time-consuming and laborious; the second is to use machine learning methods to predict material properties. The training set trains the model, and then uses the trained model to predict the new test data. According to the different methods of selecting the training set, it can be divided into two categories, one is to select all the original data as the training set, and the other is to select The data of the same cluster is used as the training set, but the prediction accuracy of the two methods has a large error. Through analysis, it is found that the reason is that the former ignores the characteristic differences between the original data, resulting in a large error in the prediction of material properties, while the latter has a large error in the prediction of the cluster. The prediction error of the test data at the edge of the class is relatively large, so the improvement of the prediction accuracy is transferred to how to select the training set. By selecting the appropriate training set, the accuracy of the model progress will be higher, and the prediction result will be more accurate.

Images