Prediction of missing values of chemical elements in glass relics and subclassification based on neural network

Abstract

In this study, the chemical composition data of ancient glass were sorted out and analyzed. Based on the training principle of BP neural network, BP neural network was established to solve the problem. After several iterations, 14 input layers, 5 neurons and the training method of radial basis function were finally determined. The data before weathering of weathered relics were finally obtained, so as to predict and restore the missing value of ancient glass chemical elements. In order to verify the rationality and sensitivity of the results, certain parameters were determined to process the data, and cluster analysis was performed again. By comparing the two results, we found that the difference between the two results was small, which verified the rationality and stability of the classification results. Then, through k-means algorithm, the types of glass were subclassified based on the different chemical composition content. For example, high-potassium glass was divided into high-potassium high-calcium glass and high-potassium low-calcium glass, and lead-barium glass was divided into lead-barium high-calcium glass and lead-barium low-calcium glass.