Construction of glass composition analysis and identification model based on K-means algorithm optimized by genetic algorithm

Abstract

Ancient glass is highly susceptible to weathering by the environment in which it is buried. During the weathering process the proportion of its composition changes, thus affecting the correct judgment of its category. In this paper, the two categories of high potassium and lead-barium glass were subclassified separately, and firstly, the elbow method was used to combine the SC contour coefficients to obtain their optimal number of clusters as 3,7 respectively; after that, the initial point selection was set as the decision variable, the interclass distance and the minimum as the fitness function, and the initial population size was set as 50, the clustering distance as Euclidean distance and other parameters, and the genetic algorithm was used to optimize the K-means clustering to achieve the subclassification delineation. Taking lead barium as an example, after optimization, CHI is increased by 4, DBI is decreased by 0.3, contour coefficient is increased by 0.2, and the clustering effect becomes better. Finally, the Kappa consistency test was performed, and the Kappa coefficient was 0.822 indicating the reasonableness of the clustering results; by changing the selected feature vector dimensions for sensitivity analysis, the Kappa coefficients were all above 0.8, which were not sensitive to the dimensionality of the feature vectors. The model effectively implements the problem of subclassifying different classes of glass for subclassification.