Prediction of thermal diffusivity of volcanic rocks using machine learning and genetic algorithm hybrid strategy

Abstract

Thermal diffusivity, a major determinant of heat transfer, has a substantial effect on rocks thermal properties. This investigation employed artificial intelligence to predict the thermal diffusivity of volcanic rocks. Models of machine learning with Genetic Algorithm are established to have a hollistic insight of thermal diffusivity analysis and prediction using elemental composition of volcanic rocks, and properties including temparature, density, specific gravity and apparent porosity. The support vector machine, ensembled learning tree, Gaussian process regression, and decision tree machine learning models are combined with a genetic algorithm to predict the thermal diffusivity of volcanic rocks. Using the genetic algorithm, feature selection and hyper-parameter optimization of machine learning methods improved the accuracy of model predictions. The SVM-GA predicts more accurately than other machine learning models, with a coefficient of determination of 0.99 and a root mean square error of 0.0046. In addition, the partial dependence plots analysis illustrates the potential effect of each parameter on thermal diffusivity. Finally, a graphical user interface based on the SVM-GA model was developed to predict thermal diffusivity without extensive machine learning knowledge. The difference between a graphical user interface and experimental yield was less than 1%.