Graph neural network coal molecule property prediction method applying KAN network

Abstract

The invention provides a graph neural network coal molecule property prediction method applying a KAN network, and the method comprises the steps: constructing a coal molecule structure data set which comprises a plurality of sample coal molecule structures; determining data characteristics corresponding to each sample coal molecular structure; according to the data features, an initial coal molecule property prediction model based on a graph convolutional neural network is constructed, and the initial coal molecule property prediction model adopts a KAN network to perform feature aggregation; and training the initial coal molecular property prediction model to obtain a target coal molecular property prediction model, the target coal molecular property prediction model being used for methane adsorbability prediction. By implementing the method disclosed by the invention, a network architecture algorithm aiming at coal molecule property prediction can be designed in combination with a leading edge map neural network technology and a KAN network technology with relatively good interpretability, so that data-driven rapid and accurate prediction of coal molecule properties is realized.

Images