RESEARCH ON THE PREDICTION MODEL OF MECHANICAL RESPONSE FOR STRUCTURES BASED ON GRAPH NEURAL NETWORK

The prediction of structural mechanical response is a crucial aspect of engineering structural analysis. In order to overcome the limitations of traditional computational analysis model and enhance the accuracy and efficiency in predicting structural mechanical responses, a structural mechanical response prediction model based on graph neural networks (GNN-SMRP) is proposed, which is suitable for predicting the mechanical response of various engineering structures. The structural analysis modeling process and graph data construction are integrated to transform any engineering structure into graph data suitable for the prediction model. Relying on the unique information transfer mechanism of graph neural networks, the correlation between data is deeply mined. By connecting different post-processing networks, various mechanical response prediction tasks are implemented. To validate the effectiveness of the model, numerical experiments were conducted using the transmission tower structure as an example to study the prediction performance of the model for component axial force and node displacement. The research shows that the model has strong prediction performance in both tasks, achieving an accuracy of over 98%. Moreover, the model exhibits better prediction performance for node displacement than for axial force.