Intelligent Optimization of Ultra-Deepwater Oil Transmission Risers Using an Enhanced Graph Neural Network

Abstract

Abstract With the continuous global demand for oil and advances in deepwater drilling technology, the development of deepwater oil and gas resources has become increasingly important. Steel catenary risers play a crucial role in connecting subsea and surface facilities. However, complex marine conditions in ultra-deepwater environments pose significant challenges to the design of SCRs. To achieve efficient and reliable operation of SCRs under these extreme conditions, this study proposes an intelligent optimization design method based on enhanced Graph Neural Networks. Initially, considering the structural characteristics of the SCR dataset, graph data features comprising various nodes were constructed. Subsequently, an enhanced GNN model was obtained through cross-validation and Bayesian hyperparameter optimization, incorporating Graph Attention Networks with pooling layers, which achieved exceptional results in predicting dynamic responses of SCRs. Utilizing this model, multi-objective intelligent optimization of SCRs in the South China Sea region was conducted, and the optimization effects were validated through simulation. The results show a reduction of 26.76% in maximum top tension and 24.15% in Max Von Mises at the touchdown point, with wave-induced fatigue life and vortex-induced motion fatigue life improved by 24.31% and 63.16%, respectively. This research demonstrates that enhanced GNNs can efficiently learn the data relationships between different node features of SCRs and realize their effective optimization under extreme marine conditions, offering a reliable optimization design solution for ultra-deepwater oil and gas resource development. This approach opens new avenues for the optimization design of other marine structures.