Optimization design of metal sealing rings for ultra-high-pressure gas well hangers based on adaptive local-global Kriging with gradient-guided hybrid optimization

The sealing performance of metal sealing rings in ultra-high pressure gas well hangers is critical to wellbore integrity and production safety, yet their optimization design faces significant challenges due to highly nonlinear responses, abrupt gradient changes, and substantial computational costs. To address the inherent trade-offs between model accuracy and efficiency, as well as global exploration and local exploitation in conventional surrogate-based methods, this paper proposes an adaptive local-global Kriging and gradient-guided hybrid optimization method integrating adaptive learning with gradient-guided hybrid optimization. An adaptive correlation length updating mechanism enhances local approximation capability in regions with sharp gradient variations. A hybrid acquisition function combining expected improvement, global variance, and gradient information enables dynamic exploration – exploitation optimization, forming a collaborative ‘optimization-driven modeling and modeling-enhanced optimization’ framework. Validated on Sphere, Rastrigin, Ackley, and Rosenbrock functions, the proposed method demonstrates superior optimization accuracy, convergence speed, and stability compared to Kriging+EI (Expected Improvement), Kriging+PSO(Particle Swarm Optimization), and Gradient+EGO (Efficient Global Optimization). Applied to the multi-variable optimization design of a metal sealing ring, the method achieves approximately 2.8% higher optimization accuracy and 13% lower computational cost than Kriging+EI, substantiating its effectiveness and generalization potential for complex engineering optimization problems.