Annealing governs the microstructure and properties of materials, while traditional empirical optimization is costly and hard to scale. To address defects of single algorithms in modeling nonlinear process-performance relations, this study establishes an accurate and stable universal optimization model for hyperelastic alloy annealing, enabling intelligent process optimization and precise mechanical property evaluation. The precise control of the annealing process is essential for tailoring the mechanical response of these alloys, particularly when they are utilized as high-performance functional fibers in advanced applications. This paper proposes a hybrid model integrating random forest and BP neural network. It screens key parameters using the feature selection and anti-overfitting strengths of random forest, and builds a two-layer prediction framework with the strong nonlinear mapping ability of BP neural network. Systematic experiments collect data of annealing temperature, holding time, cooling rate and mechanical properties, with validity verified by yield strength, shape recovery rate and other indicators. Experimental results show that the fused model greatly improves prediction accuracy, with 15%-20% error reduction over single algorithms and a correlation coefficient of 0.94 between predicted and measured values. The Ti-45Nb alloy reaches 1200 MPa tensile strength and excellent hyperelastic recovery under the optimal process window. Furthermore, this research provides a reliable optimization strategy for hyperelastic component processing, with great value for the development of intelligent textile materials and flexible structural elements, laying a solid foundation for the industrial application of hyperelastic alloys.

hyperelastic alloy, annealing process, random forest, mechanical properties, functional fibers