难熔高熵合金具有优异的高温稳定性和力学性能，在航空航天等高温结构领域具有广阔的应用前景。然而，这类合金普遍存在高强度与低塑性难以兼顾的“强塑性矛盾”。传统的成分设计主要依赖经验和试错，研发周期长且成本高。本文提出了一种面向抗压屈服强度(YS)与断裂应变(FS)协同优化的机器学习驱动难熔高熵合金成分设计框架，构建了包含298组由Al、Hf、Mo、Nb、Ta、Ti、V、Zr和W元素组合构成的合金体系压缩性能的数据集，通过特征池与模型池筛选获得适用于YS和FS预测的最优特征子集与机器学习模型。特征重要性分析表明，局域尺寸错配与模量错配主导屈服强度，而电负性差与结构稳定性参数γ对断裂应变具有显著影响。在此基础上，设计期望改进效用函数并结合NSGA-II多目标优化算法，在九元成分空间中搜索强度-塑性权衡的Pareto最优解。模型预测结果表明，YS与FS的预测精度分别达到R2=0.94和0.87。依据协同优化结果，确定并制备了Mo18Nb26Ti25V11Zr20合金，具有单一BCC相结构，在保持36.5%断裂应变的同时，抗压屈服强度达到1713 MPa，与模型预测结果高度一致。

Refractory high-entropy alloys (RHEAs) exhibit outstanding high temperature stability and mechanical strength, making them promising candidates for aerospace and other high temperature applications. However, their widespread application is constrained by the inherent trade-off between high strength and limited ductility, while conventional trial-and-error compositional design strategies remain inefficient and costly. To address this challenge, a Machine Learning (ML)-driven compositional design framework is proposed for the synergistic optimization of compressive Yield Strength (YS) and Fracture Strain (FS) in RHEAs. The data comprising 298 reported literature on compressive property of Al-Hf-Mo-Nb-Ta-Ti-V-Zr-W alloys was constructed. By systematically developing feature pools and model pools, optimal feature subsets and ML models were identified for accurate prediction of YS and FS. Feature importance analysis reveals that local atomic size mismatch and modulus mismatch dominate the yield strength, whereas electronegativity difference and the structural stability parameter γ play critical roles in governing the fracture strain. An expected improvement (EI)-based utility function was further integrated with the NSGA-II multi-objective optimization algorithm to efficiently explore the nine-element compositional space and identify the Pareto-optimal strength-ductility trade-off. The developed models achieve high predictive accuracy with R2 values of 0.94 for YS and 0.87 for FS. Guided by the optimization results, a promising alloy composition, Mo18Nb26Ti25V11Zr20, was prepared. The alloy exhibits a single-phase BCC structure, delivering a high compressive yield strength of 1713 MPa, while maintaining a fracture strain of 36.5%, in excellent agreement with model predictions. These results demonstrate that the proposed machine learning framework enables efficient and synergistic optimization of strength and ductility in RHEAs.