针对存在球状禁飞区约束的高超声速飞行器再入轨迹规划问题，提出了一种基于动态目标重构和混合步长控制的序列凸优化方法，使飞行器在规避禁飞区的基础上主动远离禁飞区。首先，引入势函数对球状禁飞区进行软约束建模，将再入轨迹规划问题转化为软硬约束耦合的凸子问题进行求解。其次，为解决势函数引发的数值溢出问题，降低算法对初值的敏感程度，构建“可行性-最优性”两阶段动态解耦框架：第一阶段专注于可行性求解，为后续优化生成高质量初始轨迹；第二阶段通过软约束势函数动态重构优化目标，引导轨迹在不显著牺牲主要性能指标的前提下最大化远离禁飞区。在此基础上，进一步设计了混合步长控制策略，将信赖域算法和线搜索算法融合，充分利用被拒绝步的下降信息，有效提升了计算效率。最后，通过数值仿真验证了所提算法能够保证飞行器与球状禁飞区保持充分安全距离，且相比传统信赖域凸优化算法求解速度提升95%，终端高度与终端速度的误差分别降低19倍和6倍；相较于高斯伪谱法，在求解速度提高3倍的同时，具有同一水平的终端高度与终端速度的精度，展现出良好的应用前景。

To address the reentry trajectory planning problem for hypersonic vehicles with spherical no-fly zone (NFZ) constraints, this paper proposes a sequential convex optimization approach based on dynamic objective reconstruction and a hybrid step-size control strategy, enabling proactive NFZ separation assurance. First, a potential function is introduced to model the spherical NFZs as soft constraints, transforming the reentry trajectory planning problem into a sequence of convex subproblems with coupled soft and hard constraints. Second, to mitigate numerical overflow issues caused by the potential function and reduce sensitivity to initial guesses, a two-phase “feasibility–optimality” dynamic decoupling framework is developed. The first phase focuses on computing a feasible solution to generate a high-quality initial trajectory for subsequent optimization, while the second phase dynamically reconstructs the optimization objective using the soft-constraint potential function to maximize the standoff distance from NFZs without significantly compromising primary performance objectives. Furthermore, a hybrid step-size control strategy is designed by integrating the trust-region method with the line-search algorithm, effectively ex-ploiting descent information from rejected steps and improving computational efficiency. Numerical simulations demonstrate that the proposed algorithm guarantees a sufficient safety margin from spherical NFZs, achieving a 95% improvement in computational speed and reducing terminal altitude and velocity errors by factors of 19 and 6, respectively, compared with the conventional trust-region convex optimization method. Moreover, relative to the Gauss pseudospectral method, the pro-posed approach achieves a threefold increase in computational speed while maintaining comparable terminal altitude and velocity accuracy, highlighting its promising potential for practical applications.