论文部分内容阅读
针对卫星系统顶层设计中广泛存在仿真耗时、设计空间大以及非线性约束的特点,提出了免梯度混合优化算法。混合算法结合树状高斯过程(treed Gaussian process,TGP)模型、广义模式搜索和过滤法的优点,通过TGP模型将设计空间划分为互不相交的子空间,在各个子空间构建独立的高斯模型代替实际模型,并根据模型预测值和预测误差生成迭代点,进而指导模式搜索进行寻优,同时结合过滤法处理非线性约束。卫星系统中多星协同观测优化设计表明,该方法能够以较少的迭代次数获得满意解,具有很好的全局搜索特性。
Aiming at the characteristics of satellite system top-level design, such as time-consuming simulation, large design space and nonlinear constraints, we propose a gradient-free hybrid optimization algorithm. The hybrid algorithm combines the advantages of the TGP model, the generalized mode search and the filtering method, divides the design space into disjoint subspaces by TGP model and constructs an independent Gaussian model in each subspace instead of the Gaussian model Actual model, and generate iterative points according to model predictive value and prediction error, and then guide the pattern search for optimization, combined with the filtering method to deal with nonlinear constraints. The multi-star collaborative observational optimization design in satellite system shows that this method can obtain satisfactory solution with fewer iterations and has good global search characteristics.