Constellation optimization using an evolutionary algorithm with a variable-length chromosome
Constellation optimization using an evolutionary algorithm with a variable-length chromosome
复制标题
DOI：
10.1109/aero.2018.8396743
复制
发表时间：
2018-03
期刊：
影响因子：
--
通讯作者：
Nozomi Hitomi;Daniel Selva
中科院分区：
文献类型：
--
作者：
Nozomi Hitomi;Daniel Selva
这项工作提出了一种新的进化算法，可以搜索卫星星座的空间，以优化与覆盖相关的指标（例如最小化平均重访时间或最大化每日可见时间），同时最小化卫星及其半长轴的数量。它利用可变长度染色体来表示卫星星座，使用两个专门的运算符来处理可变长度染色体，以及在优化过程中调整搜索策略的自适应运算符选择器。所提出的染色体用 m 个 n 元组对 m 个卫星星座进行编码，其中每个卫星由一个 n 元组（例如轨道元素的子集）定义。目前，优化星座的进化算法采用固定长度的染色体，其中：1）星座中卫星的最大数量是预先指定的，并且每颗卫星都有一个额外的布尔变量，用于指示是否在星座中显示卫星；或者2）算法运行多次，每次都探索具有特定数量卫星的星座。前一种方法中的染色体表示局部性低且冗余，这使得进化算法的搜索变得更加困难。后一种方法没有利用针对给定大小的星座发现的良好的部分解决方案或图式，这些解决方案或图式可以为具有不同数量卫星的另一个星座的设计提供信息。相比之下，当算法探索具有不同数量卫星的星座时，可变长度染色体会增长和收缩，并显着减少染色体表示中的冗余。使用所提出的可变长度染色体表示的优化运行的功效和效率以多目标星座设计问题上的固定长度染色体为基准，该问题的目标是同时最小化全局平均重访时间、星座中的卫星数量以及卫星的平均半长轴。结果表明，与使用固定长度染色体的搜索相比，使用可变长度染色体进行的搜索使用少数千次的函数评估即可获得高质量的解集。
This work presents a new evolutionary algorithm that searches over the space of satellite constellations to optimize coverage-related metrics (e.g. minimizing average revisit time or maximizing daily visibility time) while simultaneously minimizing the number of satellites and their semi-major axes. It utilizes a variable-length chromosome to represent a satellite constellation, two specialized operators to handle the variable-length chromosomes, and an adaptive operator selector that adjusts the search strategy during the optimization. The proposed chromosome encodes an m satellite constellation with m n-tuples, where each satellite is defined by an n-tuple (e.g. a subset of the orbital elements). Currently, evolutionary algorithms that optimize constellations employ a fixed-length chromosome where either 1) the maximum number of satellites in the constellation is pre-specified and each satellite has an extra Boolean variable that dictates whether to manifest the satellite in the constellation or 2) the algorithm is run several times, each time exploring constellations with a specific number of satellites. The chromosome representation in the former method has low-locality and is redundant, both of which makes the search more difficult for an evolutionary algorithm. The latter method does not take advantage of good partial solutions or schemata discovered for a constellation of a given size that could inform the design of another constellation with a different number of satellites. In contrast, a variable-length chromosome grows and contracts as the algorithm explores constellations with different numbers of satellites and significantly reduces the redundancy in the chromosome representation. The efficacy and efficiency of an optimization run using the proposed variable-length chromosome representation is benchmarked against a fixed-length chromosome on a multiobjective constellation design problem whose goal is to simultaneously minimize the global average revisit time, the number of satellites in the constellation, and the average semi-major axis of the satellites. Results show that the search conducted with the variable-length chromosome reaches a high-quality solution set using thousands of fewer function evaluations than a search with a fixed-length chromosome.