Low Earth Orbit Regional Satellite Constellation Design via Self Organization Feature Maps

Satellite constellation design is one kind of typical Multiobjective Optimization Problem (MOP). In this paper, aim at feature of high-dimensional decision space, an model-based multiobjective Evolutionary Algorithm(EA) via Self Organization Feature Map(SOM) is put forward for reducing decision space dimension of satellite constellation design: internal topology of input training set in the population can be kept by neurons of SOM network, which makes topology similar points between in high-dimensional input data space remains neighboring relationship in low-dimensional data space after mapping, so as to achieve the dual purpose to maintain diversity of population and accelerate convergence of algorithm. Experiment on low earth orbit regional satellite constellation design shows that the optimization results can meet the demand of coverage performance for specific region, but running speed of the algorithm should be improved, the reason is maybe that SOM modeling is timeconsuming with increasing of decision space dimension.