摘要
常规粒子群算法PSO对低敏感参数较难进行最优估计,本文提出联合自适应惯性权重和变异的混合粒子群算法ADPSO.结果表明:最佳拟合效果不随粒子数和变异概率的增大而加强.就四参数估计而言,粒子数取400个,变异概率小于0.03,可实现最优拟合.采用最佳粒子数和较小的变异概率,利用新近月球重力场模型GL0990d和LRO激光测高数据,对月球南半球部分高地的物理参数进行最优估计.发现研究区域的模型导纳谱很好地拟合了观测值,模型重力异常与观测重力异常的残差较小,符合参数估计的要求,验证了算法的有效性.ADPSO适合于月球物理参数估计,为大规模的反演运算提供了参考.
Ordinary particle swarm optimization PSO fails frequently in estimation of low sensitivity selenophysics parameters.With an adaptive inertia weight and a mutation factor,an admixed particle swarm optimization ADPSO is proposed.It is found that misfit is not reduced as increasing number of particles and probability of mutation.For four-parameter inversion,the best-fitting parameters can be estimated as considering about 400 particles and a low probability（0.03） of mutation.With employment of gravity filed model GL0990 d and altimeter data from LRO,we apply the method into selenophysical parameter inversion on southern highland of the moon.It shows a best-fit between modeled admittance spectral and observed values.Small residual of gravity anomaly verifies success of parameter inversion and validity of the method.The approach can be used in selenophysics research and could provide a reference for large-scale estimation of parameters.
出处
《计算物理》
CSCD
北大核心
2017年第6期740-746,共7页
Chinese Journal of Computational Physics
基金
国家自然科学基金(41404021)
贵州省科学技术基金(黔科合J字[2014]2128)
贵州师范大学博士科研项目资助
关键词
月球物理
参数估计
混合粒子群
导纳谱
selenophysics
parameter estimation
admixed particle swarm
admittance spectral
