摘要
大地电磁反演问题通常表述为目标函数最优化,难点是多参数、非线性和不适定性,局部和全局方法都不能实现快速全局优化[4]。针对局部线性方法易使解陷入局部极值,严重依赖初始模型,而传统的遗传算法在优化应用中存在局部搜索能力弱、早熟收敛等问题。这里引进一种求解一维大地电磁测深反演问题的实数编码广义遗传算法。该算法利用拟网格法初始种群和综合交叉策略,克服了早熟收敛现象,从而提高了遗传寻优的效率。理论模型反演与其它方法比较,结果说明遗传算法具有不依赖初始模型,不容易陷入局部极小,多点多路径概率搜索,以及隐含并行性等优点。
The inverse problem of MT is usually described as objective function optimization,which has many difficulties such as multi-parameter,nonlinear in the parameter and ill-posed problem,etc..Both local-optimization and global-optimization methods are unable to achieve fast global optimization[4].Due to the fact that the method based on local linearization is usually lost in local minimum values and the optimization solution is dependent on the selection of initial solution,the standard genetic algorithm has poor local searching ability and premature convergence.In this paper,a one-dimensional MT inversion of the real-coded generalized genetic algorithm has been presented.By the initial grid of cross-species and integrated strategy,the phenomenon of premature convergence has been overcome and the efficiency of genetic optimization has been improved.Inversion and comparison results from theoretical model has shown that this algorithm is not dependent on the initial model and is not easy to a local minimum,and has the advantages of more multi-path probability searching and the implied parallel.
出处
《物探化探计算技术》
CAS
CSCD
2009年第4期314-318,288,共5页
Computing Techniques For Geophysical and Geochemical Exploration
基金
国家"863"计划资助项目(2006AA06Z110)
国家自然科学基金资助项目(40674035)
关键词
广义实码遗传算法
初始种群
交叉策略
全局优化
一维MT反演
generalized real-coded genetic algorithm
the initial population
cross-strategy
global optimization
one-dimensional inversion MT