用于重力反演的“自适应多段交换”遗传算法

Self-adapting and multi-parts crossing genetic algorithm for gravitational inversion

1. 同济大学海洋地质教育部重点实验室,上海200092;2. 江苏石油勘探局地质科学研究院,江都225261

Key Laboratory of Marine Geology, Tongji University, Shanghai 200092

用重力异常反演密度界面是一个非线性优化问题,以往多是在一定条件下将其转化成线性或拟线性问题,然后采用局部优化算法求解。局部优化算法对初始模型要求较高,容易陷入局部极小。鉴于此原因,人们开始利用具有全局寻优功能的优化算法,遗传算法就是其中之一。但是,目前多使用传统遗传算法,没有充分考虑重力异常反演本身的特点,计算代价比较高。

The inversion of gravimetric interface is a non linear problem. The traditional local searches are prone to trapping in local minimum if the initial model is not suitable. In recent years, the global optimization methods such as genetic algorithms (GA) are used to invert gravimetric interface. But the character of gravity inversion is not considered when GA is used to invert gravimetric interface. Self adapting and multi parts crossing GA is presented in this paper by combining the characters of gravitational inversion with the crossing process of GA. It ensures the evolution towards the optimum direction, and also accelerates the computing speed.