地球物理资料非线性反演方法讲座(三) 模拟退火法

Simulated Annealing Method

模拟退火法源于统计热力物理学,它模拟熔融状态下物体缓慢冷却达到结晶状态的物理过程。模拟退火反演算法的基本思想是:生成一系列参数向量模拟粒子的热运动,通过缓慢地减小一个模拟温度的控制参数,使模拟的系统最终冷却结晶达到系统能量最小值的过程。模拟退火反演算法实质是利用了地球物理反演问题求解过程与熔化固体退火过程的相似性,开辟了地球物理反演的新途径,是非线性反演算法中一种最常用的算法。本讲座概要地介绍了模拟退火法的基本原理,模型搜索及解的接受准则,模拟退火法的分类及实现方法,并给出了模拟退火法在地震资料、电法资料及其他地球物理资料反演中的实例,最后总结和归纳了拟退火法的特点以及该方法的局限性。

Simulated annealing originates from statistical thermal physics, which simulates cooling process from melting state at high temperature to the crystallization state at low temperature. The idea of this physical process can be used to develop a new nonlinear optimization method- simulated annealing algorithm in which the model parameters are viewed as state vector and the objective function is viewed as energy function of the physical system. The essence of simulated annealing is a heuristic Monte Carlo method with higher efficiency and effectiveness. This paper not only introduces the principle, classification, diagram, application, advantages and disadvantages of the simulated annealing method, but also points out the necessity of the research of the improved simulated annealing method.