一种用于波形反演的改进差分进化算法

Waveform inversion with an improved differential evolution algorithm

当波形反演中待求参数维数较高时,传统的全局优化方法逐渐失去其有效性和优点。本文在差分进化算法的选择算子中借鉴协同进化法分解—协调的思想,将复杂问题分解为若干子问题,并为每个子问题引入局部适应度的概念,提出一种改进的差分进化算法(DE-CCS)。此方法首先根据局部适应度选择出准下一代;考虑到子问题之间的协调优化,最终的下一代则仍根据全局适应度选取。改进差分进化算法同时利用局部适应度和全局适应度引导进化方向,提高了收敛速度,且对高维问题更加有效。将该方法应用于正演速度慢、局部极值多的高维波形反演,模型和实际数据算例结果证明了该方法的有效性。

While the number of the parameters in waveform inversion is large,the conventional global optimization methods often lose their effectiveness and advantages.In this paper,an improved differential evolution(DE) algorithm is proposed.For DE operator selection,in the light of co-evolutionary,we decompose the complex problem into some subcomponents and introduce a local fitness function for each subcomponent.Then a quasi-next generation is selected one subcomponent by one subcomponent according to the local fitness values.However,considering the interdependence among subcomponents,the final next generation is still selected according to the global fitness values.Therefore,the evolution direction of the problem is guided by the local fitness values and the global fitness value simultaneously.This improved DE algorithm with a co-evolutionary selection operator is called DE-CCS,which has fast convergence rate and is effective for high-dimensional optimization problems.DE-CCS is applied to the complex waveform inversion problems.Model and real data experiments demonstrate the effectiveness of DE-CCS.