论文部分内容阅读
地震波形反演是非线性问题,其目标函数都是多极值的函数,若用线性化的反演方法求解,常会遇到迭代收敛于目标函数局部最优等困难。本文研究能求得目标函数全局最优解的遗传算法训练人工神经网络的地震波形反演方法。考虑到遗传算法训练神经网络地震波形反演的未知参数量大,而通常的二进制编码遗传算法占用计算机内存量大,不能在较小内存的计算机上实现,故以可节省内存的0-1编码遗传算法训练神经网络,提出了加速网络收敛的方法。
The inversion of seismic waveforms is a non-linear problem, and the objective functions are all functions of multi-extremum. If the linearized inversion method is used to solve the problem, the convergence of the iteration to the local optimality of the objective function is often encountered. In this paper, we study the seismic waveform inversion method of artificial neural network trained by genetic algorithm which can get the global optimal solution of objective function. Taking into account the genetic algorithm training neural network waveform retrieval unknown large amount of parameters, and the usual binary-coded genetic algorithm occupies a large amount of computer memory can not be implemented in a small memory computer, it can save the memory 0-1 encoding Genetic algorithm training neural network, put forward to accelerate the network convergence method.