基于遗传算法优化的BP神经网络在密度界面反演中的应用

Application of BP Neural Network Based on Genetic Algorithm in the Inversion of Density Interface

BP神经网络方法在二维密度界面的反演中取得了较好的效果,但在反演三维界面时,由于模型更复杂、参数更多,BP神经网络的收敛速度和反演精度都有一定程度的下降。为了改善反演效果,本文利用遗传算法对BP神经网络的权值、阈值选择过程进行优化,获得了更好的网络模型;并将此模型应用于密度界面模型的反演中,预测误差从上百米减小到数十米,同时迭代计算步数减少了近2/3,有效减少了计算时间,反演结果更准确。利用基于遗传算法优化的BP神经网络反演了法国某地区莫霍面深度,预测相对误差仅为1.8%,取得了较好的应用效果。基于遗传算法优化的BP神经网络在密度界面的反演中具有良好的应用价值和研究前景。

The method of BP neural network has achieved good results in the inversion of 2D density interface,however,the converging speed and inversion accuracy would decrease when it is used to inverse 3D density interface due to more complicated model and more parameters.Genetic algorithm is used to optimize the process of choosing weights and thresholds of BP network in this paper in order to improve inversion results.Then a better network model is obtained and this model will be utilized in the inversion of density interface model.This method could increase inversion accuracy as well as reduce calculation time,and better inversion results would be achieved.At last the method is utilized to inverse the depth of Moho in some region in France and the application effect is good.It is illustrated that BPneural network based on genetic algorithm has benign application value and research prospect in the inversion of density interface.