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

发布时间：2018-05-07 15:28

本文选题：BP神经网络 + 遗传算法　；参考：《吉林大学学报(地球科学版)》2017年02期

【摘要】：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。
[Abstract]:The BP neural network method has achieved good results in the inversion of two-dimensional density interface, but the convergence rate and inversion accuracy of the BP neural network with more parameters are decreased to some extent because of the more complex model and the more parameters in the inversion of the three-dimensional interface. In order to improve the inversion effect, the genetic algorithm is used to optimize the weight and threshold selection process of BP neural network, and a better network model is obtained, and the model is applied to the inversion of the density interface model. The prediction error is reduced from hundreds of meters to tens of meters, and the number of iterative calculation steps is reduced by nearly 2 / 3, which effectively reduces the calculation time and the inversion results are more accurate. A BP neural network based on genetic algorithm is used to inverse the Moho depth in a certain area of France. The relative error of prediction is only 1.8, and a good result is obtained. BP neural network optimized by genetic algorithm has good application value and research prospect in the inversion of density interface.
【作者单位】：吉林大学地球探测科学与技术学院;
【基金】：国家高技术研究发展计划(“863”计划)项目(2014AA06A613)~~
【分类号】：P631;TP18

【参考文献】