摘要
瞬变电磁反演是高维非凸的复杂非线性反演问题。利用传统的BP(back propagation)神经网络可以有效缓解瞬变电磁反演的过拟合现象,但是BP算法收敛速度慢、易陷入局部最优。为了解决这些问题,提出了一种基于小波包分解(wavelet packet denoising,WPD)和遗传算法(genetic algorithm,GA)优化BP神经网络的方法(WPD-GA-BP),并应用于瞬变电磁反演中。首先,采用基于硬阈值和Daubechies系列中Db13的WPD方法降低观测磁场数据中的噪声成分,同时提出一种剔除冗余特征的样本采集策略。然后,引入具有全局性的GA优化BP神经网络初始权重,提升BP算法的学习能力和求解精度。最后,基于中心回线源一维瞬变电磁正演理论,构建层状地电模型,经WPD预处理后进行反演,并比较GA-BP与传统Occam、单一BP、PSO-BP(particle swarm optimization-BP)、DE-BP(differential evolution-BP)等算法的反演结果。理论模型与实测数据反演结果表明:在瞬变电磁层状地电模型反演中,WPD-GA-BP比其他算法具有更高的精度以及更强的稳定性和正演数据拟合能力,可有效应用于电磁探测反演解释中。
Transient electromagnetic inversion is a complex nonlinear problem with high-dimensional non-convexity.The traditional BP neural network can effectively alleviate the over-fitting phenomenon for transient electromagnetic inversion.However,the BP method has the disadvantage of converges slowly and easily falls into local optimum.In order to solve these problems,an approach based on wavelet packet denoising(WPD)and genetic algorithm(GA)to optimize BP neural network(WPD-GA-BP)was proposed and applied to transient electromagnetic inversion.A wavelet packet denoising method based on hard threshold and Db13 was used to reduce noise signal from observed magnetic field data.And a sample collection strategy was proposed to remove redundant features.Additionally,the global GA algorithm was introduced to optimize the BP initial weight,which improved the learning ability and solution accuracy for BP.Finally,based on the 1-D transient electromagnetic forward theory with center loop source,a layered geoelectric model was established,and then inversion was performed after WPD processing,in which the inversion results by GA-BP algorithm were compared with that of the traditional Occam,BP,particle swarm optimization-BP(PSO-BP)and differential evolution-BP(DE-BP).The results of theoretical model and measured examples show that the proposed method is superior to others algorithm in the accuracy,stability and higher forward data fitting ability,which can be effectively applied to the inversion interpretation for electromagnetic exploration.
作者
李瑞友
白细民
张勇
汪靖
朱亮
丁小辉
李广
Li Ruiyou;Bai Ximin;Zhang Yong;Wang Jing;Zhu Liang;Ding Xiaohui;Li Guang(School of Software and Internet of Things Engineering,Jiangxi University of Finance and Economics,Nanchang 330013,China;Jiangxi Institute Co.,Ltd.of Survey and Design,Nanchang 330095,China;Power Supply Service Management Center of State Grid Jiangxi Electric Power Co.,Ltd.,Nanchang 330000,China;School of Geophysics and Measurement Control Technology,East China University of Technology,Nanchang 330013,China)
出处
《吉林大学学报（地球科学版）》
CAS
CSCD
北大核心
2024年第3期1003-1015,共13页
Journal of Jilin University：Earth Science Edition
基金
国家自然科学基金项目(41904076)
江西省教育厅科学技术项目(GJJ2200528)
南昌市水文地质与优质地下水资源开发利用重点实验室开放基金(20231B22)。
关键词
瞬变电磁法
小波包分解
BP神经网络
遗传算法
反演
transient electromagnetic method
wavelet packet denoising
BP neural network
genetic algorithm
inversion
