基于优化Adam算法的GPR全波形反演OA

Ground-penetrating radar(GPR)time-domain two-parameter full-waveform inversion has been plagued by problems such as poor inversion accuracy,parameter interferences,huge computational volume,and easy to fall into local minima.To address these problems,the Adam algorithm is introduced into the GPR time-domain full-waveform inversion.Since this method does not require iterative step-size calculation,it dramatically reduces memory consumption and computational complexity,thereby improving the efficiency and accuracy of the inversion.Firstly,the influence of parameter adjustment factors in the inversion is analyzed,and then the advantages and shortcomings of the single-scale and multi-scale serial inversion strategies are discussed.To obtain Adam hyperparameters suitable for full waveform inversion(FWI),the effects of different hyperparameters on the Adam algorithm's inversion performance are tested,and optimized parameters more suitable for FWI are proposed and evaluated using synthetic data from the Overthrust model.The experimental results show that,compared with full-waveform inversion using the Adam algorithm with preset parameters and the L-BFGS algorithm,the Adam algorithm with optimized hyperparameters achieves faster convergence and higher inversion accuracy.Finally,for the inversion of the road model with cavitation disease,the parameter-optimized Adam algorithm still slightly outperforms the L-BFGS algorithm when the uniform medium is used as the initial model.