论文详情
非线性共轭梯度(nonlinear conjugate gradient,NLCG)反演法具有较好的稳定性和反演精度,在二维反演中得到了广泛的应用,但是正则化因子选取困难以及计算量较大等问题限制了其在一维反演中的应用。针对上述问题,结合一维反演实际,提出了一种改进的NLCG优化算法。该算法以Bostick反演结果为初始模型,采用正则化因子自适应迭代调整方案,每次迭代时根据目标函数自动调整正则化因子,无需再对正则化因子进行不断尝试反演;简化了预处理因子和最优步长的计算,用单步线性搜索法代替迭代法求最优步长;对反演结果进行优化并做二次NLCG反演,进一步提高了反演精度,使非线性共轭梯度一维反演算法更高效。通过几个模型算例分析了优化算法的有效性。实际大地电磁数据NLCG反演结果与钻井资料吻合好、精度高；与Occam反演结果相比分层更加清晰，未出现异常高阻层,对地下电性结构的划分更加可靠。
Nonlinear conjugate gradient (NLCG) inversion method has been widely used in the 2D magnetotelluric inversion because of its stability and accuracy.However,this method is rarely used in the 1D inversion because it is difficult to find regularized factor and large amount computation.An improved 1D NLCG inversion algorithm is presented to overcome above problems.Firstly,Bostick inversion results is regarded as initial model and then we proposed an adaptive iterative adjustment procedure for regularized factors.The regularized factors are adjusted automatically according to the objective function in each iteration.Then,the calculation of optimal step size and precondition factors are simplified,finding the most suitable step size by a linear single step search instead of Gauss-Newton iteration.At last,the inversion model is optimized to do secondary NLCG inversion to improve the inversion precision and make the 1D NLCG algorithm more efficient.Several examples are given to illustrate the validity of the improved 1D algorithm,the processing results of actual magnetotelluric data further validate the reliability of the algorithm by contrasting with the drilling data and Occam inversion results.