摘要
Nonlinear resistivity inversion requires efficient artificial neural network(ANN)model for better inversion results.An evolutionary BP neural network(BPNN)approach based on differential evolution(DE)algorithm was presented,which was able to improve global search ability for resistivity tomography 2-D nonlinear inversion.In the proposed method,Tent equation was applied to obtain automatic parameter settings in DE and the restricted parameter Fcrit was used to enhance the ability of converging to global optimum.An implementation of proposed DE-BPNN was given,the network had one hidden layer with 52 nodes and it was trained on 36 datasets and tested on another 4 synthetic datasets.Two abnormity models were used to verify the feasibility and effectiveness of the proposed method,the results show that the proposed DE-BP algorithm has better performance than BP,conventional DE-BP and other chaotic DE-BP methods in stability and accuracy,and higher imaging quality than least square inversion.
Nonlinear resistivity inversion requires efficient artificial neural network （ANN） model for better inversion results. An evolutionary BP neural network （BPNN） approach based on differential evolution （DE） algorithm was presented, which was able to improve global search ability for resistivity tomography 2-D nonlinear inversion. In the proposed method, Tent equation was applied to obtain automatic parameter settings in DE and the restricted parameter Fcrit was used to enhance the ability of converging to global optimum. An implementation of proposed DE-BPNN was given, the network had one hidden layer with 52 nodes and it was trained on 36 datasets and tested on another 4 synthetic datasets. Two abnormity models were used to verify the feasibility and effectiveness of the proposed method, the results show that the proposed DE-BP algorithm has better performance than BP, conventional DE-BP and other chaotic DE-BP methods in stability and accuracy, and higher imaging quality than least square inversion.
基金
Project(20120162110015)supported by the Research Fund for the Doctoral Program of Higher Education,China
Project(41004053)supported by the National Natural Science Foundation of China
Project(12c0241)supported by Scientific Research Fund of Hunan Provincial Education Department,China
