IRI: An intelligent resistivity inversion framework based on fuzzy wavelet neural network

Abstract

To acquire high-quality resistivity inversion results, an intelligent resistivity inversion (IRI) framework based on a fuzzy wavelet neural network (FWNN) is proposed in this paper. In the IRI framework, first, the FWNN is applied to build an interpretable inversion model for analyzing the apparent resistivity data. Takagi–Sugeno–Kang (TSK) fuzzy model is introduced to FWNN to explain the rules of the inversion results, and the wavelet neural network (WNN) is applied to construct the consequent part for each fuzzy rule and enhance the high-order nonlinear fitting ability of TSK fuzzy model. Then, to enhance the generalization and robustness, an elastic gradient descent (EGD) method is designed, which is used to update the linear parameters of the FWNN. Next, a differential and adaptive whale optimization algorithm (DAWOA) is introduced to train the nonlinear parameters of the FWNN for avoiding local optimum. Moreover, in the proposed DAWOA, a differential foraging strategy and an adaptive predation strategy are introduced to improve global exploration and local exploitation. All these measures can improve the inversion accuracy and accelerate the training process of FWNN for ERI inversion. Several simulation results demonstrate the feasibility and applicability of the IRI framework.