Feature Extraction and Intelligent Identification of Induced Polarization Effects in 1D Time-Domain Electromagnetic Data Based on PMI-FSVM Algorithm

Yanqi Wu,Jingya Shao,Yanju Ji,Fei Teng,Binyuan Ma

doi:10.1109/access.2020.3016828

Yanqi Wu, Jingya Shao + Show 3 more

Open Access

https://doi.org/10.1109/access.2020.3016828

Copy DOI

Abstract

The transient electromagnetic method can obtain resistivity and chargeability simultaneously in polarizable medium detection. Typically, we assume that the earth may contain a chargeable medium if the electromagnetic (EM) data appear negative values or sign reversals. Unfortunately, with barely perceptible characteristics, some EM responses with the induced polarization (IP) effects are considered to be non-polarizable responses. Insufficient understanding of features and inaccurate identification of the IP responses limits the use of the IP effects for broader purposes. For these reasons, we perform 1D forward modeling to discuss the degree of EM response affected by the IP effects and to extract polarization characteristics. To identify the IP effects, we combine partial mutual information (PMI) and the fuzzy support vector machine (FSVM) methods to complete the intelligent identification algorithm. We verify the efficiency and practicality of the algorithm by building Debye loops in field experiments. From the analysis, we distinguish the strong and weak IP effects by introducing the impact ratio. The strong IP responses manifest fast decays and sign reversals, and the weak IP responses primarily show fast decays or outward concavity. The identification algorithm validation results show that the recognition accuracy reaches 90.7%. In the field experiment verification, the Debye loop successfully simulates the IP effects of different intensity, and the identification results indicate that the algorithm has potential in the measured data. With this intelligent identification algorithm, the measurements can provide access to the weak polarizable medium when the impact ratio exceeds 30%.

Highlights

The induced polarization (IP) effect is a crucial electrochemical phenomenon that exists primarily in the dispersive medium with high economic value, such as disseminated sulfide minerals, some aqueous media, and hydrocarbon resources
INFLUENCE OF CONDUCTIVITY DISTRIBUTION OF THE 1D SYNTHETIC MODEL We found that the algorithm can successfully identify more than 80% of weak IP effects and 100% of strong IP effects
If a resistive nonpolarizable medium appears in the medium depth (200 m - 400 m), the EM data show a fast decay in the late time, which may be erroneously recognized as IP responses

Summary

INTRODUCTION

The induced polarization (IP) effect is a crucial electrochemical phenomenon that exists primarily in the dispersive medium with high economic value, such as disseminated sulfide minerals, some aqueous media, and hydrocarbon resources. In the TEM detection, the IP effects raised attention because the EM data revealed sign reversals or negative values in later time [5]–[7]. The influences of polarization parameters on the time of sign reversals and the maximum absolute value of the negative response were studied intensively in [8]–[9]. Principal component analysis, and Schmidt transform can extract critical features from all variables to reduce dimension and eliminate correlations [23]–[25] These methods perform poorly in solving nonlinear problems. The decay curves of type B and C are less affected by the IP effects and show polarization characteristics indistinctly; we call them weak IP responses. Apart from the characteristic parameters about sign reversals such as time of sign reversal, the maximum absolute value

SLOPE FITTING BASED ON PIECEWISE NONLINEAR LEAST-SQUARES METHOD

Findings

DISCUSSION

CONCLUSIONS