Muhammad Abdulkarim

Infomax and FASTICA using principle component analysis as preprocessor for airwave removal in seabed logging

This research aims to apply the FASTICA and Infomax algorithm in the field of seabed logging, by utilizing the Principal Component Analysis (PCA) as preprocessor. All the three algorithms are statistical algorithms used for signal deconvolution and are respectively in the field of Independent Component Analysis (ICA). In seabed logging (SBL) implies the marine controlled source electromagnetic (CSEM) technique for the detection of hydrocarbons underneath the seabed floor. The results from SBL, indicate the presence of Hydrocarbon, but due to the presence of noise, in the form of airwaves, interfere with the signals from the subsurface and tend to dominate the receiver response. Hence, the Infomax and FASTICA de-convolution algorithms are used, considering PCA as a pre-processor to filter out the airwaves which disrupt the subsurface signals within the receiver response. The results obtained from simulations and their comparative analysis, indicate that the results from the infomax algorithm are better.

Subsurface exploration of seabed using electromagnetic waves for the detection of hydrocarbon layers

This paper focuses on the detection of hydrocarbon layers under the seabed using Electromagnetic methods and to prove the relationship between the thickness and resistivity constrast of the hydrocarbon. Simulations have been carried out by varying the depth of seawater from 1000m to 100m and the resistivity contrast and thickness for each level of depth is also varied. The electric field is also measured using various simulation models and graphs over different offsets. The results obtained prove that the resistivity property of Hydrocarbon is directly proportional to the thickness, and at particular points the presence of hydrocarbon layer is clearly significant.

A Simulation Study on Factors Affecting Airwaves Using Factorial Design

In shallow water Sea Bed Logging (SBL) survey, air layer response from the Electro-Magnetic (EM) signals creates a disturbance known as the source-induced airwaves. The airwaves commonly denote the energy that propagates from the EM source via the atmosphere to the receiver on the seabed. As a result, the airwaves dominate the measured survey data, so that the sought-after signals from possible hydrocarbon layers in the subsurface can be totally masked. In this study, a 5x5 factorial design is used to analyze the effect of frequency, seawater conductivity, sediment conductivity, seawater depth and offset on the magnitude of airwaves. The result based on F-statistics, indicates that frequency has higher significant effect on the magnitude of the airwaves followed by the seawater depth, offset, seawater conductivity and sediment conductivity in that order.

Airwaves estimation in shallow water CSEM data: Multi-layer perceptron versus multiple regression

In this study, a Multi-Layer Perceptron Neural Network and Multiple Regression techniques are used to estimate airwaves associated with shallow water Controlled-Source Electro-Magnetic (CSEM) data. Both techniques are appropriate for the development of estimation models. However, multiple regression models make some assumptions about the underlying data. These assumptions include independence, normality and homogeneity of variance. Conversely, neural network based models are not constrained by such assumptions. The performance of the two techniques is calculated based on coefficient of determination (R2) and mean square error (MSE). The results indicate that MLP produced better estimate for the airwaves with MSE of 0.0113 and R2 of 0.9935.

Performance Comparison between MLP Neural Network and Exponential Curve Fitting on Airwaves Data

This study aims at comparing the performance of a Multi-Layer Feed-Forward Neural Network and exponential curve fitting Models for the estimation of airwaves associated with shallow water Controlled Source Electro-Magnetic (CSEM) data. The performance measure is based on Mean Square Error (MSE), Sum of Squares Error (SSE) and coefficient of determination (R2). The MLP-NN network produced better and superior results with low MSE of 1.13e-7, SSE of 0.00017 and higher R2 of 99.35%.

A multiple regression model for predicting airwaves in shallow water sea bed logging data

This paper focuses on formulating a multiple regression model using matrix notation that can be used to predict the magnitude of airwaves in Shallow Water Sea Bed Logging (SBL) Data. The term airwaves refer to the propagated EM signals from the source antenna via atmosphere that is induced along air/sea surface and interferes with the subsurface signal. In shallow water, the airwaves have the ability to mask other subsurface responses possibly containing valuable information about subsurface resistive structure such as hydrocarbon reservoir. A fair representation of SBL environments was simulated to generate the airwaves data. Magnitude of airwaves at selected offset is used as the dependent variable. Whereas the predictor variables (independent variables) for the proposed multiple regression model are the frequency, seawater depth, seawater conductivity, sediment conductivity and offset. Akaikes Information Criterion (AIC) is used for selecting the multiple regression models. The formulated regression m...

A simulation study to verify the effects of air waves in Sea Bed Logging for shallow water

Detection of hydrocarbons (HC) by a Controlled Source Electromagnetic (CSEM), based on resistivity contrast, makes electromagnetic (EM) waves convincing method for HC detection in deep water exploration. However, HC survey done in shallow water is difficult due to a phenomenon called “air wave effect”. The waves that are produced by EM transmitter interact with air-sea interface to generate air waves that diffuse from the sea surface to the receivers. These air waves dominate the measured EM data such that the presence of the HC may not be detected. This work is a verification of the effect of air waves in shallow water environment. Data with hydrocarbon at 500m depth and data without hydrocarbon were simulated using CST EM Studio for this study for sea water depths from 1000m to 100m. Results have shown that the presence of hydrocarbon in shallow water is shielded by air waves.

Comparison on performance of radial basis function neural network and discriminant function in classification of CSEM data

Classification of Controlled Source Electro-Magnetic data into dichotomous groups based on the observed resistivity contrast measures is presented. These classifications may indicate the possible presence of hydrocarbon reservoir. Performance of Radial Basis Function of Neural network and Discriminant Function models were analyzed in this study. Both models classification accuracy, Sensitivity and Specificity are compared and reported. Gaussian basis function was used for the hidden units in the RBF neural network, while quadratic form is used for the discriminant functions. The Controlled Source Electro-Magnetic data used for this study were obtained from simulating two known categories of data with and without hydrocarbon using COMSOL Multiphysics simulation software. The preliminary result indicates that the radial basis function neural network display superior accuracy, sensitivity and specificity in classifying CSEM data when compared to discriminant functions model.

A preliminary study on application of discriminant analysis for classification of Controlled Source Electromagnetics data

Marine Controlled Source Electro-Magnetic (CSEM) for hydrocarbon exploration survey data can be classified into two groups. The ability to classify the raw survey data is crucial since this classification may indicate the presence of hydrocarbon. This paper presents the preliminary results of applying discriminant analysis in classifying CSEM data into dichotomous groups based on their electric field (E-field) and magnetic field (B-field) values. Two types of data, with and without hydrocarbon, were simulated and used to develop the discriminant model. Statistical analysis is carried out to test the significance of the discriminant function as a whole, the discrimination between groups and the extent to which that variable makes a unique contribution to the prediction of group membership. The results obtain indicates the potential of the discriminant analysis in classifying the data.