Shear Wave Prediction in Carbonate Reservoirs: Can Artificial Neural Network Outperform Regression Analysis?

Hadi, F. A. (Baghdad University) | Nygaard, R. (Oklahoma State University)

OnePetro 

ABSTRACT: Formulating a prediction tool that can estimate the shear wave velocity (VS) is of particular importance for many applications related to petrophysics, seismic, and geomechanics. Shear wave data can be measured from both in-situ field and laboratory tests. However, they are often not measured during well logging for cost and time-saving purposes. For this reason, various prediction methods including regression analysis and artificial neural network (ANN) can be used for predicting the shear wave velocity. This study was conducted on dataset taken from a producing section in SE Iraq in which simple systematic equations have been demonstrated to predict VS from measurable well logs. The results reveal that the compressional wave velocity (VP) is more conservative in predicting VS rather than bulk density. Considering those parameters together can increase the performance metrics of the predictive methods. Although the results of regression analysis and ANN resemble to be closely, the higher value of determination coefficient (0.96) and the lower value of mean square error (0.0011) of ANN demonstrated that the ANN is more precise than regression analysis. An empirical model with high performance using ANN has been also developed to estimate VS from measurable well logs. Comparison of the developed models with the literature is then presented. The validity of the proposed models was successfully checked with data from another field study. This study presents efficient and cost-effective methods for predicting VS by incorporating measurable well logs as long as the rock tests and shear log measurements are not available.

1. INTRODUCTION

Borehole-based rock mechanical properties are not directly measured in the wellbore, and thus the shear wave velocity becomes essential to relate with conventional well logs. Shear wave velocity has a large number of applications in petrophysical, seismic, and geomechanical studies (Greenberg and Castagna, 1992; Kumar, 1976; Omnes, 1978). In geomechanical characterization, shear wave data is beneficial to obtain parameters such as Poisson’s ratio and Young module. However, well logs often lack shear wave data. There are four basic methods for determining the shear wave values in a reservoir: well logs, laboratory measurements, and theoretical or statistical approaches. Conventional well logs often measure compressional sonic waves within a section of interest, but shear sonic data is rarely measured because of the high cost and time taken to acquire the data. By nature, carbonate reservoirs can have a large variance in mechanical properties due to their depositional environment and complex diagenetic processes (Hadi et al., 2017). As a result of these variances, sufficient quantities of high quality core samples are rarely recovered from these reservoirs because they might be weak, thinly bedded, or fractured rocks (Ceryan et al., 2013). Reservoir characteristics such as fluid pore pressure and stress values are another inherent challenge in predicting shear wave values (Vs) since they cannot be accurately simulated through laboratory measurements (Maleki et al., 2014). There is also a lack of theoretical models to describe rock elastic properties (Ameen et al., 2009). These difficulties can lead to an inadequate understanding of reservoir properties and potentially cause inaccuracies during wellbore stability analysis. To overcome these difficulties, on the other hand, there is a demand for a simple, inexpensive, time-saving, and high performance predication model in which shear wave velocities can be determined from measurable well logs.