Well tops guided prediction of reservoir properties using modular neural network concept: A case study from western onshore, India

Abstract

This paper proposes a complete framework consisting pre-processing, modeling, and postprocessing stages to carry out well tops guided prediction of a reservoir property (sand fraction) from three seismic attributes (seismic impedance, instantaneous amplitude, and instantaneous frequency) using the concept of modular artificial neural network (MANN). The dataset used in this study comprising three seismic attributes and well log data from eight wells, is acquired from a western onshore hydrocarbon field of India. Firstly, the acquired dataset is integrated and normalized. Then, well log analysis and segmentation of the total depth range into three different units (zones) separated by well tops are carried out. Secondly, three different networks are trained corresponding to three different zones using combined dataset of seven wells and then trained networks are validated using the remaining test well. The target property of the test well is predicted using three different tuned networks corresponding to three zones; and then the estimated values obtained from three different networks are concatenated to represent the 1 predicted log along the complete depth range of the testing well. The application of multiple simpler networks instead of a single one improves the prediction accuracy in terms of performance evaluators-correlation coefficient, root mean square error, absolute error mean and program execution time. Then, volumetric prediction of reservoir properties is carried out using calibrated network parameters. This stage is followed by post-processing to improve visualization. Thus, a complete framework, which includes pre-processing, model building and validation, volumetric prediction, and post-processing, is designed for successful mapping between seismic attributes and a reservoir characteristic. The proposed framework outperformed a single artificial neural network in terms of reduced prediction error, program execution time and improved correlation coefficient as a result of application of the MANN concept.

A Novel Preprocessing Scheme to Improve the Prediction of Sand Fraction From Seismic Attributes Using Neural Networks

“…The improvement in mapping with introduction of the regularization step is observed from the performance analysis. The comparison among the results achieved in this work and existing literatures [17], [57] reveals the superiority of the proposed regularization step to obtain improved performance in terms of the performance evaluators. In the present study, synthetic SF logs are generated over the study area from available seismic information using the validated network parameters.…”

“…The present work differs from the work done in [17] in terms of preprocessing stage, division of training-testing dataset, adopted machine learning technique, and postprocessing algorithm. In addition, the well tops and horizon information are used [17] to carry out zone-wise division of the overall dataset and modular ANN is applied to model SF from multiple seismic attributes.…”

“…For example, normalization, relevant attributes selection, importance sampling, etc. have been part of the preprocessing technique in different fields such as environmental modeling [53], chemistry [54], biomedical [55], [56], and reservoir engineering [17], [57], [58]. Generally, CC values above 80%, RMSE, AEM values below 0.15, and SI value below 0.35 can be considered as good fit in such studies.…”

“…Comparison between the validation performance (prediction using data not part of training) reported in [57] and those achieved in this study has revealed that the regularization stage improves the prediction result in terms of CC, AEM, and RMSE. The preprocessing step involving a single ANN in [17] is similar to the case in the present paper while modeling with original SF as target. Comparison among the ANN performances with and without the regularization step reveals that the results of ANN have improved in terms of evaluators in the former case.…”

“…For example, ANN has been used in climatological studies [7], ocean engineering [8], telecommunications [9], text recognition [10], financial time series [11], reservoir characterization (RC) [12]- [17], etc. A diverse dataset containing information assembled from multiple domains can be used for learning and validation of ANN.…”