| Popis: |
Permeability is one of the most important petrophysical parameter required for reservoir evaluation and monitoring. Reliable permeability data derived from core measurements conditioned to well test permeability is scarce for most reservoirs as there is low percentages of cored wells due to technical and economical reasons. Therefore, suitable approach to estimate the permeability values in the non-cored wells of any reservoir with an acceptable accuracy, unquestionably, deemed very necessary. Deriving permeability using log data is the approach that has been followed in the industry. This requires, however, prior effort of modeling relationships between log responses and core permeability in the cored wells. These relationships are normally modeled using conventional empirical techniques and recently Artificial Intelligence (AI) technologies claimed their shares in this area. This study employs hybrid of both AI and conventional approaches to improve the prediction of permeability for very heterogeneous carbonate reservoir with tarmat layers. To come up with more appropriate permeability prediction model, this study has been conducted in four main phases. In the initial phase, reservoir layers were grouped considering data availability and layers’ lithology, and petrophysical data; separate grouping has been considered for layers with tarmat. Then each group was divided into up to five clusters depending on porosity and core permeability cross-plots. In the second phase, Adaptive Neuro Fuzzy Inference System (ANFIS) has been designed to predict the rock cluster index based on wells’ GR, Rxo, and Porosity logs along with volume of anhydrite, calcite and dolomite. In the third phase, conventional curve fitting technique has been implemented for each rock cluster identified in the previous phase. Then, further improvement has been done for each rock cluster by introducing correction formula depending on the core versus predicted permeability cross-plots. This final phase involved a statistical analysis of the prediction quality for the overall and the blind test data. Statistical analysis was performed to assess the quality of permeability predictions. An accuracy of about 85% was achieved considering all the data involved in the study including the tarmat layers and 72% for the blind test data including part of the tarmat layers as well. The accuracy is improved even more when the tarmat layers are excluded from the analysis. The accuracy of the tarmat layers on the other hand, was found to be 83.33% for all the data and around 65% for the blind test data. This improved permeability prediction is depicted more clearly when core and predicted permeability were plotted versus depth, showing very good to impressive match for all wells considered in this study. This study investigated the effect of level of the porosity and permeability cutoffs on the accuracy of the permeability prediction. The overall data accuracy for lower cutoffs is slightly better than the higher cutoff case (85% vs. 82.7%). On the other hand, the blind test data showed the higher cutoff case is more accurate (72.5% vs. 71.8%). Based on the results and considering the uncertainty in low permeability measurements, this study recommends applying certain levels of porosity and permeability cutoffs. |
