炭酸塩岩油層における浸透率の推測

抄録

Permeability is one of important factors for a reservoir simulation model because flow in a model is affected a lot by this parameter. Whether “reservoir characterization” method or conventional reservoir evaluating method would be taken to construct a simulation model, permeability distribution along the wells is important, because the distribution is used as a base to predict field-wide permeability distribution.
In this study, two permeability prediction methods were evaluated for estimating uncored well permeability from wire line well logs. One is method using artificial neural network (ANN), which is suitable to analyze complicated non-linear problem. The other is multiple linear regression analysis (MLRA) method. Also classical empirical porosity-permeability relationship method was evaluated to compare the accuracy or validity of ANN and MLRA methods.
ANN showed little improvement and MLRA showed no improvement, compared to conventional porosity-permeability relationship result. Especially, better results were seen when all data were fed in to the ANN system, assuming that no geological information was provided. However, even ANN, which showed the best result, is not recommendable method for this reservoir because of lower quality than required. In conclusion, these trials showed difficulty of estimating uncored well permeability in this reservoir because of poor log quality and the nature of this complicated carbonate reservoir.