Reservoir lithofacies modeling using well logs and seismic data based on Sequential Indicator Simulations and Probability Perturbation Method in a Bayesian framework

In this paper, an inverse framework based on Bayes' theorem is suggested for integrating well logs and seismic data into the reservoir lithofacies modeling process. The proposed method is based on the combination of the Sequential Indicator Simulation (SIS), and a stochastic optimization method (i.e. the Probability Perturbation Method (PPM)). SIS is used to calculate the conditional probability of presence or absence of lithofacies indicators in each grid-block, and PPM is applied to update (perturb) the conditional probability used in SIS. A notable innovation presented in this study is using the Genetic algorithm' crossover operator to increase the PPM exploitation capability. To demonstrate the efficiency of our proposed approach, the results of its application on a 3D test model is compared with outcomes of two commonly- used constraining approaches on SIS technique. Qualitative and quantitative analysis of the obtained results on 3D test model reveals a (23.8)% and (16.98)% (on average) improvement in consistency of the lithofacies models generated using the proposed approach with the reference lithofacies model over the employed Vertical Probability Trend and the Seismic Probability Trend constraining approaches on SIS, respectively. Besides, the obtained results show that implementing the crossover operator leads to a 4.56% improvement in matching of the constructed lithofacies models with the reference model.

Automated summary

An inverse framework based on Bayes' theorem is proposed for integrating well logs and seismic data into the reservoir lithofacies modeling process in this paper. The method combines Sequential Indicator Simulation (SIS) with a stochastic optimization method (Probability Perturbation Method (PPM)), and utilizes a Genetic Algorithm's crossover operator to enhance the PPM exploitation ability. Application of this approach on a 3D test model shows significant improvement in consistency and matching of lithofacies models compared to traditional constraining approaches.