Data driven description of reservoir petrophysical properties

In this study, a data driven reservoir description procedure is proposed. Four types of heuristic combinatorial algorithms, including simulated annealing (SA), greedy algorithm (GRA), genetic algorithm (GA) and a modified stochastic hillclimbing (SHC) are applied to seismic inversion by integrating the well data defined information directly into the inversion process, and compared according to their computational performances. The modified stochastic hillclimbing is chosen for further application due to its better performance. The sensitive factors are studied, including wavelet, noise and objective function sensitivity. A hybrid algorithm is also proposed by combining the greedy algorithm and stochastic hillclimbing. The new algorithm is more flexible to preserve or drift away the well data defined information. The combinatorial algorithms show their capability to recover low frequency components. The SHC is also modified for the inversion based on the geological model.; A co-simulation technique has been applied to determine the porosity distribution by integrating seismic and well data. A new co-simulation method is proposed based on the correlation scatter cloud using SHC. The porosity is simulated by honoring the spatial relationship and correlation relation with the shape. Two perturbation schemes are tested: one is the perturbation within the outline of the scatter cloud, and another one is the perturbation by sampling the cdf of the cloud for each interval of the densely sampled variable. The synthetic case studies show the two perturbations can capture more heterogeneities compared to correlation only.; Based on Fourier transform, the reservoir properties are transferred into the frequency domain. According to sampling law, the relation between the variogram and frequency is studied. Also, the highest heterogeneity the upscaling can capture is studied based on filtering. Different data sources have different spectral ranges. The data sources can be merged based on the frequency domain by pasting the spectrum. The geological model and geostatistical simulation results are merged by preserving the low frequency from the geological model, and the high frequency from the geostatistical modeling. The dynamic data can be merged with geological model and geostatistical simulated results by repairing the high frequency components.; Based on the porosity from the seismic or geostatistical simulation and the correlation between porosity and permeability, a new method is proposed to merge the static and dynamic data. That is, the permeability is only perturbed in the correlation scatter cloud. By applying this method, the porosity which honors the static data, such as seismic data and well data, will not be changed. The permeability matches the dynamic information, and the correlation with the porosity which is the reality is not changed pressure data can be reasonably matched in each well.; A field case study using the Frio sand reservoir in south Texas has been conducted. The work demonstrates the application of the entire procedure to field data. Through the data driven reservoir description, the generated porosity and permeability can match the static and dynamic data. It is believed that this is a practical way to generate a more reasonable description of the reservoir properties.