A step forward to closing the loop between static and dynamic reservoir modeling

Abstract

The current trend for history matching is to find multiple calibrated models instead of a single set of model parameters that match the historical data. The advantage of several current workflows involving assisted history matching techniques, particularly those based on heuristic optimizers or direct search, is that they lead to a number of calibrated models that partially address the problem of the non-uniqueness of the solutions. The importance of achieving multiple solutions is that calibrated models can be used for a true quantification of the uncertainty affecting the production forecasts, which represent the basis for technical and economic risk analysis. In this paper, the importance of incorporating the geological uncertainties in a reservoir study is demonstrated. A workflow, which includes the analysis of the uncertainty associated with the facies distribution for a fluvial depositional environment in the calibration of the numerical dynamic models and, consequently, in the production forecast, is presented. The first step in the workflow was to generate a set of facies realizations starting from different conceptual models. After facies modeling, the petrophysical properties were assigned to the simulation domains. Then, each facies realization was calibrated separately by varying permeability and porosity fields. Data assimilation techniques were used to calibrate the models in a reasonable span of time. Results showed that even the adoption of a conceptual model for facies distribution clearly representative of the reservoir internal geometry might not guarantee reliable results in terms of production forecast. Furthermore, results also showed that realizations which seem fully acceptable after calibration were not representative of the true reservoir internal configuration and provided wrong production forecasts; conversely, realizations which did not show a good fit of the production data could reliably predict the reservoir behavior. Thus a statistical approach was confirmed to be the only way to reduce the uncertainty inherent to reservoir modeling and should be adopted as a standard in reservoir studies.