Recent developments in computational intelligence, particularly in machine learning, have strongly improved empirical modeling. The field which encompasses these new techniques and approaches is known as data-driven modeling. This is based on analyzing particular system data in order to find links between the system variables (input, internal and outputs) with no explicit knowledge of the system’s physical behavior .
Data-driven reservoir modeling is a new technique (under development) used to build models which represent hydrocarbon flow in porous media. It is based on field measurements rather than the physics principles introduced in a reservoir simulator, such as partial differential equations. Such field measurements are based on well configurations, well completion, well logs, core analysis, well tests, and production/injection history.
According to , data-driven reservoir modeling does not rely on the principle that all modeling of natural phenomena must start with physics to have accuracy and credibility. The process starts with the premise that data carries information. Therefore, all data collected during drilling, reservoir, and production operations include marks in time and space regarding flow in porous media.
By using this approach, the construction of a cohesive and accurate reservoir model is conditioned by key factors. First, it is crucial to organize and assemble data using a proper strategy, especially if there is a large amount of available data. Additionally, it is crucial to have a set of appropriate tools and experienced petroleum engineers to deal with and interpret such large volume of data.
The data-driven modeling technique is also known as “Top-Down Modeling (TPM)” and it is an alternative, or even a complement, to numerical reservoir simulation in cases where simulation costs are unfeasible. Unlike traditional modeling approaches, TDM attempts to build a realization of the reservoir by starting with its history (well production behavior), complemented by other types of field measurements – static and dynamic variables. In addition, the small computational effort required by TDM makes it an ideal tool for reservoir management, uncertainty quantification, and field development planning .
TDM relies on the use of Artificial Intelligence (AI) and Pattern Recognition in order to develop consistent reservoir models based on measurements rather than mathematical formulations and physics principles of flow through porous media. According to , these techniques provide the means for finding patterns among non-linear and independent parameters in field development planning. Therefore, such unique and elegant approach consists of an integration of traditional reservoir engineering methods with pattern recognition capabilities.
Among its main advantages, it is worth mentioning: flexible data requirement, short development time, and minimal computational overhead. In addition, the high speed calculation allows for more a optimized analysis and decision-making process. On the other hand, the major constraint is that it can only be applied when a reasonable amount of field data is available.
Taking all the mentioned factors into account, data-driven modeling can be considered a viable technology, particularly for unconventional assets, where the physics of production is not completely well understood yet. To further illustrate the general use of this method, the three major steps in the development of a Top-Down shale reservoir model are as follows :
- Spatio-temporal database development: an extensive data mining and analysis process should be carried out to fully understand the data in the database. The activities of data compilation, curation, quality control and preprocessing are crucial but time consuming steps in Artificial Intelligence reservoir modeling.
- Simultaneous training and history matching of the reservoir model: the model development and history matching in an AI-based reservoir model are performed simultaneously during the training process. Thus, the goal is to make sure that this reservoir model represents and encompasses the fluid flow behavior being modeled. , It is also imperative to have a robust strategy to validate the capability of prediction in an AI-based model, by using completely blind data.
- Sensitivity analysis and uncertainties quantification: both developed and history matched models are confronted against changes in reservoir parameters and operational constraints in order to perform a sensitivity analysis and quantify inherent uncertainties.
- Deployment of the model in predictive mode: similarly to conventional reservoir simulation models, an AI-based reservoir model is deployed in predictive mode for reservoir management and decision-making goals.
In summary, data-driven reservoir technology is a new way to model and understand a reservoir and its flow behavior. However, because it is still at an early stage, it requires effort from major companies, engineers, academia and other organizations in order to further develop it until it reaches a maturity level which will allow it to become more widely used and recognized as a powerful tool for assessing reservoir modeling data. The enhancement of this technology looks promising as an innovative contribution to reservoir management. ESSS’s reservoir simulation data management platform, Kraken, features a dedicated Python library and interface to perform top-down modeling to allow data scientists focused on reservoir engineering to benefit from this type of capability.
 D. Solomatine, L.M. See and R.J. Abrahart. Practical Hydroinformatics. Computational Intelligence and Technological Developments in Water Applications. Chapter 2: Data-Driven Modelling: Concepts, Approaches and Experiences. 2008.
 Esmaili, Soodabeh, and Shahab D. Mohaghegh. “Full field reservoir modeling of shale assets using advanced data-driven analytics.” Geoscience Frontiers 7.1 (2016): 11-20.
 Mohaghegh, S.D., West Virginia University & Intelligent Solutions, Inc., Gaskari, R. and Maysami, M., Intelligent Solutions, Inc., Khazaeni, Y. Boston University. Data-Driven Reservoir Management of a Giant Mature Oilfield in the Middle East. SPE, 2014.