Integrating a Data Driven Model into a Multilayer Pattern Waterflood Simulator

Abstract

This work proposes the integration of a data driven model into a multilayer pattern analytical simulator. This model allows quantifying areal distribution injection in each layer and models the delay in the producer's response through two kind of parameters used in the simulation. This paper deals with several problems appearing in the parameters’ identification process in real waterflood projects with large amount of unknowns and practical situations that lead to ill conditioned matrices. The waterflood simulator consists of multi - layer patterns. These injection patterns contain a series of flow elements that link each injector well with the neighboring producing wells in each layer. The patterns’ construction takes into account the geometry involved in the injection and producer wells configuration in each layer. Patterns change in time as wells/layers are opened or shut in or wells are converted. A complete algorithm able to support the dynamic changes of the patterns was developed to identify the set of parameters in the data driven model (distribution coefficients and time constants) to fit the total production/layer injection history. Then, each flow element, characterized by its pore volume and mobile oil saturation, can be swept by the water injected in the layer, areally prorated by the distribution coefficients towards each producer well. The developed identification algorithm proved to be fast and robust to identify distribution coefficients and time constants in several synthetic cases and real multilayer reservoirs. The algorithm was able to handle the high number of unknowns generated in big multilayer multistage waterflood projects. Problems related to ill conditioned matrices, such as collinearities due to multilayer injection and insufficient data in the identification process, were successfully solved using different strategies. The data driven model reduces the effort and uncertainty associated with the proper distribution of injection each time a pattern changes simultaneously modeling the delay in producers’ response, improving the whole simulation process. Besides, it is capable to detect low connection zones highlighting the presence of useless flow elements due to low permeability regions, sealing faults or operational problems. The impact of the integration of the data driven model into the simulation process was evaluated quantifying history matching and computational run time for real waterflood projects. A complete methodology to perform simulations in mature fields is presented. For the first time, a data driven model is integrated into a waterflood pattern simulator in order to quantify connectivities between wells once the patterns have been constructed in time. In this way, the data driven model incorporates geometric considerations and geological information. The methodology is especially useful when production comes from multilayer reservoirs or from old fields where the lack of information impedes performing traditional numerical simulations.

A State-of-the-Art Literature Review on Capacitance Resistance Models for Reservoir Characterization and Performance Forecasting

“…In addition to the previously mentioned references, recent developments and applications of ML-CRMs can be found in [36][37][38][39].…”

“…Equations (34)- (36) set the basis of the Koval model. However, they are expressed in terms of saturation which is not directly measured in the field, and µ o , µ s and H may be unavailable.…”