No abstract

No abstract

In recent times, several machine learning algorithms have been utilized to analyze massive amount of data and provide fast and effective smart solution. To monitor the saturation over time, reservoir saturation tool is run in each well to provide the cased hole saturations. A single well may have as many has 15-20 runs over time which provides an estimate of present saturation, depletion across reservoirs or any other changes in saturation profile to EOR system in place and its impact. This paper demonstrates the idea of the use of machine learning algorithm to perform a time lapse saturation analysis and present a saturation forecast based on class-based machine learning combined with the time series modelling using multiple time lapse runs in a single well for a field. The first step involves using an unsupervised class-based machine learning model that classifies the input petrophysical data into set of classes with distinct similarity. The next step of the workflow involves using time series modeling on each of the obtained classes. Different methods were studied and eventually "Prophet" was used for time series modeling to forecast saturation over time. A way forward and game changer is propagating the classes to multiple wells with multi runs and predict the changes in saturation and pressure for a formation across a field to make holistic interpretation. Reversing the time series modeling can also provide an estimate of OH conditions and original hydrocarbon in place. The use of CBML and time series modelling for petrophysical saturation evaluation will open new doors to understanding of fields already under production and introduce smart decision-making capabilities. The idea presented in this paper can be implemented in different fields across the globe with different formation settings and EOR system in place and can be further advanced in terms of providing deep insight and interpretation like forecasting the change in reservoir pressure.

The present oil industry is more challenged than ever to develop novel methods for oil exploration and production, while reducing costs at the same time. This necessity changes the need of logging tools for reservoir characterization. Saturation height modeling (SHM) is an important aspect of determining the production capability of an oilfield. This is often performed by taking core samples, which is pivotal for such analysis, but expensive and challenging. Further, cores are usually taken in the zones of interests in the well. This calls for an alternate analysis, which is not only available for the entire interval of the well but is also less expensive than the traditional coring techniques. Nuclear Magnetic Resonance (NMR) applications have proved promising over the years to perform SHM, without using cores. NMR, however, has a shallow depth of investigation and using wireline measurements is even more challenging due to longer time after bit and increased mud filtrate invasion. Consequently, its use is restricted to quantifying porosity. This makes it imperative to remove the effect of any filtrate or hydrocarbons from NMR logs to be able to use them for any advance analysis. A novel methodology is presented in this paper to perform SHM analysis in carbonates. It uses NMR data along with modern processing techniques like factor analysis (Jain et al. 2013) and fluid substitution (Minh et al. 2016) and integrated workflow to define hydrocarbon uncontaminated pseudo capillary pressure curves and saturation height functions for different rock facies observed in the formation. The results are validated on five wells in the same field, and further confirmation is also done with testing results.

No abstract