Author: Azucena Gomez
Global Business Manager for Digital Transformation
Author: Samiran Roy
Principal Consultant, Big Data Center of Excellence
One of the most important factors in successfully identifying reservoirs and estimating realistic reservoir parameters is to understand play characteristics within the associated geology. The first steps are to identify geobodies with anomalous seismic behavior and then visualize them without any artifacts such as multiples, smiles or frowns resulting from migration anomalies, offset gathers which are not flat, etc.
The second step is to rank the geobodies based on the likelihood of hydrocarbon occurrence in the sub-surface versus the estimated risks associated with recovering it. Determining the types of reservoir fluids is the most important factor in this workflow. The third step is to quantify other reservoir properties, such as risked volumes of gas in identified prospects.
At each step, uncertainty can be introduced by human interpretation and bias. What if there was a way to minimize these uncertainties?
Artificial Intelligence (AI) and Machine Learning (ML) can help you overcome the myriad challenges in making reservoir fluid predictions. By applying AI to estimate fluid likelihood in identified geoanomalies, you can focus on creating business value for your operation.
Direct Hydrocarbon Indicators (DHI) play an important role in identifying hydrocarbon reservoirs by using patterns in seismic signatures based on both pre-stacked and post-stacked data. Amplitude versus angle (AVA) and amplitude versus offset (AVO) are often used to estimate reservoir fluid. However, the workflows are time-consuming and there are higher levels of uncertainty especially in the presence of coal and residual gas. Estimation of reservoir fluids through AVO/AVA analysis of seismic offset gathers requires conditioning the gathers and domain expertise.
A data driven ML project ingests all this data and extracts geoanomalies using automated algorithms to generate AVA/AVO plots. It then develops models to categorize each anomaly with its probabilistic fluid presence. Once the ML model is properly trained, the resulting likelihood score of prospects by AI and ML can reduce your interpretation time cycle, adding further business value.
Build a Better Model to Predict Fluid Likelihood
The schematic below shows the process of fluid estimation modeling using AI/ML. Raw subsurface data is ingested, its geospatial location is recorded, and the data is sent to Exploratory Data Analysis (EDA) and preprocessing tools. At that point one copy of the data is stored for future use and another copy goes to ML core workflows, where an iterative process trains and optimizes the model.
Figure: The workflow includes EDA and preprocessing components that perform outlier analysis, noise removal, and apply attributions. The data is then sent to a stable and robust ML modeling scheme.
As the model is trained, the Individual Analyzer allows users to analyze each prospect individually with its seismic gathers behavior and estimates the probable fluid presence with its likelihood of occurrence. Location-wise or trace-wise response is the key for the Individual Analyzer to investigate the feature’s importance for predicting reservoir fluid. Then all the information in the form of model features are stored in ASCII format for further updates and the model output can be consumed in any visualization solution, enabling universal adaptability between multiple solutions or software.
Bias-free Explainable Predictions, Faster Results
In the E&P industry, AI is often brought in to help make complex, high-stakes decisions which involve people, safety, or huge sums of money. But E&P end users can be averse to entrusting those decisions to black-box algorithms where cause-and-effect relationships between data inputs and model outputs are not clear to the decision makers.
Unlike conventional black box AI, our solution is explainable: it provides insight into the steps and features the AI uses to draw its conclusions, so you know how solutions are derived. The result is faster and bias-free fluid likelihood prediction.
The clarity offered by our data-driven AI approach can reduce uncertainties introduced by human interpretation and bias, helping you to mitigate the risks associated with those uncertainties. When you can overcome the challenges in making reservoir fluid likelihood predictions, you can accelerate your decision-making process.
As part of our SmartDigital® co-innovation service, we can collaborate alongside you to help you learn how AI and ML can enhance your exploration portfolio management and make a demonstrable impact on your E&P business. Contact us today at firstname.lastname@example.org.