Title : Machine learning application for joint rock physics model optimization, facies classification and compaction modeling: A North Sea example
Abstract:
Petrophysical facies interpretation and Rock Physics modelling are two fundamental stages in any quantitative interpretation workflow. To perform these stages accurately and efficiently, data exchange and cooperation with different subject matter experts are critical. As collaboration tends to increase complexity, reliable automation is a needed tool for fast and smooth delivery. Automation for petrophysical facies interpretation translates to machine learning algorithms that, at the current state-of-the-art, are not able to fully capture compaction trends. The error added to such automated ML interpretation may propagate in the Rock Physics modelling and ultimately in the inversion.
To overcome these current ML limits, we have developed the Rock Physics Machine Learning (RPML) toolkit that takes compaction trends into account in the inversion for petrophysical facies and rock physics model parameters. The toolkit outputs the interpreted petrophysical facies logs, the petro-elastic depth trends and the rock physics parameters as optimized by the inversion. The RPML toolkit allows saving of both the prior, regionally bound models, as well as the posterior calibrated rock physics models, thus optimizing the data exchange among teams. In fact, a regional prior model can be tested, approved, and applied to different areas in the same region, assuring consistency and accuracy across different projects. The case study presented here is set in the Central North Sea, within the Forties Field and illustrates the flexibility of the application, consistency with manual facies interpretation, and value of cross-disciplinary integration.
Audience Takeaway Notes:
- The novel RPML toolkit blends machine learning and rock physics to provide the necessary guidance to petrophysical interpretation of facies Rock Physics model calibration, depth trends identification. It provides balance with automation for more consistent and less biased interpretations.
- The novel Rock Physics driven machine learning algorithm demonstrates great efficiency in facies prediction for large dataset, while keeping true to the pressure trends, since it is depth and pressure aware.
- The novel toolkit RPML automates key QI workflow steps: petrophysical interpretation, classification, depth trend identification per facies and Rock Physics Models calibration.
