Predicting Bound Fluid Volumes Using Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional reservoir characterization methods fail to accurately identify movable water in subsurface regions, leading to challenges in hydrocarbon production and field development, particularly due to the lack of magnetic resonance (MR) logs in horizontal wells due to budget constraints.
Innovation Solution
A machine-learning based system that predicts magnetic resonance logs (NMR) using resistivity-density-neutron-GR logs, leveraging data on calcite, dolomite, and uninvaded zone water to develop a predictive model for bound fluid volume, enabling accurate identification of non-hydrocarbon fluids and optimizing well drilling operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If magnetic resonance (MR) logs are collected to accurately identify bound fluid volumes, then measurement precision is improved, but cost and device complexity increase significantly
Solution Approach 1:
The patent creates a predictive copy of NMR log data by training a machine learning model on wells that have both basic logs and NMR logs. The trained model then generates predicted NMR responses for wells with only basic logs, effectively copying the valuable NMR information without requiring actual NMR log acquisition. This resolves the contradiction by providing measurement precision through prediction while avoiding the device complexity and cost of actual NMR logging.
2Reliability
If magnetic resonance (MR) logs are acquired to identify water bearing intervals, then reliability of fluid identification is improved, but loss of time and increased cost occur
Solution Approach 1:
The patent performs preliminary action by pre-training the machine learning model on a database of wells with complete data (including NMR logs) before deployment. Once trained, the model can rapidly predict NMR responses for new wells without requiring actual NMR log acquisition time. This resolves the contradiction by establishing reliability through preliminary model training while eliminating the time loss associated with actual NMR logging operations.
3Ease of operation
If basic resistivity-density-neutron-GR logs are used without NMR logs, then ease of operation and cost are improved, but measurement precision and reliability of bound fluid volume identification deteriorate
Solution Approach 1:
The patent introduces a machine learning model as an intermediary that bridges the gap between basic logs and NMR log information. The model takes easily acquired basic logs as input and generates predicted NMR responses as output, effectively mediating between simple data acquisition and precise fluid volume measurement. This resolves the contradiction by maintaining ease of operation through basic log acquisition while achieving measurement precision through the intermediary ML model prediction.
Data Source
AI summary
Methods and systems, including computer programs encoded on a computer storage medium are described for implementing a system that predicts bound fluid volumes for use in well drilling operations at a subsurface region. The system derives inputs from log data generated for one or more wells. A predictive model of the system processes each of the inputs based on algorithms used to train the predictive model. Based on the processing, the model computes correlations between data points in the log data and reference parameters that are indicative of a fluid volume at the subsurface region. Based on the computed correlation, the model generates a prediction that includes a bound fluid volume for the subsurface region. The system determines a characteristic of a non-hydrocarbon fluid at a first zone of the subsurface region based on the bound fluid volume.