Mud Logging Hydrocarbon Prediction via ML Regression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing analysis methodologies in the oil and gas industry for identifying hydrocarbon resources through mud gas analysis are primarily qualitative and struggle to provide reliable, quantitative data due to variations in drilling parameters, environmental changes, and fluid changes.
Innovation Solution
A method and system utilizing machine learning techniques, specifically statistical classification and regression analysis, to predict unknown characteristics of hydrocarbon resources from input data on measured properties of gas samples extracted from drilling fluid. This involves building models using preexisting databases containing known fluid properties of reservoir fluids to identify fluid types, estimate heavier hydrocarbon fractions, and predict properties like gas-oil ratio (GOR) and stock tank oil (STO) density.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If traditional qualitative analysis methodology is used for mud gas analysis, then the analysis process is simple and easy to implement, but the measurement precision and reliability of hydrocarbon resource identification deteriorates
Solution Approach 1:
The patent replaces traditional qualitative analysis methods with machine learning-based quantitative prediction models. The system uses statistical classification and regression analysis to transform mud gas chromatography data into quantitative predictions of fluid properties (GOR, STO density, fluid type), substituting manual interpretation with automated computational algorithms that provide both ease of operation and high measurement precision
Solution Approach 2:
The patent changes the analytical parameters from qualitative categories to quantitative continuous variables. By using regression analysis to predict specific numerical values for gas-oil ratio, stock tank oil density, and other fluid properties, the system transforms the nature of the output parameters while maintaining operational simplicity through automated processing
2Measurement precision
If machine learning models are implemented for quantitative prediction, then the measurement precision and reliability of hydrocarbon resource identification improves, but the device complexity and computational requirements increase
Solution Approach 1:
The patent applies preliminary action by pre-training machine learning models using extensive databases of known fluid properties before deployment. The statistical classification and regression models are built in advance using historical mud gas analysis data and corresponding PVT laboratory results, allowing the system to provide accurate quantitative predictions without requiring complex real-time computation during actual drilling operations
Solution Approach 2:
The patent introduces an intermediary layer of pre-built prediction models that act as mediators between raw mud gas chromatography data and final hydrocarbon resource characterization. These models, trained on comprehensive databases, serve as computational intermediaries that translate simple input data into accurate quantitative predictions without requiring the end system to handle the full complexity of the analysis
3Reliability
If comprehensive databases of known reservoir fluids are used for model training, then the reliability and accuracy of predictions improves, but the time and resources required for model building increases
Solution Approach 1:
The patent performs preliminary action by building and training comprehensive machine learning models in advance using extensive databases of known reservoir fluid properties. The statistical classification and regression models are developed beforehand using historical mud gas analysis data paired with laboratory-measured PVT properties, enabling rapid and reliable predictions during actual drilling operations without requiring time-consuming analysis at that stage
Solution Approach 2:
The patent transforms the model building process by changing from traditional method-by-method development to a unified statistical learning approach. By using machine learning algorithms that can simultaneously learn multiple fluid property predictions from comprehensive databases, the system reduces overall model building time while improving reliability through the use of diverse training data covering various reservoir conditions
Data Source
AI summary
Methods (and related apparatus) include obtaining data regarding a measured property. The measured property includes an amount of each of predetermined hydrocarbons in a gas sample extracted from drilling fluid exiting a wellbore having a hydrocarbon resource. An unknown characteristic of an investigated fluid property of the hydrocarbon resource is predicted utilizing the obtained input data and one or more predetermined models each built via statistical classification and regression analysis of a preexisting database containing records. Each record includes known characteristics of fluid properties of a different one of known reservoir fluids. The fluid properties include the investigated fluid property and the measured property. The investigated fluid property includes a fluid type of the hydrocarbon resource, an amount of at least one additional hydrocarbon, gas-oil ratio, or stock tank oil density.