ML-Based PVT Fluid Clustering for Streaming Reservoir Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing reservoir characterization techniques face challenges in efficiently monitoring and predicting hydrocarbon fluid properties in large reservoirs where samples become available piecemeal after production has commenced, particularly due to the vast amount and streaming nature of geophysical data that overwhelm human analysis capabilities.
Innovation Solution
Utilizing machine learning (ML) techniques to segregate and classify hydrocarbon fluid samples based on earlier acquired PVT data, leveraging existing equation of state (EoS) models to compute fluid properties iteratively and generate realistic renderings of production courses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional reservoir characterization techniques are used to analyze hydrocarbon fluid samples, then fluid properties can be determined, but the vast amount and streaming nature of geophysical data overwhelm human analysis capabilities and reduce efficiency
Solution Approach 1:
The patent replaces manual human analysis of reservoir fluid samples with an automated computerized system that uses machine learning algorithms. The system automatically clusters compositional measurements, classifies new samples into clusters, and drives thermodynamic models to determine fluid properties, eliminating the bottleneck of human analysis capability and enabling efficient processing of large volumes of streaming geophysical data.
2Measurement precision
If thermodynamic models are developed for each individual hydrocarbon fluid sample, then accurate fluid properties can be determined, but the process becomes computationally intensive and time-consuming when samples become available piecemeal
Solution Approach 1:
The patent segments the reservoir into multiple clusters based on compositional measurements from different locations. Instead of developing thermodynamic models for each individual sample, the system groups similar samples into clusters and develops a single thermodynamic model for each cluster. This segmentation approach maintains measurement precision by preserving the unique characteristics of each cluster while dramatically reducing the time and computational resources needed compared to individual sample modeling.
Solution Approach 2:
The patent creates universal thermodynamic models that can be applied to multiple samples within a cluster. Each cluster model serves as a universal representation for all samples in that cluster, allowing the system to determine fluid properties for numerous samples using a single model. This multi-functionality approach maintains accuracy by capturing the essential characteristics of each group while eliminating redundant model development efforts.
3Productivity
If machine learning techniques are used to cluster and classify hydrocarbon fluid samples, then data processing efficiency is improved, but system complexity increases
Solution Approach 1:
The patent introduces machine learning algorithms as intermediary components that bridge the gap between raw compositional measurements and thermodynamic model application. The ML system acts as a mediator that automatically performs clustering and classification tasks, transforming unprocessed sample data into organized clusters that can be efficiently processed by thermodynamic models. This intermediary layer automates complex data processing while maintaining a clear workflow structure.
Data Source
AI summary
A method includes: accessing a first dataset comprising: (i) a first plurality of records of compositional measurements, and (ii) a data structure encoding a plurality of thermodynamic models developed using pressure-volume-temperature (PVT) data; accessing a second dataset comprising a second plurality of records of compositional measurements; analyzing the first plurality of records to generate a plurality of clusters; classifying the second plurality of records into one or more clusters generated from the first plurality of records; driving a thermodynamic model from the plurality of thermodynamic models that corresponds to a given cluster to determine a fluid property of portions of the hydrocarbon fluid samples that correspond to portions of the second plurality of records classified into the given cluster; and presenting a rendering of the fluid property as time elapses and additional records become available from the second dataset.