Fracture Driven Interaction Detection via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting and mitigating fracture-driven interactions in hydrocarbon production fields are inefficient, relying on heuristic models and legacy data of poor quality, which are expensive and lack precision in predicting frac interference and identifying effective mitigative actions.
Innovation Solution
A method using machine learning algorithms to analyze well completion and production data, identifying parent-child well pairs, generating event labels based on dynamic pressure exceptions and fluid ratio behavior, and training a model to characterize fracture-driven interactions, enabling precise prediction and mitigative actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If legacy data such as well head pressure is used to detect fracture interference, then detection capability is provided, but the data quality is poor and insufficient to distinguish between valid frac-event and operational noise
Solution Approach 1:
The patent combines multiple data sources including well head pressure, production data, and operational data into a unified analysis framework. By merging these diverse data streams, the system overcomes the limitations of individual legacy data sources and achieves both reliable detection and precise measurement of fracture interference events.
Solution Approach 2:
The patent introduces machine learning models as intermediary components that process and interpret legacy data. These models act as mediators between the raw low-quality data and the detection output, extracting meaningful signals while filtering out operational noise, thereby improving both reliability and measurement precision.
2Reliability
If heuristic models are used to model fracture propagation, then prediction capability is provided, but extensive computing resources are required and precision is not achieved
Solution Approach 1:
The patent replaces complex heuristic modeling approaches with machine learning models that are better suited for the available data. This substitution reduces computing resource requirements while maintaining or improving prediction precision, as the ML models can efficiently process the actual operational data without requiring extensive computational simulations.
Solution Approach 2:
The patent changes the approach from physics-based heuristic parameters to data-driven parameters that better reflect actual field conditions. By using parameters derived from real operational data rather than theoretical models, the system achieves higher precision with reduced computational complexity.
3Measurement precision
If machine learning methods are used with secondary effects such as long term changes in parent well production, then fracture interference detection is provided, but the methods face challenges with noisy primary data and require extensive time windows
Solution Approach 1:
The patent performs preliminary processing and feature extraction on primary data before applying machine learning models. By pre-processing the noisy data to extract relevant features and patterns, the system reduces the impact of noise without requiring excessively long time windows, thereby improving detection accuracy while minimizing time loss.
Solution Approach 2:
The patent segments the analysis into multiple components including primary data analysis, secondary effect analysis, and feature integration. This segmentation allows the system to process short-term primary data effects separately from long-term secondary effects, improving detection accuracy without requiring the entire long time window to be analyzed as a single unit.
4Adaptability or versatility
If current broad and regional methods are used to address fracture interference, then general applicability is provided, but the methods are unable to reference direct data related to the phenomenon that varies by region, reservoir zone, and historical production history
Solution Approach 1:
The patent implements dynamic, adaptive machine learning models that can adjust to different regions, reservoir zones, and production histories. Rather than using static broad methods, the system learns from local data patterns and adapts its detection parameters, achieving both general applicability across different fields and high precision for region-specific conditions.
Data Source
AI summary
A method is described to detect, analyze, and characterize fracture driven interaction (FDI) events in unconventional resources using production data from a well and its nearest neighbors. The method provides a rigorous statistical analysis of the production data of a well and its nearest neighbors, and utilizes a combination of signals including pressure, rate, water-oil ratio (WOR), and fluid production to identify and characterize FDI events. The method is executed by a computer system.