Hydraulic Fracturing Job Plan Revision With Time-Series Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current hydraulic fracturing (HF) job planning lacks data-driven and automated methods for optimizing well production, relying heavily on subjective decisions and manual event detection, which leads to inefficiencies and suboptimal outcomes.
Innovation Solution
Employing data-driven statistical modeling and machine learning algorithms to automate HF event detection and integrate multiple data sources for real-time job plan adjustments, using ensemble modeling techniques to enhance well productivity predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If simple bilinear or multivariate linear models are used to model well production, then the model complexity is low, but the model accuracy is insufficient to adequately explain the effect of multiple input features on well productivity outcomes
Solution Approach 1:
The patent segments the modeling problem into two distinct components: a geospatial model that handles uncontrollable features (well location, reservoir properties) and a predictive model that handles controllable features (stimulation parameters, completion features). This segmentation allows each model to focus on specific aspects, improving overall accuracy without requiring a single overly complex model to capture all relationships.
Solution Approach 2:
The patent introduces an intermediary approach by using geospatial modeling techniques (such as kriging or universal kriging) to estimate the effect of uncontrollable features, then using these estimates as inputs or corrections to the predictive model. This intermediary step allows the system to account for spatial variability and uncontrollable factors without making the primary predictive model excessively complex.
2Measurement precision
If controllable and uncontrollable features are incorporated in the same statistical model, then the model includes all relevant features, but the uncontrollable features (such as well location) dominate the model accuracy and make it difficult to tease out the effect of controllable features on production outcomes
Solution Approach 1:
The patent explicitly segments features into controllable (stimulation and completion parameters) and uncontrollable (well location, reservoir properties) categories, and builds separate modeling components for each. The geospatial model handles uncontrollable features while the predictive model focuses on controllable features, allowing clear isolation and analysis of the effects of each feature type on production outcomes.
Solution Approach 2:
The patent extracts and separates the effect of uncontrollable features through geospatial modeling, removing their dominant influence from the predictive model. This extraction allows the predictive model to focus specifically on the relationships between controllable features and production outcomes without being overshadowed by the variability introduced by uncontrollable factors.
3Productivity
If manual event detection and subjective decisions are used in HF job planning, then the process is simple to implement, but the efficiency is low and the outcomes are suboptimal
Solution Approach 1:
The patent implements an automated job planning system that uses machine learning models to independently analyze data, detect events, and generate optimization recommendations without requiring manual intervention. The system processes stimulation and completion data, automatically identifies key events and patterns, and provides actionable insights for improving well production, thereby eliminating the inefficiencies of manual event detection and subjective decision-making.
Data Source
AI summary
The disclosure is directed to methods to design and revise hydraulic fracturing (HF) job plans. The methods can utilize one or more data sources from public, proprietary, confidential, and historical sources. The methods can build mathematical, statistical, machine learning, neural network, and deep learning models to predict production outcomes based on the data source inputs. In some aspects, the data sources are processed, quality checked, and combined into composite data sources. In some aspects, ensemble modeling techniques can be applied to combine multiple data sources and multiple models. In some aspects, response features can be utilized as data inputs into the modeling process. In some aspects, time-series extracted features can be utilized as data inputs into the modeling process. In some aspects, the methods can be used to build a HF job plan prior to the start of work at a well site. In other aspects, the methods can be used to revise an existing HF job plan in real-time, such as after a treatment cycle, a pumping stage, or a time interval.