AI Well Data Prediction for Plug and Abandonment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge lies in automatically determining attributes related to plugging and abandonment (P+A) or re-purposing activities for client wells, given the complexities of data volume, relationships, and format variations across different wells and data storage methods.
Innovation Solution
A system utilizing an artificial intelligence (AI) system with machine learning processes to gather and process client and non-client data, derive computer-generated models, and provide recommendations for P+A activities, including probability of success, cost estimates, and risk identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are trained on historical well data to predict P+A attributes, then prediction accuracy and decision-making reliability improve, but data processing complexity and computational requirements increase
Solution Approach 1:
The system performs preliminary actions by pre-processing and normalizing historical well data before model training, creating standardized data structures and feature sets in advance. This reduces the complexity of real-time processing while maintaining prediction accuracy, as the computationally intensive data preparation work is completed beforehand.
Solution Approach 2:
The patent introduces intermediary components including data normalization layers, feature extraction modules, and standardized data formats that mediate between raw historical data and the machine learning models. These intermediaries simplify the data-processing interface and reduce computational complexity while preserving predictive accuracy.
2Reliability
If comprehensive historical data from multiple wells is analyzed to improve prediction reliability, then the quality of P+A recommendations improves, but data integration difficulty and processing time increase
Solution Approach 1:
The system segments the comprehensive historical data into distinct well-specific datasets, each processed independently through standardized pipelines. This segmentation allows parallel processing of multiple wells' data, reducing overall processing time while maintaining the reliability benefits of analyzing comprehensive historical information across the portfolio.
Solution Approach 2:
The patent applies parameter changes by transforming raw historical data into standardized features and normalized parameters that are optimized for machine learning processing. This transformation reduces processing time by converting unstructured data into formats that require less computational effort while preserving the reliability-enhancing information content.
3Measurement precision
If proprietary data sharing networks are established to enhance model training data quality, then prediction accuracy improves, but data security risks and privacy concerns increase
Solution Approach 1:
The system introduces intermediary mechanisms including federated learning architectures, encrypted data transmission protocols, and anonymization layers that enable proprietary data sharing while maintaining security. These intermediaries allow multiple operators to contribute data for enhanced model training without exposing sensitive proprietary information, thus improving prediction accuracy while mitigating security risks.
Solution Approach 2:
The patent employs copying by creating anonymized replicas and synthetic representations of proprietary well data that can be shared across the network without exposing actual sensitive information. These copies preserve the statistical properties and predictive value needed for accurate modeling while eliminating security risks associated with sharing real proprietary data.
Data Source
AI summary
Techniques for determining one or more attributes in relation to transitional activities for a client well are provided. Embodiments include receiving structured data and unstructured data related to a well from one or more data sources and extracting, from the structured data and the unstructured data related to the well, numerical features, text features, and categorical features. Embodiments include generating embeddings based on the text features and generating encoding vectors based on the categorical features. Embodiments include providing, as inputs to one or more machine learning models, the numerical features, the embeddings, and the encoding vectors. Embodiments include receiving, as outputs from the one or more machine learning models in response to the inputs, one or more predicted values related to one or more transitional activities for the well. Embodiments include generating, based on the one or more predicted values, a transitional recommendation for the well.