Machine Learning Pore Pressure Prediction Using Offset Well Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting formation pore pressure ahead of drilling rely heavily on seismic data with low vertical resolution and wireline logs, which are not accurate and lack real-time capabilities, posing risks to safety and efficiency during drilling operations.
Innovation Solution
A method utilizing machine learning models trained with surface drilling parameters and mud gas data from offset wells to predict formation pore pressure profiles, excluding data from the closest offset well, and updating predictions in real-time during drilling using real-time mud gas data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If seismic data is used for formation pore pressure prediction, then prediction coverage is improved, but measurement precision deteriorates due to low vertical resolution
Solution Approach 1:
The patent segments the prediction task by using different data sources for different purposes: seismic data provides broad spatial coverage while machine learning models trained on wireline logs provide high-precision predictions at specific well locations. The well-log data is segmented and used to train models that can then predict pore pressure with high vertical resolution at multiple locations.
Solution Approach 2:
The patent introduces machine learning models as an intermediary between the low-resolution seismic data and the high-precision wireline log data. The ML models learn the relationship between seismic features and pore pressure from wireline-log-labeled data, then apply this learned relationship to predict pore pressure with high precision at locations where only seismic data is available.
2Measurement precision
If wireline logs are used for formation pore pressure prediction, then measurement precision is improved, but productivity deteriorates as data is only acquired after drilling is completed
Solution Approach 1:
The patent performs preliminary action by training machine learning models during the drilling planning phase using wireline log data from offset wells. These pre-trained models are then deployed to make real-time pore pressure predictions during drilling operations, enabling proactive drilling plan optimization before actual drilling begins and real-time adjustments during drilling.
Solution Approach 2:
The patent creates a virtual copy of the wireline log analysis capability through machine learning models. Instead of requiring actual wireline logs from each well being drilled, the system copies the predictive power of wireline log analysis by training models on offset well data and applying them to predict pore pressure in real-time during drilling operations.
3Productivity
If seismic-while-drilling approach is used, then productivity is improved with real-time predictions, but device complexity and cost increase
Solution Approach 1:
The patent enables the drilling operation to serve itself by using standard drilling data (weight on bit, rate of penetration, mud pressure) that is already collected during normal drilling operations. Machine learning models process this self-generated data to provide real-time pore pressure predictions, eliminating the need for separate specialized measurement systems while still achieving real-time prediction capability.
Solution Approach 2:
The patent makes the machine learning models universal by training them on multiple data sources (seismic data, wireline logs, and drilling data) that can all be processed by the same model framework. This multi-functional approach allows a single system to handle various data types and provide predictions across different drilling scenarios without requiring separate specialized systems.
4Reliability
If traditional drilling planning is performed without accurate pore pressure predictions, then device complexity is reduced, but reliability deteriorates due to safety risks from overpressures
Solution Approach 1:
The patent applies beforehand cushioning by making pore pressure predictions during the drilling planning phase using machine learning models trained on offset well data. These advance predictions allow drilling parameters such as mud weight to be optimized before drilling begins, cushioning against potential safety risks from overpressures by having preventive measures already in place before the drilling operation starts.
Data Source
AI summary
A method for facilitating drilling of a prospect involves, for a multitude of offset wells associated with the prospect, obtaining offset well data, the offset well data including surface drilling parameters, mud gas data, and formation pore pressure data. The method further involves training, using the offset well data, a machine learning (ML) model to make formation pore pressure predictions, where the offset well data used for training the ML model excludes the offset well data of an offset well in closest proximity to the prospect. The method also involves generating a formation pore pressure profile prediction for the prospect prior to drilling the prospect by making formation pore pressure predictions for the offset well in closest proximity to the prospect using the ML model operating on the offset well data of the offset well in closest proximity to the prospect.