Synthetic Production Log Generation for Well Productivity Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The oil and gas industry faces challenges in accurately determining the productivity level and potential of oil wells downhole at the reservoir level without relying on expensive and time-consuming production logging tools, which often require well interventions that are not operationally or economically viable.
Innovation Solution
A method using a machine learning trained model to predict oil flow values at perforated intervals of wells by accessing historical data from databases, including well production, completions, and reservoir properties, and generating a synthetic production log that includes predicted oil flow values, thereby providing a cost-effective and accurate estimate of downhole productivity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If production logging tools are used to determine downhole productivity levels, then measurement precision is improved, but device complexity and cost increase
Solution Approach 1:
The patent creates a synthetic production log that copies the essential information from actual production logging tool measurements. By training a machine learning model on historical data including production log measurements, the system generates synthetic logs that replicate the productivity information without requiring physical logging tools to be deployed in target wells.
Solution Approach 2:
The patent replaces the mechanical/physical production logging tools with a computational machine learning system. Instead of physically deploying spinner flow meters or other downhole measurement devices, the system uses trained models that process surface-level and historical data to predict downhole productivity levels.
2Reliability
If production logging tools are deployed to obtain downhole measurements, then reliability of productivity data is improved, but loss of time and operational viability deteriorate
Solution Approach 1:
The patent performs preliminary action by training the machine learning model on historical production log data before deploying it to target wells. This pre-training phase captures the relationship between various well parameters and actual productivity measurements, so that when the model is applied to new wells, it can immediately generate reliable predictions without requiring time-consuming on-site logging operations.
Solution Approach 2:
The synthetic production log copies the essential productivity information that would otherwise require time-consuming field measurements, enabling rapid assessment of multiple wells without repeated physical interventions.
3Measurement precision
If conventional production logging methods are used, then measurement precision is improved, but ease of operation and economic viability worsen
Solution Approach 1:
The patent substitutes complex mechanical production logging operations with a computational system that processes data through trained machine learning models. This replacement maintains measurement precision while dramatically simplifying operations, as the synthetic log generation requires only data input and model execution rather than physical tool deployment and retrieval.
Solution Approach 2:
The synthetic production log provides a simplified copy of the information obtained through complex logging operations, making the process easier to perform while maintaining the essential measurement accuracy needed for operational decisions.
Data Source
AI summary
A method for predicting oil flow rates is provided. The method includes accessing historical data from a plurality of databases, accessing historical perforation data and historical reservoir properties data from a simulation model, and determining fluid flow values and rock quality index values associated with perforated intervals of the plurality of wells. The method further includes corresponding the fluid flow values and rock quality values to the well production data, training, using the plurality of input values, a machine learning model for predicting oil flow values at perforated intervals of a plurality of target wells, predicting, using the trained machine learning model, the oil flow values at the perforated intervals of the plurality of target wells, and generating a synthetic production log that includes the predicted oil flow values at the perforated intervals of the plurality of target wells.