Machine Learning Sidetrack Prediction for Oil Gain
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting oil gains from horizontal sidetracking of producer wells are inefficient, relying on reservoir simulation or statistical analogs, which are time-consuming and prone to bias, and do not effectively evaluate sidetrack candidates or reduce the need for continuous drilling evaluations.
Innovation Solution
A computer-implemented method using machine learning, specifically a multilayer perception (MLP) neural network, that predicts oil gains and water reduction by selecting relevant independent variables from production, design, and subsurface data, providing a quick and unbiased evaluation of sidetrack performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If reservoir simulation or statistical analogs are used to predict sidetrack performance, then prediction accuracy is maintained, but evaluation time increases significantly and continuous drilling evaluations are required
Solution Approach 1:
The patent applies preliminary action by pre-training a machine learning model on historical sidetrack data and reservoir characteristics before actual sidetrack evaluation. The model learns patterns and relationships from past performance, enabling rapid prediction of oil gains and water reduction for new sidetrack candidates without requiring time-consuming reservoir simulations at the time of evaluation.
Solution Approach 2:
The patent uses copying by creating a simplified machine learning model that replicates the predictive capabilities of complex reservoir simulations. Instead of running full reservoir simulations for each sidetrack candidate, the system uses a trained ML model that copies the essential prediction functionality, dramatically reducing evaluation time while maintaining acceptable accuracy.
2Adaptability or versatility
If engineers manually evaluate sidetrack candidates considering multiple variables, then comprehensive assessment is possible, but objectivity is lost due to biases and limitations in the number of variables considered
Solution Approach 1:
The patent applies universality by designing a machine learning model that can handle multiple types of input variables simultaneously - geological parameters, reservoir properties, well completion data, and production history. This universal approach allows comprehensive assessment of sidetrack candidates while maintaining objectivity, as the model consistently applies the same evaluation criteria to all candidates without human bias.
Solution Approach 2:
The system applies self-service by automatically evaluating sidetrack candidates using the trained machine learning model without requiring manual engineer intervention for each evaluation. The model independently processes input data, applies learned patterns, and generates predictions, eliminating human biases while maintaining comprehensive variable consideration.
3Reliability
If continuous drilling evaluations and screening are performed, then thorough candidate assessment is achieved, but capital expenditures increase and the process becomes inefficient
Solution Approach 1:
The patent applies taking out by extracting the essential prediction functionality from complex, time-consuming reservoir simulations and encapsulating it in a streamlined machine learning model. This extracted model retains the core predictive capabilities needed for thorough assessment while removing the computational overhead, enabling efficient evaluation of multiple sidetrack candidates without sacrificing assessment quality.
Solution Approach 2:
The patent applies parameter changes by transforming the evaluation process from requiring full reservoir simulation parameters to using a reduced set of key input parameters that the machine learning model has learned are most predictive of sidetrack performance. This parameter reduction maintains assessment thoroughness while dramatically improving drilling efficiency and reducing capital expenditures.
Data Source
AI summary
Systems and methods include a computer-implemented method for displaying incremental values of average rate variations over a cumulative time window. A subset of independent variables corresponding to production variables of the oil well are selected using statistical analysis from a set of independent variables corresponding to production features of an oil well. The production parameters include performance variables for production of the oil well, design variables of the design of the oil well, and modeled independent variables of the oil well. Using the subset of independent variables and machine learning, predicted values of dependent variables are determined including an oil gain (ΔQo) and a water reduction (ΔQw) associated with production of the oil well. A display specifying incremental values of average rate variations over a cumulative time window is generated in a graphical user interface. The display is based at least in part on the predicted values of dependent variables.