Machine Learning Drilling Model for ROP Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current drilling technologies face challenges in accurately predicting the rate of penetration (ROP) in subterranean formations due to the dynamic and complex interactions between various geological and operational factors, limiting the efficiency and productivity of well-drilling operations.
Innovation Solution
A method involving the selection of analog wells based on similarity criteria to generate a training data set, which is then used with a machine-learning algorithm to create a drilling model that predicts the ROP profile for a target well, enabling more accurate drilling operation planning and execution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional drilling modeling methods are used, then the drilling operation can be performed, but the accuracy of ROP predictions is insufficient due to dynamic and complex interactions between geological and operational factors
Solution Approach 1:
The patent creates virtual copies of actual drilling operations by generating synthetic training data sets that replicate the complex interactions between geological formations and drilling parameters. These synthetic data sets are created by perturbing actual drilling data with realistic variations, allowing the machine learning model to learn from multiple simulated scenarios without requiring additional physical drilling operations. This copying approach enables accurate ROP predictions while avoiding the need for complex physical modeling.
Solution Approach 2:
The patent performs preliminary data preparation and model training before actual drilling operations begin. Training data sets are generated in advance by collecting historical drilling data, applying perturbations to create synthetic scenarios, and training the machine learning model beforehand. This preliminary action allows the model to be ready for deployment with accurate predictions from the start of new drilling operations, rather than attempting to model complex interactions in real-time during drilling.
2Measurement precision
If more drilling parameters and geological factors are considered in the model, then the prediction accuracy improves, but the data processing and model training complexity increases
Solution Approach 1:
The patent performs extensive data processing, feature engineering, and model training in advance before drilling operations begin. Historical drilling data is collected, cleaned, and processed beforehand. Synthetic training data sets are generated by applying perturbations to historical data, and the machine learning model is trained on these prepared data sets before deployment. This preliminary action reduces real-time processing requirements during actual drilling operations.
Solution Approach 2:
Instead of processing complex geological and operational data in real-time, the patent creates synthetic copies of drilling scenarios by perturbing historical data. These synthetic data sets replicate the complexity of real drilling conditions without requiring actual real-time data processing. The model learns from these pre-generated copies, enabling fast predictions during drilling operations without the computational burden of processing raw sensor data in real-time.
3Measurement precision
If a machine-learning based approach is implemented, then the ROP prediction accuracy improves by capturing dynamic interactions, but the initial setup and training process becomes more complex
Solution Approach 1:
The patent simplifies machine learning implementation by using synthetic data copies instead of requiring complex data collection infrastructure. Historical drilling data is perturbed to create synthetic training data sets that replicate real drilling conditions without needing additional sensors or data collection systems. This copying approach makes machine learning more accessible and easier to implement while maintaining high prediction accuracy.
Solution Approach 2:
The patent implements self-service mechanisms where the system automatically generates its own training data by perturbing historical data sets. The machine learning model trains on synthetic data that it generates itself, reducing the need for manual data collection, cleaning, and preparation. This self-service approach simplifies the overall implementation process while enabling the model to capture complex drilling dynamics.
Data Source
AI summary
The disclosure relates to a method for performing a drilling operation in a subterranean formation of a field. The method includes obtaining, prior to the drilling operation, a target well data set specifying a target well to be drilled, selecting, from a set of existing wells, a number of analog wells that satisfy a pre-determined similarity criterion with respect to the target well, generating, from a number of analog well data sets of the analog wells, a training data set for the target well, where the training data set includes a rate-of-penetration (ROP) profile for each analog well, generating, using a machine-learning algorithm and based on the training data set, a drilling model that predicts the ROP profile of the target well, and performing, based on the drilling model, modeling of the drilling operation to generate a predicted ROP profile of the target well.