Preference Learning Model Selection for Drilling ROP Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current drilling models in the oil and gas industry face challenges in accurately predicting rate of penetration (ROP) due to varying assumptions and lengthy learning processes, leading to potential delays and safety issues when selecting between different models based on sensor signals.
Innovation Solution
A model-based optimization technique using preference learning to select between linear and non-linear models, adjusting biases based on user preferences and real-time inputs, to optimize controllable parameters like weight on bit (WOB) and rotations per minute (RPM) for improved ROP prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If model selection is learned over a course of time through iterative matching of predictions to outcomes, then the model selection accuracy can be improved, but the learning process takes considerable time and introduces substantial room for error
Solution Approach 1:
The system pre-processes historical drilling data to pre-train multiple ROP prediction models with different assumptions and characteristics before actual drilling operations begin. This preliminary preparation allows the system to have ready-to-use models instead of needing to learn and train models in real-time during drilling, thus reducing the learning process time while maintaining model selection accuracy.
2Adaptability or versatility
If multiple models are used to predict ROP values, then the prediction coverage and adaptability are improved, but the complexity of model selection and management increases
Solution Approach 1:
The system introduces a sensor signal-based model selection mechanism that uses readily available sensor data (WOB, RPM, vibration, temperature) as an intermediary to automatically select the most appropriate ROP prediction model. This intermediary approach simplifies the complex task of choosing among multiple models by using objective, real-time sensor readings rather than requiring complex comparative analysis of multiple model predictions.
Solution Approach 2:
The system changes the selection criterion from comparing model prediction accuracies to using sensor signal thresholds and patterns. By monitoring parameters like vibration levels, temperature, and operational conditions, the system dynamically selects models based on current drilling conditions, reducing the complexity of model management while maintaining adaptability across different drilling scenarios.
3Measurement precision
If a lengthy iterative learning process is used to update model selection, then the model accuracy can be improved, but errors can occur during iteration causing delays and safety issues
Solution Approach 1:
The system uses real-time sensor data from the drilling operation itself to automatically select and update model preferences without requiring external intervention or lengthy iterative learning processes. The drilling operation's own sensor signals (vibration, temperature, WOB, RPM) serve as the basis for model selection, allowing the system to self-adjust and maintain accuracy while ensuring operational reliability and avoiding delays.
Data Source
AI summary
A model optimizer for predicting a drill bit variable can select a model from multiple models based on a learned preference. The preference may be updated according to preference indicator received from a user in response to an output model selection.