Deep Learning Neural Networks for Well Production Forecasting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional well productivity forecasting methods are either too simplistic or excessively complex and time-consuming, failing to provide accurate and efficient real-time estimates due to their limitations in incorporating realistic oilfield operation constraints.
Innovation Solution
The implementation of deep learning neural networks, specifically using long-short term memory (LSTM) or gated recurrent unit (GRU) and convolutional neural networks (CNN), to forecast well production rates by analyzing temporal and spatial data, optimizing well development and completion strategies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If numerical simulation of fluid flow through porous media by solving partial differential equations is used, then measurement precision is improved, but loss of time increases
Solution Approach 1:
The patent creates a simplified copy of the complex numerical simulation model by training a neural network on historical simulation data. The trained network serves as a surrogate model that replicates the behavior of the full numerical simulator but executes much faster, enabling real-time forecasting without repeated PDE solutions.
Solution Approach 2:
The patent performs preliminary numerical simulations to generate training data before deployment. By pre-computing a comprehensive dataset covering various reservoir conditions and well configurations, the system prepares all necessary information in advance, allowing the neural network to make rapid predictions without performing full numerical simulations during operation.
2Productivity
If deep learning neural networks are used, then productivity is improved, but device complexity increases
Solution Approach 1:
The patent divides the forecasting task into separate neural network components, each handling specific aspects of reservoir simulation. By segmenting the model into modular units that can be trained and deployed independently, the system achieves high productivity while managing complexity through organized, reusable components.
Data Source
AI summary
Disclosed are systems and methods for obtaining an input sequence of input data features associated with a well for at least one time stamp during a period of time including well production rates for the well and well operation constraints for the well, dividing the input data features into a training data subset, a validation data subset, and a test data subset, building a well production model for the well using machine learning based on the training data subset, the validation data subset, and the test data subset, and generating a forecast for the well for a future period of time using the well production model, the forecast comprising a future well production rate for the well including at least one of an oil rate, a gas rate, and a water rate.