Generative Adversarial Network for Gas Rate Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Accurate determination of hydrocarbon flow rates, particularly gas flow rates, during flowback operations is costly and has varying accuracies, which affects the economic viability and efficiency of well evaluations and ongoing production.
Innovation Solution
Employing a Generative Adversarial Network (GAN) to estimate gas rates by training a generator and discriminator neural networks to iteratively improve the generation and classification of gas rate data, using historical data from previously drilled wells.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional methods are used to determine hydrocarbon flow rates, then measurement accuracy is achieved, but cost increases
Solution Approach 1:
The patent creates synthetic gas rate data samples using a generator neural network that mimics the statistical properties and distributions of real measured data. These synthetic copies are then used to train the measurement system, eliminating the need for expensive physical measurements during the training phase while maintaining measurement accuracy.
Solution Approach 2:
The patent replaces the mechanical/physical measurement system with an artificial intelligence-based estimation system. Instead of using physical flow meters and measurement equipment that incur high operating costs, the system uses trained neural networks to estimate gas rates from well test data, significantly reducing operating costs while maintaining acceptable accuracy.
2Measurement precision
If more real measured data is collected for training, then model accuracy improves, but cost and time increase
Solution Approach 1:
The patent performs preliminary action by generating synthetic training data before the actual training process begins. The generator network is pre-trained to create realistic synthetic gas rate data that captures the underlying distributions and relationships, which then serves as training data for the measurement model, eliminating the need to wait for extensive real data collection.
Solution Approach 2:
The patent creates synthetic copies of real measured data through the generator neural network. These synthetic data samples replicate the statistical properties, distributions, and relationships of actual gas rate measurements, providing sufficient training material without requiring extensive real data collection in the field.
3Loss of energy
If a simple estimation model is used, then cost decreases, but measurement precision deteriorates
Solution Approach 1:
The patent employs a dynamic two-network architecture where the generator and discriminator networks iteratively adapt to each other during training. The generator becomes increasingly sophisticated at creating realistic synthetic data, while the discriminator becomes increasingly accurate at distinguishing real from synthetic data, resulting in a powerful estimation system that achieves high precision without excessive computational cost.
Solution Approach 2:
The discriminator network serves as an intermediary that validates the quality of synthetic data generated by the generator. This intermediary mechanism ensures that the synthetic training data maintains the statistical properties and relationships of real data, thereby preserving measurement precision while avoiding the need for complex direct measurement models.
Data Source
AI summary
A computer implemented method for estimating a gas rate using a generative adversarial network is described. The method includes inputting training data to a generator. The method includes providing input to a discriminator, the input comprising the generated data samples and real data samples, wherein the discriminator outputs a binary classification of the input. Additionally, the method includes training the discriminator and the generator by evaluating the output of the discriminator, and estimating gas rates using the trained generator.