A Natural Gas Energy Metering Method Based on Density-Sound Velocity Correlation.

Area of Science:

• Energy Science
• Mechanical Engineering
• Data Science

Background:

• Accurate natural gas energy metering is crucial for commercial transactions.
• Existing methods face challenges in precision under varying operating conditions.

Purpose of the Study:

• To develop a novel, high-precision method for natural gas energy metering.
• To leverage physical property correlations and machine learning for improved accuracy.

Main Methods:

• Constructed a 10,000-sample dataset for natural gas properties.
• Employed machine learning algorithms to build predictive models for compression factor and calorific value.
• Developed a model-switching strategy based on input feature ranges to enhance prediction accuracy.

Main Results:

• Achieved high accuracy for compression factor (MAE: 0.00118, R²: 0.9987) and calorific value (MAE: 0.1583, R²: 0.9736) models.
• Validated the method on a real-flow test bench with maximum prediction errors of 0.061% (compression factor) and 1.19% (calorific value).
• Demonstrated a maximum relative energy error of 1.21% across various natural gas samples.

Conclusions:

• The proposed method effectively achieves accurate natural gas energy metering under practical operating conditions.
• Integration of physical property correlations with machine learning offers a robust solution for energy measurement.
• The model-switching strategy significantly enhances the predictive performance for natural gas properties.