Flow Measurement of Natural Gas in Pipeline Based on 1D-Convolutional Neural Network

• Main Author: Tianjiao Zhang
• Format: Article
• Language: English
• Published: Atlantis Press 2020-08-01
• Series: International Journal of Computational Intelligence Systems
• Subjects: Flow measurement; Ultrasonic signal; Arrival time; Feature recognition; Convolutional neural network
• Online Access: https://www.atlantis-press.com/article/125943389/view

Time-difference method is a vitally significant algorithm for measuring natural gas flow with ultrasonic gas flowmeter. The key of this algorithm is to accurately measure the arrival time of ultrasonic signal. However, it is difficult to determine the feature points corresponding to the arrival time stably and accurately. To solve this problem, based on great feature recognition ability of deep learning, one-dimensional-convolutional neural network (1D-CNN) is utilized to determine the arrival time of ultrasonic signal according to the feature of the arrival time. First of all, a dataset, which includes different features such as different arrival time, different signal-to-noises (SNRs), etc., is used as a training set to train the 1D-CNN. Then, based on the size of the training set, an 1D-CNN is designed which includes three convolution and pooling layers and one fully connected layer to determine the arrival time, and the gas flow rate is calculated. To verify this method, an experimental ultrasonic gas flowmeter system is developed. By comparing with the typical method of determining arrival time, most of the deviations distribute close to zero and less than ±5 us using the proposed 1D-CNN, which verifies the effectiveness of the proposed 1D-CNN method.