Machine learning model with Bayesian optimization for ultrasonic flowmeter in-use measurement

The ultrasonic flowmeter is one of the most widely used flowmeters in natural gas trade. An efficient in-use measurement method to guarantee the accuracy of ultrasonic flowmeter in service is needed. A comprehensive machine learning model with Bayesian optimization for ultrasonic flowmeter in-use measurement is presented. The approach was demonstrated on Random Forest (RF). The test was fulfilled on the gas flow standard facility, and the meter under test is a 4-path DN200 ultrasonic flowmeter. The consistency of the labelled data and predictions of the model was evaluated. After Bayesian optimization, the performance of RF model improved significantly. Based on the evaluation criteria, the increase of model scores is up to 5 %, indicating that Bayesian optimization can efficiently improve the machine learning based in-use measurement method for ultrasonic flowmeter.