Using machine learning to adjust ultrasonic flow meter performance

The article proposes and describes an algorithm for adjusting the operation of an ultrasonic flow meter. Unlike conventional gas meters, ultrasonic flow meters have such advantages as contactless operation, minimal pressure losses and high accuracy. The use of integration methods in the theory of ultrasonic flow meters imposes its own requirements on the velocity profiles of the gas flowing through the flow meter. Strongly curved profiles lead to large errors in measuring gas flow, so it is so important to be able to adjust the flow meter readings when they operate on a curved flow near various local resistances. The authors describe in detail the machine learning algorithm based on Gaussian processes for efficient prediction of flow meter correction factors. Block diagrams of the proposed method implementation are provided. The method is based on the use of a training sample. In turn, the training sample is a set of parameters calculated using three-dimensional computer modeling. The calculations were performed for more than 100 classes of local resistances (single bends, double bends in one and different planes, diffusers, confusers and other combinations of the described local resistances), as well as for more than 30 000 conditions of the ultrasonic flowmeter installation relative to local resistances. for the entire set of flow converter installation conditions in the Reynolds number range of 1E6-2E7. The analysis of the efficiency of the method for correcting the ultrasonic flowmeter readings was performed. The diagrams of the distribution of the results of predicting the K-factor by error are shown. It is shown that in more than 70 % of the variants of the ultrasonic flowmeter location after the local resistance, the algorithm determines the correction factor with an accuracy of less than 0,1 %. The developed and described algorithm for predicting the correction factors for correcting the values of the measured flow of natural gas allows calculating the correction factor with good accuracy and thereby eliminates the need to use flow shapers and straighteners in front of ultrasonic flowmeters.