Abstract: The ultrasonic flow meter for trade handover needs to be verified by qualified legal metrological verification institutions before it can be used. Affected by many factors such as working condition temperature, pressure, temperament composition and installation conditions, there are differences between the real flow verification results of ultrasonic flow meters and the factory test correction Settings. In this paper, the key factors affecting the indication error of ultrasonic flowmeter are analyzed, and a real flow indication error prediction model of ultrasonic flowmeter is established based on the integrated machine learning principle. The operating parameters such as signal-to-noise ratio and gain of each sound channel of the flowmeter and the indication error of air calibration are taken as the model input, and the real flow indication error at the corresponding flow point is taken as the output. The test of DN80, DN200 and DN400 ultrasonic flowmeters with different diameters is carried out. The average prediction accuracy of the ultrasonic flowmeter is better than that of the test results, and the small flow points of Q_t and below can also be predicted.