Prediction of Cavitation Characteristics of Throttle Orifice Plate Based on Improved BP Neural Network

In order to obtain the cavitation characteristics near throttle orifice plate in nuclear pipeline efficiently, a reliable improved backpropagation (BP) neural network prediction model was constructed. Firstly, the geometric feature parameters of the throttle orifice plate were extracted, and the sample data base of these parameters was generated by using the Latin Hypercube Sampling approach. Then, the minimum cavitation number corresponding to each sample is obtained by computational fluid dynamics (CFD) method, and the dimensionless parameter is used as the output response. Finally, in view of the deficiency of the original BP neural network prediction model, an improved prediction model of throttle orifice plate cavitation characteristics is established by combining genetic algorithm. The results show that the orifice diameter and front opening angle have strong global sensitivity to the minimum cavitation number; the prediction accuracy of the BP neural network prediction model optimized by genetic algorithm has been greatly improved, and the root mean square error is reduced by about 36.4%.