基于改进BP神经网络的节流孔板空化特性预测

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

为高效地获取核级管道中节流孔板附近的空化特性,构建了可靠的改进反向传播(BP)神经网络预测模型。首先提取了节流孔板的几何特征参数,并使用拉丁超立方抽样(LHS)方法生成了上述几何特征参数的样本库;然后通过计算流体力学(CFD)方法得到了各个样本对应的最小空化数,以该无量纲参数作为输出响应;最后针对原始BP神经网络预测模型的不足,结合遗传算法建立了节流孔板空化特性的改进预测模型。结果表明,孔板开孔直径和前开角度对最小空化数具有较强的全局敏感度;通过遗传算法优化后的BP神经网络预测模型的预测精度得到了大幅提升,误差均方根降低约36.4%。

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%.