中文摘要：

建立BP和GA-BP神经网络预测离心泵反转作液力透平的压头和效率.用63台离心泵作透平的参数作为训练两个网络的样本,以泵的叶轮出口直径、叶片包、出口宽度、出口安放角、叶片数和比转速作网络输入层,透平压头和效率作输出层.用6组样本外的数据测试经训练后的两个网络的预测能力,并做网络预测值与试验值的相关性和线性回归分析.结果表明,BP网络对压头和效率预测的平均误差为5.33%和0.86%,GA-BP网络对压头和效率预测的平均误差为3.56%和0.46%.GA-BP网络预测精度高,预测结果与实验值相关性强,预测所用时间为BP网络的1/3,更适合做泵反转作液力透平的性能预测.

英文摘要：

BP and GA-BP neural networks were established to predict the head and efficiency of centrifugal pumps working as hydraulic turbine.The parameters of 63 centrifugal pumps as turbines were used as sample for training above-mentioned two neural networks.The outlet diameter of impeller,blade wrap angle,blade outlet width,blade outlet angle,number of blade,and specific speed of pump were taken as input layer of the network and the head and efficiency of the turbine were taken as output layer.Six groups of extra data beyond the samples were used to test the prediction ability of the two already trained neural networks.The analysis of the correlativity and linear regression of predicted values to experimental values was made.The result showed that the average error of head and efficiency predicted with BP network was respectively 5.33%and 0.85%and the average error of head and efficiency predicted with GA-BP network was 3.56% and 0.46%,respectively.GA-BP network had higher predictive accuracy,close correlation of predicted results to experimental results,and its time spent on prediction was 1/3of that with BP network.GA-BP network was more suitable for performance prediction of a pump working as hydraulic turbine.