改进的BP神经网络对飞机换热器结垢厚度预测

Prediction of Fouling Thickness of Aircraft Heat Exchanger by Modified BP Neural Network

为了节约飞机维修成本,准确预测换热器结垢厚度,通过利用改进的BP神经网络预测模型,利用25组数据,建立了换热器结垢厚度与四个因素(环境温度、空调系统进口压力、初级换热器出口温度、次级换热器出口温度)之间的网络预测模型。模型包括4个输入神经元,9个隐含层神经元和1个输出层神经元。训练结果表明,改进之后的BP神经网络模型不仅克服了原始BP神经网络收敛速度慢,稳定性差的特点,还可以以较高的精度预测换热器的结垢厚度。

In order to save the cost of aircraft maintenance,by using the modified BP neural network prediction model and using 25 sets of data,the network prediction model between the scaling thickness of heat exchanger and four factors(ambient temperature,inlet pressure of air conditioning system,outlet temperature of primary heat ex-changer and outlet temperature of secondary heat exchanger)was established.The model consists of 4 input neurons,9 hidden layer neurons and one output layer neuron.The training results show that the modified BP neural network model can not only overcome the characteristics of slow convergence rate and poor stability of the original BP neural network,but also predict the scaling thickness of the heat exchanger with high accuracy.