基于神经网络的地埋管换热器出水温度预测

Export Temperature Prediction of Buried Pipe Heat Exchanger based on Neural Network

本文利用人工神经网络对地埋管换热器出口水温的预测,出口水温对热泵的有效操作以及能源节约都至关重要。在系统模拟以及系统识别中,人工神经网络应用广泛。为了训练人工神经网络预测模型,采用有限的实验方法组为训练、测试数据。在此研究中,在输入层里,含有土壤导热系数、埋管间距、钻孔深度、管内水流量、地埋管换热器入口水温、输入热流量的大小及时间;地埋管换热器出口水温在输出层。网络中,反向传播学习三种算法分别为:Levenberg-Marquardt算法(LM),比例共轭梯度算法(SCG)以及动量批梯度下降函数(GDM),同时运用切线非传递函数,从而得出最佳方法。预测结果显示,最合适的演算法以及隐藏神经元的数量为LM-10。训练之后,均方根(RMS)为1%,方差值R2的绝对分数为99.9%,最大cov的变异系数比为25.7%。说明人工神经网络可以对地埋管换热器出口水温精确预测。

This paper shows the applicability of Artificial Neural Networks （ANNs） to predict export water temperature of heat exchanger. Export water temperature of heat exchanger is important for the optimal control and energy saving operation of heat pump systems. ANNs have been used in varied applications and they are useful in system modelling and system identification. In order to train the ANNs, limited experimental measurements were used as training data and test data. In this study, in input layer, there are soil coefficient of thermal conductivity, buried tube spacing, drilling depth, water flow, buried tube heat exchanger entrance temperature and the size and the time input of heat flow; export water temperature is in output layer. The back propagation learning algorithm with three different variants, namely Levenberg-Marguardt（LM）, Gradient Descent Method （GDM）, and Scaled Conjugate Gradient（SCG）.The best algorithm and neuron number in the hidden layer are found as LM with ten neurons. After the training, it is found that RMS is 1%, and R2 is 99.999% and coefficient of variation （coy） in percent value is 25.7%. So ANNs can be used for prediction of exoort water temoerature of heat exchanger as an accurate method.