%0 Journal Article
%T 基于PCA的EC管外壁阻垢率预测模型研究<br>Research on Prediction Model of Scale Resistance Rate of EC Pipe Outer Wall Based on PCA
%A 陈杰
%A 陈永保
%A 刘婧楠
%A 赵顺安
%A 章立新
%J Modeling and Simulation
%P 76-87
%@ 2324-870X
%D 2022
%I Hans Publishing
%R 10.12677/MOS.2022.111007
%X 由于蒸发式冷凝器(EC)换热管外阻垢率不易准确测试，搭建了一个在喷淋水中加入柠檬酸(CA)阻垢的实验装置并获取了180组实验数据。本文利用BP、GRNN两种神经网络对阻垢率进行预测并结合主成分分析(PCA)对实验输入参数进行降维并对降维前后的预测结果进行对比。降维前两种模型的输入参数为实验进行时间、换热管外壁温度、喷淋水温度、pH值、电导率，降维后两种模型的输入参数为Y1和Y2，输出参数均为阻垢率。对比降维前后的预测结果得出：经过PCA降维后，BP的网络运行时间从20s降为15s，误差指标RMSE、MAPE分别从2.45%、3.6%降为1.44%、2.38%，相关系数R从0.9745升高到0.9885；GRNN的运行时间从0.8s降为0.4s，误差指标RMSE、MAPE分别从1.78%、2.85%降为1.04%、1.98%，相关系数R从0.9853升高到0.9966，并且降维后GRNN模型比BP模型预测时间超短、误差指标小、相关性更高，更适合本领域。<br />
Because it is difficult to accurately measure the external scale inhibition rate of evaporative con-denser heat exchange tube, an experimental device for adding citric acid into spray water was built and 180 groups of experimental data were obtained. In this paper, BP and GRNN neural networks are used to predict the scale inhibition rate, and combined with PCA, dimension reduction of experimental input parameters is carried out, and the prediction results before and after dimension reduction are compared. Before dimensionality reduction, the input parameters of the two models are experiment time, outer wall temperature of heat exchange tube, spray water temperature, pH value and conductivity. After dimensionality reduction, the input parameters of the two models are Y1 and Y2, and the output parameters are scale inhibition rate. By comparing the prediction results before and after dimension reduction, the running time of BP network decreased from 20s to 15s, error index RMSE and MAPE decreased from 2.45% and 3.6% to 1.44% and 2.38% respectively, and correlation coefficient R increased from 0.9745 to 0.9885. The running time of GRNN decreased from 0.8s to 0.4s, the error index RMSE and MAPE decreased from 1.78% and 2.85% to 1.04% and 1.98% respectively, and the correlation coefficient R increased from 0.9853 to 0.9966. Moreover, GRNN model after dimensionality reduction has shorter prediction time, smaller error index and higher correlation than BP model, which is more suitable for this field.
%K 蒸发式冷凝器，阻垢，主成分分析，神经网络预测<br>Evaporative Condenser
%K Discaling
%K PCA
%K Neural Network Prediction
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=47801