BP神经网络预测架空覆冰导线气动系数  

Aerodynamic Coefficients Prediction of Overhead Iced Conductor by BP Neural Network

作　　者：刘小会[1,2] 周顺 汪顺 蔡萌琦 严波[4] 闵光云 LIU Xiaohui;ZHOU Shun;WANG Shun;CAI Mengqi;YAN Bo;MIN Guangyun(School of Civil Engineering,Chongqing Jiaotong University,Chongqing 400074,China;State Key Laboratory of Bridge and Tunnel Engineering in Mountain Areas,Chongqing Jiaotong University,Chongqing 400074,China;School of Architecture and Civil Engineering,Chengdu University,Chengdu 610106,China;College of Aerospace Engineering,Chongqing University,Chongqing 400044,China;SinoFrench Institute of Nuclear Engineering and Technology,Sun Yatsen University,Zhuhai 519082,Guangdong,China)

机构地区：[1]重庆交通大学土木工程学院,重庆400074 [2]重庆交通大学省部共建山区桥梁及隧道工程国家重点实验室,重庆400074 [3]成都大学建筑与土木工程学院,成都610106 [4]重庆大学航空航天学院,重庆400044 [5]中山大学中法核工程与技术学院,广东珠海519082

出　　处：《实验室研究与探索》2022年第3期42-49,共8页Research and Exploration In Laboratory

基　　金：国家自然科学基金资助项目(51308570,51507106);重庆市自然科学基金面上项目(cstc2021jcyj-msxmX0166)。

摘　　要：为了解决风洞试验求覆冰输电线在风荷载下的气动系数成本高、耗时长的问题,提出了一种预测气动系数的BP神经网络方法。使用试凑法对BP神经网络模型进行调参训练,选择8×8、10×8、10×16的BP神经网络模型。以训练好的BP神经网络模型为基础,先预测了新月形、扇形覆冰单导线在不同风速下的气动系数,再预测了无覆冰、新月形、扇形覆冰四分裂导线在不同风速下的气动系数以及新月形覆冰四分裂导线在不同覆冰厚度下的气动系数,最后与风洞试验所测结果以及线性插值的结果对比。结果表明:BP神经网络的预测结果与风洞试验结果拟合程度较线性插值结果好,相似度很高,BP神经网络预测方法具有可行性。In order to solve the high-cost and time-consuming problem in obtaining the aerodynamic coefficients of iced transmission lines under wind load, a BP neural network method for predicting aerodynamic parameters is presented. The cut-and-try method is used to train the parameters of BP neural network model, and the BP neural network models of 8×8, 10×8 and 10×16 are selected. Firstly, the aerodynamic coefficients of crescent-shaped and fan-shaped iced single conductor under different wind speeds are predicted by the BP neural network method. And then the aerodynamic parameters of no icing, crescent-shaped and fan-shaped iced quad bundle conductors under different wind speeds are predicted, and the aerodynamic coefficients of crescent-shaped iced quad bundled conductors at different thicknesses are predicted. In addition, the results are compared with those measured by wind tunnel test and linear interpolation. The results show that the fitting degree between the prediction results of BP neural network and wind tunnel test results is better than that of linear interpolation, and the similarity is very high, which shows the feasibility of BP neural network prediction method.

关 键 词：BP神经网络 气动系数 单导线 四分裂导线 

分 类 号：TM752[电气工程—电力系统及自动化]