基于BP神经网络的接触网绝缘子干冰清洗研究  

Research on Dry Ice Cleaning for Insulator of Catenary Based on BP Neural Network

作　　者：王国志[1] 侯杰文 张元彬 WANG Guozhi;HOU Jiewen;ZHANG Yuanbin(Advanced Drive Energy Saving Technology Engineering Research Center of Ministry of Education,Southwest Jiaotong University,Chengdu 610031)

机构地区：[1]西南交通大学先进驱动节能技术教育部工程研究中心,成都610031

出　　处：《计算机与数字工程》2022年第7期1609-1614,共6页Computer & Digital Engineering

摘　　要：为了解决目前接触网绝缘子干冰清洗过程中缺乏全面的理论指导,难以实现清洗效果与成本节约的平衡,提出一种CFD技术与BP神经网络相结合的研究方法。参考实际情况以及相关安全标准,确定清洗参数压缩空气压强、干冰质量流量、清洗角度、清洗靶距的范围,采用拉丁超立方抽样方法抽样50组参数组合进行CFD仿真试验,其结果作为神经网络训练集,再另抽取10组作为测试集来验证训练好的模型的精度。根据误差分析表明,该模型为高精度模型,能有效地反映各清洗参数之间的交互作用。这种方法能对清洗过程进行合理指导,提高清洗效率,降低干冰消耗,对清洗工程作业具有实际意义。In order to solve the current lack of theoretical guidance in the dry ice cleaning process of insulators of catenary,it is difficult to achieve a balance between cleaning effect and cost savings,a research method combining CFD technology and BP neu⁃ral network is proposed.The actual situation and relevant safety standards is referred to determine the range of cleaning parameters compressed air pressure,dry ice mass flow rate,cleaning angle,and cleaning target distance.Latin hypercube sampling method is used to sample 50 sets of parameter combinations for CFD simulation test,and the results are used as neural network training and then another 10 groups are extracted as a test set to verify the accuracy of the trained model.According to the error analysis,the model is a high-precision model,which can effectively reflect the interaction between various cleaning parameters.This method can reasonably guide the cleaning process,improve cleaning efficiency,and reduce dry ice consumption,which is of great significance to cleaning engineering operations.

关 键 词：干冰清洗 接触网绝缘子 CFD BP神经网络 拉丁超立方抽样 

分 类 号：U226.8[交通运输工程—道路与铁道工程]