摘要
为了研究考虑高海拔多环境因素影响下输电线路可听噪声的预测问题,在海拔2400 m高度点的500 kV同塔双回线路下,搭建了边相外20、30、35 m三处可听噪声观测站,同时利用气象站进行多环境因素指标的数据采集。文中提出了一种基于多头注意力机制(multi⁃head attention,MHA)的卷积神经网络(convolutional neural network,CNN)—双向长短期记忆网络(bi⁃directional long short term memory,BiLSTM)模型进行可听噪声预测。首先,采用皮尔逊相关性分析对多种环境因素数据进行相关程度计算比较与剔除;然后,为充分挖掘可听噪声数据中的时序特征,使用CNN对多环境因素数据进行特征提取;再将提取的特征向量输入到BiLSTM中进行训练,并通过在BiLSTM端引入多头注意力机制,使模型学习权重更高的数据特征,从而提升模型预测精度;结果表明,该方法构建的组合模型可以提升考虑多因素特征可听噪声数据的预测精度,且具有较好的泛化性。
To study the prediction of audible noise of transmission lines under the influence of multiple environmen⁃tal factors at high altitudes,such three audible noise observation stations as 20 m,30 m and 35 m away from the side phase are set up under the 500 kV double⁃circuit line on the same tower at the altitude of 2400 m.At the same time,the meteorological station is used to collect data on multiple environmental factors.In this paper,a kind of convolu⁃tional neural network(CNN)⁃bi⁃directional long short⁃term memory(BiLSTM)model based on the multi⁃head atten⁃tion(MHA)mechanism is proposed for audible noise prediction.Firstly,the Pearson correlation analysis is used for calculation,comparison and elimination of correlation degree of various environmental factors data.Then,in order to fully mine the temporal features of the audible noise data,CNN is used for feature extraction of the multiple environ⁃mental factors data.Then,the extracted feature vectors are input to BiLSTM for training,and the multi⁃head atten⁃tion mechanism is introduced at the BiLSTM side to make the model learn data features with higher weight,thus im⁃proving the prediction accuracy of the model.The results show that the combined model constructed by this method can improve the prediction accuracy of the audible noise data considering multi⁃factor features and has good general⁃ization.
作者
黄悦华
张子豪
陈庆
刘兴韬
涂金童
HUANG Yuehua;ZHANG Zihao;CHEN Qing;LIU Xingtao;TU Jintong(College of Electrical Engineering and New Energy,China Three Gorges University,Hubei Yichang 443002,China)
出处
《高压电器》
CAS
CSCD
北大核心
2024年第12期160-169,共10页
High Voltage Apparatus
基金
国家自然科学基金资助项目(52007103)
国家电网有限公司总部科技项目(SGSCDK00HBJS2100038)。
关键词
输电线路可听噪声
多环境因素
卷积神经网络
双向长短期记忆网络
多头注意力机制
audible noise of transmission line
multi⁃environmental factors
convolutional neural network
bi⁃directional long short⁃term memory
multi⁃head attention
