基于多方面信息学习的电网启动送电方案特征词抽取模型  

Feature Word Extraction Model for Power Grid Start-Up Transmission Operation Schemes Based on Multi-Aspect Information Learning

作　　者：于洋 谢民 邵庆祝 张骏 张沛[2] 饶国政[3] YU Yang;XIE Min;SHAO Qingzhu;ZHANG Jun;ZHANG Pei;RAO Guozheng(State Grid Anhui Electric Power Company,Hefei 230061,China;School of Electrical Engineering,Beijing Jiaotong University,Beijing 100044,China;College of Intelligence and Computing,Tianjin University,Tianjin 300350,China)

机构地区：[1]国网安徽省电力有限公司,安徽合肥230061 [2]北京交通大学电气工程学院,北京100044 [3]天津大学智能与计算学部,天津300350

出　　处：《电力信息与通信技术》2022年第9期25-33,共9页Electric Power Information and Communication Technology

基　　金：国家自然科学基金项目(61832014);国网安徽省电力有限公司科技项目(B31200200007)。

摘　　要：现有中文信息抽取存在中文词中丰富的信息被忽略的问题。为提高特征词抽取的质量,文章提出一种基于多方面信息学习的电网启动送电方案特征词抽取模型。所提模型首先从方案的词嵌入信息、实体类型信息、字符表示及位置嵌入信息等多个方面学习尽可能丰富的特征信息,生成具有多个方面信息的特征向量;然后卷积神经网络自动加权的卷积操作对特征向量进行学习,并进一步采用双向长短记忆网络和注意力机制进行处理;最后,通过相邻数据信息的线性链条件随机场进行解码,实现对特征词的抽取。实验结果表明,所提基于多方面信息学习的双向长短记忆网络及注意力机制的模型对不同电网设备类型、状态和动作特征词抽取的各项性能均有较大提升。The existing Chinese information extraction methods have the problem of ignoring the rich information in Chinese words.A feature word extraction model for power grid start-up transmission operation schemes based on multi-aspect information learning is proposed to improve the quality of feature word extraction.We propose a method to learn as rich feature information as possible from the scheme,such as word embedding information,entity type information,character representation,and location embedding information.Feature vectors with multi-aspect information are generated.The feature vectors are learned by the convolutional neural network with the automatically weighted convolution operation.Furthermore,it is processed using bidirectional long short-term memory and attention mechanism.Finally,the processed results are decoded by linear conditional random fields of adjacent data information to realize the extraction of feature words.The experimental results show that the bidirectional long short-term memory and attention mechanism model proposed in this paper based on multi-aspect information learning has improved the performance of different grid device types,states,and action feature extraction.

关 键 词：电网启动送电方案 双向长短记忆网络 自注意力 信息抽取 

分 类 号：TM86[电气工程—高电压与绝缘技术]