摘要
随着电网结构日渐复杂,电网新设备启动愈来愈多,依赖电网运行人员手工编制启动方案无法满足电网智能管理需求。基于大量历史电网新设备启动方案数据,智能编制电网新设备启动方案成为研究的重点。但电网历史新设备启动方案为非结构化数据,无法直接利用。为将非结构化数据转化为结构化数据、提高命名实体识别准确率,文章提出多分类BiLSTM-CRF模型,通过word2vec将编码启动方案利用余弦相似度生成相似度矩阵,将启动方案按相似度分类,利用Bi LSTM-CRF训练多个模型,最后标注出识别结果,将非结构化数据转化为结构化数据。通过小规模数据集验证,结果表明,文章所提算法在准确率、训练时间、F1值上均取得较好结果,能够准确将历史电网新设备启动方案转化为结构化数据,给研究人员提供数据基础。
With the increasing complexity of power grid structure,there are more and more new power grid equipment startup.Relying on the manual preparation of startup scheme by power grid operators can not meet the needs of power grid intelligent management.Based on a large number of historical power grid new equipment startup scheme data,the preparation of smart grid new equipment startup scheme has become the focus of research.However,the power grid historical new equipment startup scheme is unstructured data,which can not be used directly.In order to transform unstructured data into structured data and improve the accuracy of named entity recognition,a multi classification BiLSTM-CRF model is proposed in this paper.The coding start-up scheme is coded by word2vec,the cosine similarity is used to generate the similarity matrix,the start-up scheme is classified according to the similarity,multiple models are trained by BiLSTM-CRF,and finally the recognition results are marked to transform the unstructured data into structured data.The experimental results show that the proposed algorithm achieves good results in accuracy,training time and F1 value.It can accurately convert the historical power grid new equipment startup scheme into structured data and provide data basis for researchers.
作者
张大波
郭怀新
储著伟
王博欣
ZHANG Dabo;GUO Huaixin;CHU Zhuwei;WANG Boxin(Anhui Province Key Laboratory of Renewable Energy Utilization and Energy Saving,Hefei University of Technology,Hefei 230009,Anhui Province,China;Huaibei Power Supply Company,State Grid Anhui Electric Power Co.,Ltd.,Huaibei 235000,Anhui Province,China;Anqing Power Supply Company,State Grid Anhui Electric Power Co.,Ltd.,Anqing 246001,Anhui Province,China;Rongcheng Power Supply Company,State Grid Shandong Electric Power Company,Rongcheng 264300,Shandong Province,China)
出处
《电力信息与通信技术》
2023年第1期54-61,共8页
Electric Power Information and Communication Technology
基金
国家自然科学基金重点项目(51637004)
中央高校基本科研业务费资助(PA2021GDSK0085)。
