摘要
针对当前分布式光伏智能运维缺乏特定知识库支撑的问题,研究提出一种改进的面向智慧运维的分布式光伏知识库构建方法,该方法从实体提取和实体关系提取两个方面进行改进。为减小文本固定句式对知识库构建效果的影响,设计了基于词向量(word embedding)改进的TextRank算法的实体提取方法。为解决现有模型分类器层与数据匹配不足的问题,设计了基于多标签问题改进的分段卷积神经网络模型的实体关系提取方法。通过实验对比分析结果表明,改进后的方法可有效提升自动化程度,减少人力成本,提升准确性。
Aiming at the lack of domain-specific knowledge base in many distributed photovoltaic maintenance systems,a framework regarding how to construct a knowledge base intelligently for smart maintenance of distributed photovoltaic was proposed,which included two aspects that were entity extraction and entity relation extraction.To relax the potential susceptibility from sentence patterns,an improved TextRank algorithm based on word embedding was applied for entity extraction.To improve the unfitness between the classifier layer and data deficiency,a Piecewise Convolutional Neural Network(PCNN)model for multi-labeling was leveraged to extract entity relations.Categories of experiments showed that the improved approaches could help to provide a higher degree of automation and reduce labor costs with better accuracy.
作者
欧一鸣
苏雍贺
邹孝付
靳健
张长志
陶飞
OU Yiming;SU Yonghe;ZOU Xiaofu;JIN Jian;ZHANG Changzhi;TAO Fei(School of Automation Science and Electrical Engineering,Beihang University,Beijing 100191,China;School of Government,Beijing Normal University,Beijing 100875,China;Electric Power Research Institute,State Grid Tianjin Electric Power Company,Tianjin 300384,China)
出处
《计算机集成制造系统》
EI
CSCD
北大核心
2020年第12期3205-3215,共11页
Computer Integrated Manufacturing Systems
基金
国家重点研发计划资助项目(2018YFB1500800)
河南省机械装备智能制造重点实验室开放基金资助项目(IM201901)
国家电网有限公司科技资助项目“分布式光伏系统智慧运维技术”(SGTJDK00DYJS2000148)。
关键词
分布式光伏
智能运维
知识库
知识图谱
自然语言处理
distributed photovoltaic
smart maintenance
knowledge base
knowledge graph
natural language processing
