NLP和推理引擎下电力基建现场风险区段识别  

Risk Section Identification of Electric Power Construction Site Based on NLP and Reasoning Engine

作　　者：钱朝军 李俊 宗震 张龙 邬桐 QIAN Chaojun;LI Jun;ZONG Zhen;ZHANG Long;WU Tong(State Grid Anhui Electric Power Co.,Ltd.Construction Company,Hefei 230071,China;College of Information Science and Engineering,Northeastern University,Shenyang 110819,China;Economic Research Institute,State Grid Liaoning Electric Power Co.,Ltd.,Shenyang 110015,China)

机构地区：[1]国网安徽省电力有限公司建设分公司,安徽合肥230071 [2]东北大学信息科学与工程学院,辽宁沈阳110819 [3]国网辽宁省电力有限公司经济技术研究院,辽宁沈阳110015

出　　处：《无线电工程》2023年第1期155-160,共6页Radio Engineering

基　　金：国网辽宁省电力有限公司2019年第一批科技项目(2019YF-25)。

摘　　要：为识别电力基建现场风险区段,降低电力基建现场风险,提出了基于自然语言处理(Natural Language Processing, NLP)和推理引擎的电力基建现场风险区段识别方法。利用NLP技术深入挖掘电力基建现场报告文本,通过分词技术分析报告内相关内容,采用词频-逆文档频率(Term Frequency-Inverse Document Frequency, TF-IDF)算法统计现场报告内容词频,获取文本报告特征。将高权重特征项输入推理引擎内,利用推理引擎确定匹配度最高的事例,通过模拟退火思想优化推理引擎机制,实现电力基建现场风险区段识别。实验结果表明,所提方法风险识别误差均值约为3.5%,且根据所提方法识别结果进行有针对性优化后,应用对象内各区段风险均有不同程度下降。In order to identify the power infrastructure site risk section and reduce the power infrastructure site risk, a power infrastructure site risk section identification method based on Natural Language Processing(NLP) and reasoning engine is proposed. NLP technology is used to deeply mine the on-site report text of power infrastructure, the relevant contents in the report are analyzed through word segmentation technology, and Term Frequency-Inverse Document Frequency(TF-IDF) algorithm is used to count the word frequency of the on-site report content to obtain the characteristics of the text report. The high weight feature items are input into the reasoning engine. The reasoning engine is used to determine the case with the highest matching degree, and the reasoning engine mechanism is optimized through simulated annealing to realize the risk section identification of power infrastructure site. The experimental results show that the average risk identification error of the proposed method is about 3.5%, and after targeted optimization according to the identification results of the proposed method, the risk of each section in the application object decreases to varying degrees.

关 键 词：自然语言处理 推理引擎 基建现场 风险区段识别 特征项 匹配度 

分 类 号：TN929.5[电子电信—通信与信息系统]