基于改进ADAM算法的变电站SCD文本分词方法  

A Word Segmentation Method for Substation SCD Text Based on Improved ADAM Algorithm

作　　者：郑翔 陈韶昱 吴俊飞 阮黎翔 骆兆军[3] 徐小俊 ZHENG Xiang;CHEN Shaoyu;WU Junfei;RUAN Lixiang;LUO Zhaojun;XU Xiaojun(Quzhou Power Supply Company of State Grid Zhejiang Electric Power Co.,Ltd.,Quzhou 324100,China;Electric Power Research Institute of State Grid Zhejiang Electric Power Co.,Ltd.,Hangzhou 310014,China;Guodian Nanjing Automation Co.,Ltd.,Nanjing 210032,China)

机构地区：[1]国网浙江省电力有限公司衢州供电公司,浙江衢州324100 [2]国网浙江省电力有限公司电力科学研究院,浙江杭州310014 [3]国电南京自动化股份有限公司,江苏南京210032

出　　处：《微型电脑应用》2025年第1期255-258,共4页Microcomputer Applications

摘　　要：针对电力领域文本数据分词准确性较低的问题,提出一种基于改进ADAM(adaptive moment estimation)算法的中文分词技术。选用Skip-Gram模型作为字嵌入模型,将字词转为分布式向量,搭建卷积神经网络-门控循环单元-条件随机场(CNN-Bi-GRU-CRF)模型实现电力领域文本语句的分割,提出一种改进的ADAM算法,通过控制不同时间窗口的学习率优化神经网络模型,提高模型训练速度。将所提算法运用于变电站SCD(system configuration description)文本数据分词的算例分析,通过与其他主流分词算法进行比较,验证所提分词技术的先进性与准确性。For the problem of low word segmentation accuracy of text data in the power field,a Chinese word segmentation technique based on improved ADAM(adaptive moment estimation)algorithm is proposed.The Skip-Gram model is selected as the word embedding model to convert the words into distributed vectors.A convolutional neural network-bidirectional-gate recurrent unit-conditional random field(CNN-Bi-GRU-CRF)model is built to realize the segmentation of text statements in the power field.An improved ADAM algorithm is proposed to optimize the neural network model by controlling the learning rate of different time windows and improve the model training speed.The proposed algorithm is applied to the example analysis of substation SCD(system configuration description)text data word segmentation,and the advancement and accuracy of the word segmentation technique proposed is verified by comparing it with other mainstream word segmentation algorithms.

关 键 词：中文分词技术 ADAM算法 CNN-Bi-GRU-CRF 变电站SCD文本 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]