| 引用本文: | 陈 旭,张 弛,刘千宽,等.基于深度语义学习的智能录波器自配置方法[J].电力系统保护与控制,2021,49(2):179-192.[点击复制] |
| CHEN Xu,ZHANG Chi,LIU Qiankuan,et al.Automatic configuration method of intelligent recorder based on deep semantic learning[J].Power System Protection and Control,2021,49(2):179-192[点击复制] |
|
| 摘要: |
| 智能录波器的基础配置工作是将全站配置描述文件(Substation Configuration Description, SCD)中智能二次设备(Intelligent Electronic Device, IED)各运行数据输出端口的地址信息分类映射至录波器不同信息组中。目前主流配置方法是针对端口的文本描述进行人工配置。在大规模高电压等级变电站内端口文本描述繁杂，人工操作耗时长，工作量大。针对该现状，提出了基于字符级TextCNN深度语义学习的智能录波器自配置方法。首先利用word2vec模型针对高维稀疏的文本样本矩阵进行降维与稠密化处理，实现字符词向量的分布式表达。之后建立TextCNN模型，基于其多层次抽象化提取样本特征的结构特点进行文本语义挖掘与分类。依据文本分类结果实现端口地址信息的分类映射。案例分析表明，基于TextCNN模型的录波器自配置方法具有分类时间短与分类精度高的优点，提高了录波器自动化配置的准确性。 |
| 关键词: 智能录波器 信息自配置 文本挖掘 词向量分布式表达 文本卷积神经网络 |
| DOI：DOI: 10.19783/j.cnki.pspc.200367 |
| 投稿时间：2020-04-08修订日期：2020-04-22 |
| 基金项目:南方电网公司科技项目资助(000000KK52180019) |
|
| Automatic configuration method of intelligent recorder based on deep semantic learning |
| CHEN Xu,ZHANG Chi,LIU Qiankuan,PENG Ye,ZHOU Daming,ZHEN Jialin |
| (1. Power Dispatching and Control Center of China Southern Power Grid, Guangzhou 510663, China;
2. Wuhan Kemov Electric Co., Ltd., Wuhan 430023, China) |
| Abstract: |
| The groundwork of an intelligent recorder is to map the address information of Intelligent Electronic Device (IED) data output ports within the Substation Configuration Description (SCD) file to different recorder information groups. At present, the mainstream mapping method is manual configuration based on the text description of the output port address. In a large-scale and high-voltage substation, output port address text descriptions are numerous and diverse. The manual operation often takes a huge amount of time. Also the labor cost is high. In order to improve the situation, this paper proposes an automatic information configuration method for an intelligent recorder based on a Text Convolutional Neural Network (TextCNN), which deeply learns the semantics of tests at the character level. First, to realize the distributive expression of text character vectors, the word2vec is introduced to thicken and reduce the dimension of the sparse and high-dimensional sample vector matrix. Secondly, this paper establishes a TextCNN model for text semantic analysis. Based on its ability to obtain multi-level abstract features of samples, the text semantic mining is therefore performed and its classification is realized. The mapping of the IED output port address information is completed based on the text classification result. The case study shows that the automatic intelligent recorder information configuration based on the TextCNN model has the characteristics of short classification time and high accuracy. This improves the accuracy of the automatic configuration for the intelligent recorder.
This work is supported by the Science and Technology Project of China Southern Power Grid Co., Ltd. (No. 000000KK 52180019). |
| Key words: intelligent recorder automatic information configuration text mining distributive expression of word vectors text convolutional neural network |
