| 引用本文: | 袁 培,王舶仲,毛文奇,等.基于多重生成对抗网络的智能开关设备状态感知与诊断研究[J].电力系统保护与控制,2021,49(6):67-75.[点击复制] |
| YUAN Pei,WANG Bozhong,MAO Wenqi,et al.Research on state perception and diagnosis of intelligent switches based ontriple generative adversarial networks[J].Power System Protection and Control,2021,49(6):67-75[点击复制] |
|
| 摘要: |
| 随着电力物联网数据驱动技术的不断发展，传感器采集的设备量测数据规模爆发式增长，海量异构的多源监测数据给智能开关设备的实时状态感知和诊断带来了新的挑战。针对上述问题，提出一种基于多重生成对抗网络和DS证据理论的开关设备状态感知方法。首先基于DS证据理论构造融合视频、温度、压力、姿态传感器等多源数据的基本信任分配，获取表征开关设备状态的特征信息。根据特征信息和状态类别，建立包含样本生成、数据分类和特征识别的多重生成对抗网络。采用比较、关联、聚类等算法，结合随机梯度下降法更新网络层间参数，最终实现对开关设备运行状态的判别和诊断。以某区域电网的开关设备为例，算例分析结果表明该方法能准确地感知设备的实时状态并对异常信息提出告警。 |
| 关键词: 智能开关设备 状态感知 异常诊断 多传感器 多重生成对抗网络 |
| DOI：DOI: 10.19783/j.cnki.pspc.200733 |
| 投稿时间：2020-06-27修订日期：2020-09-29 |
| 基金项目:国家自然科学基金项目资助(51877072)；国家电网有限公司总部科技项目资助(5216A0180002) |
|
| Research on state perception and diagnosis of intelligent switches based ontriple generative adversarial networks |
| YUAN Pei1,WANG Bozhong2,MAO Wenqi3,JIANG Yizhou3,LI Peng2,WANG Lide2,YI Jin3,DUAN Haoran4 |
| (1. State Grid Hunan Electric Power Corporation Research Institute, Changsha 410007, China; 2. State Grid Hunan Electric
Power Corporation Maintenance Company, Changsha 410004, China; 3. State Grid Hunan Electric Power Company
Limited, Changsha 410004, China; 4. Hunan Key Laboratory of Intelligent Information Analysis and Integrated
Optimization for Energy Internet Hunan University, Changsha 410082, China) |
| Abstract: |
| With the continuous development of data driving technology in the power Internet of Things, the scale of device measurement data collected by sensors has grown enormously. Massive and heterogeneous multi-source monitoring data has brought new challenges to real-time state perception and diagnosis of intelligent switches. To tackle these challenges, a switches state perception method based on a Triple Generative Adversarial Network (TGAN) and DS evidence theory is proposed. First, based on DS evidence theory, basic belief assignment that combines multi-source data such as video, temperature, pressure, and attitude sensors are constructed to obtain characteristic information that shows the state of switches. Based on this characteristic information and state classification, a TGAN that includes sample generation, data classification, and feature recognition is established. Comparison, association, clustering and other algorithms are combined with stochastic gradient descent to update layer parameters. Finally, the state perception and diagnosis of switches are achieved. Switches of a certain regional power grid are taken as an example. Analytical results show that the method can accurately perceive the real-time status of switches and raise an alarm for abnormal information.
This work is supported by the National Natural Science Foundation of China (No. 51877072) and the Science and Technology Project of the Headquarters of State Grid Corporation of China (No. 5216A0180002). |
| Key words: intelligent switches state perception abnormity diagnosis multi-sensor triple generative adversarial network |
