| 引用本文: | 张振宇,张明龙,高 源,等.基于多域特征的扰动辨识方法研究[J].电力系统保护与控制,2021,49(22):137-144.[点击复制] |
| ZHANG Zhenyu,ZHANG Minglong,GAO Yuan,et al.Power disturbance identification research based on multi-domain features[J].Power System Protection and Control,2021,49(22):137-144[点击复制] |
|
| 摘要: |
| 扰动波形的辨识是基于扰动开展特征提取和信息挖掘等研究工作的前提，而噪声、扰动间干扰以及特征提取方法的影响，都有可能导致针对不同扰动提取出的同一域下典型特征间边缘重叠，进而影响扰动辨识的准确性。提出一种利用多域典型特征来识别扰动类型的辨识方法。首先，利用多域特征样本和单域特征样本先后训练神经网络，进而结合DS证据理论融合各域输出以建立面向多域特征的辨识算法。其次，在对三种因素影响下的单域特征开展分析的基础上，对所提出的辨识算法与各种传统的基于单域特征辨识算法的正确率进行对比，以论证所提出辨识算法的有效性。该方法克服了待辨识扰动单域下异常特征对辨识精度的影响，受噪声影响小，算法稳定性好。 |
| 关键词: 扰动辨识 多域特征样本 改进dropout算法 DS证据理论 配电网 |
| DOI：DOI: 10.19783/j.cnki.pspc.210085 |
| 投稿时间：2021-01-21修订日期：2021-05-13 |
| 基金项目:国家电网有限公司总部科技项目资助“基于物联网技术的配电开关一二次深度融合与精益运维关键技术研究及应用”(52130421000S) |
|
| Power disturbance identification research based on multi-domain features |
| ZHANG Zhenyu,ZHANG Minglong,GAO Yuan,LUO Xiang,ZHU Ke |
| (1. Electrical Power Research Institute, State Grid Fujian Electric Power Co., Ltd., Fujian 350007, China;
2. School of Electrical Engineering, Shandong University, Jinan 250061, China) |
| Abstract: |
| The identification of the disturbance waveform is the precondition of the research work of feature extraction and information mining based on the disturbance. However, noise, interference between disturbances and the effect of feature extraction method may lead to edge blurring of features of the same domain extracted from different disturbances, thus affecting the accuracy of disturbance recognition. Therefore, this paper proposes an identification method based on multi-domain typical features to identify disturbance types. First, the neural network is trained successively with multi-domain and single-domain features, and then the output of network in each domain is fused based on DS evidence theory to set up a multi-domain features oriented identification algorithm. Next, on the basis of analyzing the influence of three factors on the single-domain features, the effectiveness of the proposed algorithm is demonstrated by comparing its identification accuracy with that of the traditional single-domain feature recognition algorithms. The method reduces the influence of anomalous features in a single domain on the identification accuracy, and is robust to noise and stable.
This work is supported by the Science and Technology Project of the Headquarters of State Grid Corporation of China “Key Technologies Research and Application of Primary and Secondary Deep Integration with Lean Operation and Maintenance of Distribution Switch Based on Internet of Things” (No. 52130421000S). |
| Key words: disturbance identification multi-domain feature sample improved dropout algorithm DS evidence theory distribution system |
