| 引用本文: | 臧海祥,郭镜玮,黄蔓云,卫志农,孙国强,赵佳伟.基于改进Wasserstein生成式对抗网络的电力系统不良数据辨识[J].电力自动化设备,2022,42(9): |
| ZANG Haixiang,GUO Jingwei,HUANG Manyun,WEI Zhinong,SUN Guoqiang,ZHAO Jiawei.Bad data identification of power system based on WGAN-GP[J].Electric Power Automation Equipment,2022,42(9): |
|
| 摘要: |
| 随着新能源并网以及大量电力电子器件的投入，电力系统的数据类型向多元复杂化的趋势发展。针对大规模电力系统中出现的不良数据辨识性能差、辨识效率低的问题，提出了一种基于改进Wasserstein生成式对抗网络(WGAN-GP)的不良数据辨识方法。基于历史数据库中的状态量得到多断面正常量测数据并训练WGAN-GP模型；将含不良数据的量测信息输入训练好的WGAN-GP模型，得到对应的量测重构数据，并得到最终的量测重构误差；为了避免人为确定阈值的主观性，提出了一种基于C4.5决策树模型的不良数据阈值确定方法，将量测重构误差输入训练好的决策树模型，即可定位1组量测信息中的不良数据位置。以IEEE标准系统和某实际省网为算例进行仿真测试，结果表明相较于已有方法，所提方法具有更好的辨识性能和更高的辨识效率。 |
| 关键词: 电力系统 不良数据辨识 数据驱动 Wasserstein生成式对抗网络 决策树模型 |
| DOI：10.16081/j.epae.202205052 |
| 分类号:TM71 |
| 基金项目:国家自然科学基金资助项目(U1966205)；中央高校基本科研业务费专项资金资助项目(B220202003) |
|
| Bad data identification of power system based on WGAN-GP |
|
ZANG Haixiang, GUO Jingwei, HUANG Manyun, WEI Zhinong, SUN Guoqiang, ZHAO Jiawei
|
|
College of Energy and Electrical Engineering, Hohai University, Nanjing 211100, China
|
| Abstract: |
| With the integration of new energy sources into the electric power grid and the input of a large number of power electronic devices, the data types of power system are becoming more and more complicated. Aiming at the problem of poor performance and low efficiency of bad data identification in large-scale power system, a bad data identification method based on WGAN-GP(Wasserstein Generative Adversarial Network with Gradient Penalty) is proposed. Based on the state quantities in the historical database, the normal measurement data of multiple sections are obtained and the WGAN-GP model is trained. The measurement information containing bad data is input into the trained WGAN-GP model to obtain the corresponding measurement reconstruction data and the final measurement reconstruction error. In order to avoid the subjectivity of determining the threshold value manually, a determination method of bad data threshold value based on C4.5 decision tree model is proposed. By inputting the measurement reconstruction error into the trained decision tree model, the location of bad data in a group of measurement information can be identified. The simulative results of IEEE standard systems and a real provincial power grid show that compared with the existing methods, the proposed method has better identification performance and higher identification efficiency. |
| Key words: electric power systems bad data identification data-driven Wasserstein generative adversarial net-work decision tree model |
