| 引用本文: | 马彬喻,杨军,彭晓涛,李蕊,申锦鹏,江克证,柳丹,曹侃.基于改进CVAE-GAN的电力系统暂态稳定评估样本增强方法[J].电力自动化设备,2025,45(9):216-224. |
| MA Binyu,YANG Jun,PENG Xiaotao,LI Rui,SHEN Jinpeng,JIANG Kezheng,LIU Dan,CAO Kan.Sample augment method for power system transient stability assessment based on improved CVAE-GAN[J].Electric Power Automation Equipment,2025,45(9):216-224. |
|
| |
|
|
| 本文已被：浏览 991次 下载 806次 |
码上扫一扫！ |
|
|
| 基于改进CVAE-GAN的电力系统暂态稳定评估样本增强方法 |
|
马彬喻1,2, 杨军1,2, 彭晓涛1,2, 李蕊1,2, 申锦鹏1,2, 江克证3, 柳丹3, 曹侃3
|
|
1.交直流智能配电网湖北省工程中心，湖北 武汉 430072;2.武汉大学 电气与自动化学院，湖北 武汉 430072;3.国网湖北省电力有限公司电力科学研究院，湖北 武汉 430077
|
|
| 摘要: |
| 实际电力系统的暂态失稳样本占比少，不平衡数据降低了数据驱动的暂态稳定评估的失稳样本识别率和可靠性。对此，提出了基于改进条件变分生成对抗网络(CVAE-GAN)的电力系统暂态稳定评估样本增强方法。通过改进输入样本组成比例提高模型对失稳样本分布的学习能力，改进模型网络结构以适应电力系统量测数据特点，采用预训练方式为模型提供良好的初始状态促进训练的收敛。利用训练完成的改进CVAE-GAN模型合成高质量失稳样本，添加到原始样本中实现样本增强。重新训练分类器，实现在线暂态稳定评估。改进的IEEE 39节点系统和改进的南卡罗莱纳州500节点电网测试结果表明，所提方法能够有效学习原始数据分布特性，实现样本增强，从而提升暂态稳定评估精度和失稳样本的识别率。 |
| 关键词: 数据增强 数据不平衡 条件变分生成对抗网络 暂态稳定评估 电力系统 |
| DOI：10.16081/j.epae.202504023 |
| 分类号: |
| 基金项目:国家电网有限公司科技项目资助(合同号：5100-202399365A-2-2-ZB) |
|
| Sample augment method for power system transient stability assessment based on improved CVAE-GAN |
|
MA Binyu1,2, YANG Jun1,2, PENG Xiaotao1,2, LI Rui1,2, SHEN Jinpeng1,2, JIANG Kezheng3, LIU Dan3, CAO Kan3
|
|
1.Hubei Engineering and Technology Research Center for AC/DC Intelligent Distribution Network, Wuhan 430072, China;2.School of Electrical Engineering and Automation, Wuhan University, Wuhan 430072, China;3.State Grid Hubei Electric Power Research Institute, Wuhan 430077, China
|
| Abstract: |
| The proportion of transient instability samples in actual power system is small, and imbalanced data reduces the recognition rate of instability samples and reliability in data-driven transient stability assessment(TSA). Aiming at this problem, a sample augment method for TSA of electric power systems based on improved conditional variational auto-encoder with generative adversarial network(CVAE-GAN) is proposed. The composition ratio of input samples is improved to enhance the model’s ability to learn distribution of instability samples, the model network structure is improved to adapt to the characteristics of electric power systems measurement data, and the pre-training is adopted to provide a good initial state for the model to promote training convergence. The high-quality instability samples are synthesized by using the improved CVAE-GAN model trained, and added to the original samples for sample augment. Then the classifier is retrained to achieve online TSA. The test results of the modified IEEE 39-bus system and the modified 500-bus power grid in South Carolina show that the proposed method can effectively learn the distribution characteristics of the original data, achieve sample augment, and thus improve the accuracy of TSA and the recognition rate of unstable samples. |
| Key words: data augment data imbalance CVAE-GAN transient stability assessment electric power systems |
|
|
|
|
