| 引用本文: | 蒋 昊,许立雄,崔晓丹,等.基于保护电路运行状态增强识别的双馈风电场等值建模[J].电力系统保护与控制,2024,52(22):129-142.[点击复制] |
| JIANG Hao,XU Lixiong,CUI Xiaodan,et al.Equivalent modeling of a DFIG farm based on enhanced recognition of protection circuit operating state[J].Power System Protection and Control,2024,52(22):129-142[点击复制] |
|
| 摘要: |
| 网侧故障期间各风电机组保护电路运行状态是影响双馈风电场并网点暂态响应特性的主要因素。针对双馈风电机组保护电路运行状态难以显式数学表征的问题，提出了基于保护电路运行状态增强识别的双馈风电场等值建模方法。首先，根据样本对模型训练过程的影响程度构建了核心样本指数用于提取核心样本，利用生成对抗网络(generative adversarial network, GAN)对核心样本进行学习生成增强。然后，采用增强样本集训练深度信念网络(deep belief network, DBN)构建双馈风电机组保护电路运行状态识别模型。最后，根据识别结果及风速对双馈风电场内机组进行分群，计算各群等值参数最终建立双馈风电场等值模型。通过对比试验以及指标评价，验证了样本增强对模型训练的提升效果以及运用所提方法构建的等值模型的准确性。 |
| 关键词: 直流卸荷保护 撬棒保护 核心样本 生成对抗网络 深度信念网络 两步分群 |
| DOI：10.19783/j.cnki.pspc.240400 |
| 投稿时间：2024-04-05修订日期：2024-06-24 |
| 基金项目:国家电网有限公司总部管理科技项目资助(5108- 202218280A-2-67-XG) |
|
| Equivalent modeling of a DFIG farm based on enhanced recognition of protection circuit operating state |
| JIANG Hao1,XU Lixiong1,CUI Xiaodan2,3,WU Jialong2,LI Linxiu1 |
| (1. College of Electrical Engineering, Sichuan University, Chengdu 610065, China; 2. Nanrui Group Co., Ltd. (State
Grid Electric Power Research Institute Co., Ltd.), Nanjing 211106, China; 3. National Key Laboratory of
Risk Prevention Technology and Equipment for Power Grid Operation, Nanjing 211106, China) |
| Abstract: |
| During a grid side fault, the operating status of each unit protection circuit in DFIG-based wind farms is the main factor affecting the transient response characteristics of point of connection. It is difficult to express the operational status of the protection circuit of DFIG mathematically. Thus an equivalent modeling method of DFIG-based wind farms based on enhanced recognition of the protection circuit operating state is proposed. First, a core sample index is constructed to extract core samples according to the influence of samples on the model training process, and the core samples are trained and enhanced by a generative adversarial network (GAN). Then, the enhanced sample set is used to train a deep belief network (DBN) to construct the operational state recognition model of the DFIG-based wind turbine protection circuit. Finally, according to the recognition results and wind speed, the units in the DFIG-based wind farms are divided into groups, and the equivalent parameters of each group are calculated to establish the equivalent model of DFIG-based wind farms. Through comparative tests and index evaluation, the effectiveness of sample augmentation in improving model training and the accuracy of the equivalent model constructed using the proposed method are verified. |
| Key words: chopper protection crowbar protection core sample generative adversarial networks deep belief networks two-step clustering |
