考虑配电网三相电压特征的IHPO-CSSVM电压暂降源识别

Voltage sag source identification using IHPO-CSSVM with consideration of three-phase voltage characteristics on the distribution network

中文关键词: 完全集合经验模态分解与自适应噪声  改进的猎人猎物优化算法  代价敏感支持向量机  配网侧电压暂降

英文关键词: complete ensemble empirical mode decomposition with adaptive noise  improved hunter-prey optimizer  algorithm-cost-sensitive support vector machine  distribution network side voltage sag sources

中文摘要:

      随着分布式新能源和电力电子设备广泛接入配电网,能源供应和负荷需求等方面呈现出新的特点。考虑到支持向量机（support vector machine,SVM）算法的超参数选择困难以及电压暂降源信号数据类别不平衡等问题,提出了一种基于完全集合经验模态分解与自适应噪声（complete ensemble empirical mode decomposition with adaptive noise,CEEMDAN）和改进的猎人猎物优化代价敏感 SVM（improved hunter-prey optimizer cost-sensitive SVM,IHPO-CSSVM）的电压暂降源识别方法。通过在 Matlab/Simulink仿真平台模拟电路,获得不同类型的电压暂降源,利用CEEMDAN从需求侧电压暂降信号中提取三相电压的特征向量,并计算其近似熵,构建新的特征向量,输入到IHPO-CSSVM分类器进行训练。与SVM、CSSVM、极限学习机进行比较,仿真结果表明IHPO-CSSVM的识别准确率最高,该方法能够准确地从复杂的电压信号中提取出有用的特征,并通过优化模型参数来提升识别准确率,可以有效解决配网侧的电压暂降源识别问题。

英文摘要:

      With the extensive integration of distributed power sources and power electronic devices into distribution networks, new characteristics are manifesting in aspects of energy supply and load demand. A voltage sag source identification method combining complete ensemble empirical mode decomposition with adaptive noise（CEEMDAN）and improved hunter-prey optimizer cost sensitive support vector machine（IHPO-CSSVM）is proposed to address the difficulties in selecting hyperparameters for support vector machine（SVM）and the imbalance of voltage sag source signal data categories. By simulating circuits on the Matlab/Simulink simulation platform, different types of voltage sag sources are obtained. The CEEMDAN is used to extract the feature vectors of the three-phase voltage of the voltage sag source signal, and its approximate entropy is calculated. A new feature vector is constructed and input into the IHPO-CSSVM classifier for training. Compared with SVM, CSSVMand extreme learning machine, simulation results show that IHPO-CSSVM has the highest recognition accuracy. This method can accurately extract useful features from complex voltage signals and improve recognition accuracy by optimizing model parameters, providing an effective solution for voltage sag problems in power systems.