考虑数据集误差下基于数据驱动的新能源设备多工况导纳获取方法  

Multi-operating-point Admittance Acquisition of Renewable Equipment Based on Data-driven Considering Measurement Error on Datasets

作　　者：李晗[1] 李萌 王垚鑫 马骏超 倪秋龙 李海盼 年珩[1] LI Han;LI Meng;WANG Yaoxin;MA Junchao;NI Qiulong;LI Haipan;NIAN Heng(College of Electrical Engineering,Zhejiang University,Hangzhou 310027,Zhejiang Province,China;Economic Research Institute Co.,Ltd.,State Grid Zhejiang Electric Company,Hangzhou 310027,Zhejiang Province,China;Electric Power Dispatching Center,State Grid Zhejiang Electric Company,Hangzhou 310027,Zhejiang Province,China)

机构地区：[1]浙江大学电气工程学院,浙江省杭州市310027 [2]国网浙江省电力有限公司电力科学研究院,浙江省杭州市310027 [3]国网浙江省电力有限公司电力调度控制中心,浙江省杭州市310027

出　　处：《中国电机工程学报》2025年第3期948-959,I0012,共13页PROCEEDINGS OF THE CHINESE SOCIETY FOR ELECTRICAL ENGINEERING

基　　金：国家电网有限公司科技项目(5211DS220008)。

摘　　要：基于数据驱动的神经网络建模方法已经广泛用于分析电力电子设备的多工况阻抗/导纳模型。然而,实际测量获取的导纳数据样本较少,并且由于测量噪声的影响导致阻抗数据质量较差,这将劣化模型的预测性能,导致模型预测值与真实导纳之间存在较大误差。针对该问题,文中提出考虑测量误差影响下基于数据驱动的多工况导纳模型获取方法。首先,以模型预测值与真实值之间的均方误差作为评价指标来指导神经网络训练;然后,分析多工况下电压电流噪声对导纳测量的影响,并建立测量误差与模型指标的关系;进一步地,通过贝叶斯算法搜索使上述指标最小的最优模型参数,进而降低噪声样本对神经网络模型训练的干扰,提高模型输出的准确度。最后,搭建基于双馈感应发电机的BP神经网络导纳模型,并在含测量误差数据集中验证所提方法的有效性。The data-driven neural network modeling approach has been widely used for analyzing the multioperating condition impedance/admittance models of power electronic devices.However,there are limited samples of admittance data obtained from actual measurements,and the quality of impedance data is compromised due to the influence of measurement noise.This degradation in data quality can negatively impact the predictive performance of the model,resulting in significant errors between the model’s predictions and the actual admittance.To address this issue,this paper proposes a data-driven approach for obtaining multi-operating condition admittance models that consider the effects of measurement errors.Initially,the guidance for neural network training is based on the mean square error(MSE)as the evaluation metric between model predictions and actual values.Then,the paper analyzes the impact of voltage and current noise under various operating conditions on admittance measurements and establishes the relationship between measurement errors and above metrics.Furthermore,Bayesian algorithms are employed to search for optimal model parameters that minimize the metrics,thereby reducing the interference of noise samples in neural network model training and improving the accuracy of model outputs.Finally,this paper constructs a back propagation neural network(BPNN)admittance model based on the doubly fed induction generator(DFIG)and validates the proposed method with a dataset containing measurement errors.

关 键 词：双馈感应发电机 数据驱动 多工况导纳 噪声分析 贝叶斯优化 

分 类 号：TM464[电气工程—电器]