摘要
风机叶片覆冰是破坏风机运行工况和电网稳定性的因素之一。传统的覆冰监测方法成本高,且对叶片原有机械结构存在潜在的损害。文中建立了一种基于气象数据和深度学习的覆冰监测模型。通过分析Makkonen模型,从热力学机理出发,针对传统监测模型在液态水含量等直接影响覆冰速率的核心参数表征方面存在的局限性,充分考量气象数据中与覆冰高度密切相关的特征量,同时引入时间序列分析方法以捕捉变量在时间维度上的变化规律。为解决跨风电场数据的分布偏移问题,设计深度自适应标准化模块对输入特征进行域不变性转换,并构建Transformer-时序卷积网络(TCN)双通道架构以同步捕获气象参数的全局时序依赖与局部突变特征。最后以某山区的实际风机数据进行实例仿真,结果表明该模型在实现风机叶片上的覆冰情况诊断方面表现出色,为风机叶片覆冰监测拓展了可用的技术手段。
Icing on wind turbine blades is one of the factors that impair the operational conditions of wind turbines and the stability of the power grid.Traditional icing monitoring methods are costly and may potentially damage the original mechanical structure of the blades.This paper establishes an icing monitoring model based on meteorological data and deep learning.By analyzing the Makkonen model from a thermodynamic mechanism perspective,and addressing the limitations of traditional monitoring models in characterizing core parameters such as liquid water content that directly affect icing rate,the model fully considers feature quantities in meteorological data that are closely related to the icing intensity.Meanwhile,time series analysis methods are introduced to capture temporal variation patterns of variables.To tackle the distribution shift problem across wind farm data,a deep adaptive normalization module is designed to perform a domain-invariant transformation on input features.A dual-channel transformertemporal convolutional network(TCN)architecture is constructed to simultaneously capture global temporal dependencies and local abrupt features of meteorological parameters.Finally,simulations are conducted using actual wind turbine data from a mountainous region.The results show that the model excellently performs in diagnosing icing conditions on wind turbine blades,thereby expanding available technical means for wind turbine blade icing monitoring.
作者
李彬
袁军
苏盛
蒙文川
杨再敏
Bin;YUAN Jun;SU Sheng;MENG Wenchuan;YANG Zaimin(State Key Laboratory of Disaster Prevention&Reduction for Power Grid(Changsha University of Science&Technology),Changsha 410014,China;Energy Development Research Institute,China Southern Power Grid Co.,Ltd.,Guangzhou 510663,China)
出处
《电力系统自动化》
北大核心
2026年第3期180-188,共9页
Automation of Electric Power Systems
基金
国家自然科学基金资助项目(51777081)
广西电网公司科技项目(046000KK52220007)。
关键词
风机
叶片
覆冰
气象数据
时间序列分析
时序卷积网络(TCN)
特征提取
深度学习
wind turbine
blade
icing
meteorological data
time series analysis
temporal convolutional network(TCN)
feature extraction
deep learning
