基于深度神经网络的风力机疲劳载荷代理模型研究  

RESEARCH ON SURROGATE MODELS FOR FATIGUE LOADS PREDICTION OF WIND TURBINES BASED ON DEEP NEURAL NETWORK

作　　者：黄国庆 刘伟杰 王彬滨 彭留留 杨庆山[1] 谭舒 Huang Guoqing;Liu Weijie;Wang Binbin;Peng Liuliu;Yang Qingshan;Tan Shu(School of Civil Engineering,Chongqing University,Chongqing 400045,China;Sichuan Electric Power Engineering Co.,Ltd.,Chengdu 610041,China)

机构地区：[1]重庆大学土木工程学院,重庆400045 [2]四川电力设计咨询有限责任公司,成都610041

出　　处：《太阳能学报》2025年第4期398-405,共8页Acta Energiae Solaris Sinica

基　　金：国家自然科学基金面上项目(52378480);重庆市自然科学基金创新发展联合基金(CSTB2024NSCQ-LZX0010);学科创新引智基地项目(B18062);中央高校基本科研业务费项目(2022CDJQY-009)。

摘　　要：提出一种基于深度神经网络的风力机疲劳载荷代理模型,旨在解决风力机场址评估效率低下的问题。开展基于深度神经网络(DNN)的风力机疲劳载荷代理模型研究。首先,根据平均风速、湍流强度、风切变、偏航误差、入流角和空气密度6维环境变量的分布和相关性进行准蒙特卡洛抽样,获得10000个环境变量样本。然后,采用TurbSim和OpenFAST对NREL 5 MW参考风力机进行仿真得到载荷时程,并通过MLife计算得到1 Hz的等效疲劳载荷(DEL)数据库。最后,运用DNN方法建立DEL的代理模型并对模型精度进行详细验证。结果表明:基于DNN的DEL代理模型具有较高的预测精度,计算效率得到显著提升。A deep neural network(DNN)-based surrogate model for wind turbine fatigue load is proposed to address the low-efficiency of site-specific suitability assessment.A surrogate model of fatigue loads of wind turbines based on deep neural network(DNN)is comprehensively investigated.Firstly,10000 samples of six environmental variable space including average wind speed,turbulence intensity,wind shear,yaw misalignment,vertical inflow angle and air density are generated by quasi-Monte Carlo method.Then,TurbSim and OpenFAST are used to simulate the load time history of the NREL 5MW reference wind turbine,and MLife is used to obtain the damage equivalent load(DEL)database at 1 Hz.Finally,the DNN method is used to establish DEL surrogate models for 7 load channels,and the accuracy of the models are extensively verified.Results show that the DNN-based DEL surrogate models have high prediction accuracy and computational efficiency.

关 键 词：风力机 疲劳载荷 OpenFAST 深度神经网络 代理模型 

分 类 号：TM315[电气工程—电机]