基于神经网络与非线性自回归算法的海上风能资源测量研究  

Research on Offshore Wind Energy Resource Measurement Based on Neural Networks and Nonlinear Autoregressive Algorithm

作　　者：曹善桥 CAO Shanqiao(Datang Renewable Energy Test and Research Institute Co.,Ltd.,Beijing 100053,China)

机构地区：[1]大唐可再生能源试验研究院有限公司,北京100053

出　　处：《自动化应用》2025年第3期146-149,共4页Automation Application

摘　　要：风力资源评估对风电场的前期建设至关重要。利用激光雷达或气象观测塔测量能保证高精度的数据,但价格昂贵且时空并不连续;利用天气研究与预报(WRF)模型等的数值模拟方法生成的时空连续数据精度相对较低。针对上述问题,提出一种混合方法,将测量与模拟相结合,用于海上风能评估。首先,使用多保真高斯过程回归(MFGPR)进行时间数据融合,将陆上位置的风力间歇性测量数据和连续模拟数据相结合。然后,利用非线性自回归(NAR)和带有外部输入的非线性自回归(NARX)的人工神经网络进行空间数据融合,通过陆上位置的风力数据对海上的风力数据进行预测。研究表明,该方法可在2%的误差范围内准确评估海上风力资源。Wind resource assessment is crucial for the early construction of wind farms.The use of laser radar or meteorological observation towers can ensure high-precision data measurement,but it is expensive and not spatially and temporally continuous.The accuracy of spatiotemporal continuous data generated by numerical simulation methods such as Weather Research and Forecasting(WRF)models is relatively low.A hybrid method is proposed to address the above issues,combining measurement and simulation for offshore wind energy assessment.Firstly,temporal data fusion is performed using Multi-Fidelity Gaussian Process Regression(MF-GPR),which combines intermittent wind measurements at onshore locations with continuous simulated data.Then,spatial data fusion is performed using Non-Linear Autoregression(NAR)and Non-Linear Autoregression with External Input(NARX)artificial neural networks to predict wind data at sea from wind data at onshore locations.It is shown that the proposed data fusion method can accurately assess the offshore wind resource within a 2%error margin.

关 键 词：人工神经网络 高斯过程回归 时空数据融合 海上风能资源评估 

分 类 号：P457.6[天文地球—大气科学及气象学]