Optimal Estimation Of Wind Speed In Front Of Wind Turbines Under All Working Conditions

Wind power is regarded as one of the key technologies to replace fossil energy because of its clean and efficient,low-carbon and large installed potential.The wind speed in front of the wind turbine is the wind speed in front of the wind rotor.Accurate estimation of the wind speed in front of the wind turbine is important to improve the power generation capacity of the wind turbine,optimize the wind turbine control strategy and save the wind farm operation cost.In this paper,the following studies are conducted to accurately estimate the wind speed in front of the turbine:Firstly,in order to improve the estimation accuracy of the wind speed in front of wind turbines,a method is proposed to identify and classify the all wind speed working condition domains of the wind turbine.This method first selects the candidate input variables for the wind speed characteristics based on the wind turbine mechanism analysis,and then verifies the reasonableness of the candidate variables using maximal information coefficient method.Considering the time delay characteristics of the wind speed characteristic input in front of the wind turbine,the akaike’s information criterion is used to determine the dynamic differential order.The feature input,wind speed in front of the wind turbine and its delay order are integrated to form a dynamic difference space tensed by the dynamic difference regression vector.After the high-dimensional parametric clustering and hyperplane estimation based on the feature extraction,the entire working space of the wind turbine is divided into multiple working condition domains.Secondly,for each working condition domains of the wind turbine,the dynamic differential regression vector of the current working condition domain is used as the input to the characteristic model of the wind speed in front of the wind turbine for that domain.and dynamic characteristics of the wind speed in front of the wind turbine are modeled in each domains using three temporal dynamic modeling methods:long short-term memory neural network,bidirectional long short-term memory neural network,and gated recurrent unit.To further measure the model performance,the model performance is evaluated from the time domain and frequency domain perspectives,respectively.The effectiveness of the model in estimating the wind speed in front of the wind turbine at all working condition domains of wind turbines is verified by simulation examples.Finally,an ultra-short-term rolling prediction model of the wind speed in front of the wind turbine is developed using the wind speed in front of the wind turbine estimation results as data samples to achieve accurate prediction of the wind speed in front of the wind turbine for the next 15 seconds.In order to further determine the uncertainty of the predicted wind speed in front of the wind turbine,the nonparametric conditional kernel density estimation method is used to establish the fluctuation interval of the predicted the wind speed in front of the wind turbine at a given confidence level,and the validity of the ultrashort-term rolling interval prediction model for the wind speed in front of the wind turbine is verified by simulation examples.