Research On Inflow Wind Speed Calculation Of Wind Turbines Based On Data Driven

As wind power technology develops rapidly,the proportion of wind energy in China’s energy structure is gradually increasing.The inflow wind speed of wind turbines is a key data for evaluating wind power characteristics and intelligent operation control of wind farms,but it is difficult to obtain and often replaced by nacelle wind speed.Due to the influence of factors such as wind turbine rotation,the nacelle wind speed cannot reflect the real inflow wind speed,so it is necessary to establish an accurate inflow wind speed calculation model.The IEC standard provides a simple and efficient method for calculating the inflow wind speed of wind turbines,but it is difficult to meet practical engineering needs due to constraints such as terrain,turbine wake,and wind speed observation cost.In response to these problems,the authors studied a data-driven method for calculating the inflow wind speed of wind turbines,considering the influence of wind farm terrain conditions,and fully mining the spatiotemporal correlation between adjacent wind turbine nacelle wind speeds.By establishing a mapping relationship between the nacelle wind speed of a certain turbine and the measured inflow wind speed,the accurate inflow wind speed of multiple turbines in the wind farm can be obtained.The main research contents include:(1)Research on the inflow wind speed calculation method of wind turbines considering the influence of terrain conditions:The distribution pattern of errors of the IEC nacelle transfer function method under the influence of terrain conditions was studied,and a calculation model for the inflow wind speed of wind turbines considering the influence of terrain conditions was established.The model used the nacelle wind speed,dynamic terrain steepness,and inclination as input parameters,and the inflow wind speed in front of the rotor as the output parameter.It was based on a time convolutional neural network(TCN)for the calculation of the inflow wind speed of wind turbines.The results of the case study showed that the TCN model considering the temporal sequence reduced the calculation error by 15.1%compared to the traditional nacelle transfer function method.Furthermore,after considering the influence of terrain conditions,the model’s calculation error was reduced by 43.7%,resulting in more accurate inflow wind speeds.(2)Research on the inflow wind speed calculation method of wind turbines based on spatiotemporal graph convolutional networks:In response to the fact that the inflow wind speed calculation method considering the influence of terrain conditions did not account for the impact of obstacle wakes,the established model was only applicable to a limited sector.Taking into account the spatiotemporal correlation characteristics of nacelle wind speeds of wind turbines,a calculation model for the inflow wind speed of wind turbines based on spatiotemporal graph convolutional networks was proposed.The model used neighboring wind speeds of multiple adjacent wind turbines and dynamic terrain coefficients as input parameters,extracted implicit spatiotemporal dependencies based on the relative positions of wind turbines,and obtained the inflow wind speed of the central turbine.Experimental results considering data from all sectors showed that compared to the TCN model considering the influence of terrain conditions,the proposed model reduced the calculation error of inflow wind speed by 56.5%and expanded the sectors for computing the inflow wind speed.(3)Research on the inflow wind speed calculation method of multiple wind turbines based on semi-supervised spatiotemporal graph convolutional networks:To address the problem of the inflow wind speed calculation model’s inability to handle multiple wind turbines in a wind farm,the similarity between nacelle wind speeds of wind turbines was considered,and a calculation model for the inflow wind speed of multiple wind turbines based on semi-supervised spatiotemporal graph convolutional networks was established.The model used the nacelle wind speeds and dynamic terrain coefficients of multiple turbines within the wind farm as input parameters and outputted the inflow wind speed for multiple turbines.The results of the case study showed that compared to directly using the nacelle transfer function extrapolation method,the proposed model reduced the calculation error by 40.9%.This method only required measuring the inflow wind speed of a single wind turbine and achieved simultaneous calculation of the inflow wind speed for multiple turbines,thereby reducing the cost of wind measurement.