Research On Intelligent Unmanned Monitoring Method For Wind Farms Using Attention Models

As wind energy is a renewable green energy source,it is gradually becoming more and more widely used around the world.With the increasing number of wind farms,some problems related to wind farms come up,among which the prediction of wind farm environmental parameters and the damage detection of wind turbine blades in wind farms are widely concerned.The establishment of wind farms makes its surrounding environment change,the prediction of wind farm environmental parameters in advance can effectively help us to explore the establishment of wind farms on the local climate and environment in the future period of time,and through the prediction of wind speed in the wind farm environmental parameters can solve the wind speed uncertainty in wind power generation,so that wind power can be better integrated into the grid system;In addition,the wind Turbine blade as a key component of the wind turbine,undetected blade damage may cause devastating damage to the entire wind power system,and will therefore bring huge maintenance costs,in the wind power system to monitor the wind turbine blade is conducive to the stable operation of the system.Therefore,this paper studies how to solve the environmental prediction problem and blade damage detection problem faced by wind farms at this stage from the perspective of attention model and multi-task learning,respectively,and the main work is as follows:(1)Attention-based modeling for environmental time series prediction in wind farms.Considering that the prediction of environmental parameters within wind farms can be influenced by inherent parameters and potential spatio-temporal drivers in the natural environment,and predicting multivariate perceptual data is challenging,this paper develops an RNN encoding-decoding based architecture for implementing sequence-series learning by embedding spatial,parametric and temporal attention layers to extract site correlation of spatial distribution,parameter dependence of multi-source variables and temporal correlation of temporal variation,and finally experimental validation shows that the model achieves good prediction performance in wind field environmental parameter prediction.(2)Multi-task learning-based image detection of wind turbine blade damage.Considering the small amount of data on blade damage in wind farms and the fact that the collected data are easily affected by complex environments,the data set used for training is expanded in this paper.In order to better discriminate the damage type of the target,an attention mechanism is introduced in the detection model to improve the expression of image features.Then,according to the specificity of the detection task,a set of neural network models is designed based on the multi-task learning approach specifically for wind farm wind turbine blade damage detection,and finally,the reliability of the model performance in wind farm wind turbine blade damage detection is demonstrated through a large number of experiments.(3)The platform of wind farm UAV monitoring system is built.In order to be able to monitor the wind farm more conveniently,this paper builds a wind farm UAV monitoring system.By introducing the way of automatic inspection by UAV,the automation degree of wind farm wind turbine blade damage detection is improved,and the UAV is controlled by the developed mobile terminal software,then a front and back-end system is built based on Web development,and the wind farm environmental parameters prediction function and wind farm wind turbine blade damage detection function are integrated in this system,and the communication between each component,UAV control,gimbal control,inspection path setting,blade detection results and wind field environmental prediction results query and other basic functions.