Research On Performance Optimization And Power Lifting Method Of Wind Turbine

Wind energy,as a renewable and clean energy source,has a rising share in primary energy.How to utilize the wind energy more effectively in the wind power generation and further optimize the performance of the unit is the core issue of wind power research.The bottleneck encountered in the development of wind power nowadays is that its performance optimization depends on multi-dimensional control parameters,and a great deal of uncertainty and nonlinearity exist in the control process due to natural factors.Making the traditional algorithm difficult to control performance and accuracy continue to improve.Machine learning method is a typical representative of intelligent methods,the core idea is to use implicit algorithm iterative learning to complete the task.An important process of machine learning is model training and verification,which is accompanied by a huge demand for data and computing power.As hardware computing power is improved and algorithms are optimized for operational performance,deep neural networks and integrated learning methods take on an important role as artificial intelligence again.Based on big data of wind farm operation,this paper studies the power forecasting,performance optimization and power improvement of wind turbine with artificial intelligence.The main research contents include:For the characteristic engineering of wind farm historical data,the data are preprocessed,feature filtered and dimensionality-reduced analyzed based on statistical algorithms.Wind turbine speed correction and end-to-end power forecasting are accomplished using time-series data and deep learning methods.In the condition that the unit is not equipped with expensive wind measurement device,more accurate wind field data can also be obtained through model calibration,so that the wind turbine can accurately track the maximum power curve.The verification results show that the model can effectively correct the wind speed error and has good generalization performance.It can be carried on different wind turbines by transfer learning.In the power prediction model,due to the particularity of the wind farm environment,the promotion of training speed will bring greater flexibility to practical application.In this paper,we trained the predictive modeling with the new structure named SRU.Compared with the speed of LSTM There is a big improvement.The test results show that the prediction tasks can be realized more accurately in the ultra-short-term power forecasting,and the accuracy of short-term power forecasting can meet the requirements of power grid resource allocation.In the task of power upgrade,a wind direction correction model based on multi-Grained Cascade forest and fuzzy control rules is proposed,which makes the unitfully utilize wind energy at each power point.This new model of ensemble learning combined with deep learning can adaptively control the scale of the model and has strong generalization performance while ensuring accuracy.The verification results show that the cabin can be yaw corrected,and its performance has certain advantages over both neural network and support vector machine.