Short Term Wind Speed Prediction Based On Stacking Hybrid Model

Accurate prediction of wind speed is very important in renewable energy manufacturing.In recent years,the issue of environmental pollution has been discussed by everyone.Industrial development and the popularization of automobiles have led to the excessive use of coal,petroleum,and other fuels for daily life and production activities,resulting in excessive carbon emissions and a series of air pollution issues.Experts from all walks of life have also made efforts to develop environmentally friendly and clean energy.Wind energy is pollution-free and belongs to natural renewable resources.Popularizing the use of this energy can greatly alleviate environmental pollution,reduce carbon emissions,mitigate the greenhouse effect,and many other benefits.Therefore,humans are encouraged to use wind energy as a green,clean,and free energy.For energy managers and power operators,reducing the uncertainty of the chaotic nature of wind is also an important task for more accurate prediction of wind speed and power.Accurate wind power prediction can be used to understand the potential of wind energy.It can be applied to wind farm design,wind farm,power grid management,and other aspects to better accurately capture wind energy,and can also achieve widespread utilization of wind energy.Therefore,accurate prediction of wind speed or power generation has become a key task with profound impact on human beings and huge benefits.The purpose of this study is to provide some technical support for the subsequent development of wind energy.Due to the highly nonlinear,intermittent,and chaotic characteristics of wind speed series,many difficulties have been brought to practical prediction.This paper proposes a short-term wind speed hybrid prediction method based on variational mode decomposition(VMD)and Stacking ensemble learning from the perspectives of wind speed signal decomposition and multiple model fusion extraction to maximize efficiency,aiming at improving the accuracy and robustness of short-term wind speed prediction performance.Firstly,wind speed data have sequence combinations of different frequencies,and are nonlinear and volatile.In order to better understand the overall pattern and movement trend of wind speed data,it is necessary to preprocess the original wind speed data.These sequences with different frequencies can be separated from chaotic wind speed data,and abnormal data can be processed to improve the accuracy of wind speed prediction.VMD is a new adaptive data decomposition method that uses a completely non recursive method to find the optimal solution of a variational model,which can better avoid modal aliasing and endpoint effects,and is used to decompose wind speed series into stationary components to solve non-stationary problems;Secondly,considering the differences in the training principles of different algorithms and the actual model effects,Stacking’s first layer base learner uses Light GBM,LSTM,and Fully Connected Network(FCN),while the second layer learner uses FCN.This combination method not only effectively utilizes the trend of time series changes,but also fully exploits sequence information and non-linear information under different fluctuation levels,achieving comprehensive extraction of wind speed series information from these two perspectives.During the experiment,the wind speed data measured by wind power equipment from NOAA was used for research,and it was decomposed into strong fluctuation wind speed segment and weak fluctuation wind speed segment for experiment.The experiment verified that the VMD Stacking hybrid model performed well,both better than the single model,and this model also had authenticity and reliability in predicting the actual sudden wind speed.In specific wind speed prediction experiments,the RMSE,MAE,and MAPE of the model are0.1772,0.1553,and 8.32%,respectively.Compared to other different decomposition methods or different combinations of base learners,the VMD-Stacking(Light GBM,LSTM,FCN)model fully utilizes the characteristics of time series and wind speed fluctuation information,and its prediction performance is good,further improving the accuracy and stability of shortterm wind speed prediction.