Research On Segmented Short-term Wind Speed Dynamic Prediction Based On Spatiotemporal Characteristic Contributio

China has vigorously developed renewable energy sources to cope with energy security and environmental protection issues,and wind power generation technology has been rapidly developed.Due to the random and fluctuating of wind power,wind power grid with large-scale has caused many challenges for the stable operation of the power system.In order to eliminate the damage to the grid caused by the volatility of wind power,the power system urgently needs accurate and fast prediction technology to assist system to adjust the power supply structure in time.In this paper,we investigate the decomposition-ensemble prediction framework in the wind speed combination prediction method from the perspectives of method practicality and test timeliness:(1)A periodic segmented feature number locking method is researched,it correctly identifies and utilizes the periodic characteristics for wind series clustering and feature components extraction,and it dynamically determines the number of feature decomposition contained in different categories of wind series.At the same time,an optimized feature compression method is designed to explore the nonlinear relationship between spatial influence factors and wind speed series,and expand the input feature factor library.Furthermore,a joint feature components contribution evaluation algorithm is designed,which scientifically divides the feature components into different joint components and marks the contribution degrees,so as to facilitate the discriminated learning of the deep learning method.(2)A novel discriminated learning neural network is constructed to create contrapuntal input interfaces according to the contribution of joint feature components to the original sequence.On this basis,the parallel operation of prediction models with different complex structures is realized,so as to enhance the operation efficiency of the model.In addition,study uses optimization algorithms for collaborative optimization of hyperparameters of the deep learning model to further improve the model robustness.Furthermore,the study fully considers the disadvantages of static model which leads to the increase of errors over time,and introduces transfer learning methods to locally adjust the prediction model structure for dynamic prediction and enhance the model prediction timeliness.(3)The study also designs a short-term wind speed prediction system to simulate wind farm operation,and reasonably embeds the prediction model into the system to transfer theoretical research into practical application.This paper uses historical data from several wind farms in the Hexi Corridor region to fully justify the projects.The results of the evaluation indexes prove that the research projects can improve the shortcomings of the decomposition-ensemble prediction framework,and the continuous operation of the system simulation illustrates the practicality of the model.