Research And Software Implementation Of Shortterm Wind Speed Forecasting Method Based On Deep Transfer Learning

With the depletion of fossil energy and global climate change,clean energy is receiving more and more attention.Wind energy is a clean and green renewable energy source with wide distribution,large storage capacity and low cost,which makes it an energy source with great potential.However,wind is highly random and uncontrollable,which has a great impact on the stability and safety of wind power generation and the scheduling of power grids.Accurate wind speed forecasting is needed to ensure the stable operation of wind turbines,to improve safety,to reduce the operating costs of wind farms,and to facilitate the scheduling and adjustment of power grids.In this paper,we investigate the short-term wind speed prediction method based on Long Short-Term Memory(LSTM)network and proposes a short-term wind speed prediction method based on wind direction decomposition.In addition,this paper proposes a short-term wind speed forecasting method based on deep transfer learning to address the problem that accurate wind speed forecasting models cannot be obtained by training a small amount of data at locations where there is a lack of historical wind speed data.The main research elements of this paper are as follows.(1)On the basis of wind speed forecasting model based on wind speed preprocessing,wind direction features are introduced,which is different from the previous direct input of wind direction as features into the model,and the short-term wind speed forecasting method based on wind direction decomposition is proposed in this paper.The proposed method is compared with the wind speed forecasting method based on wind speed preprocessing and the wind speed forecasting method with direct input of wind direction features.The experimental results show that the proposed short-term wind speed forecasting method based on wind direction decomposition has the best performance,with the root mean square error,mean absolute error and mean absolute percentage error of single-step forecasting being 0.1033,0.0814 and 0.0208,respectively.(2)For wind speed forecasting at locations with insufficient historical wind speed data,a short-term wind speed forecasting method based on deep transfer learning is proposed in this paper.The effect of wind speed preprocessing and wind direction decomposition,the number of freezing layers in fine-tuning,and the location of the target site on the transfer effect is investigated through wind speed forecasting comparison experiments,including the comparison experiments with the target site model obtained by training with a small amount of historical wind speed data at the target site.The comparison experiments with the model trained with a small amount of historical wind speed data at the target site are included.The experimental results show that the prediction accuracy of this method is higher than that of the target site model,and the root mean square error,mean absolute error and mean absolute percentage error of its single-step forecasting are 0.1595,0.0713 and 0.0143,respectively,which are lower than that of the target site model.This shows that the method in this paper is an effective transfer and can improve the accuracy of wind speed forecasting for locations lacking historical wind speed data.(3)Wind speed forecasting software was designed and implemented.After requirement analysis and architecture design,the Express.js framework was used to implement the server side of the software,and the Vue.js and Electron.js frameworks were used to implement the client side of the software.The software provides users with a visual interface for training and fine-tuning of forecasting models and wind speed forecasting experiments,as well as model and experiment record management functions,providing a simple,practical and convenient platform for relevant personnel.