Short-term Wind Direction Prediction Based On Deep Learning

Accurate wind direction prediction in wind farms is crucial for effectively improving the conversion efficiency of wind energy and ensuring the safe operation of yaw systems.According to the characteristics of wind direction data,this thesis studies data conversion,intelligent optimization and signal decomposition related to the wind direction prediction.The established wind direction prediction models can be effectively applied to the short-term wind direction prediction,and provide a solid theoretical foundation for the wind direction prediction in wind farms.The main work of this thesis is as follows.For the characteristics of randomness and discontinuity of the wind direction data,sine and cosine functions are firstly used to convert the wind direction data into two time series.Then an echo state network(ESN)model is established to predict each subsequence of wind direction.Finally,an interval prediction model of wind direction is proposed based on ESN and Bootstrap.The experimental results show that the trigonometric function transformation improves the prediction accuracy of machine learning models,and the proposed interval prediction model achieves good performance.To optimize the network structure of deep learning model,the sparrow search algorithm(SSA)is introduced to optimize the gated recurrent unit(GRU).This method significantly enhanced the deep features,and further mines the temporal characteristics of the wind direction components.The key parameters are optimized in GRU through SSA to achieve the wind direction prediction.Experimental results show that the proposed model can effectively improve the accuracy of short-term wind direction prediction.In order to improve the accuracy of short-term wind direction forecasting,a hybrid model,named EEMD-CNN-GRU,is proposed based on ensemble empirical mode decomposition(EEMD),convolutional neural network(CNN)and gated recurrent unit(GRU).In view of the characteristics of randomness and unsteadiness of wind direction series,EEMD is firstly used to decompose the data into multiple components.Then,the local connection and weight sharing of CNN are harnessed to extract the potential features in components.Finally,GRU is adopted to further construct features by the extracted potential features from CNN,and the predicted values of each component are superposed to obtain the ultimate prediction results.The experimental results show that the proposed prediction method achieves good performance compared with other models such as BP neural network and long short-term memory(LSTM).At the end of the thesis,the research work of this paper is summarized and the feasibility analysis of the future research direction is made.