A Research On Uncertainty Prediction Of Wind Speed Based On Deep Learning And Quantile Regression

Wind energy as a clean energy is the main growth point of new energy development in China,but the wind speed has large fluctuation and random characteristics,which affects the stability of wind power generation to a certain extent.Accurate wind speed prediction is of great significance to the grid connection,safety and dispatch of wind power generation.The wind speed prediction has a lot of research work from the initial single-point prediction to the interval prediction.This article combines deep learning and quantile regression,and introduces the squared loss and Huber loss into Convolutional Neural Networks(CNNs)and Long Short-Time Memory neural networks(LSTMs).Two network models can effectively perform interval prediction and apply them to wind speed interval prediction.First,the wind speed data is upgraded,and the convolutional neural network is used to extract the wind speed features.Two CNN and quantile combined interval prediction methods are proposed.The first is to replace the original loss function of the quantile with a squared loss.Then,it is combined with the convolutional neural network to form CNN-squared.The second is to fuse the quantile loss and Huber loss to construct a new loss,and then combined with the convolutional neural network to form a CNN-Huber interval prediction model.The methods all solve the problem that there is an underivable point when the quantile is directly combined with the neural network.Secondly,LSTM is a classic network for processing time series.It combines LSTM with quantile squared loss and quantile Huber loss,and proposes LSTM-squared and LSTM-Huber interval prediction models.The combination of two deep learning algorithms and quantiles for wind speed interval prediction and the existing QRNN(Quantile Regression Neural Networks)wind speed interval prediction results show that: the interval prediction model combined with deep learning and quantiles proposed in this paper is based on wind speed data.Good results have been obtained on the set,and the experiment proves that the model of interval prediction proposed in this paper has good prediction ability and can be applied to the task of interval prediction.