Research On Wind Speed Prediction Based On Deep Learning

As a green,environmentally friendly and renewable energy source,wind energy has become a key area of energy development in various countries.Among them,wind power generation is the main way to utilize wind energy.However,since wind speed is both regular and random,the uncertainty in the actual process of grid connection of wind power generation will affect the operation,scheduling and cost of the power system.Therefore,accurate wind speed prediction is of great importance to the development and utilization of wind energy,as well as the safe and stable operation of power systems.In order to achieve high-precision wind speed prediction,this thesis uses the long-shortterm memory artificial neural network(LSTM)as the basis,introduces the convolutional neural network(CNN),integrates the attention mechanism,and designs a hybrid integrated prediction model combined with data preprocessing.The main work of this thesis is as follows.(1)For the wind speed data,this thesis combines improved singular spectrum analysis and Empirical Wavelet Transform(EWT)to preprocess the wind speed data,firstly,the original wind speed data set is noise-reduced using the improved SSA,and then decomposed into multiple time series components using EWT.It is also decomposed into multiple timeseries components by EWT.Then the Pearson coefficients between each component are calculated,and the wind speed subseries is obtained by merging the related components based on this.(2)Using convolutional neural network for feature extraction,the constructed feature extractor can reduce the complexity of the network model while also reducing the learning of network parameters,making the deep learning model easier to train.In this thesis,based on the characteristics of wind speed data,the 1D-CNN structure is selected so that the feature learning capability of CNN and the time series memory function of LSTM can be combined,i.e.,the feature extraction of wind speed.(3)In response to the traditional model LSTM does not have the ability to focus on important information,this thesis introduces an attention mechanism based on the long shortterm memory neural network(LSTM)so that the model can focus on the information and important parts of the signal,which is called the wind speed prediction model of convolutional long short-term memory network(CLSTM-ATT)based on the attention mechanism.The advantage of the combined prediction model proposed in this thesis is that it not only improves the prediction accuracy but also preserves relatively good stability.Finally,the prediction effect of this prediction model is verified and its superiority is demonstrated by comparing it with four groups of models.(4)The wind speed data of the actual wind farms are used to verify the proposed model in terms of model prediction accuracy,model complexity,model stability,and DieboldMariano C DM test.The best experimental results show that the model proposed in this thesis shows its superiority in all evaluation indexes.