Ultra-Short-Term Wind Speed And Wind Power Forecasting Based On LSTM

Wind energy is a kind of renewable and clean energy.Wind power generation powered by wind energy is one of the vigorous development targets for optimizing the energy structure in China and accelerating the realization of the "dual carbon" goal.However,the randomness and uncontrollability of wind energy will affect the stability of the output of wind turbines,which will adversely affect the reliable operation of the power system,and the prediction of wind speed and wind power is an effective way to alleviate this adverse impact.In this paper,based on the long short-term memory network(LSTM),the ultra-short-term wind speed and wind power prediction method is studied.The main contents are as follows:Aiming at the problem that the uncertainty of wind speed is strong and it is difficult to track its changes for accurate prediction,an ultra-short-term wind speed prediction model combining the secondary decomposition method and LSTM is proposed.Firstly,the historical wind speed sequence is decomposed once by variational mode decomposition(VMD),and the residual components after primary decomposition are decomposed twice by complete empirical mode decomposition with adaptive noise(CEEMDAN);then,the subsequences are respectively input to LSTM for training and predicting;finally,the prediction results of the subsequences are superimposed to obtain the wind speed prediction results.Through case analysis and method comparison,the feasibility and superiority of the proposed VMD-CEEMDAN-LSTM prediction method are verified.Wind power of wind farms is easily affected by wind speed,temperature and other meteorological factors.Based on this,a multivariate ultra-short-term wind power prediction model combining self-attention time convolution(SATCN)and LSTM is proposed.The SATCN in the proposed model is an improved temporal convolutional network,which is used to extract the temporal features of meteorological factors and the correlation features between variables,and SA enables it to focus on features that contribute more.A certain wind farm data set is used for example analysis.Starting from different seasons,the pre-processed meteorological data is used as the input and the wind power data is used as the output,and the SATCN-LSTM model is constructed for training and predicting respectively.The comparative analysis results of different models on the error index suggest that the proposed wind power prediction method can effectively extract the characteristics of meteorological data,and has higher accuracy and better applicability.In order to further improve the wind power prediction effect,an error correction model based on decomposition and reconstruction is used to correct the power prediction results.First,the error sequence is decomposed by VMD,and the subsequences with low degree of correlation are eliminated according to the calculated Spearman correlation coefficient between the decomposed subsequence and the error sequence,and the remaining subsequences are superimposed and reconstructed;then the reconstructed sequence is input into the LSTM model for training and prediction;finally,the reconstructed sequence prediction results and the power prediction results are superimposed to correct the power results.The results of example analysis and model comparison show that this method can better correct the original prediction results and improve the wind power prediction effect.