Ultra-short-term Wind Power Prediction Based On Two-layer Decomposition And LSTM

Wind power generation is an efficient and clean way of power generation.No matter at home or abroad,the installed capacity of wind power is expanding year by year.Now,all countries are actively carrying out research on the optimization of wind power integration.Improving the accuracy of power prediction will help coordinate power generation,accelerate the process of wind power integration,and benefit the development of power grid.This paper forecasts wind power based on statistical method.The main contents are as followsFirstly,the significance of research on wind power prediction technology is introduced,and the research status and main technology of wind power prediction at home and abroad are introduced.The data of Spain wind farm Sotavento is selected for research and experiment.Firstly,the original data were preprocessed,and to create a single Back Propagation(BP)neural network prediction model,Random Forests(RF)prediction model,Long Short Term Memory(LSTM)neural network forecast model.The wind power sequence is predicted with 10 minutes as the step length,and the super short term wind power is predicted with 1,3 and 6 steps.Results show that,the effect of LSTM neural network forecasting model is superior to other two models.By making predictions based on a single model,the potential physical meaning of power series is explored.The components are reconstructed by fast ensemble empirical mode decomposition(feemd)and sample entropy,and the length of input variable is determined by partial autocorrelation coefficient(PACF).Each component is predicted and the predicted results are finally superimposed.The first mock exam shows that the prediction performance of the model has a higher prediction performance than the single model.The decomposition and integration model has a higher prediction performance than the single model.In order to reduce the spatial complexity of wind power time series,a wind power prediction model based on Two-layer decomposition technology and particle swarm optimization LSTM neural network is constructed.The original wind power data is decomposed by feemd,and the high-frequency complex components with high frequency and large fluctuation are decomposed by VMD to further reduce the spatial complexity and temporal chaos of th e sequence.The length of each component input variable is also determined by partial autocorrelation coefficient.By particle swarm optimization LSTM deep learning respectively forecast each component and residual component,and use adaptive learning strategies of particle swarm optimization(PSO)algorithm choose LSTM neural network parameters,which saves the steps of setting parameters manually.Finally,the prediction results of each component are superimposed to improve the prediction accuracy.Experi ments show that the method is feasible and effective.