Research On Multi-step Prediction Technology Of Minute Wind Power

In order to solve the problem of decreasing non-renewable energy reserves,polluting environment and wasting resources,the demand for wind power generation technology is increasing day by day.However,the instability of wind power itself has a certain impact on wind power prediction,and further has an impact on the safe and stable operation of the power grid.Aiming at the low accuracy of existing models,a wind power forecasting model based on Bidirectional Long short-term Memory(BiLSTM)is proposed.BiLSTM can solve the nonlinear relationship among feature vectors in wind power sequence.Then,the improved Intelligence Particle Swarm Optimization(IPSO)algorithm is used to optimize BiLSTM’s hyperparameters.The weight of the model was trained by the Attention Mechanism(AM).Finally,the historical data of a wind farm in Inner Mongolia Autonomous Region were used for 0-15 minutes in advance test.The results show that the proposed IPSO-BILSTM-AM model has high prediction accuracy and can provide scientific reference for wind farm power dispatching and control.Time series prediction of wind power prediction is also an important research field.With the progress of neural network,time series prediction has reached a new level.This paper proposes a time series prediction Model based on Empirical Mode Decomposition(EMD),Long short-term Memory network(LSTM)and LittleBit Model(LB).This model is suitable for predicting time series with large data volume and large fluctuation,and solves the assignment problem of decomposed subseries.First,EMD decomposed the original sequence into a series of sub-sequences and innovatively proposed a distribution function,which divided the extreme and non-extreme sub-sequences according to the fluctuation state of the sub-sequence.The extreme subsequence is put into LSTM for training prediction,and the non-extreme subsequence is put into LittleBit Model for training prediction.Finally,all subsequences are integrated into the final prediction result.Compared with some existing model methods,the prediction accuracy of each data set is improved.