Abstract: In order to accurately predict wind power,an ultra-short-term wind power prediction model was proposed based on improved arithmetic optimization algorithm（IAOA）,variational modal decomposition（VMD）and long short-term memory network（LSTM）. The IAOA algorithm was used to optimize the key decomposition parameters k and α of VMD,and the inherent modal functions（IMF）obtained were periodic,which could improve the prediction accuracy of LSTM. Meanwhile,the IAOA algorithm was used to optimize the LSTM network parameters. Through the prediction analysis of wind power data,the results show that the IAOA-VMD-LSTM prediction model has higher prediction accuracy than other models.