Aiming at the problem of low prediction accuracy caused by the difficulty of capturing time series laws and the highly non-stationary characteristics,this paper proposes a time series prediction model based on quadratic decomposition and GRU-attention.First,the complete ensemble empirical mode decomposition(CEEMDAN)is used to decompose the time series into several modal components with different characteristics,and the complexity of each component is quantified according to the sample entropy.Secondly,variational mode decomposition(VMD)is used to weaken the non- stationarity characteristics of high entropy components.TheGRU prediction model is then optimized using the attention mechanism.Finally,a GRU-attention model is established for each component to predict,and the prediction results of each component are added to obtain the final result.The experimental analysis proves that compared with other models, the model proposed in this paper can better capture the complex laws of the sequence,reduce the non-stationarity of the sequence,and have higher prediction performance,and its average absolute percentage error reached 2.9%,the coefficient of determination reached 0.891.