In order to consider the impact of dynamic features of the fermentation process on stage division and improve the prediction accuracy, a quality prediction method based on attention long short-term memory(LSTM) is proposed. Firstly, the original 3D data are unfolded along the batch direction. Partial least square(PLS) analysis is performed on each time slice matrix to obtain the score matrix of process variables and quality variables. The joint score matrices are clustered using the affinity propagation(AP) algorithm. Then the encoder-decoder model is used to extract the dynamic characteristics of the process dynamics, and the AP algorithm is used for the second division. Finally, the production process is divided into different stable phases and transition phases through the comprehensive analysis of the two-step division results. The LSTM integrated quality prediction model is established in each stage after the division. Penicillin fermentation simulation data and E. coli production data are tested, and the results demonstrate the feasibility and effectiveness of the proposed method.