Abstract: To address the problems that bottleneck units in unmanned aerial vehicle (UAV) discrete manufacturing workshops exhibit significant dynamic migration and that traditional static identification methods are difficult to support production prediction and scheduling optimization, a bottleneck unit prediction method based on Convolutional Neural Network and Bidirectional Long Short-Term Memory with Dynamic Attention Mechanism (CNN-BiLSTM-DAM) is proposed. First, considering the characteristics of UAV workshops, such as multi-variety and small-batch production, complex process routes, and frequent disturbances, a multidimensional bottleneck feature system composed of production load, equipment utilization, buffer queue length, and average active time is constructed. Second, CNN is used to extract local spatial correlation features among production units, BiLSTM is adopted to capture bidirectional temporal dependencies in bottleneck evolution, and a dynamic attention mechanism is introduced to enhance the representation of key time-step features. Finally, UAV workshop operation data are generated through Python-based discrete event simulation, and model training, testing, and comparative experiments are carried out. The results show that the RMSE (root mean square error), MAE (mean absolute error), and R² of the CNN-BiLSTM-DAM model are 0.0202, 0.0139, and 0.8976, respectively, outperforming CNN-BiLSTM, BiLSTM-DAM, and BiLSTM models. The bottleneck unit prediction accuracy on the test set reaches 76.74%. The results indicate that the proposed method can effectively learn the dynamic migration pattern of bottleneck units in UAV workshops and provide support for production scheduling optimization, resource allocation, and intelligent decision-making in manufacturing systems.