Abstract: To address the challenge of accurate positioning and identification of blockages in long-distance coal gangue slurry transportation pipelines, a 513m-long industrial test system for Distributed Acoustic Sensing (DAS)-based monitoring was established. It simulates four operating conditions: normal slurry flow, and 20%, 40%, and 60% pipeline blockages. A dual-branch 1DCNN-LSTM pipeline blockage identification model with a cross-attention mechanism is proposed. It takes the time-domain energy features and frequency-domain power spectral density features of vibration signals from all monitoring points as dual-branch inputs. Through the One-Dimensional Convolutional Neural Network (1DCNN) and Long Short-Term Memory Network (LSTM), the model extracts deep time-frequency features and nonlinear inter-sequence relationships respectively. The cross-attention mechanism effectively fuses these time and frequency domain features, and the blockage status at the current pipeline monitoring position is output via the fully connected layer. Experimental results show that the model achieves an accuracy of 91.27%, recall of 93.10%, and F1-score of 92.28% for normal transportation status. For different blockage states, the average accuracy, average recall, and average F1-score reach 92.72%, 92.06%, and 92.39% respectively. Compared with three existing models (MCNN-HMM, CNN-Bi-LSTM-CTC, DSN-KSMDD-GRU), the three indicators are improved by 1.46-3.00, 1.47-3.08, and 1.63-3.17 percentage points respectively, with an identification speed of 0.21s. The model demonstrates effectiveness and robustness, providing technical support for the safety monitoring and blockage identification of slurry transportation pipelines.