In the primary support stage of tunnel construction. The wet spray trolley pumping system is prone to pipeline blockage, which can lead to unplanned downtimes, construction delays and other issues. Therefore, it is urgent to enhance the capabilities of equipment failure prediction and operation maintenance. To address challenges including strong noise interference and insufficient complex feature extraction from long-term sequential data in fault prediction method of pipeline blockage, a pumping system pipeline blockage fault prediction model based on the CNN-LSTM-Attention model was proposed. Pumping pressure data were took as the prediction target and processed using interquartile range for outlier removal and Kalman filtering for data smoothing, ensuring the stability of input data under high-noise conditions. The model employed CNN (convolutional neural network) to extract local spatiotemporal features from pumping pressure data, integrated LSTM (long short-term memory) network to capture the long-term dynamic characteristics of pumping processes, and incorporated the Attention mechanism to adaptively weight the critical nodes of the fluctuating pressure, achieving high-precision pressure trend prediction. The experimental results demonstrated that the prediction performance of the proposed model was significantly superior to that of traditional models such as CNN, LSTM, and CNN-LSTM. Based on the model prediction results, a health status evaluation system for wet spray trolley pumping system was established, and its health status prediction platform was developed, effectively supporting the decision-making at the construction site.