This paper addresses the needs of big data multimodal artificial intelligence analysis and safety management and control in major livelihood water conservancy projects in Chongqing. By fine-tuning the CLIP model, a model suitable for intelligent analysis of big data in water conservancy construction is established. Low-rank matrices are inserted into the key layers of the text encoder and image encoder, and the cross-modal feature interaction is optimized by combining the EFCA (Enhanced Feature Channel Attention) mechanism. By integrating computer vision (YOLOv8), multimodal understanding (CLIP), and natural language processing (VERT), intelligent analysis and question-answering capabilities for construction site scenarios are achieved. After the model is constructed, it is validated based on a large water conservancy project in western China. The experimental results show that the improved model increases the accuracy by 12.3% compared with the baseline CLIP in zero-shot prediction tasks, the F1-score reaches 89.7% in 5-shot few-shot learning scenarios, and the cross-modal feature alignment is improved by 18.5%. Its zero-shot generalization performance and few-shot transfer efficiency are significantly better than traditional multimodal models, providing an intelligent analysis solution with both precision and generalizability for safety management and control of water conservancy projects.