| 杨阳蕊,潘世峰,刘雪梅,马威,冯岭.多模态知识图谱与大模型协同的水利工程风险应对决策推荐[J].水利学报,2025,56(4):519-530 |
| 多模态知识图谱与大模型协同的水利工程风险应对决策推荐 |
| Multimodal knowledge graph collaborated with large model for decision recommendation of water projects risk response |
| 投稿时间：2024-09-08 |
| DOI：10.13243/j.cnki.slxb.20240575 |
| 中文关键词: 多模态知识图谱 多模态大模型 风险应对 南水北调工程 |
| 英文关键词: multimodal knowledge graph multimodal large model risk response South-to-North Water Diversion Project |
| 基金项目:国家自然科学基金项目(72271091);河南省高等学校重点科研项目(25A520006);河南省科学院科技开放合作项目(220901008);华北水利水电大学硕士研究生创新能力提升工程项目(NCWUYC-202416098) |
|
| 摘要点击次数: 196 |
| 全文下载次数: 109 |
| 中文摘要: |
| 随着水利工程发展和信息化需求的增加，有效处理并利用水利工程巡检数据做出风险应对决策变得更加重要。南水北调工程风险巡检文本涉及诸多影响因素、复杂的相互关系以及多源异构数据。为了充分利用该文本以保证对工程风险处置的及时有效性，本研究基于巡检文本数据提出一种结合多模态知识图谱与多模态大模型(MLLM)的方法，用于对南水北调工程运行中的风险做出针对性应对决策推荐。首先，选用更加关注区域位置的开源多模态大模型VIP-LLaVA作为对话模型;其次，基于水调工程巡检数据，构建领域多模态知识图谱;最后，以水利领域多模态知识图谱作为多模态大模型增强知识，采用检索与生成协同迭代的方式寻找应对风险的最佳方案。结果表明，由该方法生成的风险应对决策在BLEU、ROUGE-L和METEOR评价指标上效果良好，能够及时对工程中的运行风险做出合适的应对决策推荐。相关研究可以有效提高工程安全维护管理的可靠性，降低工程风险对供水安全的影响。 |
| 英文摘要: |
| With the development of water projects and the increasing demand for informatization，it becomes increasingly crucial to effectively process and use inspection data from water project to make risk response decisions.The risk inspection texts of the South-to-North Water Diversion Project involve numerous influencing factors，complex interrelationships，and multi-source heterogeneous data.To effectively utilize the inspection text and ensure the timeliness and effectiveness of the engineering risk response，a method that integrates multimodal knowledge graph with multimodal large model (MLLM) based on inspection text data is proposed，aiming to provide targeted decision-making recommendations for risks in the operation of the South-to-North Water Diversion Project.Firstly，the open-source multimodal large language model VIP-LLaVA，which focuses more on regional locations，is chosen as the dialog model.Secondly，a domain-specific multimodal knowledge graph is constructed based on water project inspection data.Lastly，the multimodal knowledge graph (MMKG) of the water conservancy domain is used as the augmented knowledge of the multimodal large model，and the retrieval and generation of collaborative and iterative approach is adopted to identify the most pertinent solutions and to formulate risk response decisions.The results indicate that the method works well on BLEU，ROUGE-L and METEOR evaluation indexes when generating risk response decisions，and can formulate corresponding response decision recommendations to the operational risks in the project in time.The related research can effectively improve the reliability of engineering safety maintenance management and reduce the impact of engineering risks on water supply safety. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |