Objective Dam safety monitoring and early warning is an important means to assess the safety status of dams through comprehensive processing and analysis of various monitoring data. It is highly abstract and fuzzy. Moreover, the dam has numerous monitoring points with disordered distribution, making it difficult for traditional monitoring methods to describe the complex relationships among monitoring points, such as spatial relationships and structural mechanism correlations. Additionally, there is a lack of suitable platforms to systematically summarize and infer these monitoring points and their various complex relationships, thereby enabling comprehensive monitoring and early warning of dam safety from a holistic structural perspective. In response to these issue, a dam safety intelligent monitoring and early warning method based on knowledge graph was proposed. Methods Firstly, according to the regulations in “Drawing standard for hydraulic structure of hydropower and water conservancy project”, monitoring points were named and distinguished in a combination of letters and numbers, forming distinguishable and independent entities. Referring to the manual determination of adjacent monitoring points, two adjacent conditions determining rules were proposed based on spatial mathematical models to form an automatic determination method for monitoring point spatial adjacency relationships. Addressing the common issues of varying lengths and time-step misalignment in data sequences of monitoring points, a rule for determining data sequence similarity based on dynamic time warping distance was proposed, achieving automatic determination of spatial adjacency relationships and data sequence similarity relationships between monitoring points. By analyzing the internal characteristics of monitoring points, the monitored physical quantity, single monitoring point warning level, and affiliated components were designated as object properties, while the three-dimensional coordinates, current monitoring data, and historical monitoring data were designated as data properties, forming intrinsic attribute information of monitoring points. Through the above arrangement, the characteristics of entities, relationships, and attributes centered around dam monitoring points were obtained, forming a knowledge system for dam safety monitoring. Secondly, after comparing the advantages and disadvantages of various graph databases, Neo4j database, known for constructing complex relationships, was selected as the base to describe the complex relationships and internal characteristics of monitoring points using a mesh-like graph structure. Based on the py2neo toolkit to batch process the fixed instructions for building the graph database, the first dam safety monitoring knowledge graph centered around monitoring points was constructed, providing a new approach for efficient organization and relationship representation of monitoring points. Subsequently, based on the knowledge guidance and multi-level relationship query capabilities of the graph database, the associated response mechanism for spatial distribution characteristics and data sequence similarity relationships of monitoring points was established. The method can automatically identify abnormal monitoring points and divide them into different groups based on the intrinsic relationships of monitoring points, achieving interconnection and communication of monitoring points, thereby addressing the difficulty of traditional methods in dealing with complex relationships among dam monitoring points. Furthermore, considering comprehensively the quantity of anomalous monitoring points, the alert level, spatial clustering degree, and similarity of data sequences, a method combining knowledge graph and mathematical models was proposed to introduce the concept of clustering degree for multiple monitoring points and to construct a spatial impact region algorithm for anomalous point groups. It ingeniously transforms the dam safety evaluation problem into a three-dimensional spatial calculation problem of anomalous monitoring point impact regions, achieving the quantification of dam safety monitoring and early warning. Then, taking the impact region of anomalous monitoring points of the dam as the core, considering the abnormal level and the data sequences similarity relationship of monitoring points, the safety operational state score of the dam was calculated, and the function of Tanh was utilized to convert the operational state score into a standardized range. Finally, referring to the safety level standards for dam failure risk in the “Guide for safety assessment of large dams for hydropower station in operation”, the operational state of the dam was divided into four levels, facilitating dam safety managers to differentiate the risk level and enabling comprehensive monitoring and management from the perspective of the overall dam structure. Results and Discussions To validate the feasibility of the proposed dam monitoring and early warning method, a section of a concrete gravity dam was taken as an example, analyzing 155 deformation and seepage monitoring points. The result was that the calculated operational state score of the dam is 76.99, indicating a second-level warning state, requiring timely measures to be taken. Conclusions The research indicates that the proposed method fully leverages the advantages of knowledge graphs in information fusion and brain-like reasoning. It integrates dam monitoring points into a data network for unified analysis, and proposes for the first time the concepts of clustering degree for anomalous monitoring points and structural impact regions. The method realizes multi-point dam safety monitoring and early warning by comprehensively considering the quantity of monitoring points, spatial distribution, and structural mechanism association characteristics, thereby addressing the problem of numerous and independent dam monitoring points. © 2024 Sichuan University. All rights reserved.
