Method and system for constructing space-time dynamic graph of engineering construction project

Abstract

The invention provides a time-space dynamic graph construction method and system for an engineering construction project, and relates to the field of railway tunnel engineering management.The method comprises the steps that construction data entities are extracted, and an entity set is generated; taking entities in the entity set as nodes to be added into the graph, and configuring attributes for each node; adding a relation edge for each node according to a logic structure between the construction data; adding a time label and a space label to each node so as to form a data graph; associating each entity in the data atlas with a time chain and a space chain to form a data atlas attached to the time-space chain; combining a node merging strategy and a graph neural network to optimize a graph structure; and dynamically modeling dimensional changes of entities and relationships in time by means of a time sequence diagram neural network. According to the method, the dynamic change graph of the relationship between the construction data entities is constructed, so that the spatial-temporal information and the entity relationship of each process are accurately expressed and efficiently associated, and the retrieval efficiency and the management level are favorably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of railway tunnel engineering management, and in particular to a method and system for constructing a spatiotemporal dynamic graph of an engineering construction project. Background Art

[0002] Railway engineering construction data is characterized by being numerous, complex, chaotic, and discrete, and lacks a unified modeling and storage method. Especially in long and complex linear projects, a large amount of heterogeneous data is involved, such as geological forecasts, construction logs, progress ledgers, maintenance records, and experimental test reports. These data contain key information such as construction progress, time, and space. Traditional data recording carriers mainly process linear data and cannot effectively integrate spatiotemporal dynamic information, resulting in information disconnection, affecting construction decisions and overall management. Although RAG (Retrieval-Augmented Generation)-based technology can store data in a vector database, it will lose time and space information, resulting in inaccurate retrieval results. Existing data graph technology can better connect multiple data through nodes and edges, but how to accurately construct an entity relationship graph containing spatiotemporal information is still a technical problem. Therefore, it is urgent to propose a lossless and structured data storage and representation method. Summary of the invention

[0003] The present invention aims to solve at least one of the technical problems existing in the prior art or related technology, discloses a method and system for constructing a spatiotemporal dynamic graph of an engineering construction project, constructs a dynamic change graph of construction data entities and the relationships between entities, so that the spatiotemporal information and entity relationships of each process can be accurately represented and efficiently associated, which can not only improve the retrieval efficiency but also accurately manage and track the dynamic changes in the construction process, providing an important basis for real-time monitoring, quality assessment and progress optimization of construction.

[0004] The first aspect of the present invention discloses a method for constructing a spatiotemporal dynamic graph of an engineering construction project, comprising: extracting a construction data entity set: extracting construction data entities from a construction process document to generate an entity set; constructing a data graph: adding entities in the entity set as nodes to the graph and configuring attributes for each node; adding relationship edges to each node according to the logical structure between construction data; adding time tags and space tags to each node to form a data graph; attaching time chains and space chains: sorting nodes in time according to time information of the construction process to form a chain in the time dimension; classifying nodes according to construction locations or mileage ranges to form a chain in the space dimension; using time tags and space tags, associating each entity in the data graph with the time chain and the space chain to form a data graph attached to the spatiotemporal chain; graph compression: calculating the similarity between nodes, identifying similar nodes according to a set similarity threshold to merge nodes, and the node merging strategy includes: aggregating according to the time attributes or space attributes of similar nodes, aggregating similar nodes as child nodes into the same parent node; if the merged child nodes have differences in some attributes, In the case of differences, the difference attributes are combined into multi-valued attributes; if the merged sub-node has multiple edges with the same node, the weight information of multiple edges is retained; based on the graph neural network, the deep relationship between nodes and edges in the data map is captured to optimize the weights of nodes and edges; construct a dynamic map: based on the time-series graph neural network, a time embedding vector is introduced for each node in the data map. The time embedding vector is generated according to the historical feature sequence of the node and updated iteratively over time; in the convolution process of each layer of the time-series graph neural network, the current state of the node is integrated with the information of the historical time point using the time series aggregation mechanism; the edge weight value is dynamically adjusted using the time-sensitive edge weight update mechanism, and the time-series graph neural network can embed the time factor in the edge weight by adding time information to the edge calculation; in order to ensure the continuity and consistency of the data map over time, the time-series graph neural network adopts a state transfer mechanism to regularly insert new nodes and edges into the map, and the information of the new node is generated based on historical similar nodes and time embedding vectors; the characteristics of the nodes and edges are retrained according to actual construction needs so that the model can reflect the real-time status of the construction process.

[0005] According to the method for constructing a spatiotemporal dynamic graph of an engineering construction project disclosed in the present invention, preferably, before the step of extracting the construction data entity set, it also includes: data preprocessing: uniformly normalizing and structuring the data documents generated during the construction process.

[0006] According to the method for constructing a spatiotemporal dynamic graph of an engineering construction project disclosed in the present invention, preferably, after the step of constructing the dynamic graph, it also includes: dynamic graph optimization: optimizing the timing graph neural network by means of time window adjustment, weight adaptive update and data incremental update to improve the performance and computational efficiency of the timing graph neural network in a dynamic environment.

[0007] The method for constructing a spatiotemporal dynamic graph of an engineering construction project disclosed in the present invention preferably also includes: recording spatial changes as time series data by parsing construction logs; mapping the status updates of spatial positions to corresponding time nodes; parsing and mapping the spatial information involved in the logs to specific time points of the time graph whenever new construction log data is input; dynamically expanding the features of each time node by adding the status updates of spatial positions to the slices of the time graph, so that the spatial changes can be represented by the updates of the time graph.

[0008] According to the method for constructing a spatiotemporal dynamic graph of an engineering construction project disclosed in the present invention, preferably, after the step of graph compression, it also includes: an efficient retrieval strategy: by traversing the time chain, locating the data graph slice at the target time point, and retrieving the construction status and material usage data contained in the data graph slice; or by selecting the start and end time points on the time chain, locating and retrieving all relevant data graph slices within the time interval, and performing aggregation analysis to generate the overall situation within the target time period; or by traversing the spatial chain and finding the target work point location, locating the relevant data graph slices, and then combining the construction log to locate the time, thereby converting the spatial point problem into a time point problem for retrieval; or setting the start point and the end point on the spatial chain, marking all work points within the range, identifying the corresponding time period of each work point on the time chain, obtaining the time slice within the spatial range, and performing aggregation operations on multiple slices to generate overall analysis data of the construction situation within the work point interval.

[0009] According to the method for constructing a spatiotemporal dynamic graph of an engineering construction project disclosed in the present invention, preferably, the construction process data documents include: a production notice, a batching notice, an ecological ledger, a mix ratio report, a smooth blasting self-evaluation form, a tunnel visualization surrounding rock list, a process description, a concrete maintenance record form, a quality and safety inspection notice, and three-dimensional section scanning data.

[0010] According to the method for constructing a spatiotemporal dynamic graph of an engineering construction project disclosed in the present invention, preferably, the construction data entities include: construction site, concrete, surrounding rock, over-excavation volume, over-limit volume, progress index, face excavation, initial support, initial support of invert, invert filling, invert lining, lining arch wall, forecast conclusion, initial support concrete, initial support concrete of invert, invert filling concrete, invert lining concrete, lining arch wall concrete, actual usage, disk release, excess consumption, drilling, charging and blasting, slag discharge, erection of frames and hanging of nets, locking feet and anchor rods, inspection report, shotcrete spraying, and cross-section excavation.

[0011] The second aspect of the present invention discloses a spatiotemporal dynamic graph construction system for engineering construction projects, including: a memory for storing program instructions; a processor for calling the program instructions stored in the memory to implement a spatiotemporal dynamic graph construction method for engineering construction projects such as any of the above-mentioned technical solutions.

[0012] The beneficial effects of the present invention include at least: based on the application of spatiotemporal dynamic graph, it is possible to accurately manage and track dynamic changes in the construction process, providing an important basis for real-time monitoring, quality assessment and progress optimization of construction; at the same time, the dynamic modeling technology of spatiotemporal graph supports the real-time update of data graphs of railway construction projects, providing efficient and flexible tools for monitoring, management and optimization of the construction process. Specifically, the graph compression algorithm is a lossless compression, which retains the core information while greatly reducing the scale of the graph, and avoids the storage and processing of redundant information. For example, when performing construction progress analysis, the spatiotemporal dynamic graph can be directly generated based on the merged node information, saving memory space and accelerating the traversal and query efficiency of the graph structure. This compression method is particularly suitable for highly repetitive scenes in railway construction, such as adjacent construction records, the same material usage records, etc., so that the system can significantly reduce the storage cost and computing resource consumption on the basis of maintaining information integrity. Temporal GNN (temporal graph neural network) captures the changes of nodes and edges over time in the time dimension through hierarchical temporal learning, supports dynamic updating of the graph, and maintains the consistency of the graph with the construction site. Based on efficient retrieval strategies, fast and accurate multi-dimensional queries can be achieved on highly compressed spatiotemporal dynamic graphs, greatly improving the management and monitoring capabilities of railway construction projects. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A schematic diagram of the overall architecture of a method for constructing a spatiotemporal dynamic graph of an engineering construction project according to an embodiment of the present invention is shown.

[0014] Figure 2 A schematic block diagram of a system for constructing a spatiotemporal dynamic graph of an engineering construction project according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0015] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention can also be implemented in other ways different from those described herein, and therefore, the present invention is not limited to the limitations of the specific embodiments disclosed below.

[0016] According to one embodiment of the present invention, a method for constructing a spatiotemporal dynamic graph of an engineering construction project is disclosed, comprising: Extract construction data entity set: extract construction data entities from construction process data documents and generate entity sets; Constructing a data graph: adding entities in the entity set as nodes to the graph and configuring attributes for each node; adding relationship edges to each node according to the logical structure between construction data; adding time labels and space labels to each node to form a data graph; Attachment of time chain and space chain: Sort the nodes in time according to the time information of the construction process to form a chain in the time dimension; classify the nodes according to the construction location or mileage range to form a chain in the space dimension; use time tags and space tags to associate each entity in the data map with the time chain and space chain to form a data map attached to the time and space chain; Graph compression: Calculate the similarity between nodes, identify similar nodes according to the set similarity threshold for node merging. The node merging strategy includes: aggregating similar nodes according to their time attributes or spatial attributes, taking similar nodes as child nodes, and aggregating them into the same parent node; if the merged child nodes have differences in some attributes, then combine the difference attributes into multi-valued attributes; if the merged child nodes have multiple edges with the same node, then retain the weight information of multiple edges; based on the graph neural network, capture the deep relationship between nodes and edges in the data graph to optimize the weights of nodes and edges; Efficient retrieval strategy: by traversing the time chain, locating the data map slice at the target time point, and retrieving the construction status and material usage data contained in the data map slice; or by selecting the start and end time points on the time chain, locating and retrieving all relevant data map slices within the time interval, and performing aggregation analysis to generate the overall situation within the target time period; or by traversing the spatial chain and finding the target work point location, locating the relevant data map slice, and then combining the construction log to locate the time, thereby converting the spatial point problem into a time point problem for retrieval; or setting the start and end points on the spatial chain, marking all work points within the range, identifying the corresponding time period of each work point on the time chain, obtaining the time slice within the spatial range, and performing aggregation operations on multiple slices to generate overall analysis data of the construction situation within the work point interval; Constructing dynamic graphs: Based on the time-series graph neural network, a time embedding vector is introduced for each node in the data graph. The time embedding vector is generated according to the historical feature sequence of the node and updated iteratively over time. In the convolution process of each layer of the time-series graph neural network, the time series aggregation mechanism is used to fuse the current state of the node with the information of the historical time point. The time-sensitive edge weight update mechanism is used to dynamically adjust the weight value of the edge. By adding time information to the calculation of the edge, the time-series graph neural network can embed the time factor in the weight of the edge. To ensure the continuity and consistency of the data graph over time, the time-series graph neural network adopts a state transfer mechanism to regularly insert new nodes and edges into the graph. The information of the new node is generated based on historical similar nodes and time embedding vectors. The characteristics of the nodes and edges are retrained according to actual construction needs to ensure that the model always reflects the real-time status of the construction process.

[0017] like Figure 1 As shown, the specific implementation of the method for constructing a spatiotemporal dynamic graph of an engineering construction project disclosed in the above embodiment includes the following aspects: S1, domain modeling: for the specific scenario of railway tunnel construction, build a construction data map that depends on time and space; S2, graph compression algorithm: It combines the node merging strategy of multi-timestamp and attribute aggregation with the graph neural network (GNN) to perform deep compression modeling on the spatiotemporal graph and efficiently optimize the graph structure; S3, efficient retrieval: Based on the construction log medium, relying on time series efficient retrieval; S4, dynamic update algorithm: relying on the temporal graph neural network (Temporal GNN), dynamically modeling the changes of entities and relationships in the time dimension.

[0018] The above S1 includes: In the complex environment of railway tunnel construction, a large amount of data is generated during the construction process and has a highly structured demand. These data include but are not limited to: production notice, batching notice, ecological ledger, mix ratio report, smooth blasting self-assessment form, tunnel visualization surrounding rock list, process description, concrete maintenance record form, quality and safety inspection notice and three-dimensional cross-section scanning, etc. First, these data are uniformly standardized and structured to meet the needs of subsequent analysis and model building. S1 specifically includes: S1.1 Key scenario analysis: Four main analysis objectives were defined for key scenarios in the construction process: Judgment on the timing of secondary lining construction: During the drilling and blasting construction of railway tunnels, the compliance of secondary lining construction is comprehensively analyzed and cross-verified in combination with surrounding rock monitoring data and construction logs to ensure the safety and stability of the structure, optimize the project progress, reduce project costs, and comply with regulatory requirements.

[0019] Concrete Usage Analysis: Accurately analyze the demand and actual usage of concrete to optimize resource allocation and cost control.

[0020] Concrete quality analysis: During the drilling and blasting construction of railway tunnels, the quality of the secondary lining concrete is tracked and evaluated by analyzing the compressive strength and rebound strength of the specimens, identifying problems in the design and construction process to improve construction quality and reduce economic costs.

[0021] Construction progress analysis: Real-time monitoring of construction progress, timely adjustment of construction plans and resource allocation, and ensuring that the project is completed on schedule.

[0022] S1.2 Entity extraction and data graph construction: For the selected scenarios and construction data under the scenarios, through expert analysis, some examples of the entity sets summarized are as follows: construction site, concrete, surrounding rock, over-excavation volume, over-limit volume, progress index, face excavation, initial support, initial support of invert, invert filling, invert lining, lining arch wall, forecast conclusion, initial support concrete, initial support concrete of invert, invert filling concrete, invert lining concrete, lining arch wall concrete, actual usage, disk release, overconsumption, drilling, charging and blasting, slag discharge, stand and net hanging, lock feet and anchor rods, inspection report, shotcrete, cross section Based on the above entity set, the construction process of the data graph includes the following steps: Create the core entity objects in the graph according to the extracted entity set. For example, entities such as "construction mileage", "excavation method", and "equipment type" will be added to the graph as nodes; then assign the corresponding attributes to each entity object. For example, the attributes of the "maintenance record" entity may include maintenance method, temperature, humidity, etc.; the attributes of the "inverted arch initial support" entity may include front mileage, footage, etc.; then, according to the logical structure in the construction data, with the guidance of experts, add relationship edges to the relevant entities; then bind the entity objects with time and space labels. For example, the start date and end date of the "inverted arch initial support concrete" entity are marked, and the work point where it is located is also marked.

[0023] S1.3 Time and space chain attachment: In view of the time and space sensitivity of railway tunnel construction data, the following steps are used to connect the graph to a one-way chain of time and space to achieve time and space binding of data to avoid loss and interference: (1) Extract all entity nodes and their associated attributes and relationships from the constructed data graph.

[0024] (2) Sort the entities in time according to the time nodes in the construction process (such as the construction start date, end date, etc.) and build a chain in the time dimension.

[0025] (3) According to the construction location or mileage range, the entities are divided into corresponding spatial nodes, for example, using work points, mileage sections, etc. as the classification basis.

[0026] (4) Use time and space labels to associate the sorted entities with their corresponding time nodes and space nodes to form a chain relationship of time and space serialization.

[0027] (5) Integrate the processed spatiotemporal correlation relationship with other information in the graph and output a data graph containing a complete spatiotemporal chain.

[0028] The above S2 includes: In railway construction projects, building graphs based on spatiotemporal chains can ensure the accuracy of the spatiotemporal information of the data and facilitate the systematic archiving and processing of the data. However, the amount of data generated during the construction process is huge, and the high similarity between different graphs also causes a waste of resources. For example, the same construction site recorded at different time points may generate multiple similar entity nodes, and the relationships between these nodes (such as material usage, construction process status, etc.) may also be highly overlapping. In order to resolve the contradiction between accuracy and practicality, a graph compression algorithm based on graph neural network (GNN) is proposed to optimize storage space, improve data processing efficiency, and accelerate the process of data query and analysis. S2 specifically includes: S2.1 Redundant data identification and removal: First, a comparison and merging algorithm based on node similarity is proposed. The core idea of ​​the algorithm is to automatically identify and merge similar nodes according to the set similarity threshold by calculating the similarity between nodes. The specific steps are as follows: (1) Node similarity calculation: The characteristics of each node can be regarded as a vector (for example, including the node's attributes such as construction location, material type, timestamp, construction status, etc.). The vector similarity calculation method is used to measure the similarity between two nodes. The similarity formula is:

[0029] in, and are two node feature vectors to be compared, is the dimension of the feature vector, and i represents the i-th node.

[0030] (2) Similar node comparison and threshold determination: Setting a similarity threshold , if the similarity between two nodes , they are considered redundant nodes. The setting of this threshold can be adjusted according to the specific characteristics of the data. For example, in a group of nodes with high similarity, it is appropriate to reduce value to enhance the merging effect.

[0031] (3) Node merging: When two nodes are determined to be redundant, the system will merge them into a new node and merge their edges. The weight of the merged edge will be weighted averaged according to the importance of the relationship between the edges to retain the representativeness of the edge. For example, if the node and nodes Both with nodes If there are edges of the same type, the weight of the merged edge can be calculated as:

[0032] in, and is the edge weight before merging, and the weight after merging is its average value.

[0033] For example: Suppose there are two nodes and , they represent the status of different work points and have the following characteristics: : Location is "Construction Site 1", timestamp is "October 1, 2024", material type is "Concrete", status is "Initial Support" node :The location is "Construction Site 1.5", the timestamp is "October 1, 2024", the material type is "Concrete", the status is "Initial Support", due to the node and They are exactly the same in time, material and status, and the construction locations are very close, so they can be regarded as redundant nodes. Assuming the similarity threshold is 0.9, and If the similarity is 0.95, the merge condition is met. and Merge into a new node , the spatial attribute can be expressed as "construction point 1-1.5" and all associated edges are merged.

[0034] Optimization effect: This node comparison and merging algorithm significantly reduces the size of the graph while retaining the core information and avoiding the storage and processing of redundant information. For example, when analyzing the construction progress, a spatiotemporal dynamic graph can be generated directly based on the merged node information, saving memory space and accelerating the traversal and query efficiency of the graph structure. This compression method is particularly suitable for highly repetitive scenarios in railway construction, such as adjacent construction records, identical material usage records, etc., which enables the system to significantly reduce storage costs and computing resource consumption while maintaining information integrity.

[0035] S2.2 Node merging strategy In the above process, in order to retain the independent information of different nodes and perform effective lossless compression at the same time, a node merging strategy of multiple timestamps and attribute aggregation will be adopted. This strategy can not only retain the specific information of each node, but also distinguish the status and attributes in different time and space segments when querying. The node merging strategy specifically includes: 1. When the core attributes of two nodes, such as spatial (temporal) position and material type, are consistent, but the time (space) stamps are slightly different, they are considered as "child nodes" in the time (space) order, rather than completely redundant nodes. The system aggregates these child nodes into a "parent node" and retains all time (space) stamp information in the "parent node" so that specific time (space) points can be distinguished when querying. For example, if the node and Both are located in "Construction Location 1", have the same material type "concrete" and status "initial support", but the timestamps are "October 1, 2024" and "October 2, 2024" respectively, then you can merge the nodes Two timestamps are retained in , expressed as [October 1, 2024, October 2, 2024].

[0036] 2. If the merged child nodes differ in some attributes, the system will retain these attributes as multi-valued attributes. For example, different material usage or different monitoring values ​​can be stored as time series data and accurately called by timestamp when querying. For the attribute of "concrete usage" in the example, if the node and The concrete consumption of the two nodes is 50 tons and 60 tons respectively. The recorded concrete consumption is [October 1, 2024: 50 tons, October 2, 2024: 60 tons].

[0037] 3. Multi-weight merging of edges: If the merged node has multiple edges with the same node, the weight information of multiple edges is retained to distinguish the relationship weights at different time points during query. For example, if the node and Respectively with nodes There are different construction activity records. After merging, the timestamp and activity type will be retained on the edge, and the corresponding activity record can be accessed by timestamp when needed.

[0038] Effect of merging strategy: Through the above multi-timestamp and multi-attribute node merging strategy, effective graph compression is achieved while ensuring the traceability and distinguishability of information in different time periods and with different attributes. This method can significantly optimize the storage efficiency of spatiotemporal dynamic graphs while providing more flexible query support.

[0039] S2.3 Feature Learning: After removing redundant information and merging similar nodes from the graph, in order to further optimize the model's expressiveness and feature weights, a graph neural network (GNN) is introduced for feature training. Through feature learning, the model can capture the deep relationships between nodes and edges in the graph, optimize the weights of nodes and edges, and thus more effectively extract key information. The specific steps include: Capturing local graph structure information: During feature training, the graph convolutional network (GCN) is first used to capture the local connection features between entities in the graph. GCN integrates the information of adjacent nodes into the current node through neighborhood aggregation operations at each layer, so that the features of each node are not only its own information, but also the contextual information of its neighborhood structure. This feature aggregation process can effectively extract the local structural features of the graph and deepen the understanding of complex graphs layer by layer. The feature update formula of GCN is

[0040] in Indicates The node feature matrix of the layer, is the adjacency matrix of the graph, is the degree matrix, is the training weight matrix, is a nonlinear activation function. Through GCN, the model can aggregate the information of neighboring nodes layer by layer, effectively capture the local features of the graph, and enable the model to better understand the connection characteristics and structural relationships between nodes.

[0041] Distribution of relationship edge weights: In addition to node feature aggregation, the model also needs to dynamically assign weights to each edge in order to identify and highlight important relationships. To this end, the Graph Attention Network (GAT) is introduced to dynamically calculate the weights of edges. GAT introduces an attention mechanism that enables the model to allocate adaptive weights between neighboring nodes, thereby enhancing the representation of important relationships. The attention weight calculation formula is:

[0042] in Is a node and The weight of the edge between is the learnable weight matrix, represents the vector concatenation operation, T represents the attention mechanism parameters, j represents the input feature vector of the jth node, iRepresents the input feature vector of the i-th node. The edge weights obtained after Softmax regularization are used to weight the aggregate node features, thereby enhancing the discrimination and representation of important relationships. Through this process, the model can distinguish different node relationships based on the edge weights and achieve refined modeling of the graph structure.

[0043] Modeling and prediction of relationships between entities: After capturing the features of nodes and edges, entity embeddings are provided as input to GNN to learn and predict the types of relationships between entities. Through feature embedding and weight optimization, the model can effectively capture the complex relationship information in the graph. This process ensures that the relationship information in the complex network structure is accurately modeled, while improving the model's predictive ability, making the relationship representation more refined and accurate. Specifically, the loss function is optimized by training the model:

[0044] in, represents the true label, Representing the predicted value, by minimizing the loss function, the model can effectively learn the potential relationship types between entities. This optimization process ensures that the model accurately models the relationship information in the complex network structure, while enhancing the prediction ability and making the relationship representation more refined and accurate.

[0045] Effect of feature training: further compressing and optimizing redundant information in spatiotemporal dynamic graphs, improving the effectiveness of graph representation and the accuracy of retrieval, and providing more efficient data support for dynamic monitoring, progress tracking and resource optimization of railway construction projects.

[0046] The above S3 includes: After lossless compression of the graph data, the data graph volume and storage space are significantly optimized, thereby speeding up the overall retrieval speed. The question retrieval process for railway construction scenarios is usually highly dependent on precise positioning in time and space. For this reason, the retrieval questions are classified into the following types in order to systematically process and improve the accuracy of the query: 1. Timing issues: Time point queries refer to data queries for a specific time point, usually used to query the construction status, material consumption or project progress at a certain moment. For example, query the initial support concrete consumption of a construction site on a specific date. Since the compressed graph retains multiple sub-node information and timestamps at each time node, the system can directly locate the node from the time chain and extract relevant information.

[0047] Implementation method: By traversing the time chain, quickly locate the atlas slice at the target time point and retrieve the construction status and material usage data contained therein. In view of the atlas compression structure, the system retains detailed spatiotemporal correlation information for each time point at the node on the time chain, which greatly improves the efficiency of the retrieval operation.

[0048] 2. Time period issue: Time period queries refer to queries for construction data within a certain time interval. This type of search is usually used to analyze construction progress and resource consumption trends within a certain period of time. For example, analyze the consumption of initial support concrete in a tunnel within a month. Time period searches can provide detailed information within a time span.

[0049] Implementation method: By selecting the start and end time points on the time chain, the system can quickly locate and retrieve all relevant atlas slices within the time interval. In each slice, detailed construction information at that time point can be extracted and aggregated and analyzed to generate the overall situation of the time period.

[0050] 3. Spatial point problem: Spatial point questions are queries for specific geographic locations, usually used to understand the specific construction status or resource usage of a construction site. For example: Analyze the initial support concrete usage of DK*69+627, the main tunnel construction point at the entrance of the ** tunnel. For this type of problem, the construction point can be located by traversing the spatial chain, but by using the construction log as a medium, the problem can be mapped to the time chain, which usually corresponds to one or more continuous slices of the time chain map. Therefore, the problem is converted into (1) or (2) problem for retrieval, which greatly improves the retrieval efficiency and accuracy. Implementation method: During the graph compression process, the spatial chain retains the spatial location and construction status information of each work point. By traversing the spatial chain and finding the target work point location, the system can quickly locate the relevant graph slices, and then combine the construction log to locate the time, thereby realizing the dual mapping of time and space.

[0051] 4. Space segment problem: The spatial segment problem is a query within a specific mileage range. For example: Analysis of the concrete usage of the initial support of the entire section of the ** tunnel entrance main tunnel working point DK*69+593-DK*69+597 mileage range. This query involves multiple time node information at multiple spatial locations, so it can be converted into a many-to-many mapping problem, and searched by traversing the space and time chains at the same time.

[0052] Implementation method: The spatial segment query first locates the starting and ending spatial points, and marks all the work points within the range through the spatial chain. Next, the corresponding time period of each work point is identified on the time chain, the time slice within the spatial range is obtained, and the multi-slice aggregation operation is performed to generate the overall analysis data of the construction situation within the mileage interval.

[0053] Effect of efficient retrieval: Through the introduction of efficient retrieval strategies, fast and accurate multi-dimensional queries can be achieved on highly compressed spatiotemporal dynamic graphs, greatly improving the management and monitoring capabilities of railway construction projects.

[0054] The above S4 includes: So far, the spatiotemporal graph set is still static and applicable to archived construction data. However, the dynamic characteristics of railway construction projects cause the data to be continuously updated and evolved over time, which makes the static graph unable to meet the needs of long-term projects under construction to continuously update construction data. Therefore, in order to enable the system to process dynamic data and capture the graph information that changes over time during the construction process, the temporal graph neural network (Temporal GNN) is introduced to realize the dynamic modeling of the spatiotemporal graph. Temporal GNN can efficiently capture the evolution characteristics of entities and relationships over time, support efficient modeling of the time dimension, thereby constructing a time-sensitive knowledge graph to ensure that the system can always reflect the latest status of the construction site. The specific steps include the following: S4.1, Timing Modeling Requirements Analysis: During railway construction, the time dependency of data is reflected in many aspects: Real-time changes in construction status: such as the progress of support structures and the use of materials, which will change over time. Continuous introduction of new entities: The addition of new construction sites or engineering projects requires the system to dynamically expand the graph. Relationship updates: As construction progresses, process relationships, construction sequences, and time dependencies are also changing. To this end, Temporal GNN uses hierarchical temporal learning to capture the changes of nodes and edges over time in the time dimension, support dynamic updates of the graph, and maintain consistency between the graph and the construction site.

[0055] S4.2, Temporal GNN dynamic feature modeling process: Node time embedding: For each node (such as construction site, material type, etc.), the system introduces a time embedding vector to represent the time sensitivity of the node. The time embedding vector is generated based on the historical feature sequence of the node and updated iteratively over time. Through time embedding, the model can reflect the characteristics of the node changing over time in the graph.

[0056] in, For a node at time Features, is the adjacency matrix, is the weight matrix at the current time, represents the aggregation process of the historical features of the node, is the history window length.

[0057] Temporal update of edge weights: Use a time-sensitive edge weight update mechanism to dynamically adjust the edge weights so that they can reflect the importance of the relationship between nodes over time. By adding time information to the edge calculation, Temporal GNN can embed time factors in the edge weights. For example, suppose the edge weight changes over time as:

[0058] Where (t) represents time t, It's time Node and nodes The edge weights between T represents the attention mechanism parameters, j represents the input feature vector of the jth node, i Represents the input feature vector of the i-th node.

[0059] State transfer and update mechanism: To ensure the continuity and consistency of the graph over time, Temporal GNN uses a state transfer mechanism to regularly insert new node and edge information into the graph. The information of the new node is generated based on historical similar nodes and time embedding, so that the newly added nodes can quickly find an adaptive position in the model. At the same time, the characteristics of the nodes and edges are retrained according to the actual construction needs to ensure that the model always reflects the real-time status of the construction process.

[0060] S4.3, Optimization of dynamic model: In order to achieve real-time and effective management of dynamic data during the construction process, the Temporal GNN model has added a variety of optimization measures in its design to improve the performance and computational efficiency of the model in a dynamic environment, including time window adjustment, weight adaptive update, and data incremental update.

[0061] 1. Dynamic adjustment of time window: Definition: Time window length It is an important parameter in Temporal GNN, which determines the amount of historical information used by the model at each state update. Different time window lengths have different effects on real-time performance and computational efficiency.

[0062] Function: In practical applications, the spatiotemporal dynamic characteristics of railway construction projects will change with the construction phase. For example, at key construction nodes (such as tunnel excavation or support structure construction), the real-time monitoring requirements are high, and the time window can be set shorter to capture construction changes more timely; in the regular construction phase, appropriately extending the time window can aggregate more historical data and improve the stability of the model.

[0063] Adjustment strategy: Short time window: suitable for key node monitoring and real-time response. Through a small time span, the system can more quickly reflect status changes in a short period of time.

[0064] Long-term window: Suitable for global monitoring, especially for macro tasks such as analyzing material consumption trends and progress forecasting. Long-term window can reduce the frequency of model updates and improve calculation efficiency.

[0065] Through the flexible time window adjustment mechanism, Temporal GNN can achieve a balance between real-time performance and efficiency in different construction stages.

[0066] 2. Weight adaptive update: Definition: Weight adaptive update means that in Temporal GNN, the model can dynamically adjust the weights of edges and nodes according to changes in the time dimension to make them more consistent with the characteristics of time series data.

[0067] Function: In the railway construction map, the strength of the relationship between different nodes and edges will change over time. For example, the relationship of certain construction parts may be particularly important in a specific stage, and the weight of the relationship will gradually weaken after the construction stage. Therefore, the model needs to update the weights of edges and nodes based on the changes in the time dimension through an adaptive mechanism to reflect the importance and time sensitivity of the relationship.

[0068] Update method: Time decay mechanism: Apply time decay to outdated information or data with a long history, gradually reducing the weights of these nodes and edges, so that the model can focus on recent data.

[0069] Adaptive learning: The model automatically adjusts weights according to the time dimension during training, so that nodes and edges with closer time have a larger proportion in the prediction.

[0070] Through the adaptive updating of weights, the model can capture important changes in nodes and relationships in the time dimension, so that the time-sensitive characteristics of the construction process are fully reflected.

[0071] 3. Data incremental partial update: Definition: Incremental update means that each time the data is updated, the system only trains the newly added nodes and edges without having to recalculate the entire graph. This update method is suitable for scenarios where construction data is constantly increasing and can effectively reduce computing costs.

[0072] Function: During the construction process, the system needs to frequently update the newly added nodes (such as new construction sites) or edges (such as new material consumption records) of the graph. However, if the entire graph is retrained for each data update, it will result in excessive computational overhead and low efficiency. Therefore, the incremental update mechanism can significantly improve the update efficiency of the graph, allowing the system to quickly adapt to changes in new data.

[0073] Implementation: Local training: In incremental updates, only newly added nodes, edges, and their adjacent nodes are trained, so that new data can be quickly integrated into the existing graph without affecting the global structure.

[0074] Dynamic graph merging: For each incrementally updated local graph, the system will perform a global merge after a period of time to ensure the structural integrity and consistency of the graph.

[0075] Through incremental data updates, the system can efficiently process newly added spatiotemporal data, avoid repeated calculations, and significantly improve the adaptability and computing efficiency to dynamic data.

[0076] S4.4 Spatiotemporal mapping strategy: In order to simplify the tracking of spatial changes and avoid frequent updates of spatial chains, a spatiotemporal mapping strategy is introduced to integrate spatial changes into the time map, so that all data updates are centralized in the time dimension for management.

[0077] Implementation mechanism: Spatial changes are recorded in construction logs: Construction logs usually record the operation activities, status updates, and resource usage of each construction site in chronological order. By systematically parsing the construction logs, spatial changes can be recorded as time series data. The status updates of these spatial locations will be automatically mapped to the corresponding time nodes.

[0078] Mapping and updating of the time map: Whenever new construction log data is input, the system will parse and map the spatial information involved in the log (such as the construction site or specific location) to a specific time point in the time map. By adding the status update of the spatial position to the slice of the time map, the system can dynamically expand the characteristics of each time node so that spatial changes can be represented by the update of the time map. For example, if the construction log records that a support structure casting operation was completed at the location "DK*69+627" on "May 1, 2024", the operation will be mapped to the "May 1, 2024" slice of the time map, enabling the system to fully track the construction status from a time perspective.

[0079] like Figure 2 As shown, according to another embodiment of the present invention, a spatiotemporal dynamic graph construction system 200 for engineering construction projects is also disclosed, including: a memory 201, used to store program instructions; a processor 202, used to call the program instructions stored in the memory to implement the spatiotemporal dynamic graph construction method for engineering construction projects as in the above-mentioned embodiment.

[0080] All or part of the steps in the various methods of the above embodiments can be completed by controlling the relevant hardware through a program, and the program can be stored in a readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other readable medium that can be used to carry or store data.

[0081] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for constructing a spatiotemporal dynamic graph of an engineering construction project, characterized in that: include: Extract construction data entity set: extract construction data entities from construction process data documents and generate entity sets; Constructing a data graph: adding entities in the entity set as nodes to the graph and configuring attributes for each node; adding relationship edges to each node according to the logical structure between the construction data; adding time labels and space labels to each node, thereby forming a data graph; Attaching the time chain and the space chain: sorting the nodes in time according to the time information of the construction process to form a chain in the time dimension; classifying the nodes according to the construction location or mileage range to form a chain in the space dimension; using the time label and the space label, associating each entity in the data map with the time chain and the space chain to form a data map attached to the time and space chain; Graph compression: Calculate the similarity between nodes, identify similar nodes according to the set similarity threshold for node merging, and the node merging strategy includes: aggregating similar nodes according to their time attributes or spatial attributes, and aggregating similar nodes as child nodes into the same parent node; if the merged child nodes have differences in some attributes, then combine the difference attributes into multi-valued attributes; if the merged child nodes have multiple edges with the same node, then retain the weight information of multiple edges; based on the graph neural network, capture the deep relationship between nodes and edges in the data graph to optimize the weights of nodes and edges; Constructing dynamic graphs: Based on the time-series graph neural network, a time embedding vector is introduced for each node in the data graph. The time embedding vector is generated according to the historical feature sequence of the node and is iteratively updated over time. In the convolution process of each layer of the time-series graph neural network, the time series aggregation mechanism is used to fuse the current state of the node with the information of the historical time point. The time-sensitive edge weight update mechanism is used to dynamically adjust the weight value of the edge. By adding time information to the calculation of the edge, the time-series graph neural network can embed the time factor in the weight of the edge. In order to ensure the continuity and consistency of the data graph over time, the time-series graph neural network adopts a state transfer mechanism to regularly insert new nodes and edges into the graph. The information of the new node is generated based on historical similar nodes and time embedding vectors. The characteristics of the nodes and edges are retrained according to actual construction needs.

2. The method for constructing a spatiotemporal dynamic graph of an engineering construction project according to claim 1, characterized in that: Before the step of extracting the construction data entity set, it also includes: Data preprocessing: Unified standardization and structuring of data documents generated during the construction process.

3. The method for constructing a spatiotemporal dynamic graph of an engineering construction project according to claim 1, characterized in that: After the step of building the dynamic graph, it also includes: Dynamic graph optimization: The time-series graph neural network is optimized by means of time window adjustment, weight adaptive update and data incremental update to improve the performance and computational efficiency of the time-series graph neural network in a dynamic environment.

4. The method for constructing a spatiotemporal dynamic graph of an engineering construction project according to claim 3, characterized in that: Also includes: By parsing the construction log, the spatial changes are recorded as time series data; the status updates of the spatial locations are mapped to the corresponding time nodes; Whenever new construction log data is input, the spatial information involved in the log is parsed and mapped to a specific time point in the time map. By adding the state update of the spatial position to the slice of the time map, the characteristics of each time node are dynamically expanded, so that the spatial changes can be represented by the update of the time map.

5. The method for constructing a spatiotemporal dynamic graph of an engineering construction project according to claim 1, characterized in that: After the graph compression step, it also includes: Efficient retrieval strategy: by traversing the time chain, locating the data map slice at the target time point, and retrieving the construction status and material usage data contained in the data map slice; or by selecting the start and end time points on the time chain, locating and retrieving all relevant data map slices within the time interval, and performing aggregation analysis to generate the overall situation within the target time period; or by traversing the spatial chain and finding the target work point location, locating the relevant data map slice, and then combining the construction log to locate the time, thereby converting the spatial point problem into a time point problem for retrieval; or setting the start point and end point on the spatial chain, marking all work points within the range, identifying the corresponding time period of each work point on the time chain, obtaining the time slice within the spatial range, and performing aggregation operations on multiple slices to generate overall analysis data on the construction situation within the work point interval.

6. The method for constructing a spatiotemporal dynamic graph of an engineering construction project according to any one of claims 1 to 5, characterized in that: The construction process data documents include: production notice, batching notice, ecological ledger, mix ratio report, smooth blasting self-evaluation form, tunnel visualization surrounding rock list, process description, concrete maintenance record form, quality and safety inspection notice and three-dimensional section scanning data.

7. The method for constructing a spatiotemporal dynamic graph of an engineering construction project according to any one of claims 1 to 5, characterized in that: The construction data entities include: construction site, concrete, surrounding rock, over-excavation volume, over-limit volume, progress index, face excavation, initial support, initial support of invert, invert filling, invert lining, lining arch wall, forecast conclusion, initial support concrete, initial support concrete of invert, invert filling concrete, invert lining concrete, lining arch wall concrete, actual usage, disk release, excess consumption, drilling, charging and blasting, slag discharge, erection of frames and hanging of nets, locking feet and anchor rods, inspection report, shotcrete, and cross-section.

8. The spatiotemporal dynamic graph construction system for engineering construction projects is characterized by: include: A memory for storing program instructions; A processor is used to call the program instructions stored in the memory to implement the method for constructing a spatiotemporal dynamic graph of an engineering construction project as described in any one of claims 1 to 7.

Images