A method for extracting the integrity of equipment connection relationships in building electromechanical systems based on graph machine learning

Abstract

This invention discloses a method for extracting the integrity of connection relationships of building electromechanical systems (EMS) equipment based on graph machine learning. First, based on preset distance and normal angle thresholds, it retrieves the nearest ports of non-identical components for missing or spurious connections, generating several first candidate edges. Then, it calculates the matching score of each first candidate edge: when the first candidate edge with the highest score meets the first matching score threshold, it is taken as the only valid connection of the component port; when the first candidate edge with the highest score does not meet the first matching score threshold, the connection relationship of the component port is repaired using an EMS equipment connection relationship link prediction model, thereby achieving the integrity extraction of the connection relationship graph of building EMS equipment in low-quality BIM models. This invention can complete the integrity extraction of connection relationships of building EMS equipment in low-quality BIM models, significantly reducing the reliance on high-quality modeling.

Description

Technical Field

[0001] This invention relates to the field of building operation and maintenance, and in particular to a method for extracting the integrity of the connection relationships of building electromechanical system equipment based on graph machine learning. Background Technology

[0002] Building electromechanical systems (EMS) are the core infrastructure supporting building operation, encompassing multiple professional subsystems such as HVAC, electrical, water supply and drainage, and fire protection. Each subsystem consists of numerous electromechanical devices connected by pipelines, forming a complete media and signal transmission network. The connections between these devices constitute the logical framework of the building EMS's operating mechanism, playing a crucial supporting role in scenarios such as operation and maintenance management, fault tracing, energy efficiency analysis, and simulation training. Especially in large public buildings, the number of EMS devices is enormous, and the systems are complex and interconnected. Manually querying and analyzing the upstream and downstream relationships of these devices is inefficient and prone to errors. Therefore, this paper employs graph theory methods to generate the connection relationships of building EMS devices within the building's Building Information Model (BIM). However, this method has limited universality and is typically only applicable to high-quality BIM models with high modeling accuracy. For low-quality BIM models, issues such as missing components, spatial offsets, or incomplete attribute definitions during modeling can lead to spurious connections between EMS components, missing ports, and default connector attributes, resulting in incomplete extraction of the building EMS device connection relationships. Summary of the Invention

[0003] The purpose of this invention is to provide a method for extracting the integrity of connection relationships of building electromechanical system equipment based on graph machine learning, so as to solve the problem that the extraction of connection relationships of building electromechanical system equipment in the automatic extraction of connection relationships of building electromechanical system equipment in low-quality BIM models based on graph theory methods is incomplete due to issues such as spurious connections, missing ports, and default connector attributes.

[0004] To address the aforementioned technical problems, the present invention provides a method for extracting the integrity of connection relationships of building electromechanical system equipment based on graph machine learning, comprising:

[0005] Step 1: Construct a vertex set V in the electromechanical equipment connection graph G=(V, E) from the low-quality BIM model using graph theory methods, and extract all connections between vertices V. i The geometric and attribute information of the component is represented; the geometric information includes at least three-dimensional coordinates, floor, minimum bounding box size, number and location of ports, and normals; the attribute information includes at least the electromechanical system to which it belongs, material, medium, specifications, and port connection information.

[0006] Step 2: Traverse the vertices Vi in the vertex set V obtained in Step 1. Using the port connection information in the attribute information of vertex Vi, obtain the set of vertices {Vj} directly connected to it. Check if vertex V exists in the electromechanical equipment connection graph G. i With vertex V j The edge (V) i V j If )∈E, and does not exist, then generate the corresponding edge (V). i V j The edge is generated and updated in the electromechanical equipment connection graph G=(V, E). If the edge already exists, it is skipped and the corresponding edge is not generated again.

[0007] Step 3: Traverse the vertices Vi in the vertex set V mentioned in Step 1. For a vertex Vi that has the characteristics of port number, position and normal information, check whether each port of the component it represents has an edge connection. If a port does not have an edge connection, it is determined that the port connection is missing or false. Wherein, missing connection means that the minimum bounding boxes of two components are connected and the port positions and normals are consistent, but the port connection information attribute is missing; false connection means that the minimum bounding boxes of two components have a small gap, or slightly intersect, or the port normals have a small deviation.

[0008] Step 4: For ports in the component represented by vertex Vi that are found to have missing or false connections in Step 3, based on the preset distance threshold [d] and normal angle threshold [δ], search for neighboring ports that are not in the same component, generate several first candidate edges, and calculate the matching degree score of each first candidate edge:

[0009] When the first candidate edge with the highest score satisfies the first matching degree score threshold [S1], the first candidate edge with the highest score is used as the only valid edge connecting the missing or spurious component ports and updated to the edge set E in the electromechanical equipment connection graph G=(V,E); the connectivity analysis of the electromechanical equipment connection graph G=(V,E) is performed, and the vertex set V is merged based on the valid edge connection to form several independent connected subsets, so as to extract the electromechanical system equipment connection subgraph {Gi} in the low-quality BIM model;

[0010] When the first candidate edge with the highest score does not meet the first matching degree score threshold [S1], the connection relationship of the component port with missing or false connection is repaired by the electromechanical system equipment connection relationship prediction model, so as to completely extract the undirected graph G' of the connection relationship of the building electromechanical system equipment in the low-quality BIM model.

[0011] Furthermore, the method for extracting the integrity of connection relationships of building electromechanical system equipment based on graph machine learning provided by the present invention also includes:

[0012] Step 5: Using graph theory methods combined with the operating mechanism rules of building electromechanical systems, identify the types of electromechanical equipment in the undirected graph G' of the connection relationship of building electromechanical system equipment, and determine the top and bottom devices of the electromechanical system. Use the shortest path algorithm to search the path from each top device to its corresponding bottom device, thereby determining the unidirectional flow direction of the medium. In the path analysis process, remove redundant loops and unnecessary connected edges introduced by redundant prediction or topology errors, and obtain the directed graph G'' of the connection relationship of building electromechanical system equipment that conforms to the operating mechanism.

[0013] Furthermore, the method for extracting the integrity of the connection relationship of building electromechanical system equipment based on graph machine learning provided by the present invention calculates the matching degree score according to formula (1) in step 4:

[0014] , (1);

[0015] In equation (1), S(i, j) is the matching score of the first candidate edge, k is the matching score index of each item, and ω k ω represents the weights of each matching score indicator, where 0 ≤ ω k ≤1 and S k (i,j) represents the matching score of each matching score index, sys represents the consistency of the electromechanical system, mat represents the consistency of the material, med represents the consistency of the medium, spec represents the conformity of the specification model, dist represents the geometric proximity, ang represents the port normal consistency, and floor represents the spatial continuity.

[0016] Furthermore, the method for extracting the integrity of the connection relationship of building electromechanical system equipment based on graph machine learning provided by the present invention includes the following steps in step 4: The method for establishing a link prediction model for the connection relationship of electromechanical system equipment.

[0017] Step 4.1: Using high-quality BIM models from several existing engineering projects as base data, extract the mechanical and electrical equipment, pipelines, and nodes, along with their geometric and attribute information, from each BIM model according to the methods in Steps 1 and 2. This process converts each high-quality BIM model into a structurally complete mechanical and electrical system equipment connection diagram {G} with accurate mechanical and electrical connections. init};

[0018] Step 4.2: For the electromechanical system equipment connection diagram from Step 4.1 {G} init The node and edge attributes of {G} are standardized within a unified coding space, including the node's geometric information, attribute information, and local topological features. Simultaneously, missing or inconsistent attribute descriptions are filled in according to a preset filling strategy, resulting in the standardized electromechanical system equipment connection diagram {G}. std};

[0019] Step 4.3: According to the preset probability and disturbance parameter set {ε} trans , θ rot , δ normal} For the standardized electromechanical system equipment connection diagram of step 4.2 {G std Random translations and small-angle rotations are applied to the node spatial coordinates of {G}, while a small angular perturbation is applied to the port normal to form a geometrically enhanced sample set {G} several times larger. geo};

[0020] Step 4.4: Apply the geometric augmentation sample set {G} from Step 4.3 std The nodes and edges of the graph are subjected to controlled random perturbations, including random fine-tuning of numerical attributes within a preset relative range or truncation within the engineering allowable range, performing intra-class replacement or approximate class substitution on categorical attributes according to a set probability, masking some attributes according to the missing simulation rate to generate missing samples, and generating an enhanced graph {G} containing attribute perturbations. attr};

[0021] Step 4.5: The enhanced atlas {G} generated in Step 4.4, containing attribute perturbations... attr As positive samples, multiple types of negative samples are constructed according to the set negative sample construction strategy, and positive and negative samples are labeled to generate a negative sample enhancement atlas {G}. neg};

[0022] Step 4.6: Enhance the negative sample atlas {G} from Step 4.5 neg The algorithm extracts f-hop neighborhood subgraphs centered on each candidate node as local structure samples. It then performs spatial translation or structural combination on the subgraphs corresponding to different regions and mechatronic systems. By integrating positive and negative samples, node / edge attribute features, subgraph representations, and label information, it generates a link prediction dataset G that is structurally sound, contextually complete, and balanced between positive and negative samples. D Where f is the number of jumps, which is an integer;

[0023] Step 4.7: Employ a small-scale subgraph-level or node-representation-based graph neural network model as the link prediction model, using the link prediction dataset G from Step 4.6. D The link prediction model is trained using the training sample data to obtain the electromechanical system equipment connection relationship link prediction model M, which meets the requirements of edge prediction accuracy.

[0024] Furthermore, the method for extracting the integrity of the connection relationships of building electromechanical system equipment based on graph machine learning provided by the present invention, in step 4, uses the electromechanical system equipment connection relationship link prediction model to analyze the electromechanical system equipment connection relationship subgraph {G}. iMethods for repairing the connection relationships of missing or tampered component ports in a document include:

[0025] Step 4.8, extract the electromechanical system equipment connection relationship subgraph {G} from the low-quality BIM model in Step 4. i The input is fed into the electromechanical system equipment connection relationship link prediction model M. The model calculates the probability scores of node pairs forming edges within and between subgraphs, and selects node pairs with link prediction probabilities greater than a preset threshold [τ] to generate a second candidate edge set. The nodes are, in particular, components represented by vertices that lack port number, position, or normal attribute features, and components represented by vertices that have ports without established edge connections and whose first candidate edges do not meet the matching degree score threshold.

[0026] Step 4.9: Calculate the second candidate edge set from Step 4.8 according to the matching score calculation method. The matching score of each second candidate edge is used to filter the set of valid predicted second candidate edges that are greater than the second matching score threshold [S2]. And will effectively predict the second candidate edge set. In the process of deduplicating the second candidate edges predicted repeatedly between the same node pairs, one second candidate edge is retained and updated to the electromechanical system equipment connection relationship subgraph {G} described in step 4. i In the process, an undirected graph G' representing the connection relationships of building electromechanical system equipment is generated.

[0027] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0028] The present invention provides a method for extracting the integrity of building electromechanical system equipment connection relationships based on graph machine learning. This method is used to extract the integrity of building electromechanical system equipment connection relationships in low-quality BIM models. When the connection ports of components in a low-quality BIM model are missing or spurious, making it impossible to extract the integrity of the building electromechanical system equipment connection relationships, the method identifies the missing or spurious ports. Then, based on a preset distance threshold [d] and normal angle threshold [δ], it retrieves the nearest ports of non-same components of the missing or spurious component ports, generating several first candidate edges. The matching degree score of each first candidate edge is calculated. When the first candidate edge with the highest score satisfies the first matching degree score threshold [S1], the first candidate edge with the highest score is taken as the only valid edge of the missing or spurious component port and updated to the edge set E in the electromechanical equipment connection relationship graph G=(V,E). The method performs connectivity analysis on the electromechanical equipment connection relationship graph G=(V,E), merging the vertex set V based on the valid edge connections to form several independent connected subsets, thereby extracting the electromechanical system equipment connection relationship subgraph {G} from the low-quality BIM model. iThis method achieves robust extraction of the connection relationships of effectively connected MEP (Mechanical, Electrical, and Plumbing) system equipment in low-quality BIM models. When the first candidate edge with the highest score does not meet the first matching degree score threshold [S1], the connection relationships of missing or spurious component ports are repaired through the MEP system equipment connection relationship link prediction model. This allows for the complete extraction of the undirected graph G' of the connection relationships of the building's MEP system equipment in the low-quality BIM model, thus achieving the repair processing of missing or spurious connections at the ports of the MEP system equipment in the low-quality BIM model and ensuring the extraction and repair of the connection relationships of missing or spurious component ports. It avoids the problems of incomplete and inaccurate extraction of MEP system equipment connection relationships in low-quality BIM models, and has the advantages of good extraction completeness and high accuracy. It can robustly extract building MEP system equipment connection relationships and achieve intelligent repair even when BIM modeling quality is uneven, attributes are missing or logical errors exist. Compared with manual repair, it has the effect of high repair efficiency and avoids human omissions. It reduces the dependence on high-quality BIM models. Through the complete extraction of building MEP system equipment connection relationships, it can provide reliable data support for fault diagnosis, energy path analysis and abnormal energy consumption location in the building operation and maintenance phase. Attached Figure Description

[0029] Figure 1 This is a flowchart of a method for extracting the integrity of connection relationships of building electromechanical system equipment based on graph machine learning.

[0030] Figure 2 This is a schematic diagram showing a case where there is a gap between the minimum bounding boxes of two components, resulting in a dummy connection at the component port.

[0031] Figure 3 This is a schematic diagram showing a case where the component ports are spuriously connected due to a slight intersection of the minimum bounding boxes of two components.

[0032] Figure 4 This is a schematic diagram illustrating a case where the normals of the ports of two components are inconsistent, resulting in a false connection at the port of a component.

[0033] Figure 5 This is a diagram showing the repaired port dummy connection. Detailed Implementation

[0034] The present invention will now be described in detail with reference to the accompanying drawings. The advantages and features of the present invention will become clearer from the following description. It should be noted that the drawings are all in a very simplified form and use non-precise proportions, and are only used to facilitate and clarify the illustration of the embodiments of the present invention.

[0035] Please refer to Figure 1 This invention provides a method for extracting the integrity of connection relationships of building electromechanical system equipment based on graph machine learning, including:

[0036] Step 1: Construct a vertex set V in the electromechanical equipment connection graph G=(V, E) from the low-quality BIM model using graph theory methods, and extract all connections between vertices V. i The representation includes the geometric and attribute information of the components; the geometric information includes at least three-dimensional coordinates, floor, minimum bounding box size, number and location of ports, and normals; the attribute information includes at least the electromechanical system to which it belongs, material, medium, specifications, and port connection information. Before constructing the electromechanical equipment connection diagram G=(V, E) from the low-quality BIM model using graph theory methods, the electromechanical equipment, pipelines, and fittings such as tees, elbows, and valves in the BIM can be uniformly abstracted.

[0037] Step 2: Traverse the vertices V in the vertex set V mentioned in Step 1. i Through vertex V i From the port connection information in the attribute information, obtain the set of vertices directly connected to it {V j Check if vertex V exists in the electromechanical equipment connection diagram G. i With vertex V j The edge (V) i V j If )∈E, and does not exist, then generate the corresponding edge (V). i V j The edge is updated in the electromechanical equipment connection graph G=(V, E). If the edge already exists, it is skipped and the corresponding edge is not generated again.

[0038] Step 3: Traverse the vertices V in the vertex set V mentioned in Step 1. i For a vertex V that has characteristics of port number, location, and normal information... i Check whether each port of the component represented by it has edge connections. If a port does not have edge connections, it is determined that the port connection is missing or spurious. Among them, missing connection means that the minimum bounding boxes of two components are connected and the port positions and normals are consistent, but the connection information attributes of the port are missing. Spurious connection means that the minimum bounding boxes of two components have small gaps, or slightly intersect, or the port normals have a small deviation. Figures 2 to 4 Examples of three types of port spoofing are provided.

[0039] Step 4: For the determination made in Step 3 by vertex V i For ports in the represented components that are missing or spuriously connected, based on preset distance thresholds [d] and normal angle thresholds [δ], the nearest neighbor ports that are not in the same component are retrieved, generating several first candidate edges, and the matching degree score of each first candidate edge is calculated:

[0040] The matching score of each first candidate edge can be calculated according to formula (1):

[0041] , (1)

[0042] In equation (1), S(i, j) is the matching score of the first candidate edge, k is the matching score index of each item, and ω k ω represents the weights of each matching score indicator, where 0 ≤ ω k ≤1 and S k (i, j) represents the matching score of each matching score index, sys represents the consistency of the electromechanical system, mat represents the consistency of the material, med represents the consistency of the medium, spec represents the conformity of the specification model, dist represents the geometric proximity, ang represents the port normal consistency, and floor represents the spatial continuity.

[0043] The matching score of the first candidate edge is calculated using seven indicators in formula (1): consistency of electromechanical system, consistency of material, consistency of medium, conformity of specifications and model, geometric proximity, port normal consistency and port normal consistency.

[0044] Among them, the consistency score of electromechanical systems S sys (i, j) is:

[0045] (2)

[0046] Among them, the material consistency score S mat (i, j) is:

[0047] (3)

[0048] In equation (3), α mat Values ​​for materials with different properties;

[0049] Among them, the media consistency score S med (i, j) is:

[0050] (4)

[0051] The specification and model conformity score S spec (i, j) is:

[0052] (5)

[0053] In equation (5), β1, β2, and β3 are the weights of the similarity of the diameter, wall thickness, and interface standard in the specifications, respectively, where:

[0054] Caliber similarity score S d (i, j) is:

[0055] (5.1)

[0056] In equation (5.1), d i d j Let i be the aperture of port i and port j;

[0057] Wall thickness similarity score S t (i, j) is:

[0058] (5.2)

[0059] Interface standard similarity score S iface (i, j) is:

[0060] (5.3)

[0061] Among them, the geometric proximity score S dist (i, j) is:

[0062] (6)

[0063] In equation (6), d is the distance between the two ports;

[0064] Port normal consistency score:

[0065] (7)

[0066] In equation (7), δ is the included angle of the port normal;

[0067] Spatial continuity score:

[0068] (8)

[0069] In equation (8), α floor This represents the value for spatial continuity between adjacent layers within the same shaft.

[0070] When the first candidate edge with the highest score satisfies the first matching degree score threshold [S1], the first candidate edge with the highest score is taken as the only valid edge connecting the missing or spurious component ports and updated to the edge set E in the electromechanical equipment connection graph G=(V,E); connectivity analysis is performed on the electromechanical equipment connection graph G=(V,E), and the vertex set V is merged based on the valid edge connection to form several independent connected subsets, so as to extract the electromechanical system equipment connection subgraph {G} from the low-quality BIM model. i}

[0071] When the first candidate edge with the highest score does not meet the first matching degree score threshold [S1], the connection relationship of the component port with missing or false connection is repaired by the electromechanical system equipment connection relationship prediction model, so as to completely extract the undirected graph G' of the connection relationship of the building electromechanical system equipment in the low-quality BIM model. Figure 5 The example provided shows a port spoofing repair.

[0072] The present invention provides a method for extracting the integrity of building electromechanical system equipment connection relationships based on graph machine learning. This method is used to extract the integrity of building electromechanical system equipment connection relationships in low-quality BIM models. When the connection ports of components in a low-quality BIM model are missing or spurious, making it impossible to extract the integrity of the building electromechanical system equipment connection relationships, the method identifies the missing or spurious ports. Then, based on a preset distance threshold [d] and normal angle threshold [δ], it retrieves the nearest ports of non-same components of the missing or spurious component ports, generating several first candidate edges. The matching degree score of each first candidate edge is calculated. When the first candidate edge with the highest score satisfies the first matching degree score threshold [S1], the first candidate edge with the highest score is taken as the only valid edge of the missing or spurious component port and updated to the edge set E in the electromechanical equipment connection relationship graph G=(V,E). The method performs connectivity analysis on the electromechanical equipment connection relationship graph G=(V,E), merging the vertex set V based on the valid edge connections to form several independent connected subsets, thereby extracting the electromechanical system equipment connection relationship subgraph {G} from the low-quality BIM model. i This method achieves robust extraction of the connection relationships of effectively connected MEP (Mechanical, Electrical, and Plumbing) system equipment in low-quality BIM models. When the first candidate edge with the highest score does not meet the first matching degree score threshold [S1], the connection relationships of missing or spurious component ports are repaired through the MEP system equipment connection relationship link prediction model. This allows for the complete extraction of the undirected graph G' of the connection relationships of the building's MEP system equipment in the low-quality BIM model, thus achieving the repair processing of missing or spurious connections at the ports of the MEP system equipment in the low-quality BIM model and ensuring the extraction and repair of the connection relationships of missing or spurious component ports. It avoids the problems of incomplete and inaccurate extraction of MEP system equipment connection relationships in low-quality BIM models, and has the advantages of good extraction completeness and high accuracy. It can robustly extract building MEP system equipment connection relationships and achieve intelligent repair even when BIM modeling quality is uneven, attributes are missing or logical errors exist. Compared with manual repair, it has the effect of high repair efficiency and avoids human omissions. It reduces the dependence on high-quality BIM models. Through the complete extraction of building MEP system equipment connection relationships, it can provide reliable data support for fault diagnosis, energy path analysis and abnormal energy consumption location in the building operation and maintenance phase.

[0073] To construct a link prediction model for the connection relationships of electromechanical system equipment, and to predict the connection edges of component ports with missing or spurious connections in low-quality BIM models, the method for extracting the integrity of connection relationships of building electromechanical system equipment based on graph machine learning provided in this embodiment of the invention includes the following steps in step 4:

[0074] Step 4.1: Using high-quality BIM models from several existing engineering projects as base data, extract the mechanical and electrical equipment, pipelines, and nodes, along with their geometric and attribute information, from each BIM model according to the methods in Steps 1 and 2. This process converts each high-quality BIM model into a structurally complete mechanical and electrical system equipment connection diagram {G} with accurate mechanical and electrical connections. init}

[0075] Step 4.2: For the electromechanical system equipment connection diagram from Step 4.1 {G} init The node and edge attributes of {G} are standardized within a unified coding space, including the node's geometric information, attribute information, and local topological features. Simultaneously, missing or inconsistent attribute descriptions are filled in according to a preset filling strategy, resulting in the standardized electromechanical system equipment connection diagram {G}. std The filling strategies include, but are not limited to, filling unknown classes, mean / median interpolation, and retaining missing masks.

[0076] Step 4.3: According to the preset probability and disturbance parameter set {ε} trans , θ rot , δ normal} For the standardized electromechanical system equipment connection diagram of step 4.2 {G std Random translations and small-angle rotations are applied to the node spatial coordinates of {G}, while a small angular perturbation is applied to the port normal to form a geometrically enhanced sample set {G} several times larger. geo}

[0077] Step 4.4: Apply the geometric augmentation sample set {G} from Step 4.3 std The node and edge attribute information of} is subjected to controlled random perturbation, including random fine-tuning of numerical attributes (such as pipe diameter, wall thickness, rated power, etc.) within a preset relative range or truncation within the engineering allowable range, performing intra-class replacement or near-class substitution of categorical attributes (such as material, interface standard, to which the electromechanical system belongs, etc.) according to a set probability, masking some attributes according to the missing simulation rate to generate missing samples, and generating an enhanced atlas {G} containing attribute perturbation. attr}

[0078] Step 4.5: The enhanced atlas {G} generated in Step 4.4, containing attribute perturbations... attrAs positive samples, multiple types of negative samples are constructed according to the set negative sample construction strategy, and positive and negative samples are labeled to generate a negative sample enhancement atlas {G}. neg The negative sample construction strategy includes randomly sampling node pairs in the graph that do not have direct connections, and screening node pairs that are spatially geometrically adjacent or have similar graph topology but should not be connected according to electromechanical system design specifications, in order to simulate easily confused non-connectivity relationships and improve the counterexamples for the link prediction model of the connection relationship of the electromechanical system equipment to be established.

[0079] Step 4.6: Enhance the negative sample atlas {G} from Step 4.5 neg The algorithm extracts f-hop neighborhood subgraphs centered on each candidate node as local structure samples. It then performs spatial translation or structural combination on the subgraphs corresponding to different regions and mechatronic systems. By integrating positive and negative samples, node / edge attribute features, subgraph representations, and label information, it generates a link prediction dataset G that is structurally sound, contextually complete, and balanced between positive and negative samples. D , where f is the number of jumps, which is an integer.

[0080] Step 4.7: Employ a small-scale subgraph-level or node-representation-based graph neural network model as the link prediction model, using the link prediction dataset G from Step 4.6. D The link prediction model is trained using training sample data to obtain a link prediction model M for the electromechanical system equipment connection relationship that meets the edge prediction accuracy requirements. Small-scale subgraph-level or node representation-based graph neural network models can be selected from SEAL (Subgraph Embedding Approach for Link Prediction), NBFNet (Neural Bellman-Ford Network), GraphSAGE (Graph Simplified and Aggregate GatE), etc. The model training can use a cross-entropy loss function or a positive / negative sample weighted loss function to iteratively optimize the model parameters, while using a pre-defined validation set and test set to evaluate the model's performance. The selection of the aforementioned graph neural network model can be based on the data scale and the structural characteristics of the building's electromechanical system to choose an appropriate model.

[0081] To achieve intelligent repair of connection relationships at component ports with missing or spurious connections, the present invention provides a method for extracting the integrity of connection relationships in building electromechanical systems based on graph machine learning. In step 4, the method uses an electromechanical system equipment connection relationship link prediction model to analyze the electromechanical system equipment connection relationship subgraph {G}. i Methods for repairing the connection relationships of missing or tampered component ports in a document include:

[0082] Step 4.8: Extract the electromechanical system equipment connection relationship subgraph {G} from the low-quality BIM model in Step 4. i The input is fed into the electromechanical system equipment connection relationship link prediction model M. The model calculates the probability scores of node pairs forming edges within and between subgraphs, and selects node pairs with link prediction probabilities greater than a preset threshold [τ] to generate a second candidate edge set. The nodes are defined as follows: nodes represented by vertices that lack port number, position, or normal attribute features; and nodes represented by vertices that have ports without established edge connections and whose first candidate edges do not meet the matching degree score threshold.

[0083] Step 4.9: Calculate the second candidate edge set from Step 4.8 according to the matching score calculation method. The matching score of each second candidate edge is used to filter the set of valid predicted second candidate edges that are greater than the second matching score threshold [S2]. And will effectively predict the second candidate edge set. In the process of deduplicating the second candidate edges predicted repeatedly between the same node pairs, one second candidate edge is retained and updated to the electromechanical system equipment connection relationship subgraph {G} described in step 4. i In the process, an undirected graph G' representing the connection relationships of the building's electromechanical system equipment is generated. The purpose of deduplication is to retain only one of the two candidate edges, ab and ba, connecting port i and port j.

[0084] To extract the media flow direction of each electromechanical system in the connection relationship of building electromechanical system equipment, thereby providing reliable data support for fault diagnosis, energy path analysis, and abnormal energy consumption location during the building operation and maintenance phase, the method for extracting the integrity of the connection relationship of building electromechanical system equipment based on graph machine learning provided in this embodiment of the invention further includes:

[0085] Step 5: Using graph theory methods combined with the operating mechanism rules of building electromechanical systems, identify the types of electromechanical equipment in the undirected graph G' of the connection relationship of building electromechanical system equipment, and determine the top and bottom devices of the electromechanical system. Use the shortest path algorithm to search the path from each top device to its corresponding bottom device, thereby determining the unidirectional flow direction of the medium. In the path analysis process, remove redundant loops and unnecessary connected edges introduced by redundant prediction or topology errors, and obtain the directed graph G'' of the connection relationship of building electromechanical system equipment that conforms to the operating mechanism.

[0086] The operating mechanisms of building electromechanical systems include, but are not limited to: in air conditioning and heating systems, the top equipment is usually a cold and heat source unit, cooling tower, etc., and the terminal equipment is usually a fan coil unit, air outlet, etc.; in fire water systems, the top equipment is usually a water pump, water tank or external water source access point, and the terminal equipment is usually a fire hydrant, sprinkler head, etc.

[0087] The graph theory algorithm is a common ancestor identification method based on the shortest path. In the tree-like electromechanical equipment connection graph, the vertex with only one connecting edge is regarded as the vertex representing the end device. Then, starting from all the end vertices, the connected paths in the graph are traversed to determine the node with the highest frequency of occurrence as the common ancestor of multiple end nodes, that is, the vertex representing the top device.

[0088] The method for extracting the integrity of building electromechanical system equipment connection relationships based on graph machine learning provided in this invention is used to robustly extract the effective connection parts of building electromechanical system equipment connection relationships in low-quality BIM models, generate a reliable equipment connection relationship sub-graph set, and then combine it with historical high-quality BIM models to construct a link prediction model. This enables the prediction of erroneous or missing equipment connection relationships and the link completion between connection relationship sub-graphs, outputting a complete directed graph of building electromechanical system equipment connection relationships. This method can be applied to intelligent operation and maintenance scenarios such as electromechanical model reconstruction, fault tracing, and energy path analysis, significantly improving the level of intelligent operation and management throughout the building's entire life cycle.

[0089] This invention is not limited to the specific embodiments described above. Obviously, the embodiments described above are only a part of the embodiments of this invention, not all of them. All other embodiments obtained by those skilled in the art based on the described embodiments of this invention are within the scope of protection of this invention. Those skilled in the art can make other modifications and variations to this invention. Therefore, if these modifications and variations of this invention fall within the scope of the claims of this invention, then this invention also intends to include these modifications and variations.

Claims

1.A method for extracting integrity of connection relationship of building mechanical and electrical system equipment based on graph machine learning, characterized in that, Comprise: Step 1: constructing vertex set V in the mechanical and electrical equipment connection relationship graph G=(V, E) from a low-quality BIM model based on a graph theory method, and extracting geometric information and attribute information of all components represented by vertex V i ; wherein the geometric information at least includes three-dimensional coordinates, a floor to which the component belongs, a minimum bounding box size, a port number position, and a normal, and the attribute information at least includes an electrical and mechanical system to which the component belongs, a material, a medium, a specification and model, and port connection information; Step 2: Traverse the vertices V in the vertex set V mentioned in Step 1. i Through vertex V i From the port connection information in the attribute information, obtain the set of vertices directly connected to it {V j Check if vertex V exists in the electromechanical equipment connection diagram G. i With vertex V j The edge (V) i V j If )∈E, and does not exist, then generate the corresponding edge (V). i V j The edge is generated and updated in the electromechanical equipment connection graph G=(V, E). If the edge already exists, it is skipped and the corresponding edge is not generated again. Step 3: traversing the vertex V in the vertex set V in step 1 i For the vertex V with the characteristics of port number, position and normal information i Check whether each port of the component represented by the vertex V has an edge connection, if a port does not have an edge connection, it is determined that the port connection is missing or false; wherein the missing connection is that the minimum bounding box of two components is connected and the port position and normal are consistent, but the connection information attribute of the port is missing; the false connection is that the minimum bounding box of two components has a small gap, or slightly intersects, or the port normal has a small deflection. Step 4: For the missing or false connected ports in the components represented by vertex V i based on the preset distance threshold [d] and the normal angle threshold [δ], the near neighbor ports of the non-same components are searched, a plurality of first candidate edges are generated, and the matching degree scores of the first candidate edges are calculated: When the first candidate edge with the highest score meets the first matching degree score threshold [S1], the first candidate edge with the highest score is taken as the only valid edge connecting the missing or false connected component port, and is updated to the edge set E in the mechanical and electrical equipment connection graph G=(V, E); the connectivity of the mechanical and electrical equipment connection graph G=(V, E) is analyzed, the vertex set V is merged based on the valid edge connection, and a plurality of independent connected subsets are formed to extract the mechanical and electrical system equipment connection relationship subgraph {G i} in the low-quality BIM model. When the highest score of the first candidate edge does not satisfy the first matching degree score threshold [S1], the connection relationship between the missing or false connection of the component port is repaired by the mechanical and electrical system equipment connection relationship link prediction model, so as to extract the connection relationship undirected graph G' of the building mechanical and electrical system equipment in the low-quality BIM model. 2.The building mechanical and electrical system equipment connection relationship integrity extraction method based on graph machine learning according to claim 1, characterized in that, Also include: Step 5: The mechanical and electrical equipment types in the building mechanical and electrical system equipment connection relationship undirected graph G' are identified by adopting the graph theory method combined with the operation mechanism rules of the building mechanical and electrical system, and the top equipment and the end equipment of the mechanical and electrical system are determined, the shortest path algorithm is used to search the path from each mechanical and electrical system top equipment to its corresponding end equipment, so as to determine the one-way flow direction of the medium, and the redundant loops and unnecessary connected edges introduced in the path analysis process due to redundant prediction or topological error are eliminated, and the building mechanical and electrical system equipment connection relationship directed graph G'' conforming to the operation mechanism is obtained. 3.The building mechanical and electrical system equipment connection relationship integrity extraction method based on graph machine learning according to claim 1, characterized in that, In step 4, the matching degree score is calculated according to formula (1): ， （1）； In formula (1), S(i, j) is a matching degree score of the first candidate edge, k is each matching degree score index, ω k is a weight of each matching degree score index, wherein 0≤ω k <1 and S k (i, j) is a matching degree score of each matching degree score index, sys is a system consistency, mat is a material consistency, med is a medium consistency, spec is a specification and model consistency, dist is a geometric proximity, ang is a port normal consistency, and floor is a spatial continuity. 4.The building mechanical and electrical system equipment connection relationship integrity extraction method based on graph machine learning according to claim 1, wherein, In step 4, the method for establishing the mechanical and electrical system equipment connection relationship link prediction model comprises: Step 4.1: Extract the mechanical and electrical equipment, pipelines and nodes and their geometric information and attribute information in each BIM model according to the method of steps 1 to 2, taking the high-quality BIM models of several existing engineering projects as the basis data, so as to convert each high-quality BIM model into a mechanical and electrical system equipment connection relationship graph {G init} with complete structure and accurate mechanical and electrical connection relationship; Step 4.2: The node and edge attributes of the mechanical and electrical system equipment connection graph {G init} of step 4.1 are standardized in a unified coding space, including the geometric information, attribute information and local topological features of the nodes, and the parts with missing or inconsistent attributes are filled according to a preset filling strategy to obtain the standardized mechanical and electrical system equipment connection graph {G std}. Step 4.3: Apply random translation and small angle rotation to the node space coordinates of the normalized mechanical-electrical system device connection graph {G std} of step 4.2 according to the preset probability and perturbation parameter set {ε trans , θ rot , δ normal}, and apply small amplitude angle perturbation to the port normal to form a several-fold geometric enhancement sample set {G geo}; Step 4.4: Apply controlled random perturbation to the attribute information of nodes and edges of the geometrically augmented sample set {G std} of step 4.3, including random fine-tuning or truncation within the engineering allowed range for numerical attributes, in-same-class replacement or approximate class replacement for categorical attributes according to a set probability, and mask processing for part of the attributes to generate missing samples according to a missing simulation rate, to generate an augmented graph set {G attr} containing attribute perturbation; Step 4.5: The enhanced graph set {G attr} containing attribute perturbation generated in step 4.4 is taken as positive samples, and multi-type negative samples are constructed according to the set negative sample construction strategy, and the positive and negative samples are labeled to generate a negative sample enhanced graph set {G neg}; Step 4.6: The negative sample enhanced graph set {G neg} of step 4.5 is extracted as a local structure sample centered on each candidate node, and the subgraphs corresponding to different regions and electronic system are spatially translated or structurally combined. The positive and negative samples, node / edge attribute features, subgraph representation and label information are integrated to generate a link prediction dataset G with complete structure specification, complete context information and balanced positive and negative samples D , where f is the number of hops, an integer; Step 4.7: Use the small-scale subgraph-level or node-based representation-based spliced graph neural network model as a link prediction model to the link prediction dataset G of step 4.6 D As training sample data, the link prediction model is trained to obtain a mechanical and electrical system equipment connection relationship link prediction model M with a required edge prediction accuracy. 5.The building mechanical and electrical system equipment connection relationship integrity extraction method based on graph machine learning according to claim 1, wherein, In step 4, the mechanical and electrical system equipment connection relationship subgraph {G i The method for repairing the connection relationship of the missing or false connected component ports in the connection relationship subgraph {G Step 4.8, the mechanical and electrical system equipment connection relationship subgraph {G i} extracted in step 4 in the low-quality BIM model is input into the mechanical and electrical system equipment connection relationship link prediction model M, the probability score of each node pair in the subgraph and between the subgraphs forming an edge is inferred and calculated, and the node pairs with a link prediction probability greater than a preset threshold [τ] are selected to generate a second candidate edge set ; wherein the node is a component represented by a vertex with missing port number, position or normal attribute characteristics, and a component represented by a vertex with an existing port that is not connected by an edge and whose first candidate edge does not satisfy the matching degree score threshold. Step 4.9, according to the matching score calculation method, the matching score of each second candidate edge in the second candidate edge set described in step 4.8 is calculated, and the effective prediction second candidate edge set is screened out, which is greater than the second matching score threshold [S2] , and the effective prediction second candidate edge set i , and the second candidate edge repeatedly predicted between the same node pair in the effective prediction second candidate edge set is de-duplicated, and one second candidate edge is retained and updated to the mechanical and electrical system equipment connection relationship subgraph {G i} described in step 4, and the building mechanical and electrical system equipment connection relationship undirected graph G' is generated.

Images