Equipment fault identification method and equipment based on graph machine learning

Abstract

The present invention provides a method for identifying equipment faults based on graph machine learning, comprising: obtaining a new building operation and maintenance fault prediction mechanism model #imgabs1# based on a directed graph #imgabs0# and a fault association prediction model M1; predicting equipment warranty work orders in daily building operation and maintenance based on the new building operation and maintenance fault prediction mechanism model #imgabs2# to obtain possible equipment fault-associated connections. The present invention can automatically construct and repair the operation model of the building electromechanical system, thereby providing technical support for the intelligent maintenance and fault repair of building facilities and equipment. The present invention automatically identifies and constructs the operation logic of the building electromechanical system by training a large-scale graph machine learning model. The present invention can construct a topological relationship model between devices to ensure that the model can accurately reflect the actual situation.

Description

Technical Field

[0001] The present invention relates to a method and device for identifying equipment faults based on graph machine learning. Background Art

[0002] The operation mechanism of the electromechanical system (MEP system) of public buildings is complex, and traditional mechanism modeling methods are difficult to meet the needs of intelligent maintenance and fault repair. At present, the operation logic of the building electromechanical system is complex and diverse, which makes modeling difficult and maintenance cost high. Summary of the invention

[0003] The purpose of the present invention is to provide a method and device for equipment fault identification based on graph machine learning.

[0004] To solve the above problems, the present invention provides a method for identifying equipment faults based on graph machine learning, comprising:

[0005] Step 1: Collect historical daily operation and maintenance data from the mechanical and electrical operation system of the building, including: operation monitoring data and work order data;

[0006] Step 2, obtaining component attributes and connection attributes in the BIM electromechanical model of the building;

[0007] Step 3: Store the components and their component attributes and connection attributes as a directed graph ;

[0008] Step 4, based on directed graph , and adopt the node classification algorithm of graph neural network to train the graph neural network and establish the connection direction classification model MS of edge nodes;

[0009] Step 5: Based on the trained node connection direction classification model MS, the undirected graph to be repaired is identified to obtain the corresponding directed graph. ;

[0010] Step 6: Based on the operation monitoring data and work order data in step 1, Establish fault repair association connections between nodes to obtain the directed graph update as ;

[0011] Step 7: Based on the directed graph , train the graph database connection prediction model M1;

[0012] Step 8, based on directed graph And the fault association prediction model M1, a new building operation and maintenance fault prediction mechanism model is obtained ; Based on a new building operation and maintenance failure prediction mechanism model , predict equipment warranty work orders in daily building operation and maintenance, and obtain possible equipment failure-related connections.

[0013] Furthermore, in the above method, step 2, obtaining component properties and connection properties in the BIM electromechanical model of the building, includes:

[0014] Step 2.1, extracting the component attributes in the BIM electromechanical model of the building, including: extracting the category attributes of the component, which is used to distinguish the attributes of pipes, pipe fittings, pipe accessories and equipment in the BIM electromechanical model; extracting the electromechanical system attributes of the component, which is used to distinguish the electromechanical system to which the component belongs, and distinguishing the system attributes of the air conditioning system, air conditioning water system and drainage system; extracting the spatial geometric attributes of the component; wherein the BIM electromechanical model of the building includes: pipes, pipe fittings, pipe accessories and equipment;

[0015] Step 2.2, extracting the connection properties of the components from the BIM electromechanical model of the building, including: upstream and downstream connection relationship properties, connection direction properties and connection quantity properties, wherein the connection relationship of the components in the BIM electromechanical model is a pair of upstream and downstream connection relationships, that is, one connection has only one upstream component and one downstream component; there are two types of upstream and downstream connection relationships of the components, components are connected through pipes and components are directly connected, and the connection direction property is temporarily assigned a default connection direction.

[0016] Furthermore, in the above method, in step 3, the components and their component attributes and connection attributes are stored as a directed graph ,include:

[0017] Step 3.1, creating different nodes according to the category attributes of the components; storing the extracted component attributes in the corresponding nodes in the connection relationship undirected graph, and storing the system attributes and spatial geometric attributes as the attributes of the corresponding nodes;

[0018] Step 3.2, converting the connection attributes of the components into edges in the connection relationship undirected graph, wherein, according to the upstream and downstream connection relationship attributes, corresponding nodes in the connection relationship undirected graph are connected to form edges, and the connection direction attribute and the connection quantity attribute are stored as attributes of the edge to form a connection relationship undirected graph containing nodes and edges;

[0019] Step 3.3, develop a visualization interface to visualize the connection direction between components, obtain the correction operation of the connection direction attribute between components input by the user in the visualization interface, obtain the direction of the corrected edge based on the correction operation, and convert the connection relationship undirected graph into a directed graph based on the corrected edge direction. ,in, is a collection of nodes, is the set of edges.

[0020] Furthermore, in the above method, step 4, based on the directed graph , and adopt the node classification algorithm of graph neural network to train graph neural network and establish the connection direction classification model MS of edge nodes, including:

[0021] Step 4.1, based on the directed graph , get each node The eigenvector of , the eigenvector is Contains: node category attributes, system attributes, spatial geometric attributes and connectivity, where i represents the node number;

[0022] Step 4.2, based on the directed graph , get each edge The eigenvector of , the eigenvector is Contains the upstream and downstream information of the connection, , Represents a connecting edge Two nodes of , j represents the node number;

[0023] Step 4.3, the directed graph The direction of the corrected edge E in is used as the training label, where the label Represents edge direction, where 1 means from node To Node , the connection direction, 0 means from the node To Node The connection direction;

[0024] Step 4.4, the directed graph By Node Split into non-overlapping training graphs And test chart ; The training graph Further divided into multiple validation folds, each validation fold is further divided into non-overlapping training and validation sets;

[0025] Step 4.5, based on the training set, validation set and test graph , apply different graph machine learning methods to train in the parameter space and evaluate the results. The model with the highest average score will win the training and obtain the winning model; based on the training set, validation set and test graph , the winning model is retrained and evaluated to obtain the connection direction classification model MS of the trained nodes.

[0026] Furthermore, in the above method, in step 5, based on the trained node connection direction classification model MS, the undirected graph to be repaired is identified to obtain the corresponding directed graph ,include:

[0027] Step 5.1: Input the undirected graph to be repaired into the trained node connection direction classification model MS to repair the edge connection direction of the building operation and maintenance mechanism model and automatically construct a directed graph of the building operation and maintenance mechanism model. .

[0028] Further, in the above method, step 6, based on the operation monitoring data and work order data of step 1, in the directed graph Establish fault repair association connections between nodes to obtain the directed graph update as ,include:

[0029] Step 6.1: Use natural language processing technology to perform semantic analysis on the operation monitoring data and work order data, extract keywords and phrases related to equipment failures, and identify specific equipment names, models, locations, etc., and define entities. , using pre-trained NER models for entity recognition ,in, For the Operation monitoring data or work order data, is the set of identified entities;

[0030] Step 6.2: Extract the device corresponding to the entity and apply it to the directed graph The corresponding nodes of ,Will Entities in , matching directed graphs The device node for reporting faults in And report the device node of the fault , by adding fault correlation connections between device nodes , to update the directed graph to ,in, Used for prediction of subsequent device node connections.

[0031] Furthermore, in the above method, step 7, based on the directed graph , training graph database connection prediction model M1, including:

[0032] Step 7.1, based on the directed graph , get the feature vector of each device node ,in, is the type of the device node, The upstream and downstream connection parameters of the device node are based on Device Failure Association Connections in , forming features to add node pairs to connect features ,Right now ; In the directed graph The node in the middle is the device node where the fault is reported Device nodes with faults reported Increase the node pair connection feature , to obtain the first directed graph;

[0033] Step 7.2: Split the first directed graph obtained in 7.1 into non-overlapping training sets With the test set ,in, , ,in, Represents a device node With device node The connection between

[0034] Step 7.3, based on the training set With the test set , apply different graph machine learning methods in the parameter space for training and evaluate the results. The model with the highest average score will win the training to obtain the winning model; based on the training set With the test set , the winning model is retrained and evaluated to obtain the trained fault association prediction model M1.

[0035] Furthermore, in the above method, step 8, based on the directed graph And the fault association prediction model M1, fault diagnosis is performed to obtain a new building operation and maintenance fault prediction mechanism model ; Based on a new building operation and maintenance failure prediction mechanism model , predict equipment warranty work orders in daily building operation and maintenance, and obtain possible equipment failure-related connections, including:

[0036] Step 8.1, for the directed graph in step 5.1 , using the fault association prediction model M1 trained in step 7, predict the directed graph in step 5.1 The fault correlation probability between different device nodes in the , select the connection between the device nodes with the highest fault correlation probability as the new connection and add it to Generate a new node association model , forming a new building operation and maintenance failure prediction mechanism model ;

[0037] Step 8.2: Use a new building failure prediction mechanism model , predict the equipment warranty work order in daily building operation and maintenance, among which, in Figure Breadth-first traversal is adopted to traverse all possible faulty device nodes of the device as the connections associated with the predicted possible device faults.

[0038] According to another aspect of the present invention, there is further provided a computer-readable storage medium having computer-executable instructions stored thereon, wherein when the computer-executable instructions are executed by a processor, the processor is enabled to: adopt any of the methods described above.

[0039] According to another aspect of the present invention, there is also provided a computer device, comprising:

[0040] Processor; and

[0041] A memory arranged to store computer executable instructions which, when executed, cause the processor to: a method as described in any one of the above.

[0042] Compared with the prior art, the present invention includes a directed graph based And the fault association prediction model M1, a new building operation and maintenance fault prediction mechanism model is obtained ; Based on a new building operation and maintenance failure prediction mechanism model Predict equipment warranty work orders in daily building operation and maintenance to obtain possible equipment failure-related connections. The present invention can automatically build and repair the operation model of the building electromechanical system, thereby providing technical support for the intelligent maintenance and fault repair of building facilities and equipment. The present invention automatically identifies and builds the operation logic of the building electromechanical system by training a large-scale graph machine learning model. The present invention can build a topological relationship model between devices to ensure that the model can accurately reflect the actual situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flowchart of a method for identifying equipment faults based on graph machine learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0045] In a typical configuration of the present application, the terminal, the device of the service network and the trusted party all include one or more processors (CPU), input / output interfaces, network interfaces and memories.

[0046] Memory may include non-permanent storage in a computer-readable medium, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0047] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include non-transitory media such as modulated data signals and carrier waves.

[0048] like Figure 1 As shown, the present invention provides a device fault identification method based on graph machine learning, the method comprising:

[0049] Step 1: Collect historical daily operation and maintenance data from the mechanical and electrical operation system of the building, including: operation monitoring data and work order data;

[0050] Here, the operation monitoring data and work order data are dynamic data of the electromechanical operation system;

[0051] Preferably, the operation monitoring data includes: operation monitoring data of connected Internet of Things devices, device sensors, BA systems (Building Automation System) and weak current systems;

[0052] The work order data includes: daily repair work orders and work order feedback data; as input to the prediction model M1;

[0053] Step 2, obtaining component attributes and connection attributes in the BIM electromechanical model of the building;

[0054] Here, the device attributes and upstream and downstream connection relationships are static data of the electromechanical operation system, and steps 2 to 5 are the process of static data processing;

[0055] Step 2.1, extracting the component attributes in the BIM electromechanical model of the building, including: extracting the category attributes of the component, which is used to distinguish the attributes of pipes, pipe fittings, pipe accessories and equipment in the BIM electromechanical model; extracting the electromechanical system attributes of the component, which is used to distinguish the electromechanical system to which the component belongs, and distinguishing the system attributes of the air conditioning system, air conditioning water system and drainage system; extracting the spatial geometric attributes of the component; wherein the BIM electromechanical model of the building includes: pipes, pipe fittings, pipe accessories and equipment;

[0056] Preferably, the Revit secondary development interface can be used to develop a script to read the component information in the RevitBIM electromechanical model, read the Category attribute of non-pipe components, and read the BoundingBox attribute of components to determine the spatial geometric attributes of the components. The Family attribute of the model is used to distinguish equipment components from general components (pipes, pipe fittings, and pipe accessories). The MEPSystem attribute of the pipe connected to the component is read and recorded as the electromechanical system attribute of the component.

[0057] Step 2.2, extracting the connection properties of the components from the BIM electromechanical model of the building, including: upstream and downstream connection relationship properties, connection direction properties and connection quantity properties, wherein the connection relationship of the components in the BIM electromechanical model is a pair of upstream and downstream connection relationships, that is, one connection has only one upstream component and one downstream component; the upstream and downstream connection relationship of the components is divided into two types, components are connected through pipes and components are directly connected, and the connection direction property is temporarily assigned a default connection direction;

[0058] The connection relationship between pipe fittings depends on the pipe connection, and the properties of the pipe are also extracted, including the length and cross-sectional size information of the pipe, which is used to define the flow in the pipe;

[0059] The connection between components is mainly the connection between equipment and pipe fittings, for example, the connection between a VRV air conditioner and a valve. This part of the connection relationship is a common connection relationship and does not have the same geometric properties as pipe connections.

[0060] The connection relationship of the components in the BIM electromechanical model is a pair of upstream and downstream connection relationships, that is, one connection has only one upstream component and one downstream component. The upstream and downstream component information is stored in the attributes of each upstream and downstream connection relationship. Due to the complexity of the BIM electromechanical model, the original BIM electromechanical model does not know the direction of the connection, so the connection direction attribute is temporarily given a default connection direction, and the number of connections for each component (with the number of connections to other structures) is also saved in the corresponding node connection attribute. Later, one component can be mapped to one node.

[0061] Specifically, the secondary development script can be used to filter the Connector connection information of all pipelines in the BIM electromechanical model, and update the connection information at both ends of the Connector to the node connection properties.

[0062] Step 3: Store the components and their component attributes and connection attributes as a directed graph ;

[0063] Step 3.1, creating different nodes according to the category attributes of the components; storing the extracted component attributes in the corresponding nodes in the connection relationship undirected graph, and storing the system attributes and spatial geometric attributes as the attributes of the corresponding nodes;

[0064] Here, different nodes can be created according to the category attributes of components, such as equipment, pipe fittings, and pipeline accessories, and all components can be converted into nodes in an undirected graph of connection relationships, and the spatial geometric attributes and system attributes can be saved as node attributes. According to the category attributes in the component attributes extracted in step 2.1, different category information Labels are assigned to the corresponding nodes to distinguish the categories of nodes in the graph, and the remaining component attributes are stored as node attributes.

[0065] Step 3.2, converting the connection attributes of the components extracted in step 2.2 into edges in a connection relationship undirected graph, wherein, according to the upstream and downstream connection relationship attributes, corresponding nodes in the connection relationship undirected graph are connected to form edges, and the connection direction attribute and the connection quantity attribute are stored as attributes of the edge to form a connection relationship undirected graph containing nodes and edges.

[0066] Here, the pairwise connections between nodes can be obtained according to the connection attributes of the nodes, edges in the graph can be created, and the edges connecting the pipelines can be assigned the attributes of the corresponding pipelines.

[0067] After completing the above two steps 3.1 and 3.2, an undirected graph of the connection relationship of the building operation and maintenance electromechanical model is obtained, and the information of the undirected graph of the connection relationship is stored.

[0068] Step 3.3, develop a visualization interface to visualize the connection direction between components, obtain the correction operation of the connection direction attribute between components input by the user in the visualization interface, obtain the direction of the corrected edge based on the correction operation, and convert the connection relationship undirected graph into a directed graph based on the corrected edge direction. ,in, is a collection of nodes, is the set of edges;

[0069] Develop a web page, visualize the component connection undirected graph according to the component attributes, modify the connection direction between components on the web page, save the direction of the edge into the original undirected graph, and convert the undirected graph into a directed graph;

[0070] Step 4, based on directed graph , and adopt the node classification algorithm of graph neural network to train the graph neural network and establish the connection direction classification model MS of edge nodes;

[0071] Step 4.1, based on the directed graph , get each node The eigenvector of , the eigenvector is Contains: node category attributes, system attributes, spatial geometric attributes and connectivity, where i represents the node number;

[0072] Step 4.2, based on the directed graph , get each edge The eigenvector of , the eigenvector is Contains the upstream and downstream information of the connection, , Represents a connecting edge Two nodes of , j represents the node number;

[0073] Step 4.3, the directed graph The direction of the corrected edge E in is used as the training label, where the label Represents edge direction, where 1 means from node To Node , the connection direction, 0 means from the node To Node connection direction.

[0074] Step 4.4, the directed graph By Node Split into non-overlapping training graphs And test chart In order to better evaluate the stability of the model and avoid overfitting, the training graph It is further divided into multiple validation folds, and each validation fold is further divided into non-overlapping training and validation sets.

[0075] Step 4.5, based on the training set, validation set and test graph , apply different graph machine learning methods to train in the parameter space and evaluate the results. The model with the highest average score will win the training and obtain the winning model; based on the training set, validation set and test graph , retrain and evaluate the winning model to obtain the connection direction classification model MS of the trained nodes;

[0076] Here, you can select the node category, node connectivity, and node connection components as the characteristic attributes of node training, split the undirected graph, select 0.4 as the test coefficient, add logistic regression model, random forest model, and multilayer perceptron model candidates to complete the model training, and retrain on the original graph based on the model with the highest score in each round.

[0077] Step 5: Based on the trained node connection direction classification model MS, the undirected graph to be repaired is identified to obtain the corresponding directed graph. ,include:

[0078] Step 5.1: Input the undirected graph to be repaired into the trained node connection direction classification model MS to repair the edge connection direction of the building operation and maintenance mechanism model and automatically construct a directed graph of the building operation and maintenance mechanism model. , avoiding manual modification of edge connection directions;

[0079] Step 6: Based on the operation monitoring data and work order data in step 1, Establish fault repair association connections between nodes to obtain the directed graph update as ;

[0080] Step 6.1: Use natural language processing technology to perform semantic analysis on the operation monitoring data and work order data, extract keywords and phrases related to equipment failures, and identify specific equipment names, models, locations, etc., and define entities. , using pre-trained NER models for entity recognition ,in, For the Operation monitoring data or work order data, is the set of identified entities;

[0081] For example, the content of the warranty work order is "2401-3 The office air conditioner does not cool". After the maintenance personnel have investigated the problem and repaired it, they send a feedback work order with the content of "The problem has been fixed, the rooftop outdoor unit is faulty". The original entity and target entity in the Chinese text are identified, and the corresponding equipment is queried from the BIM data.

[0082] Step 6.2: Extract the device corresponding to the entity and apply it to the directed graph The corresponding nodes of ,Will Entities in , matching directed graphs The device node for reporting faults in And report the device node of the fault , by adding fault correlation connections between device nodes , to update the directed graph to ,in, Used for prediction of subsequent device node connections.

[0083] Step 7: Based on the directed graph , train the graph database connection prediction model M1;

[0084] Step 7.1, based on the directed graph , get the feature vector of each device node ,in, is the type of the device node, The upstream and downstream connection parameters of the device node are based on Device Failure Association Connections in , forming features to add node pairs to connect features ,Right now ; In the directed graph The node in the middle is the device node where the fault is reported Device nodes with faults reported Increase the node pair connection feature , to obtain the first directed graph;

[0085] Step 7.2: Split the first directed graph obtained in 7.1 into non-overlapping training sets With the test set ,in, , ,in, Represents a device node With device node The connection between

[0086] Step 7.3, based on the training set With the test set , apply different graph machine learning methods in the parameter space for training and evaluate the results. The model with the highest average score will win the training to obtain the winning model; based on the training set With the test set , the winning model is retrained and evaluated to obtain the trained fault association prediction model M1.

[0087] Here, in the directed graph In the example, the connections between devices are filtered according to the connection relationship of the equipment fault work order to form a subgraph, and the upstream and downstream connection components of the equipment are added as feature vectors. The test ratio is set to 0.25 and the training ratio is set to 0.6 to perform graph connection prediction training.

[0088] Step 8, based on directed graph And the fault association prediction model M1, fault diagnosis is performed to obtain a new building operation and maintenance fault prediction mechanism model ; Based on a new building operation and maintenance failure prediction mechanism model , predict equipment maintenance work orders in daily building operation and maintenance, and obtain connections associated with possible equipment failures;

[0089] Step 8.1, for the directed graph in step 5.1 , using the fault association prediction model M1 trained in step 7, predict the directed graph in step 5.1 The fault correlation probability between different device nodes in the , select the connection between the device nodes with the highest fault correlation probability as the new connection and add it to Generate a new node association model , forming a new building operation and maintenance failure prediction mechanism model .

[0090] Step 8.2: Use a new building failure prediction mechanism model , predict the equipment warranty work order in daily building operation and maintenance, among which, in Figure Adopting breadth-first traversal, all possible equipment failure nodes are traversed, and the predicted possible equipment failure-related connections are used for feedback to provide quick decision-making for equipment maintenance.

[0091] Here, an exhaustive search can be used to traverse all possible connections, that is, to check all pairs of nodes that have not yet been connected by a relationship. For each such pair of nodes, a trained model is used to predict whether they should be connected by a link, and an exhaustive search will find all the best connections. The predicted connection is updated into the directed graph, and the next time a maintenance work order for equipment failure is initiated, all possible fault connections can be directly queried from the graph and feedback can be provided.

[0092] Step 9, optimization of the prediction model;

[0093] Step 9.1, in the daily maintenance of the building's electromechanical equipment, collect the work orders and feedback work orders for daily equipment maintenance in step 6, use the methods in steps 6 to 7 to continuously update the fault association prediction model M1, and use new data to continuously optimize the model to improve its recognition accuracy and modeling efficiency.

[0094] The present invention can automatically construct and repair the operation model of the building's electromechanical system, thereby providing technical support for the intelligent maintenance and fault repair of building facilities and equipment. The present invention automatically identifies and constructs the operation logic of the building's electromechanical system by training a large-scale graph machine learning model. The present invention can construct a topological relationship model between devices to ensure that the model can accurately reflect the actual situation.

[0095] According to another aspect of the present invention, there is further provided a computer-readable storage medium having computer-executable instructions stored thereon, wherein when the computer-executable instructions are executed by a processor, the processor is enabled to: adopt any of the methods described above.

[0096] According to another aspect of the present invention, there is also provided a computer device, comprising:

[0097] Processor; and

[0098] A memory arranged to store computer executable instructions which, when executed, cause the processor to: a method as described in any one of the above.

[0099] The detailed contents of the device embodiments of the present invention can be found in the corresponding parts of the method embodiments, which will not be described in detail here.

[0100] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

[0101] It should be noted that the present invention can be implemented in software and / or a combination of software and hardware, for example, it can be implemented using an application-specific integrated circuit (ASIC), a general-purpose computer or any other similar hardware device. In one embodiment, the software program of the present invention can be executed by a processor to implement the steps or functions described above. Similarly, the software program of the present invention (including related data structures) can be stored in a computer-readable recording medium, for example, a RAM memory, a magnetic or optical drive or a floppy disk and the like. In addition, some steps or functions of the present invention can be implemented using hardware, for example, as a circuit that cooperates with a processor to perform various steps or functions.

[0102] In addition, a part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. The program instruction for calling the method of the present invention may be stored in a fixed or removable recording medium, and / or transmitted through a data stream in a broadcast or other signal-bearing medium, and / or stored in a working memory of a computer device that runs according to the program instruction. Here, according to an embodiment of the present invention, a device is included, the device including a memory for storing computer program instructions and a processor for executing program instructions, wherein, when the computer program instruction is executed by the processor, the device is triggered to run the method and / or technical solution based on the aforementioned multiple embodiments of the present invention.

[0103] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-restrictive, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any figure mark in the claims should not be regarded as limiting the claims involved. In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the device claim can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any particular order.

Claims

1. A method for identifying equipment faults based on graph machine learning, characterized in that: include: Step 1: Collect historical daily operation and maintenance data from the mechanical and electrical operation system of the building, including: operation monitoring data and work order data; Step 2, obtaining component attributes and connection attributes in the BIM electromechanical model of the building; Step 3, storing the components and their component attributes and connection attributes as a directed graph G; Step 4: Based on the directed graph G, the graph neural network is trained using the node classification algorithm of the graph neural network to establish a connection direction classification model MS of the edge nodes; Step 5: Based on the trained node connection direction classification model MS, the undirected graph to be repaired is identified to obtain the corresponding directed graph G′; Step 6: According to the operation monitoring data and work order data of step 1, a fault maintenance association connection between nodes is established in the directed graph G′ to obtain an updated directed graph G″; Step 7: Based on the directed graph G″, training a fault association prediction model M1; Step 8: Based on the directed graph G′ and the fault association prediction model M1, a new building operation and maintenance fault prediction mechanism model G″′ is obtained; based on the new building operation and maintenance fault prediction mechanism model G″′, equipment warranty work orders in daily building operation and maintenance are predicted to obtain possible equipment fault-associated connections; Step 7, based on the directed graph G″, training the fault association prediction model M1, including: Step 7.1, based on the directed graph G″, obtain the feature vector x′ of each device node i =[C, L], where C is the type of the device node, L is the upstream and downstream connection parameters of the device node, and the device failure association connection e" in G" ij , forming a feature adding node pair connecting feature HADAMARD, that is, f ij =[C i ⊙C i , L i ⊙L j ]; in the directed graph G″, it is the device node v that reports the fault i Device node v with fault report j Adding the node pair connection feature HADAMARD to obtain a first directed graph; Step 7.2: Split the first directed graph obtained in 7.1 into non-overlapping training sets D train With the test set D test , where D train ={(x′ i , x′ j , f ij ,y ij )|(v i , v j )},D test ={(x′ i , x′ j , f ij ,y ij )|(v i , v j )}, where y ij Represents the device node v i With device node v j The connection between Step 7.3, based on the training set D train With the test set D test , apply different graph machine learning methods in the parameter space for training and evaluate the results. The model with the highest average score will win the training to obtain the winning model; based on the training set D train With the test set D test , retrain and evaluate the winning model to obtain the trained fault association prediction model M1; Step 8: Based on the directed graph G′ and the fault association prediction model M1, a fault diagnosis is performed to obtain a new building operation and maintenance fault prediction mechanism model G″′; based on the new building operation and maintenance fault prediction mechanism model G″′, equipment warranty work orders in daily building operation and maintenance are predicted to obtain possible equipment fault-associated connections, including: Step 8.1, for the directed graph G′ in step 5.1, use the fault association prediction model M1 trained in step 7 to predict the fault association probability between different device nodes in the directed graph G′ in step 5.1, select the connection between the device nodes with the highest fault association probability as a new connection, add it to the directed graph G′, and form a new building operation and maintenance fault prediction mechanism model G″′; Step 8.2, using the new building failure prediction mechanism model G″′, predict the equipment warranty work orders in daily building operation and maintenance, wherein the breadth-first traversal is used in the building failure prediction mechanism model G″′ to traverse the failure device nodes of all possible equipment as the connection associated with the predicted possible equipment failure.

2. The device fault identification method based on graph machine learning according to claim 1, characterized in that: Step 2: Obtain the component properties and connection properties in the BIM electromechanical model of the building, including: Step 2.1, extracting the component attributes in the BIM electromechanical model of the building, including: extracting the category attributes of the component, which is used to distinguish the attributes of pipes, pipe fittings, pipe accessories and equipment in the BIM electromechanical model; extracting the electromechanical system attributes of the component, which is used to distinguish the electromechanical system to which the component belongs, and distinguishing the system attributes of the air conditioning system, air conditioning water system and drainage system; extracting the spatial geometric attributes of the component; wherein the BIM electromechanical model of the building includes: pipes, pipe fittings, pipe accessories and equipment; Step 2.2, extracting the connection properties of the components from the BIM electromechanical model of the building, including: upstream and downstream connection relationship properties, connection direction properties and connection quantity properties, wherein the connection relationship of the components in the BIM electromechanical model is a pair of upstream and downstream connection relationships, that is, one connection has only one upstream component and one downstream component; there are two types of upstream and downstream connection relationships of the components, components are connected through pipes and components are directly connected, and the connection direction property is temporarily assigned a default connection direction.

3. The device fault identification method based on graph machine learning according to claim 2, characterized in that: Step 3: Store the components and their component attributes and connection attributes as a directed graph G, including: Step 3.1, creating different nodes according to the category attributes of the components; storing the extracted component attributes in the corresponding nodes in the connection relationship undirected graph, and storing the system attributes and spatial geometric attributes as the attributes of the corresponding nodes; Step 3.2, converting the connection attributes of the components into edges in a connection relationship undirected graph, wherein, according to the upstream and downstream connection relationship attributes, corresponding nodes in the connection relationship undirected graph are connected to form edges, and the connection direction attribute and the connection quantity attribute are stored as attributes of the edge to form a connection relationship undirected graph containing nodes and edges; Step 3.3, develop a visualization interface to visually display the connection directions between components, obtain the modification operation on the connection direction attributes between components input by the user in the visualization interface, obtain the modified edge direction based on the modification operation, and convert the connection relationship undirected graph into a directed graph G = (V, E) based on the modified edge direction, where V is a set of nodes and E is a set of edges.

4. The device fault identification method based on graph machine learning according to claim 3, characterized in that: Step 4: Based on the directed graph G=(V, E), the graph neural network is trained using the node classification algorithm of the graph neural network to establish a connection direction classification model MS of the edge nodes, including: Step 4.1, based on the directed graph G = (V, E), obtain each node v i The eigenvector of ∈V is x i , the eigenvector is x i Contains: node category attributes, system attributes, spatial geometric attributes and connectivity, where i represents the node number; Step 4.2, based on the directed graph G = (V, E), obtain each edge e ij =(v i , v j The eigenvector of )∈E is e ij , the eigenvector is e ij Contains the upstream and downstream information of the connection, v i 、v j Represents the connecting edge e ij Two nodes of , j represents the node number; Step 4.3, the direction of the corrected edge E in the directed graph G = (V, E) is used as a training label, where the label y ij ∈{0, 1} represents edge e ij direction, where 1 represents the direction from node v i To node v j , the connection direction, 0 means from node v j To node v i Connection direction; Step 4.4: Calculate the directed graph G = (V, E) by node v i Splitting into non-overlapping training graphs Tr and test graphs Te; further dividing the training graph Tr into multiple verification folds, and each verification fold is further divided into non-overlapping training sets and verification sets; Step 4.5, based on the training set, validation set and test graph Te, apply different graph machine learning methods in the parameter space for training, and evaluate the results, where the model with the highest average score will win the training and obtain the winning model; based on the training set, validation set and test graph Te, retrain and evaluate the winning model to obtain the connection direction classification model MS of the trained nodes.

5. The device fault identification method based on graph machine learning according to claim 1, characterized in that: Step 5, based on the trained node connection direction classification model MS, the undirected graph to be repaired is identified to obtain the corresponding directed graph G′=(V, E′), including: Step 5.1, input the undirected graph to be repaired into the trained node connection direction classification model MS, which is used to repair the edge connection direction of the building operation and maintenance mechanism model, and automatically construct a directed graph G′(V, E′) of the building operation and maintenance mechanism model.

6. The method for identifying equipment faults based on graph machine learning according to claim 5, characterized in that: Step 6, based on the operation monitoring data and work order data of step 1, establish a fault maintenance association connection between nodes in the directed graph G′=(V, E′) to obtain a directed graph update G″=(V, E″), including: Step 6.1: Use natural language processing technology to perform semantic analysis on the operation monitoring data and work order data, extract keywords and phrases related to equipment failure, and identify specific equipment names, models, and location information, define entity E = {equipment name, equipment model, equipment location}, and use the pre-trained NER model to perform entity recognition E i =NER(T i ), where T i is the operation monitoring data or work order data of the ith item, E i is the set of identified entities; Step 6.2: associate the device corresponding to the extracted entity with the corresponding node of the directed graph G′. For the directed graph G′=(V, E′), E i The entity in the directed graph G′ matches the device node v that reports the fault. i And report the faulty device node v j , by adding fault correlation connections between device nodes e″ ij , to update the directed graph to G″=(V,E″), where E″=E′∪|e″ ij |e″ ij It is a fault-correlated connection} used for prediction of subsequent device node connections.

7. A computer-readable storage medium having computer-executable instructions stored thereon, wherein: When the computer executable instructions are executed by a processor, the processor is caused to: adopt the method according to any one of claims 1 to 6.

8. A computer device, wherein: include: processor; as well as A memory arranged to store computer executable instructions which, when executed, cause the processor to: a method as claimed in any one of claims 1 to 6.

Images