A topology enhanced multi-agent collaborative operation method and system for a smart building

Abstract

The application discloses a topology-enhanced multi-agent collaborative operation and maintenance method and system for a smart building: a smart building alarm task package is received, and a standard operation and maintenance event containing asset information and alarm information is constructed; two types of descriptors are generated based on the event, a topology query descriptor is traversed in a building logical topology graph according to a corresponding relationship, and a knowledge query descriptor searches relevant rules, work orders and manuals in an operation and maintenance knowledge base; the two types of evidence are unified into a set of traceable evidence segments, the three types of intelligent agents, i.e., monitoring, topology reasoning and dispatching, process the set of traceable evidence segments in sequence, and parameters and handover tokens are transmitted according to an agreement; finally, an operation action graph containing root cause investigation, risk prevention, dispatching notification and the like is generated, and a closed-loop operation is performed after verification. According to the scheme, in an existing building lacking a complete BIM model, multi-agent reasoning is constrained by a logical topology and a knowledge base, and the illusion of a large language model and the risk of cross-system misdispatching are reduced.

Description

Technical Field

[0001] This invention relates to the fields of smart building operation and maintenance, knowledge graph, enhanced retrieval generation, multi-agent collaboration, and property work order linkage, and particularly to a topology-enhanced multi-agent collaborative operation and maintenance method and system for smart buildings. Background Technology

[0002] Large commercial complexes, office buildings, and industrial parks typically deploy multiple systems, including building automation, power supply, power distribution, water supply and drainage, environmental control, access control, video security, and property management office automation (OA). These systems generate a large amount of location data and alarm records, but cross-system fault handling still heavily relies on human experience.

[0003] Existing smart building operation and maintenance solutions typically employ fixed rules or single models for alarm classification, fault diagnosis, or work order dispatch. While such solutions can handle single-point alarms, they often lack comprehensive reasoning regarding building spatial relationships, equipment service relationships, and tenant occupancy relationships in cross-space and cross-system scenarios such as water immersion affecting lower-level electrical equipment, smoke detection impacting tenant safety, and HVAC failures affecting multiple tenant areas.

[0004] Existing solutions for knowledge graph-based fault diagnosis, digital twin alarm processing, and RAG knowledge retrieval typically address knowledge organization, visualization modeling, or text retrieval issues separately. They fail to establish a unified mechanism for the exchange of topological and textual evidence for building operations and maintenance, and lack a technical solution for verifying the consistency of evidence outputs from multiple agents before creating property management OA work orders or generating control recommendations. Therefore, a method is needed to enable collaborative operations and maintenance of multiple agents in existing buildings lacking complete BIM models, based on logical topology and building knowledge base constraints. Summary of the Invention

[0005] According to embodiments of the present invention, a topology-enhanced multi-agent collaborative operation and maintenance method and system for smart buildings is provided, comprising the following steps: Receive smart building alarm task packages and convert them into standard operation and maintenance events that include asset identifier, space unit, subsystem type, alarm type, alarm level, sampling time and triggering evidence; Generate topology query descriptors and knowledge query descriptors based on standard operation and maintenance events; Input the topology query descriptor into the building logical topology map to obtain a set of topology evidence fragments; Input the knowledge query descriptor into the building operation and maintenance knowledge base to obtain a set of knowledge evidence fragments; Evidence normalization is performed on the topological evidence fragment set and the knowledge evidence fragment set to generate a traceable evidence fragment set; The monitoring agent, topology reasoning agent, and dispatch agent process the evidence in sequence according to the evidence handover protocol. An operation and maintenance action map is generated based on the output of each intelligent agent; Perform evidence consistency verification on the operation and maintenance action map. If the action node lacks evidence reference, the handover token is mismatched, or the confidence level is lower than the threshold, block the automatic creation of property OA work orders and the automatic generation of equipment control instructions. When the evidence consistency verification passes, a property OA work order is created according to the operation and maintenance action plan, a notification is sent and / or a building equipment control suggestion that requires manual confirmation is generated, and the execution result is written back to the closed-loop feedback record.

[0006] This invention utilizes a dual-channel evidence retrieval system combining a building logical topology map and a building operations and maintenance knowledge base. This transforms the logical relationships between spaces, equipment, tenants, and work teams into traceable evidence fragments. The system then constrains the sequential output of monitoring agents, topology reasoning agents, and dispatch agents through input summaries, evidence citations, and handover tokens. Because operations and maintenance action maps require evidence consistency verification before execution, the system can prevent the creation of property management OA work orders and the invocation of building automation system interfaces lacking on-site evidence. This reduces the risk of mis-dispatching and mis-control in cross-space and cross-system fault scenarios.

[0007] It should be understood that both the preceding description and the following detailed description are exemplary and intended to provide further illustration of the claimed technology. Attached Figure Description

[0008] Figure 1 This is a flowchart of a topology-enhanced multi-agent collaborative operation and maintenance method for smart buildings according to an embodiment of the present invention; Figure 2 This is a system architecture diagram of a topology-enhanced multi-agent collaborative operation and maintenance system for smart buildings according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the evidence transfer protocol and action diagram state machine according to an embodiment of the present invention. Detailed Implementation

[0009] This invention is applicable to smart building operation and maintenance scenarios such as existing commercial complexes, office buildings, parks and data centers, and is especially suitable for existing buildings that have been connected to multi-source alarms, property OA and equipment ledgers, but lack a complete 3D BIM model.

[0010] In step S1, the system receives a smart building alarm task package. This package can originate from the alarm center, equipment and facility management module, energy management module, or manual input module. The system converts it into a standard maintenance event, which includes at least the asset identifier, space unit, subsystem type, alarm type, alarm level, sampling time, and triggering evidence.

[0011] In step S2, the system generates a topology query descriptor based on standard operation and maintenance events. The topology query descriptor constrains the traversal range of the building's logical topology graph, and includes the starting spatial unit, starting asset identifier, alarm type, topology edge type set, traversal direction, and traversal depth. For different alarm types, the system selects different topology edge type sets.

[0012] In step S3, the system generates a knowledge query descriptor based on standard operation and maintenance events. The knowledge query descriptor includes alarm type, device type, space usage, alarm level, and handling intent. The handling intent may include root cause investigation, emergency response, secondary risk prevention, work order dispatch, or manual confirmation of control recommendations.

[0013] In step S4, the system performs a dual-channel recall. On one hand, the system inputs the topology query descriptor into the building logical topology map to obtain a set of topology evidence fragments. On the other hand, the system inputs the knowledge query descriptor into the building operation and maintenance knowledge base to obtain a set of knowledge evidence fragments.

[0014] The building logical topology map includes spatial nodes, equipment nodes, tenant nodes, system nodes, and work group nodes. Spatial nodes can be constructed according to buildings, floors, rooms, equipment areas, and tenant areas; equipment nodes can be constructed according to subsystems such as power, low-voltage, HVAC, water supply and drainage, environmental, and security. Edge types include vertical below, horizontal adjacency, power supply to, water supply to, drainage to, control to, service to, and occupied. This map does not require a complete 3D BIM model, but rather constructs a logical digital twin through building spatial relationships, electromechanical system relationships, and operational service relationships.

[0015] When the alarm type is water immersion, the system selects the vertical down, lateral adjacency, power to, and occupied edge types to identify important tenants, electrical equipment, and adjacent equipment areas on the lower level; when the alarm type is smoke or fire, the system selects the lateral adjacency, service to, and occupied edge types to identify adjacent areas and tenant notification targets; when the alarm type is HVAC anomaly, the system selects the control to, service to, and lateral adjacency edge types to identify affected air-conditioned areas and related equipment.

[0016] The building operations and maintenance knowledge base includes alarm classification rules, emergency plans, equipment manuals, historical work orders, and reporting rules. The knowledge evidence fragment set can be obtained through a two-stage retrieval process: the first stage recalls candidate knowledge sources based on subsystem type and alarm level; the second stage performs vector rearrangement of the candidate knowledge sources based on equipment type, space usage, and disposal intent, while retaining the knowledge source version and original citation identifier for each knowledge evidence fragment.

[0017] In step S5, the system performs evidence normalization on the topological evidence fragment set and the knowledge evidence fragment set to generate a traceable evidence fragment set. Each traceable evidence fragment includes the evidence source, evidence type, evidence content summary, version identifier, and citation identifier. Through this structure, the conclusions of subsequent agents can be traced back to the topological path or knowledge fragment.

[0018] In step S6, the monitoring agent determines whether an alarm needs to enter the topology reasoning stage based on the triggering evidence and knowledge evidence fragments. If it is determined to be a sensor anomaly, a maintenance expected alarm, or a reported alarm, the monitoring agent outputs an action node that blocks topology reasoning and the corresponding evidence reference; if it is determined to be a real alarm, it outputs a handover token to enter the topology reasoning stage.

[0019] In step S7, the topology reasoning agent outputs an impact domain list and a secondary risk list based on the topology evidence fragment set. The impact domain list includes affected spaces, affected devices, and affected tenants. The secondary risk list includes risk type, risk level, propagation path, and preventive actions. For example, a water flooding alarm in a 14-story server room can identify a critical tenant's server room on the 13th floor through the vertical lower edge type and generate a secondary risk preventive action that is simultaneously notified to the tenant manager.

[0020] In step S8, the dispatching agent generates a dispatch plan based on the action nodes, team nodes, and knowledge evidence fragments in the operation and maintenance action diagram. The dispatch plan includes the responsible team, collaborating teams, notification recipients, work order level, work order summary, and processing suggestions. The same operation and maintenance action diagram can simultaneously generate property OA work orders and tenant early warning notifications.

[0021] The monitoring agent, topology reasoning agent, and dispatch agent exchange data according to an evidence handover protocol. This protocol requires each agent to output its agent ID, input digest, evidence reference, confidence level, action node, and handover token. If the input digest of a subsequent agent differs from the output digest of a previous agent, or if the action node lacks an evidence reference, the system will block automatic execution.

[0022] In step S9, the system generates an operation and maintenance action map based on the output of each agent. The operation and maintenance action map includes root cause investigation actions, secondary risk prevention actions, notification actions, work order actions, and manually confirmed control suggestions. The manually confirmed control suggestions include fields for suggested control objects, suggested control actions, expected impact, risk warnings, evidence citations, and confirmers. The system will not invoke the building automation system to execute the suggested control actions before obtaining confirmation from the manually confirmer.

[0023] The system performs evidence consistency checks on the operation and maintenance action map. Evidence consistency checks include: verifying whether each action node references at least one topological evidence fragment or knowledge evidence fragment; verifying whether the input summary of the dispatch agent matches the output summary of the topological reasoning agent; verifying whether the work group for the work order action matches the work group node's qualifications; and verifying whether the control suggestion is marked for manual confirmation. When the checks fail, the system blocks the automatic creation of property management OA work orders and the automatic generation of equipment control instructions.

[0024] In step S10, when the evidence consistency verification passes, the system creates a property OA work order, sends a notification, and / or generates building equipment control suggestions requiring manual confirmation based on the operation and maintenance action plan, and writes the execution result back to the closed-loop feedback record. The closed-loop feedback record includes the property OA work order number, notification status, manual confirmation result, actual handling team, actual handling result, closure time, and review tag. The system updates the historical work order index and the hit weight of knowledge evidence fragments based on the closed-loop feedback record.

[0025] As one example, the system receives a low-temperature alarm from the cooling tower water tray in Building T1, with a current value of -50.1℃. The monitoring agent, based on knowledge evidence fragments, determines that this value does not conform to the physical laws of April in Beijing, outputs an abnormal sensor action node, and blocks the topology reasoning agent from further propagating analysis. The dispatch agent then generates a work order for inspecting the low-voltage group sensors.

[0026] In another embodiment, the system receives a water flooding alarm from the 14th-floor server rack area. The topology reasoning agent identifies the important tenant's server room on the 13th floor based on the type of the vertically lower edge, and generates root cause investigation actions, a power supply emergency handling work order, and a tenant management alert notification based on emergency plan knowledge fragments. After the evidence consistency verification passes, the system creates a property management OA emergency work order and sends an alert notification to the tenant.

[0027] In another embodiment, the system receives a smoke alarm from parking lot B3. The monitoring agent searches the reporting rules and historical work orders, and finds no hot work or testing reports; the topology reasoning agent identifies adjacent parking spaces and surveillance cameras; the dispatching agent generates a level-one alarm work order based on the security team's qualifications, and simultaneously generates a large-screen pop-up notification.

[0028] Above, refer to Figures 1-3 This invention describes a topology-enhanced multi-agent collaborative operation and maintenance method and system for smart buildings, based on embodiments of the present invention. By utilizing a building logical topology map, a building operation and maintenance knowledge base, an evidence handover protocol, an operation and maintenance action diagram, and evidence consistency verification, this invention solves the problems in existing smart building operation and maintenance, such as the difficulty in reasoning about cross-system fault impact relationships, the lack of evidentiary constraints on multi-agent conclusions, and the tendency for work orders and control recommendations to be detached from on-site evidence.

[0029] It should be noted that, in this specification, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0030] Although the present invention has been described in detail through the preferred embodiments above, it should be understood that the above description should not be considered as a limitation of the present invention. Various modifications and substitutions to the present invention will be apparent to those skilled in the art after reading the above description. Therefore, the scope of protection of the present invention should be defined by the appended claims.

Claims

1. A topology-enhanced multi-agent collaborative operation and maintenance method and system for smart buildings, characterized in that, Includes the following steps: S1. Receive smart building alarm task packages and convert them into standard operation and maintenance events; S2. Generate a topology query descriptor and a knowledge query descriptor based on the standard operation and maintenance events; S3. Obtain a set of topological evidence fragments based on the building logical topology map and the topology query descriptor, and obtain a set of knowledge evidence fragments based on the building operation and maintenance knowledge base and the knowledge query descriptor; S4. Normalize the topological evidence fragment set and the knowledge evidence fragment set to generate a traceable evidence fragment set; S5. The monitoring agent, topology reasoning agent, and dispatch agent are scheduled to process the standard operation and maintenance event in sequence according to the evidence handover protocol. The subsequent agent generates the output based on the traceable evidence fragment set and the input summary, evidence reference, and handover token output by the previous agent. S6. Generate an operation and maintenance action diagram based on the output of each intelligent agent. The operation and maintenance action diagram includes at least three action nodes among root cause investigation action, secondary risk prevention action, notification action, work order action, and manual confirmation control suggestion. S7. Perform evidence consistency verification on the operation and maintenance action diagram, and block the automatic creation of property OA work orders and building automation system interface calls when the action node lacks evidence references, the input digest does not match, or the handover token is mismatched. S8. When the evidence consistency verification is passed, create a property OA work order, send a notification and / or generate building equipment control suggestions that require manual confirmation according to the operation and maintenance action plan, and write the execution result back to the closed-loop feedback record.

2. The method as described in claim 1, characterized in that, The standard operation and maintenance event includes at least asset identifier, spatial unit, subsystem type, alarm type, alarm level, sampling time, and triggering evidence; the topology query descriptor includes starting spatial unit, starting asset identifier, alarm type, topology edge type set, traversal direction, and traversal depth; the knowledge query descriptor includes alarm type, device type, spatial purpose, alarm level, and handling intent.

3. The method as described in claim 1, characterized in that, The building logical topology map does not rely on a complete 3D BIM model. The building logical topology map includes spatial nodes, equipment nodes, tenant nodes, system nodes, and work group nodes, as well as at least four edge types: vertically below, horizontally adjacent, power supply to, water supply to, drainage to, control to, serving, and occupying.

4. The method as described in claim 1, characterized in that, The topology query descriptor selects the set of topology edge types based on the alarm type: when the alarm type is water immersion, it selects vertically below, horizontally adjacent, power to, and occupied edge types; when the alarm type is smoke or fire, it selects horizontally adjacent, service to, and occupied edge types; when the alarm type is HVAC anomaly, it selects control to, service to, and horizontally adjacent edge types.

5. The method as described in claim 1, characterized in that, The set of topological evidence fragments includes affected spaces, affected assets, affected tenants, secondary risk types, propagation paths, and path weights. The path weights are calculated based at least on edge type weights, spatial distances, hierarchical relationships, and device importance levels.

6. The method as described in claim 1, characterized in that, The knowledge evidence fragment set is obtained through a two-stage retrieval process: the first stage recalls candidate knowledge sources based on subsystem type and alarm level; the second stage performs vector rearrangement of candidate knowledge sources based on device type, space usage, and disposal intent, while retaining the knowledge source version and original text citation identifier for each knowledge evidence fragment.

7. The method as described in claim 1, characterized in that, The monitoring agent is used to determine whether an alarm needs to enter the topology reasoning stage based on trigger evidence and knowledge evidence fragments; if it is determined to be a sensor abnormality, maintenance expected alarm, or reporting hit alarm, it outputs the action node, evidence reference, and handover token that block topology reasoning.

8. The method as described in claim 1, characterized in that, The topology reasoning agent is used to output an impact domain list and a secondary risk list based on a set of topology evidence fragments. The impact domain list includes at least the affected space, affected equipment, and affected tenants, and the secondary risk list includes at least the risk type, risk level, propagation path, and preventive action.

9. The method as described in claim 1, characterized in that, The dispatching agent generates a dispatching plan based on the action nodes, team nodes, and knowledge evidence fragments in the operation and maintenance action diagram. The dispatching plan includes the responsible team, collaborating teams, notification recipients, work order level, work order summary, and processing suggestions. Moreover, the same operation and maintenance action diagram can simultaneously generate property OA work orders and tenant early warning notifications.

10. The method as described in claim 1, characterized in that, The manual confirmation control suggestion includes fields for suggested control object, suggested control action, expected impact, risk warning, evidence reference, and confirmer; before obtaining a manual confirmation signal, the system blocks the building automation system interface call corresponding to the suggested control action.

11. The method as described in claim 1, characterized in that, The evidence consistency verification includes: verifying whether each action node references at least one topological evidence fragment or knowledge evidence fragment; verifying whether the input summary of the dispatch agent is consistent with the output summary of the topological reasoning agent; verifying whether the work group of the work order action matches the qualifications of the work group node; and verifying whether the control suggestion is marked as manual confirmation.

12. The method as described in claim 1, characterized in that, The closed-loop feedback record includes the property OA work order number, notification status, manual confirmation result, actual handling team, actual handling result, closure time, and review tag; the system updates the historical work order index and the hit weight of knowledge evidence fragments based on the closed-loop feedback record.

13. A topology-enhanced multi-agent collaborative operation and maintenance system for smart buildings, characterized in that, include: The event standardization module is used to convert smart building alarm task packages into standard operation and maintenance events; The query descriptor generation module is used to generate topology query descriptors and knowledge query descriptors; The topological evidence acquisition module is used to output a set of topological evidence fragments based on the building's logical topology map; The knowledge evidence acquisition module is used to output a set of knowledge evidence fragments based on the building operation and maintenance knowledge base; The evidence normalization module is used to generate a set of traceable evidence fragments; The intelligent agent collaborative orchestration module is used to schedule monitoring intelligent agents, topology reasoning intelligent agents, and order dispatching intelligent agents according to the evidence handover protocol; The Operation and Maintenance Action Map Generation Module is used to generate operation and maintenance action maps; The evidence consistency verification module is used to block automatic execution when an action node lacks evidence references, the input digest does not match, or the handover token is mismatched. The closed-loop feedback module is used to write back the status of property OA work orders, notification status, manual confirmation results, and review tags.

14. An electronic device comprising a processor and a memory, wherein the memory stores a computer program, characterized in that, When the computer program is executed by the processor, it implements the method according to any one of claims 1 to 12.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method according to any one of claims 1 to 12.

Images