Park-level operation method and system based on multi-agent collaboration and medium

Abstract

The invention discloses a park-level operation method and system based on multi-agent cooperation and a medium, and relates to the technical field of smart parks. The method comprises the following steps: acquiring data information of at least one agent from a digital twin model; constructing a spatial function diagram based on the data information of the at least one agent; in response to triggering the first event, determining a first agent associated with the first event; determining a second agent associated with the first agent based on the spatial function diagram; and determining a target execution strategy based on the scene information of the first event, and performing cooperative control on the first agent and the second agent based on the target execution strategy. Therefore, deep collaboration with the digital twinborn model can be realized, and park decision control is optimized, so that redundant communication traffic is reduced, and the accuracy of park decision is improved.

Description

Technical Field

[0001] This application relates to the field of smart park technology, and in particular to a park-level operation method, system and medium based on multi-agent collaboration. Background Technology

[0002] Existing smart park management systems generally adopt a centralized architecture, which poses a risk of single point of failure. Furthermore, device communication lacks spatial semantic awareness, meaning that communication objects are not filtered based on spatial functional associations, leading to an increase in redundant communication. In addition, the collaborative decision-making of existing smart park management systems relies solely on the status data of a single device, without integrating historical operation and maintenance association rules and dynamic adjustment mechanisms for multi-scenario weights, resulting in low decision-making accuracy. Summary of the Invention

[0003] This application provides a campus-level operation method, system, and medium based on multi-agent collaboration to solve the following technical problems: excessive redundant communication and low decision-making accuracy.

[0004] In a first aspect, embodiments of this application provide a campus-level operation method based on multi-agent collaboration, the method comprising: Obtain data information of at least one of the intelligent agents from the digital twin model; Based on the data information of at least one of the intelligent agents, a spatial function graph is constructed; In response to triggering a first event, determine a first agent associated with the first event; Based on the spatial function map, a second intelligent agent associated with the first intelligent agent is determined; Based on the scenario information of the first event, a target execution strategy is determined, and based on the target execution strategy, the first intelligent agent and the second intelligent agent are coordinated and controlled.

[0005] Secondly, embodiments of this application also provide a campus-level operating system based on multi-agent collaboration, the system comprising: An acquisition module is used to acquire data information of at least one of the intelligent agents from the digital twin model; The construction module is used to construct a spatial function graph based on data information from at least one of the intelligent agents; A determination module is configured to determine a first agent associated with the first event in response to the triggering of the first event; The determining module is further configured to determine a second intelligent agent associated with the first intelligent agent based on the spatial function map; The control module is used to determine the target execution strategy based on the scenario information of the first event, and to perform collaborative control of the first intelligent agent and the second intelligent agent based on the target execution strategy.

[0006] Thirdly, embodiments of this application also provide a campus-level operating device based on multi-agent collaboration, the device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method as described above.

[0007] Fourthly, embodiments of this application also provide a computer storage medium storing computer-executable instructions, which, when executed, implement the methods described in any of the above claims.

[0008] The campus-level operation method, system, and medium based on multi-agent collaboration provided in this application have the following beneficial effects: First, by acquiring data from at least one agent in the digital twin model, real-time state and attribute data of the agent can be obtained, avoiding redundant data collection and improving the efficiency and accuracy of data acquisition, thus achieving deep integration of the digital twin model and the control system. Next, based on the data from at least one agent, a spatial function map is constructed, providing a more intuitive understanding of the dependencies between agents. Then, in response to a triggered first event, the first agent associated with the first event is identified, enabling rapid location of agents related to the event and improving response speed and processing efficiency. Following this, based on the spatial function map, the second agent associated with the first agent is identified, allowing the identification of other agents collaborating with the first agent, improving the efficiency and accuracy of collaborative decision-making. Finally, based on the scenario information of the first event, a target execution strategy is determined, and based on this strategy, the first and second agents are collaboratively controlled. This allows for the formulation of reasonable execution strategies according to the specific circumstances and objectives of the event, achieving collaborative control among multiple agents and effectively enhancing the level of intelligence in park management. Attached Figure Description

[0009] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart of a campus-level operation method based on multi-agent collaboration is provided for embodiments of this application; Figure 2 This is a schematic diagram of the structure of a campus-level operating system based on multi-agent collaboration provided in an embodiment of this application; Figure 3 This is a schematic diagram of a campus-level operating device based on multi-agent collaboration, provided as an embodiment of this application. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0011] The technical solutions proposed in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0012] Figure 1 This document presents a flowchart illustrating a multi-agent collaborative park-level operation method as described in one or more embodiments. This method can be applied to different types of smart park management, such as the management of related equipment and systems within a park, and the collaborative control between different parks. Certain input parameters or intermediate results in the process can be manually adjusted to help improve accuracy.

[0013] This application provides a campus-level operation method based on multi-agent collaboration. It should be noted that the execution entity in these embodiments can be a server or any terminal device with data processing capabilities. For example, the server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. The terminal device can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, in-vehicle terminal, etc., but is not limited to these.

[0014] like Figure 1 As shown in the figure, the campus-level operation method based on multi-agent collaboration provided in this application embodiment specifically includes the following steps: Step 101: Obtain data information of at least one agent from the digital twin model.

[0015] It's important to note that a digital twin model is a virtual copy of a real-world entity or system. It creates a highly similar digital model through simulation, analysis, and prediction. A digital twin model typically includes the entity's physical characteristics, behaviors, performance, and rules, and can be updated in real time to reflect the entity's or system's current state and changes. Intelligent agents can be categorized into equipment (e.g., elevator equipment, photovoltaic equipment), systems (e.g., work order systems, carbon accounting systems), and collaborative systems (e.g., regional collaboration, responsible for cross-domain collaboration across multiple parks). Each intelligent agent corresponds to an entity in the digital twin model.

[0016] As an example, firstly, it is necessary to clarify the type of intelligent agent to which the data is to be acquired, such as elevators, photovoltaic panels, work order systems, etc. Here, the digital twin entity of the target intelligent agent can be located through the identifier or other attributes in the digital twin model. Subsequently, relevant attributes of the target intelligent agent are extracted from the digital twin entity, such as spatial unit coordinates, device ID, perception range (i.e., the range within which the device can perceive information about the surrounding environment), k-hop threshold (i.e., the maximum number of hops that the device can communicate with other intelligent agents in the spatial function graph (SFG), state parameters (i.e., the current state information of the device, such as temperature, humidity, operating status, etc.) and other attributes (such as carbon emission coefficient, energy consumption benchmark value, etc.) in order to construct the spatial function graph based on the data information of the intelligent agent.

[0017] Step 102: Construct a spatial function graph based on data information from at least one agent.

[0018] It should be noted that a Spatial Function Graph (SFG) is a graph structure model used to describe the spatial relationships and functional associations between devices in a smart park. SFG nodes represent devices or resources within the park, such as elevators, photovoltaic panels, and work order systems. SFG edges represent the spatial relationships and functional associations between devices, such as spatial distance and functional coupling coefficients. SFG node attributes include the device's spatial coordinates, ID, and status parameters; SFG edge attributes include spatial distance weights, functional coupling coefficients, and communication latency thresholds.

[0019] In some embodiments, the data information of the intelligent agent includes the agent's entity attributes, newly added energy and carbon-specific attributes, spatial distance weights, functional coupling coefficients, and communication latency thresholds. Step 102 can be implemented as follows: the agent's entity attributes and newly added energy and carbon-specific attributes are used as node attributes of the spatial functional graph; the agent's spatial distance weights, functional coupling coefficients, and communication latency thresholds are used as edge attributes of the spatial functional graph. This clearly describes the spatial relationships between devices, thereby quickly locating faulty devices and improving fault diagnosis efficiency.

[0020] Here, the entity attributes of the intelligent agent include information such as spatial unit coordinates, device ID, and state parameters.

[0021] As an example, assuming a smart park scenario, the spatial unit coordinates (x, y, z), device ID (Elevator-001), state parameters (operating status, location, etc.), carbon emission coefficient (0.05 tons / hour), and energy consumption baseline value (500kWh / hour) of an elevator intelligent agent A are used as node attributes of the SFG. The spatial distance weight (1.0), functional coupling coefficient (0.8), and communication latency threshold (10ms) of elevator intelligent agent A are used as edge attributes of the SFG. Subsequently, the SFG graph structure is constructed based on the node attributes and edge attributes. Each node in the SFG represents an intelligent agent, and each edge represents the spatial relationship and functional association between two intelligent agents.

[0022] Step 103: In response to triggering the first event, determine the first agent associated with the first event.

[0023] It should be noted that, in response to the conditions or states upon which the performed operation depends, one or more operations can be performed in real time or with a set delay when the conditions or states are met; unless otherwise specified, there is no restriction on the order in which multiple operations are performed. The first event refers to the event that triggers the agent to perform collaborative control, such as equipment malfunction (elevator malfunction, air conditioner malfunction, etc.), abnormal states (excessively high equipment temperature, abnormal humidity, etc.), security events (fire, intrusion, etc.), environmental changes (changes in light intensity, changes in air quality, etc.), and user requests (control requests submitted by users through a mobile app).

[0024] As an example, when an elevator malfunction (first event) is detected, the system identifies the event as an equipment malfunction event and determines the elevator equipment (first agent) directly associated with the elevator malfunction.

[0025] Step 104: Based on the spatial function map, determine the second intelligent agent associated with the first intelligent agent.

[0026] In some embodiments, step 104 above can be implemented as follows: based on the communication rules of the first agent, at least one candidate agent that satisfies the communication rules is selected from the spatial function graph; based on an attention network, the association weight between the first agent and each candidate agent is determined, and based on the association weight, a second agent associated with the first agent is selected from the candidate agents. In this way, the second agent associated with the first agent can be determined, and the association relationship between them can be used for operations such as device collaboration, fault diagnosis, and emergency response, thereby achieving efficient management and operation of the smart park.

[0027] It should be noted that communication rules are the conditions and constraints for information exchange and collaborative decision-making between device agents. They ensure efficient and orderly communication between devices and avoid redundant communication and conflicts. Communication rule types include spatial distance constraints, functional association constraints, logical association constraints, and communication latency constraints. For example, a spatial distance constraint could be a k-hop communication rule; when k is 1, a device agent can only communicate with adjacent agents in the SFG (Signal-First Group). Attention networks can be implemented using various neural network models, such as graph attention networks, self-attention networks, Transformer models, and multi-head attention mechanisms. The specific model can be determined based on the specific application scenario and is not specifically limited here.

[0028] As an example, assuming the k-hop communication threshold is 1, agents that are no more than k hops away from the first agent (e.g., the elevator) are selected from the SFG as candidate agents (e.g., fire escape access control, fire alarm system, monitoring system, etc.). Then, the spatial distance and functional association between the elevator and the candidate agents are analyzed through an attention network to obtain the association weight. Based on the association weight, the second agent associated with the first agent is selected from the candidate agents. For example, the candidate agent with the highest association weight is selected as the second agent.

[0029] In some embodiments, the above-described determination of the association weights between the first agent and each candidate agent based on the attention network can be achieved as follows: For each candidate agent, the following processes are performed: feature extraction is performed on the data information of the first agent to obtain the first feature of the first agent, and feature extraction is performed on the data information of the candidate agents to obtain the second feature of the candidate agents; the first feature, the second feature, the spatial distance between the first agent and the candidate agents, and the rule weights are fused to obtain the fused feature; the fused feature is activated, and the activation result is multiplied by a preset parameter vector to obtain the initial association score; the initial association score is normalized to obtain the association weights between the first agent and the candidate agents. Thus, by calculating the association weights between the first agent and each candidate agent, the second agents associated with the first agent are determined, and the association relationships between them are used for operations such as device collaboration, fault diagnosis, and emergency response, thereby achieving efficient management and operation of the smart park.

[0030] Here, a Spatial Graph Attention Network (SGA) is used to determine the association weights between two agents. Specifically, it can be calculated according to formula (1).

[0031] (1) in, and The feature vectors of agents i and j (e.g., the current power of the air conditioner, the real-time power generation of the photovoltaic system, etc.) are respectively transformed by linear transformation (W) to extract the key features (reflecting the current state of the agents). Let be the spatial distance between agent i and agent j; 'a' represents the rule weights, which are preset values ​​and are time-series patterns mined from time-series data, such as the overlap between peak air conditioning electricity consumption and peak photovoltaic power supply; 'a' represents the preset parameter vector, which is a set of learnable parameters.

[0032] Feature extraction is performed on the data information of the first intelligent agent, such as spatial coordinates, device ID, state parameters, carbon emission coefficient, etc., to obtain the first feature of the first intelligent agent. Similarly, feature extraction is performed on the data information of the candidate intelligent agents to obtain the second feature of the candidate intelligent agents. Next, the first feature, the second feature, the spatial distance between the first intelligent agent and the candidate intelligent agents, and the rule weights are fused to obtain the fused feature. Here, the fusion method can be simple concatenation or weighted summation, which is not specifically limited here. Subsequently, the fused feature is activated, for example, by using the ReLU function, to map the fused feature to a nonlinear space. Then, the activated fused feature and the preset parameter vector are subjected to dot product processing to obtain the initial association score. Here, the preset parameter vector is a set of learnable parameters used to measure the influence of different features on the association weight. Finally, the initial association score is normalized, for example, by using the softmax function, to convert the initial association score into a probability distribution to obtain the association weight between the first intelligent agent and the candidate intelligent agents.

[0033] As an example, suppose the first agent is elevator device A, and a candidate agent is fire escape access control B. We extract the spatial coordinates, device ID, state parameters, carbon emission coefficients, and other features of elevator device A and fire escape access control B respectively. We then concatenate the features of elevator device A (first feature), the features of fire escape access control B (second feature), the spatial distance between them, and the rule weights to obtain the concatenated features (fusion features). We then perform ReLU activation on the concatenated fusion features to obtain the activated fusion features. Finally, we perform a dot product operation between the activated fusion features and a preset parameter vector to obtain the initial association score. Finally, we normalize the initial association score using the softmax function to obtain the association weight between elevator device A and fire escape access control B.

[0034] Step 105: Determine the target execution strategy based on the scenario information of the first event.

[0035] It should be noted that the scenario information of the first event refers to the background and context information related to the first event. The scenario information of the first event may include event type (such as equipment failure, security incident, environmental change, user request, etc.), event level (such as urgency, importance, etc.), event time, event location, equipment status, environmental parameters (such as comfort level, temperature, light intensity, etc.).

[0036] In some embodiments, step 105 above can be implemented as follows: based on the scenario information of the first event, determine the comprehensive score of each candidate execution strategy; and select the candidate execution strategy with the highest comprehensive score as the target execution strategy. In this way, the target execution strategy can be accurately determined by combining different scenario information, which can effectively improve the accuracy of park management decisions.

[0037] Based on the scenario information of the first event, analyze the cause, scope of impact, and potential risks of the event, and score each candidate execution strategy according to the scenario information to obtain a comprehensive score for each candidate execution strategy. The candidate execution strategy with the highest comprehensive score is selected as the target execution strategy.

[0038] As an example, suppose the first event is an elevator malfunction. The scenario information for the elevator malfunction is "The elevator is on the 3rd floor, the malfunction time is 8 pm, and there are 5 passengers." The candidate execution strategies include "Activate the backup elevator," "Notify elevator maintenance personnel," and "Activate the emergency exit to evacuate passengers." Based on the scenario information, the comprehensive score of each candidate execution strategy is calculated. The comprehensive score of the candidate execution strategy "Activate the backup elevator" is 0.8, the comprehensive score of "Notify elevator maintenance personnel" is 0.7, and the comprehensive score of "Activate the emergency exit to evacuate passengers" is 0.6. Finally, the candidate execution strategy with the highest comprehensive score is selected as the target execution strategy, that is, "Activate the backup elevator" is selected as the target execution strategy.

[0039] In some embodiments, determining the comprehensive score of each candidate execution strategy based on the scenario information of the first event can be achieved as follows: For each candidate execution strategy, the following processing is performed: Comfort indicators, energy consumption indicators, equipment wear indicators, and carbon emission indicators of the scenario corresponding to the first event are collected in real time using sensors; based on preset weight parameters of the candidate execution strategy, the comfort indicators, energy consumption indicators, equipment wear indicators, and carbon emission indicators are weighted and summed to obtain the comprehensive score of the candidate execution strategy. In this way, the comprehensive score of each candidate execution strategy can be calculated based on the comfort indicators, energy consumption indicators, equipment wear indicators, and carbon emission indicators, and the optimal execution strategy can be selected, thereby achieving efficient management and operation of the smart park.

[0040] It should be noted that the preset weight parameters of the candidate execution strategy can be a set of preset weight parameters for each candidate execution strategy, or all candidate execution strategies can share a set of preset weight parameters; no specific limitation is made here.

[0041] First, sensors collect real-time comfort, energy consumption, equipment wear and tear, and carbon emission indicators for the scenario corresponding to the first event. Comfort indicators can be calculated based on ambient temperature, humidity, light intensity, and noise levels; energy consumption indicators can be calculated based on equipment and building energy consumption; equipment wear and tear indicators can be calculated based on equipment operating time and failure rate; and carbon emission indicators can be calculated based on equipment and building carbon emissions. Next, a set of weight parameters is preset for each candidate execution strategy to measure the influence of different indicators on the overall score. Based on the preset weight parameters, the comfort, energy consumption, equipment wear and tear, and carbon emission indicators are weighted and summed to obtain the overall score of the candidate execution strategy.

[0042] As an example, suppose the first event is an elevator malfunction. The scenario information for the elevator malfunction is "The elevator is on the 3rd floor, the malfunction time is 8 pm, and there are 5 passengers." The candidate execution strategies include "Activate the backup elevator," "Notify elevator maintenance personnel," and "Activate the emergency exit to evacuate passengers." The comfort index for "Activate the backup elevator" is 90, the energy consumption index is 80, the equipment wear index is 70, and the carbon emission index is 80. The comfort index for "Notify elevator maintenance personnel" is 80, the energy consumption index is 70, the equipment wear index is 80, and the carbon emission index is 70. The comfort index for "Activate the emergency exit to evacuate passengers" is 70, the energy consumption index is 60, the equipment wear index is 90, and the carbon emission index is 60. The preset weight parameters are comfort weight 0.4, energy consumption weight 0.3, equipment wear weight 0.2, and carbon emission weight 0.1. Finally, by weighted summation, the comprehensive score for "Activate the backup elevator" is 78, the comprehensive score for "Notify elevator maintenance personnel" is 76, and the comprehensive score for "Activate the emergency exit to evacuate passengers" is 72.

[0043] Step 106: Based on the target execution strategy, perform collaborative control on the first and second intelligent agents.

[0044] As an example, suppose the first intelligent agent is elevator equipment A, and the second intelligent agent includes access control B and air conditioner C. The target execution strategy is "force elevator equipment to land on the first floor, open access control, and shut down air conditioner". Then, according to the target execution strategy, elevator equipment A will be forced to land on the first floor, and access control B will be opened and air conditioner C will be shut down to achieve coordinated control of the first and second intelligent agents.

[0045] In some embodiments, the collaborative control data of the first and second intelligent agents is stored on a blockchain, and the digital twin model is updated based on the collaborative control results of the first and second intelligent agents. Thus, through blockchain storage and digital twin model updates, the authenticity and traceability of data in the smart park management system can be ensured, and the real-time status of equipment and environmental changes within the park can be reflected, providing a reliable data foundation for park management and decision-making.

[0046] First, identify the collaborative control data that needs to be stored, such as collaborative control commands, execution results, status information, timestamps, and agent identifiers. Next, format the selected data, for example, by converting it to JSON format. Then, encrypt the formatted data, for example, using a hash algorithm, and write the encrypted data to a blockchain network, such as Ethereum or Hyperledger Fabric. The blockchain network writes the data into blocks and generates unique block hash values, ensuring the immutability and traceability of the data for blockchain-based notarization. Simultaneously, analyze the collaborative control results of the first and second agents, such as changes in device status, environmental parameters, and personnel status. Then, based on the collaborative control results, update the relevant data in the digital twin model, such as updating device status, environmental parameters, and personnel status, and synchronize the updated digital twin model to all nodes to ensure data consistency.

[0047] As an example, after an elevator malfunctions, the elevator intelligent agent sends an opening command to the emergency exit access control intelligent agent. Upon receiving the command, the emergency exit access control intelligent agent opens the emergency exit access control and formats and encrypts information such as the collaborative control command, execution result, timestamp, and intelligent agent identifier. The encrypted data is then written to a blockchain network, such as Ethereum. The blockchain network generates a unique block hash value to ensure the immutability and traceability of the data, thus completing blockchain notarization. At the same time, the collaborative control result is analyzed, and it is found that the emergency exit access control has been opened. The status of the emergency exit access control is updated to "open" and synchronized to all nodes to achieve the update of the digital twin model.

[0048] In some embodiments, data information of a third agent is obtained from a digital twin model; in response to the third agent not belonging to a node in the spatial function graph, the spatial function graph is incrementally updated based on the data information of the third agent to obtain an updated spatial function graph. Thus, using an incremental update method can improve the construction efficiency of the spatial function graph and avoid the problem of repeated data collection.

[0049] First, data information of the third agent, such as spatial unit coordinates, device ID, and state parameters, is obtained from the digital twin model. It is then checked whether the third agent already exists as a node in the spatial functional graph (SFG). If the third agent is already a node in the SFG, no update is needed. If the third agent is not a node in the SFG, its data information is added to the SFG as a new node. Based on the spatial location and functional association of the third agent, edges are added between it and other adjacent agents, and corresponding edge attributes are set, such as spatial distance weight, functional coupling coefficient, and communication latency threshold. Simultaneously, the attributes of other nodes in the SFG are updated, such as the spatial distance weight and functional coupling coefficient of nodes adjacent to the third agent. After completing node addition, edge addition, and attribute updates, an updated spatial functional graph is obtained, which contains the data information of the third agent and reflects the latest spatial relationships and functional associations between devices within the park.

[0050] As an example, assuming the third agent is a newly installed camera, the spatial coordinates, device ID, and state parameters of the new camera are obtained from the digital twin model. If it is determined that the new camera does not exist in the nodes of the SFG, the data information of the new camera is added to the SFG as a new node, and edges are added between the new camera and other agents adjacent to it, such as edges with nearby elevators, fire escape access control devices, etc. Finally, the attributes of other nodes in the SFG are updated, such as updating the spatial distance weight and functional coupling coefficient of the nodes adjacent to the new camera, so as to obtain the updated spatial function graph. The updated spatial function graph contains the data information of the new camera and reflects the latest spatial relationships and functional associations between devices in the park.

[0051] In some embodiments, taking a multi-park energy and carbon collaborative operation and maintenance scenario as an example, firstly, the data of photovoltaic panels (ID: CN-North-OP-PV-001-BJ) in City A Park and energy storage equipment (ID: CN-North-OP-ESS-001-TJ) in City B Park are imported from the entities in the digital twin model, generating SFG—the functional coupling coefficient of photovoltaic node N102 and energy storage node N205 is 0.9, and the communication latency is 80ms; the intelligent agent is configured to reuse the digital twin logical domain attributes (k-hop threshold = 1, photovoltaic conversion efficiency 18%, energy storage capacity 100kWh) for photovoltaic / energy storage intelligent agents; then, on 2024.05.20 At 12:30, the digital twin reports the real-time status of City A's photovoltaic system: "Power generation 60kW, irradiance 6500lux, carbon emissions 0.0144 tons," and the energy storage status: "Remaining capacity 90%." The agent communication rule is that the photovoltaic agent communicates only with the City B energy storage agent based on the SFG 1-hop rule, avoiding redundant interactions. Subsequently, engine calculations are performed. By calling the ST-Apriori rule "Photovoltaic power generation > 50kW and remaining energy storage capacity > 80% can indicate surplus power storage," and using Spatial-GAT to calculate an attention weight of 0.85, the output strategy is obtained. "20kW of surplus photovoltaic power from City A is dispatched to energy storage in City B." Simultaneously, the execution feedback indicates that the energy storage device receives the instruction and starts charging, with the execution result showing "20kW stored, 95% remaining capacity," which is then synchronized to the digital twin update status. Finally, due to the cross-park energy flow dispatching involved, the blockchain notarization unit records the hash value of "dispatch time 12:32, SFG coupling coefficient 0.9, carbon emission reduction 0.0048 tons," and completes notarization by connecting to the trusted data of the digital twin. The implementation effect is an 80% improvement in cross-park energy flow dispatching efficiency, a reduction in SFG construction time from 8 hours to 2 hours, and a 70% improvement in carbon emission accounting efficiency.

[0052] In other embodiments, taking an overseas campus fire emergency scenario as an example, the SFG construction module adopts the WGS84 coordinate system, imports elevator (ID: ME-OP-ELEV-001-NEOM), access control, and air conditioning data from the digital twin model, and generates SFG—the coupling coefficient between elevator node E008 and access control node N012 is 0.95, and the communication latency is 50ms; the elevator intelligent agent reuses the corresponding entity attribute "k-hop threshold = 1, high temperature resistance threshold 60℃" in the digital twin model; on June 10, 2024. At 09:05, the smoke sensor triggered a fire alarm. The elevator's intelligent agent read the SFG's one-hop data: "Access control N012 closed, air conditioning running," and, in conjunction with the decision engine, output the strategy: "Elevator forced to land on the first floor, access control opened, air conditioning shut down." After execution, the digital twin model was fed back. Subsequently, maintenance personnel attempted remote intervention (to temporarily postpone the forced landing). The conflict resolution unit made a decision based on the "safety > maintenance" rule. Afterward, the blockchain evidence storage unit stored the "conflict time 09:06, elevator location X2Y2, SFG association status" on a local node, meeting data localization requirements. This reduced fault location time from 20 minutes to 40 seconds and conflict tracing time to 3 minutes, making it suitable for overseas scenarios.

[0053] In some embodiments, taking a single-campus conference scenario as an example, the SFG building module associates the conference room's air conditioning, lighting, and human body sensor nodes (coupling coefficient 0.9); subsequently, the intelligent agent reuses the data "15 people in the conference room, current temperature 28℃" from the digital twin model; then, the collaborative decision engine dynamically adjusts the weights. (Comfort level) = 0.4 (Energy consumption) = 0.3, output strategy: "Set air conditioning to 25℃ and turn on local lighting". After execution, comfort level increases by 22% and energy consumption decreases by 18.7%, forming a closed loop with digital twin health data.

[0054] The following beneficial effects can be achieved through the embodiments of this application: (1) By reusing digital twin spatiotemporal domain data and adopting the "incremental update (trigger type) + full verification (daily early morning)" mechanism, the SFG construction efficiency is improved by 50%, the spatial error is ≤0.5 meters, and the energy and carbon property error is ≤3%, avoiding the problem of repeated data collection in traditional systems; (2) The device intelligent agent realizes decentralized interaction based on SFG k-hop communication rules (k=1-3 configurable), reducing redundant communication by 40%; the collaborative decision engine integrates Spatial-GAT and ST-Apriori rules, improving the decision accuracy to over 88%, increasing the carbon emission accounting efficiency in energy and carbon scenarios by 70%, and improving the cross-park energy flow scheduling efficiency by 80%. (3) The blockchain evidence storage unit connects to the trusted data of digital twins, and the fault tracing time is shortened from the traditional 2 hours to within 5 minutes; it supports the deployment of lightweight nodes overseas, meets the GDPR and Saudi data localization requirements, and shortens the international implementation cycle by 6 months; in emergency scenarios, the fault location time is shortened from 20 minutes to 40 seconds, and the conflict resolution efficiency is improved by 90%.

[0055] The above are embodiments of the method proposed in this application. In addition, embodiments of the present invention also include a campus-level operating system based on multi-agent collaboration, such as... Figure 2 As shown, the campus-level operating system 200 based on multi-agent collaboration specifically includes: The acquisition module 210 is used to acquire data information of at least one of the intelligent agents from the digital twin model; Construction module 220 is used to construct a spatial function graph based on data information of at least one of the intelligent agents; The determination module 230 is configured to determine a first intelligent agent associated with the first event in response to the triggering of the first event; The determining module 230 is further configured to determine a second intelligent agent associated with the first intelligent agent based on the spatial function map; The control module 240 is used to determine a target execution strategy based on the scenario information of the first event, and to perform coordinated control of the first intelligent agent and the second intelligent agent based on the target execution strategy.

[0056] In some embodiments, the data information of the agent includes the agent's entity attributes, newly added energy and carbon-specific attributes, spatial distance weights, functional coupling coefficients, and communication delay thresholds. The construction module 220 is further configured to use the agent's entity attributes and newly added energy and carbon-specific attributes as node attributes of the spatial functional graph; and to use the agent's spatial distance weights, functional coupling coefficients, and communication delay thresholds as edge attributes of the spatial functional graph.

[0057] In some embodiments, the determining module 230 is further configured to: filter at least one candidate agent that satisfies the communication rules of the first agent from the spatial function graph; determine the association weight between the first agent and each of the candidate agents based on the attention network; and filter a second agent associated with the first agent from the candidate agents based on the association weight.

[0058] In some embodiments, the determining module 230 is further configured to perform the following processing for each candidate agent: extracting features from the data information of the first agent to obtain a first feature of the first agent, and extracting features from the data information of the candidate agent to obtain a second feature of the candidate agent; fusing the first feature, the second feature, the spatial distance between the first agent and the candidate agent, and the rule weights to obtain a fused feature; activating the fused feature and performing a dot product operation between the activation result and a preset parameter vector to obtain an initial association score; and normalizing the initial association score to obtain the association weight between the first agent and the candidate agent.

[0059] In some embodiments, the determining module 230 is further configured to determine a comprehensive score for each candidate execution strategy based on the scenario information of the first event; and to select the candidate execution strategy with the highest comprehensive score as the target execution strategy.

[0060] In some embodiments, the determining module 230 is further configured to perform the following processing for each candidate execution strategy: real-time collection of comfort index, energy consumption index, equipment wear index and carbon emission index of the scenario corresponding to the first event through sensors; and weighted summation of the comfort index, energy consumption index, equipment wear index and carbon emission index based on the preset weight parameters of the candidate execution strategy to obtain a comprehensive score of the candidate execution strategy.

[0061] In some embodiments, the campus-level operating system 200 based on multi-agent collaboration further includes an update module 250, which is used to store the collaborative control data of the first agent and the second agent on a blockchain, and update the digital twin model based on the collaborative control results of the first agent and the second agent.

[0062] In some embodiments, the update module 250 is further configured to obtain data information of a third agent from the digital twin model; in response to the third agent not belonging to a node of the spatial function graph, to incrementally update the spatial function graph based on the data information of the third agent, thereby obtaining an updated spatial function graph.

[0063] Furthermore, based on the same inventive concept, embodiments of this application also provide a campus-level operating device based on multi-agent collaboration, the structure of which is as follows: Figure 3 As shown.

[0064] Figure 3 This is a schematic diagram of a campus-level operating device based on multi-agent collaboration, provided as an embodiment of this application. Figure 3 As shown, the device includes: At least one processor 301; And a memory 302 that is communicatively connected to at least one processor; The memory 302 stores instructions that can be executed by at least one processor. The instructions are executed by at least one processor 301 to enable at least one processor 301 to perform the steps of the method corresponding to any of the above embodiments.

[0065] Some embodiments of this application provide corresponding to Figure 1 A non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured as follows: The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for IoT devices and media are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0066] The systems, devices, media, and methods provided in this application are one-to-one correspondences. Therefore, the systems and media also have similar beneficial technical effects to their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and media will not be repeated here.

[0067] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, devices, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0068] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0069] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0070] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0071] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0072] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0073] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, 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 technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0074] It should also be noted that 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 a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0075] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A park-level operation method based on multi-agent cooperation, characterized in that, The method comprises: obtaining data information of at least one of the agents from a digital twin model; constructing a space function graph based on the data information of at least one of the agents; in response to triggering a first event, determining a first agent associated with the first event; based on the space function graph, determining a second agent associated with the first agent; based on the scene information of the first event, determining a target execution strategy, and based on the target execution strategy, cooperatively controlling the first agent and the second agent.

2. The method of claim 1, wherein, The data information of the agent includes entity attributes, new energy and carbon exclusive attributes, spatial distance weights, function coupling coefficients, and communication delay thresholds. Based on the data information of at least one of the agents, the space function graph is constructed, including: the entity attributes and the new energy and carbon exclusive attributes of the agent are taken as node attributes of the space function graph; the spatial distance weights, the function coupling coefficients, and the communication delay thresholds of the agent are taken as edge attributes of the space function graph.

3. The method of claim 1, wherein, Based on the space function graph, the second agent associated with the first agent is determined, including: based on the communication rules of the first agent, at least one candidate agent satisfying the communication rules is screened out from the space function graph; based on an attention network, the association weight of the first agent and each candidate agent is determined, and based on the association weight, the second agent associated with the first agent is screened out from the candidate agents.

4. The method of claim 3, wherein, Based on the attention network, the association weight of the first agent and each candidate agent is determined, including: for each candidate agent, the following processing is performed respectively: feature extraction is performed on the data information of the first agent to obtain first features of the first agent, and feature extraction is performed on the data information of the candidate agent to obtain second features of the candidate agent; the first features, the second features, the spatial distance between the first agent and the candidate agent, and the rule weight are fused to obtain fused features; activation processing is performed on the fused features, and the activation result is dot multiplied with a preset parameter vector to obtain an initial association score; the initial association score is normalized to obtain the association weight of the first agent and the candidate agent.

5. The method of claim 1, wherein, Based on the scene information of the first event, the target execution strategy is determined, including: based on the scene information of the first event, determining the comprehensive score of each candidate execution strategy; the candidate execution strategy with the highest comprehensive score is taken as the target execution strategy.

6. The method of claim 5, wherein, Based on the scene information of the first event, the comprehensive score of each candidate execution strategy is determined, including: for each candidate execution strategy, the following processing is performed respectively: real-time collection of comfort indicators, energy consumption indicators, equipment wear indicators, and carbon emission indicators corresponding to the scene of the first event through sensors; Based on a preset weight parameter of the candidate execution strategy, the comfort index, the energy consumption index, the equipment wear index and the carbon emission index are weighted and summed to obtain a comprehensive score of the candidate execution strategy.

7. The method of claim 1, wherein, After the collaborative control of the first agent and the second agent, the method further comprises: The collaborative control data of the first agent and the second agent are stored in a block chain, and the digital twin model is updated based on the collaborative control result of the first agent and the second agent.

8. The method of claim 1, wherein, The method further comprises: Obtaining data information of a third agent from the digital twin model; In response to the third agent not belonging to the node of the space function graph, based on the data information of the third agent, the space function graph is incrementally updated to obtain an updated space function graph.

9. A park-level operation system based on multi-agent collaboration, characterized in that, The system comprises: An obtaining module, configured to obtain data information of at least one agent from a digital twin model; A constructing module, configured to construct a space function graph based on the data information of at least one agent; A determining module, configured to determine a first agent associated with a first event in response to the triggering of the first event; The determining module is further configured to determine a second agent associated with the first agent based on the space function graph; A control module, configured to determine a target execution strategy based on the scene information of the first event, and to collaboratively control the first agent and the second agent based on the target execution strategy.

10. A computer storage medium storing computer-executable instructions, which, when executed by a processor, cause the processor to perform acts comprising: The computer executable instructions, when executed, implement the method of any one of claims 1-8.

Images