AGENT equipment health assessment system and method based on knowledge graph

Abstract

The invention relates to the technical field of agent assessment, and discloses an AGENT equipment health assessment system and method based on a knowledge graph, and the method comprises the following steps: constructing a resource topological graph, loading equipment state and resource capacity data, and providing a structured knowledge basis for collaborative assessment; the equipment state comprises a health degree and a priority; quantizing the equipment resource demand, generating an equipment-resource demand matrix, and determining the dependency relationship of each piece of equipment on each type of resources; and the maintenance agent predicts the equipment resource demand, the resource agent performs deadlock detection in combination with the demand and the resource constraint, and a deadlock risk probability and a fault equipment list are output. According to the invention, through collaborative optimization of the maintenance agent and the resource agent, generation of the dynamic scheduling strategy, real-time risk monitoring and rollback mechanism triggering, and formation of a closed-loop knowledge graph updating mechanism, the system has a sustainable evolution adaptive capability, the operation and maintenance cost is effectively reduced, and faults caused by system decision conflicts are prevented.

Description

Technical Field

[0001] The present invention relates to the field of agent evaluation, and more specifically, to an AGENT device health evaluation system and method based on a knowledge graph. Background Art

[0002] With the development of industry and intelligent manufacturing, the health management of complex equipment systems faces challenges such as difficult integration of multi-source heterogeneous data, poor real-time evaluation, and weak cross-device collaboration.

[0003] Traditional single-agent (AGENT) models rely on centralized decision-making and are difficult to adapt to the dynamic coupling characteristics of distributed devices. Static knowledge graphs lack the ability to respond in real time to the evolution of device states. In particular, when the decision-making layer of the agent model outputs multiple conflicting instructions simultaneously, the optimization falls into a local optimum, causing the system to enter an infinite loop, resulting in decision-making delays or errors. Summary of the Invention

[0004] The present invention provides an AGENT device health evaluation system and method based on a knowledge graph, which solves the technical problems in related technologies of over-reliance on centralized decision-making, inability to analyze conflicts in decision-making during decision-making, resulting in decision-making delays or errors.

[0005] The present invention provides an AGENT device health evaluation method based on a knowledge graph, including the following steps: S100, Knowledge Graph Initialization: Construct a resource topology graph, load device status and resource capacity data, and provide a structured knowledge basis for collaborative evaluation; The device status includes health degree and priority; S200, Resource Requirement Modeling: Quantify device resource requirements, generate a device-resource requirement matrix, and clarify the dependence relationship of each device on various resources; S300, Collaborative Deadlock Prediction: The maintenance agent predicts device resource requirements, and the resource agent combines the requirements with resource constraints for deadlock detection, and outputs the deadlock risk probability and the list of faulty devices; S400, Deadlock Avoidance Decision: The maintenance agent optimizes the resource scheduling model, and the resource agent quantifies the dynamic deadlock risk, and collaboratively generates an optimal scheduling strategy and maintenance instructions that take into account reliability and risk; S500, Resource Coordination Execution: Allocate resources according to the scheduling sequence, monitor the deadlock risk in real time and trigger a rollback mechanism, and finally update the knowledge graph and feedback the execution result.

[0006] Further, in S100, it specifically includes the following steps: S110, Sensor Data Acquisition: Collect real-time sensor data from edge devices, including physical quantities such as vibration, temperature, and current; S120, Sensor self - check: Perform anomaly detection on each sensor channel to identify failed sensors; S130, Construct an anti - interference mask matrix: Generate a mask matrix to shield the data of failed sensors; S140, Anti - interference feature extraction: Extract robust features through masked graph convolution; S150, Feature compression and transmission: Perform dimensionality reduction processing and send it to the subsequent agent.

[0007] Furthermore, in S200, it specifically includes the following steps: S210, Heterogeneous graph spectrum construction: Integrate the compressed feature matrix output by the scheduling agent with the resource status to construct a device - resource heterogeneous graph spectrum; S220, Spatial - dependence convolution: Aggregate neighbor information through graph convolution to capture the device - resource coupling relationship; S230, Temporal evolution modeling: Integrate the changes in the historical state time window to model dynamic evolution; S240, Event - triggered momentum update: Respond to external events and correct the graph spectrum in real - time.

[0008] Furthermore, in S300, it specifically includes the following steps: S310, Device health assessment: Based on a multi - objective optimization graph attention network, evaluate the device health status; S320, Resource demand dynamic prediction: Combine historical data with the current state to predict future resource demands; S330, Deadlock risk assessment: Construct a resource competition graph to evaluate the system deadlock risk; S340, Resource dynamic adjustment: Based on a multi - objective optimization resource re - allocation strategy.

[0009] Furthermore, in S330, the calculation formula for deadlock risk quantification is as follows: Competition graph construction: ; where, is the set of resource nodes, is the set of resource competition relationship edges, is the set of edge weights, is the resource competition graph; Edge weight calculation: ; where, represents that device occupies resource and requests resource , is the total number of devices, is the resource Available quantity of Indicates the edge weight between the requested resource and the occupied resource ; Deadlock risk score: ; ; Among them, are the high-risk threshold and the low-risk threshold respectively, is the web page ranking algorithm, used to evaluate the importance of nodes, is the deadlock risk level, divided into three levels: High, Medium, and Low, is the overall deadlock risk score of the system; Furthermore, in S400, it specifically includes the following steps: S410, Resource scheduling optimization modeling: Construct an integer programming model to optimize the resource scheduling plan; S420, Deadlock risk quantification: Calculate the resource competition degree and quantify the deadlock probability; S430, Multi-objective decision-making optimization: Combine reliability and deadlock risk to solve the Pareto optimal solution; S440, Dynamic scheduling strategy generation: Generate a resource scheduling sequence and preventive maintenance instructions.

[0010] Furthermore, in S420, the calculation formula for deadlock risk quantification is as follows: Resource competition degree: ; Among them, is the competition degree coefficient of resource k, is the fault flag of device i, is the indicator function, which is 1 when the condition is satisfied and 0 otherwise, is the device and the resource demand matrix, is the total number of resource types, is the resource type index; Deadlock risk coefficient: ; Among them, is the Sigmoid activation function, which maps the output to the interval [0,1], is the deadlock risk coefficient, approaching 1 indicates high risk, and approaching 0 indicates low risk.

[0011] Furthermore, in S430, the calculation formula for multi-objective decision-making optimization is as follows: Multi-objective function: ; ; wherein, is the reliability score of device i, is the deadlock risk coefficient, is the resource scheduling decision vector, is the health score of device i, is the priority of device i, are the weight coefficients of health and priority respectively; When constraint adjustment: ; wherein, is the deadlock risk threshold, is the resource slack factor, is the predicted resource demand vector; Solve the optimization: ; ; wherein, is the total system reliability score, is the optimal decision variable combination.

[0012] Furthermore, in S500, it specifically includes the following steps: S510, resource scheduling execution: Dynamically allocate resources according to the resource scheduling sequence of S440; S520, deadlock detection after allocation: Real-time detect the deadlock risk after each allocation; S530, abnormal rollback mechanism: If the deadlock risk is detected, immediately roll back the allocation operation; S540, execution feedback update: Generate a resource scheduling report and update the knowledge graph.

[0013] The present invention also proposes a knowledge graph-based AGENT device health assessment system for performing the steps in the foregoing knowledge graph-based AGENT device health assessment method, including: Knowledge graph initialization module: Construct a resource topology graph to describe the physical / logical connection relationship between devices and resources, load device status data, load resource capacity data, establish a structured knowledge base, and support collaborative decision-making; Resource demand modeling module: Quantify the demand of devices for various resources, generate a device-resource demand matrix, clarify the resource dependence relationship between devices, and mark the key resource paths; Collaborative Deadlock Prediction Module: The maintenance agent is used to predict the sequence of device resource requirements, and the resource agent is used to detect deadlock risks by combining resource constraints, output the deadlock risk probability, and generate a list of faulty devices. Deadlock Avoidance Decision Module: The maintenance agent is used to optimize the resource scheduling model, and the resource agent quantifies the dynamic deadlock risk. They collaborate to generate the optimal scheduling strategy and output the maintenance instruction sequence. Resource Coordination Execution Module: Allocate resources to devices according to the scheduling sequence, monitor the deadlock risk in real time, trigger the rollback mechanism, update the status of the knowledge graph, and feedback the execution result.

[0014] The beneficial effects of the present invention are as follows: Through the accurate modeling of resource requirements and the forward-looking prediction of deadlock risks, and through the collaborative optimization of the maintenance agent and the resource agent, the present invention generates a dynamic scheduling strategy that takes into account both safety and efficiency. During the execution process, it monitors risks in real time and triggers the rollback mechanism, and finally forms a closed-loop knowledge graph update mechanism, enabling the system to have the adaptive ability of continuous evolution, effectively reducing the operation and maintenance cost, and preventing failures caused by system decision conflicts. Description of the Drawings

[0015] Figure 1 is the flowchart of the AGENT device health assessment method based on the knowledge graph proposed by the present invention; Figure 2 is the module block diagram of the AGENT device health assessment system based on the knowledge graph proposed by the present invention; Figure 3 is the multi-agent collaboration sequence diagram based on the health assessment method proposed by the present invention.

[0016] In the figure: 101, Knowledge Graph Initialization Module; 102, Resource Requirement Modeling Module; 103, Collaborative Deadlock Prediction Module; 104, Deadlock Avoidance Decision Module; 105, Resource Coordination Execution Module. Detailed Embodiments

[0017] Now, the subject matter described herein will be discussed with reference to exemplary embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the scope of protection of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described in relation to some examples can also be combined in other examples.

[0018] As Figure 1 shown, the AGENT device health assessment method based on the knowledge graph includes the following steps: S100, Knowledge graph initialization: Construct a resource topology graph, load device status (health, priority) and resource capacity data, providing a structured knowledge basis for collaborative evaluation; In one embodiment of the present invention, it specifically includes the following steps: S110, Sensor data acquisition: Use a scheduling agent (SA) to collect real-time sensor data from edge devices, including physical quantities such as vibration, temperature, and current; Original sensor data matrix: ; Among them, is the total number of device nodes, such as the number of wind turbine gearboxes, is the total number of sensor channels, including different types of sensors such as temperature, vibration, and current, represents the actual reading value of the j-th type of sensor of the i-th node at time t, represents the real number matrix space of N rows and D columns, represents the sensor data matrix at time t; S120, Sensor self-check (chi-square test): Perform anomaly detection on each sensor channel to identify failed sensors; Chi-square statistic calculation: ; Among them, is the actual observation value of the i-th type of sensor in the k-th time window, is the expected value calculated based on historical data, using the moving average method, is the detection time window length, default set to 10 time units, represents that the detection range covers all D sensor channels, represents the chi-square statistic of the i-th sensor, is the time window index, is the sensor channel index; Sensor failure determination: ; Among them, is the failure determination threshold, taking the value of 3.84 at a 95% confidence level, is the failure flag of the i-th sensor, 1 indicates failure, and 0 indicates normal; S130, Construct an anti-interference mask matrix: Generate a mask matrix to mask the failed sensor data; Node-level mask vector: ; Among them, is the mask identifier of the j-th node, taking values of 0 or 1, represents the failure state of the i-th type of sensor on node j, is the product symbol, and the result is 1 when all sensors are normal, otherwise it is 0, is the total number of sensor channels, represents an N-dimensional mask vector, containing the sensor working state information of each node; Adjacency mask matrix: ; Among them, M is the mask matrix, represents the binary matrix space of N×N; S140, anti-interference feature extraction: extract robust features through masked graph convolution; ; ; Among them, is the adjacency matrix with self-connections added, is the original device topology matrix, is the N-dimensional identity matrix, is the Hadamard product operator, indicating element-wise multiplication of matrices, is the trainable weight matrix, , is the target feature dimension, is the ReLU activation function, used to introduce non-linearity, is the extracted feature matrix, with a dimension of N×F, is the degree matrix, representing the connection degree of each node; S150, feature compression and transmission: perform dimensionality reduction and send it to the subsequent intelligent agent; ; ; Among them, is the compressed feature dimension, is the max pooling operation, used for dimensionality reduction, F is the original feature dimension, represents the compressed feature matrix, is the extracted feature matrix; S200, resource demand modeling (executed by the Resource Agent (RA)): quantify the device resource requirements, generate a device-resource demand matrix, and clarify the dependence relationship of each device on various resources; In an embodiment of the present invention, it specifically includes the following steps: S210, Heterogeneous Graph Construction: Fuse the compressed feature matrix output by SA with the resource status to construct a device-resource heterogeneous graph; Define the node set: ; Among them, is the device node set, containing N device nodes, is the resource node set, containing resource nodes (such as spare parts, engineers, etc.), N is the total number of devices, is the total number of resource types, is the complete node set, containing all device and resource nodes; Adjacency matrix partitioning: ; Among them, is the topological adjacency matrix between devices, from system configuration, with dimension N×N, is the device-resource requirement relationship matrix, with dimension N× , is the resource-device relationship matrix, equal to , is the relationship matrix between resources, with dimension × , is the complete adjacency matrix; Initial node embedding: ; Among them, is the compressed feature matrix output by SA, with dimension N× , is the initial embedding matrix of resource nodes, with dimension × , randomly initialized, is the compressed feature dimension, is the initial node embedding matrix, with dimension ; S220, Spatial Dependence Convolution: Aggregate neighbor information through graph convolution to capture device-resource coupling relationships; ; ; Among them, is the adjacency matrix with self-loops added, I is the identity matrix, is the degree matrix, is the spatial convolution weight matrix, with dimension × , is the hidden layer dimension, , ReLU is the rectified linear unit activation function, is the feature matrix after spatial dependence convolution; S230, Temporal evolution modeling: Incorporate historical states changes in the time window to model dynamic evolution; ; Among them, is the spatial feature matrix before time, i.e., the historical cache, is the GRU network parameter, including the weights of the update gate and the reset gate, is the time window length, GRU is the gated recurrent unit network, is the feature matrix after temporal enhancement, with dimensions × ; S240, Event-triggered momentum update: Respond to external events (such as maintenance work orders, environmental mutations) and correct the graph in real time; Event encoding: Among them, type is the event type encoding (e.g., maintenance = 1, failure = 2, etc.), priority is the event priority (an integer from 1 to 5), target is the one-hot encoded vector of the target device ID, is the multi-layer perceptron parameter, is the event vector, with dimensions ; Node state correction: ; Among them, is the event impact factor. When it is an emergency event, when it is an ordinary event, , is the indicator function, which is 1 when i is the target device / resource node and 0 otherwise, is the feature vector of node i; S300, Cooperative deadlock prediction (cooperation between Maintenance Agent (abbreviated as MA) and RA): The Maintenance Agent (abbreviated as MA) predicts the device resource requirements, and RA combines the requirements with the resource constraints to perform deadlock detection, and outputs the deadlock risk probability and the list of faulty devices; In an embodiment of the present invention, it specifically includes the following steps: S310, Device health assessment (performed by the Diagnostic Agent (abbreviated as DA)): Based on the graph attention network for multi-objective optimization, evaluate the device health status; Attention weight calculation: ; Among them, is the attention weight matrix, is the bias term, is the node 's eigenvector, is vector concatenation, is the leaky rectified linear unit activation function; Multi-head attention aggregation: ; ; Among them, is the number of attention heads (default 8), is the transformation matrix of the -th head, is the node 's neighbor set, is the activation function, is the -th attention head's attention weight of node to node ; Health score: ; Among them, is the multi-layer perceptron, is the aggregated eigenvector of node ; is the normalized health score; Fault determination threshold: ; Among them, is the health threshold (default 0.6), is the change rate threshold (default 0.1), is the change amount of the health score; S320, Dynamic prediction of resource requirements (RA execution): Combining historical data and the current state, predict future resource requirements; Demand prediction model: ; Among them, is the resource requirement of the faulty device, is the historical resource utilization rate, is the seasonal feature, is the GRU network parameter, is the gated recurrent unit network; Resource gap calculation: ; Among them, is the resource buffer capacity (default 10%), is the currently available resource amount, is the rectified linear unit activation function; S330, Deadlock Risk Assessment (DA→RA): Construct a resource competition graph to evaluate the system deadlock risk; Competition graph construction: ; Among them, is the set of resource nodes, is the set of edges of resource competition relationships, is the set of edge weights, is the competition graph of resources; Edge weight calculation: ; Among them, represents device occupying resource and requesting resource , is the total number of devices, is the available amount of resource , represents the edge weight between the requesting resource and the occupied resource ; Deadlock risk scoring: ; ; Among them, are the high and low risk thresholds respectively (default 0.8, 0.4), is the PageRank algorithm, used to evaluate the importance of nodes, is the deadlock risk level, divided into three levels: High, Medium, and Low, used to guide resource allocation decisions, is the overall deadlock risk score of the system, with a value range of [0, 1], and the larger the value, the higher the deadlock risk; S340, Resource Dynamic Adjustment (RA execution): A resource reallocation strategy based on multi-objective optimization; Objective function: ; Constraint conditions: ; Among them, is the resource scheduling decision vector, is the weight coefficient, is the set of faulty devices, is the device priority, is the device-resource demand matrix; Resource release execution: ; Among them, is the resource release flag of device , 1 means release, 0 means keep, is the current available resource vector, with dimension ; S400, Deadlock avoidance decision (MA and RA cooperation): MA optimizes the resource scheduling model, RA quantifies the dynamic deadlock risk, and cooperatively generates the optimal scheduling strategy and maintenance instructions that take into account both reliability and risk; In an embodiment of the present invention, it specifically includes the following steps: S410, Resource scheduling optimization modeling (executed by MA): Construct an integer programming model to optimize the resource scheduling plan; Decision variables: ; Among them, is the resource scheduling flag of the i-th device, is the total number of devices; Objective function (maximize system reliability): ; Among them, are the weight coefficients of health and priority respectively, controlling the influence of health and priority respectively (default ), is the health score of device i, with a value range of [0,1], is the priority of device i, with an integer value of 1-5, is the resource scheduling decision vector, indicating whether to allocate resources to device i; Resource constraints: ; Among them, is the device-resource demand matrix, with dimension , is the predicted resource demand vector, with dimension , is the current available resource vector, with dimension , is the total number of resource types, is the resource type index; S420, Deadlock Risk Quantification (RA Execution): Calculate the degree of resource competition and quantify the deadlock probability; Degree of resource competition: ; Among them, is the competition coefficient of resource k, is the failure flag (0 or 1) of device i, is the indicator function, which is 1 when the condition is satisfied and 0 otherwise; Deadlock risk coefficient: ; Among them, is the Sigmoid activation function, which maps the output to the interval [0, 1], is the deadlock risk coefficient, approaching 1 indicates high risk, and approaching 0 indicates low risk, is the competition coefficient of resource k; S430, Multi-objective Decision Optimization (MA and RA Collaboration): Combine reliability and deadlock risk to solve the Pareto optimal solution; Multi-objective function: ; ; Among them, is the reliability score of device i, is the deadlock risk coefficient, is the resource scheduling decision vector, is the priority of device i; Constraint adjustment (when ) : ; Among them, is the deadlock risk threshold, is the resource relaxation factor, is the predicted resource demand vector, is the current available amount of resource type k; Solve the optimization: ; ; Among them, is the total system reliability score, is the optimal decision variable combination; S440, Dynamic Scheduling Policy Generation (MA Execution): Generate resource scheduling sequences and preventive maintenance instructions; Allocation strategy: ; Among them, : A set of devices that need to allocate resources, : Optimized decision variable values Maintenance instructions: ; Among them, is the maintenance instruction type of device i, is the health score of device i, is the system deadlock risk coefficient; Resource scheduling sequence: ; Among them, is the scheduling weight coefficient (default ), is the device sequence for resource scheduling, is the returned sorted index sequence S500, Resource coordination execution (RA execution): Allocate resources according to the scheduling sequence, monitor the deadlock risk in real time and trigger the rollback mechanism, and finally update the knowledge graph and feedback the execution result; S510, Resource scheduling execution (RA execution): Dynamically allocate resources according to the resource scheduling sequence of S440; Traverse the scheduling sequence: ; ; Among them, is the scheduling sequence, is the total number of devices that need to allocate resources, is the device index being processed currently; Resource scheduling operation: ; Among them, is the resource requirement vector of device , is the available resource vector, is the assignment operator; Allocation status flag: ; Among them, is the resource scheduling status of device , 1 means allocated, 0 means not allocated; S520, Deadlock detection after allocation (RA execution): Detect the deadlock risk in real time after each allocation; Resource contention degree update: ; Among them, is the latest competitiveness of the resource , is the set of devices that need to allocate resources is the device 's demand for the resource , is the current available quantity of the resource , is the total number of resource types; Deadlock risk re - assessment: ; Among them, is the updated deadlock risk coefficient is the Sigmoid activation function, used to normalize the risk value is the indicator function, which is 1 when the competitiveness is greater than 1, otherwise 0 Risk determination: ; Among them, is the deadlock risk flag, 1 indicates there is a risk, 0 indicates no risk is the deadlock risk threshold is the indicator function; S530, Abnormal rollback mechanism (RA execution): If a deadlock risk is detected, immediately roll back the allocation operation; Rollback condition: ; Among them, is the deadlock risk flag; Resource recovery operation: ; Among them, is the available resource vector is the device 's resource demand vector; Status reset and sequence update: ; Among them, is the allocation status flag of the device is the operation of removing the device from the scheduling sequence ; S540, Execution feedback update (RA execution): Generate a resource scheduling report and update the knowledge graph; Allocation result statistics: ; Among them, is the vector of the total allocated resources For the allocation status of the device ; For the resource requirement vector of the device ; Remaining resource calculation: ; Wherein, is the system remaining resource vector, is the current available resource vector; Knowledge graph update: ; Wherein, is the knowledge graph sub - graph related to resources, is the update operation function, is the remaining resource data for update.

[0019] As Figure 2 shown, based on the steps in the above - mentioned evaluation method, an AGENT device health assessment system based on a knowledge graph is also proposed, including the following modules: Knowledge graph initialization module 101: Construct a resource topology graph to describe the physical / logical connection relationship between devices and resources, load device status data (health degree, priority), load resource capacity data (total available resources, type), establish a structured knowledge base to support collaborative decision - making; Resource requirement modeling module 102: Quantify the demand of devices for various resources, generate a device - resource requirement matrix (devices as rows, resources as columns), clarify the resource dependency relationship between devices (mutual exclusion / sharing), and mark the critical resource path (highly competitive resources); Collaborative deadlock prediction module 103: Maintain an agent (MA) for predicting the device resource requirement sequence, and a resource agent (RA) for detecting deadlock risks in combination with resource constraints, output the deadlock risk probability (high / medium / low), and generate a list of faulty devices (devices that may be stalled due to resource competition); Deadlock avoidance decision - making module 104: Optimize the resource scheduling model (priority scheduling / resource reservation) through the maintenance agent (MA), and the resource agent (RA) quantifies the dynamic deadlock risk (real - time risk coefficient), collaboratively generate an optimal scheduling strategy (balancing reliability and risk), and output a maintenance instruction sequence (device execution order + resource scheduling plan); Resource coordination execution module 105: Allocate resources to devices according to the scheduling sequence, monitor the deadlock risk in real - time (dynamically detect resource competition), trigger a roll - back mechanism (revoke the allocation when the risk exceeds the threshold), update the knowledge graph status (device health degree / resource margin), and feedback the execution result (success / failure / roll - back event).

[0020] As Figure 3As shown above, based on the above methods and systems, the following example is given: a wind farm maintenance scenario, where Figure 3 is the multi-agent collaboration sequence diagram based on the method in this scenario; Scenario description: A certain wind farm needs to maintain 3 wind turbines (WT1, WT2, WT3), and the resources involved include: power resources (to drive the maintenance robot), maintenance tool kits (robotic arms, detectors), and spare parts (gears, bearings); Agent division of labor: Maintenance agent (MA): formulates the maintenance plan and evaluates the health of the wind turbines; Resource agent (RA): manages resource scheduling and monitors the risk of deadlock; Execution steps and content: Step1: Knowledge graph initialization (executed by RA) Build a wind farm knowledge graph, including: Device nodes: wind turbines (WT1, WT2, WT3), maintenance robots (Robot1, Robot2); Resource nodes: power, tool kits, spare parts; Relationships: WT1 requires a tool kit, and Robot1 requires power; Step2: Resource requirement mapping (executed by MA) MA generates a resource requirement matrix schematic table according to the maintenance plan, as shown in Table 1: Table 1: Resource requirement matrix schematic table Step3: Collaborative deadlock detection (collaboration between MA and RA) RA detects resource competition: the tool kit is required by both WT1 and Robot1 at the same time (total quantity = 1, requirement = 2); MA calculates the health of the wind turbine: WT1 has a low health level (priority maintenance is required); Output deadlock risk: The competition for the tool kit leads to a high risk; Step4: Deadlock avoidance decision (collaboration between MA and RA) MA optimizes the allocation: preferentially allocate the tool kit to WT1 (low health level); RA quantifies the risk: the risk coefficient of tool kit competition > threshold; Collaborative decision: Robot1 delays execution and releases the tool kit resources; Step5: Resource coordination execution (executed by RA) RA allocates in sequence: tool kit → WT1; Deadlock monitoring: After allocation, the competition risk drops to a safe range; Update the graph: remaining tool kits = 0, mark WT1 maintenance completed.

[0021] Based on the above steps, the competition risk of the toolkit is actively resolved, the WT1 maintenance task is completed without interruption, the delayed execution of Robot1 avoids resource waste, the power resources are reallocated to WT2 maintenance, no deadlock occurs during the whole process, and the maintenance tasks are advanced orderly according to the health priority.

[0022] The embodiments of the present invention have been described above, but the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms, all of which fall within the protection scope of the present invention.

Claims

1. A method for AGENT device health assessment based on a knowledge graph, characterized in that, It includes the following steps: S100, Knowledge graph initialization: Construct a resource topology graph, load device status and resource capacity data, and provide a structured knowledge basis for collaborative evaluation; The device status includes health and priority; S200, Resource demand modeling: Quantify the device resource demand, generate a device-resource demand matrix, and clarify the dependency relationship of each device on various resources; S300, Collaborative deadlock prediction: The maintenance agent predicts the device resource demand, and the resource agent combines the demand with the resource constraints for deadlock detection, and outputs the deadlock risk probability and the list of faulty devices; S400, Deadlock avoidance decision-making: The maintenance agent optimizes the resource scheduling model, and the resource agent quantifies the dynamic deadlock risk, and collaboratively generates an optimal scheduling strategy and maintenance instructions that take into account reliability and risk; S500, Resource coordination execution: Allocate resources according to the scheduling sequence, monitor the deadlock risk in real time and trigger a rollback mechanism, and finally update the knowledge graph and feedback the execution result.

2. The method for evaluating the health of an AGENT device based on a knowledge graph according to claim 1, wherein, In S100, it specifically includes the following steps: S110, Sensor data acquisition: Collect real-time sensor data from edge devices, including physical quantities such as vibration, temperature, and current; S120, Sensor self-check: Perform anomaly detection on each sensor channel to identify failed sensors; S130, Construct an anti-interference mask matrix: Generate a mask matrix to mask the failed sensor data; S140, Anti-interference feature extraction: Extract robust features through masked graph convolution; S150, Feature compression and transmission: Perform dimensionality reduction processing and send it to the subsequent agent.

3. The method for evaluating the health of an AGENT device based on a knowledge graph according to claim 2, wherein In S200, it specifically includes the following steps: S210, Heterogeneous graph construction: Integrate the compressed feature matrix output by the scheduling agent with the resource status to construct a device-resource heterogeneous graph; S220, Spatial dependence convolution: Aggregate neighbor information through graph convolution to capture the device-resource coupling relationship; S230, Temporal evolution modeling: Integrate the changes in the historical state time window to model the dynamic evolution; S240, Event-triggered momentum update: Respond to external events and correct the graph in real time.

4. The method for evaluating the health of an AGENT device based on a knowledge graph according to claim 3, wherein In S300, it specifically includes the following steps: S310, Device health assessment: Evaluate the device health status based on a multi-objective optimization graph attention network; S320, Resource demand dynamic prediction: Combine historical data with the current state to predict future resource demand; S330, Deadlock risk assessment: Construct a resource competition graph to evaluate the system deadlock risk; S340, Resource dynamic adjustment: Based on a multi-objective optimization resource reallocation strategy.

5. The method for AGENT device health assessment based on a knowledge graph according to claim 4, wherein In S330, the calculation formula for quantifying the deadlock risk is as follows: Competition graph construction: ； Among them, is a set of resource nodes, is a set of resource competition relationship edges, is a set of edge weights, is a competition graph of resources; Edge weight calculation: ； Among them, represents the device occupying resources and requesting resources , is the total number of devices, is the available amount of resources of, represents the requested resources and the occupied resources the edge weight between them; Deadlock risk score: ； ； Among them, are the high-risk threshold and the low-risk threshold respectively, is a web page ranking algorithm used to evaluate the importance of nodes, is the deadlock risk level, which is divided into three levels: High, Medium, and Low, is the overall deadlock risk score of the system.

6. The method for AGENT device health assessment based on a knowledge graph according to claim 5, wherein In S400, it specifically includes the following steps: S410, Resource scheduling optimization modeling: Construct an integer programming model to optimize the resource scheduling scheme; S420, Deadlock risk quantification: Calculate the resource competition degree and quantify the deadlock probability; S430, Multi-objective decision-making optimization: Combine reliability and deadlock risk to solve the Pareto optimal solution; S440, Dynamic scheduling strategy generation: Generate a resource scheduling sequence and preventive maintenance instructions.

7. The method for evaluating the health of an AGENT device based on a knowledge graph according to claim 6, wherein In S420, the calculation formula for quantifying the deadlock risk is as follows: Resource competition degree: ； Among them, is the competition degree coefficient of resource k, is the fault flag of device i, is the indicator function, which is 1 when the condition is satisfied and 0 otherwise, is the device and resource demand matrix, is the total number of resource types, is the resource type index; Deadlock risk coefficient: ； Among them, is the Sigmoid activation function that maps the output to the interval [0, 1], is the deadlock risk coefficient. Approaching 1 indicates a high risk, and approaching 0 indicates a low risk.

8. The method for evaluating the health of an AGENT device based on a knowledge graph according to claim 7, wherein In S430, the calculation formula for multi-objective decision optimization is as follows: Multi-objective function: ； ； Among them, is the reliability score of device i, is the deadlock risk coefficient, is the resource scheduling decision vector, is the health score of device i, is the priority of device i, are the weight coefficients of health and priority respectively; When Constraint adjustment: ； Among them, is the deadlock risk threshold, is the resource slack factor, is the predicted resource demand vector; Solve for optimization: ； ； Among them, is the total system reliability score, is the optimal decision variable combination.

9. The method for evaluating the health of an AGENT device based on a knowledge graph according to claim 8, wherein In S500, it specifically includes the following steps: S510, Resource scheduling execution: Dynamically allocate resources according to the resource scheduling sequence in S440; S520, Deadlock detection after allocation: Detect the deadlock risk in real time after each allocation; S530, Abnormal rollback mechanism: If the deadlock risk is detected, immediately roll back the allocation operation; S540, Execution feedback update: Generate a resource scheduling report and update the knowledge graph.

10. An AGENT device health assessment system based on a knowledge graph, characterized in that, For performing the steps in the method for evaluating the health of an AGENT device based on a knowledge graph as described in any one of claims 1-9, including: Knowledge graph initialization module: Construct a resource topology graph to describe the physical / logical connection relationship between devices and resources, load device status data, load resource capacity data, establish a structured knowledge base, and support collaborative decision-making; Resource demand modeling module: Quantify the demand for various resources by devices, generate a device-resource demand matrix, clarify the resource dependency relationship between devices, and mark the critical resource path; Collaborative deadlock prediction module: Maintain an agent for predicting the device resource demand sequence, and a resource agent for detecting the deadlock risk in combination with resource constraints, output the deadlock risk probability, and generate a list of faulty devices; Deadlock avoidance decision module: Optimize the resource scheduling model through the maintenance agent, quantify the dynamic deadlock risk by the resource agent, collaboratively generate an optimal scheduling strategy, and output a sequence of maintenance instructions; Resource coordination execution module: Allocate resources to devices according to the scheduling sequence, monitor the deadlock risk in real time, trigger the rollback mechanism, update the knowledge graph status, and feedback the execution result.

Images