Power operation risk factor data association retrieval method based on graph topological structure

Abstract

The invention belongs to the technical field of knowledge maps, and particularly relates to a power operation risk factor data association retrieval method based on a map topological structure, and the method comprises the steps: accessing a power plant business system, an intelligent two-ticket system and an intelligent analysis platform, obtaining structured business data and unstructured perception data, and constructing an empowerment heterogeneous risk topological map; evaluating the active risk quality of the node by using the static service level and the dynamic perception feature; calculating a mixed semantic distance in the empowerment heterogeneous risk topological graph; according to the active risk quality and the mixed semantic distance of the node, calculating a total risk tension vector borne by the node; and driving intelligent agent search based on the total risk tension vector, executing a discrete mapping mechanism, counting adsorption times, and when the adsorption times exceed a high-risk aggregation threshold, positioning the adsorption times as a high-risk aggregation node and outputting a result. According to the method, the non-linear association retrieval capability of the hidden superposition risk in the multi-source heterogeneous data is enhanced.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph technology. More specifically, this invention relates to a method for data association and retrieval of power operation risk factors based on knowledge graph topology. Background Technology

[0002] In the modern safety production system of power energy enterprises, with the in-depth advancement of smart grid construction and digital transformation, the core control targets have fully focused on the refined control of work permits and operation permits, as well as outsourced teams and external personnel management. At present, power plants have generally built high-definition video monitoring systems covering the entire production area, as well as comprehensive business management systems including intelligent two-ticket systems for work permit management and power plant business systems, and structured business data such as equipment ledgers and personnel qualifications. These systems have realized data collection and basic applications in their respective fields, providing basic support for standardized operations in power production.

[0003] However, despite the relatively mature hardware and basic software, existing risk retrieval and control models are still hampered by the technical bottleneck of a severe disconnect between physical perception data and logical business data when facing complex and ever-changing power operation scenarios. Specifically, although intelligent analysis platforms utilize image recognition technology to capture on-site violations in real time, the data they generate is treated as isolated visual alarm events and fails to establish a deep semantic mapping with structured data such as high-voltage equipment attributes, personnel special operation qualifications, and live-line operation task status in the power plant's business system. This data silo phenomenon prevents the system from understanding the full picture of on-site operations from a unified perspective, making it difficult for managers to assess the systemic consequences of a single violation in a specific business context.

[0004] More importantly, existing retrieval technologies generally rely on linear threshold judgment rules, lacking in-depth exploration of the complex interaction mechanisms between risk entities. In actual operations, risks are often not isolated, but rather the result of nonlinear superposition of multiple heterogeneous entities coupled in a specific spatiotemporal context. For example, an outsourced worker whose qualifications are about to expire holds a work permit involving a high-risk area and exhibits abnormal loitering behavior near critical equipment. Individually, these indicators may not reach the high-level alarm threshold, but when they are closely aggregated in spatiotemporal and logical terms, the resulting qualitative risk is often fatal. Existing technologies lack a mathematical mechanism to describe the mutual attraction effect between risk entities caused by the logical proximity of their operational relationships or the physical proximity of their spatial locations, making such implicit superposition risks easily filtered or ignored by existing linear rules. Summary of the Invention

[0005] To address the technical problem in existing power operation risk retrieval technologies—namely, the lack of an effective mechanism to quantify the nonlinear superposition risk between heterogeneous entities due to the separation of physical and logical data dimensions—this invention provides a power operation risk factor data association retrieval method based on graph topology. This method includes: accessing power plant business systems, intelligent two-ticket systems, and intelligent analysis platforms to obtain multi-source heterogeneous data; constructing a weighted heterogeneous risk topology graph based on structured business data and unstructured perception data from the multi-source heterogeneous data; and using a graph embedding algorithm to map nodes in the weighted heterogeneous risk topology graph into coordinate vectors in a high-dimensional continuous vector space; processing static business levels using an exponential function and utilizing a logarithmic function. The system processes dynamic sensing features to assess the proactive risk quality of nodes; it calculates the mixed semantic distance between the shortest path hop count and the Euclidean distance between the coordinate vectors of two nodes in a weighted heterogeneous risk topology graph; based on the proactive risk quality and the mixed semantic distance, it calculates the total risk tension vector of a node; it randomly initializes multiple intelligent agents in the vector space, drives the intelligent agent heuristic search based on the total risk tension vector, and performs a discrete mapping mechanism by calculating the Euclidean distance between the intelligent agent and the coordinate vectors of nodes in the weighted heterogeneous risk topology graph; it counts the number of times each node is attracted by intelligent agents, and when the number exceeds the high-risk aggregation threshold, it is located as a high-risk aggregation node and the search results are output.

[0006] This invention establishes a multi-level diagnostic framework that integrates data fusion, risk assessment, and nonlinear correlation retrieval. It can not only uniformly represent multi-source heterogeneous factors such as personnel qualifications, equipment status, work tasks, and on-site events, but also effectively identify implicit superimposed risks that are closely coupled in time, space, and logic but whose single-point indicators have not exceeded limits through gravitational field modeling in vector space and intelligent agent aggregation behavior. It can clearly distinguish between isolated low-risk events and high-risk combination scenarios, significantly improving the earlyness, accuracy, and interpretability of power operation risk identification, and reducing safety hazards and operation and maintenance costs caused by missed and false alarms.

[0007] Preferably, the step of constructing the weighted heterogeneous risk topology map includes: constructing personnel entities, equipment entities, and task entities based on structured business data; constructing visual event entities and environmental data entities based on unstructured perception data; establishing holding relationship edges from personnel entities to task entities, and association relationship edges from task entities to equipment entities, thereby forming a logical bridge connecting personnel and equipment; establishing spatial ownership edges from environmental data entities to area or equipment entities, and dynamic status edges from visual event entities to personnel entities; thereby constructing the weighted heterogeneous risk topology map.

[0008] This invention establishes personnel task holding edges, task equipment association edges, environmental area affiliation edges, and visual event personnel status edges to uniformly model the data of the power plant business system, intelligent two-ticket system, and intelligent analysis platform into a structured heterogeneous graph. This not only preserves the authorization relationships in the business logic but also integrates the positional constraints in the physical space, solving the problem of risk association failure caused by the inability to align entities and semantic fragmentation between multi-source systems.

[0009] Preferably, the active risk quality expression of the node is: In the formula, For nodes exist Proactive risk quality at all times; This is the dynamic-static balance coefficient; It is a natural exponential function; For nodes The static reference value; Let k be the natural logarithm function; k is Index of perceived events; To connect directly to the node A set of perceived events; For time t, the first Abnormal characteristic values ​​of a perceived event; For the first The model credibility coefficient for a perceived event.

[0010] Preferably, the expression for the hybrid semantic distance is: In the formula, For nodes With nodes Mixed semantic distance between them; To assign the shortest path hop count between two nodes in a heterogeneous risk topology graph; Let be the coordinate vector of node i in the high-dimensional continuous vector space; Let be the coordinate vector of node j in the high-dimensional continuous vector space; This represents the Euclidean distance.

[0011] Preferably, the total risk tension vector experienced by the node satisfies the expression: In the formula, For nodes exist The total risk tension vector at any given moment; The set of all nodes in the weighted heterogeneous risk topology graph; for The gravitational coefficient at time t; For node i in Proactive risk quality at all times; Node j in Proactive risk quality at all times; For nodes With nodes Mixed semantic distance between them; The distance decay exponent; To prevent tiny constants with a denominator of zero; Let be the coordinate vector of node i in the high-dimensional continuous vector space; Let be the coordinate vector of node j in the high-dimensional continuous vector space; This represents the Euclidean distance.

[0012] Preferably, the gravitational coefficient satisfies the expression: In the formula, for The gravitational coefficient at time t; The maximum initial gravitational constant; It is a natural constant; This is the time decay factor; The current system time; For the target node The timestamp of the most recent update to proactive risk quality.

[0013] This invention introduces a gravity coefficient, which allows the influence of risk to decay naturally over time. This not only ensures that newly occurring anomalies maintain high weight, but also automatically weakens the interference of historically expired data, effectively improving the adaptability and timeliness of risk assessment to the dynamic evolution of operations.

[0014] Preferably, the step of driving the intelligent agent heuristic search includes: randomly initializing multiple intelligent agents in the vector space, multiplying the velocity vector of the previous moment by a random coefficient and superimposing it with the acceleration vector of the current position, and driving the intelligent agents to accelerate and gather towards the region with the greatest risk tension.

[0015] This invention randomly initializes multiple intelligent agents in the vector space and superimposes the velocity vector of the previous moment by multiplying it by a random coefficient with the acceleration vector of the current position, thereby driving the agents to accelerate and gather towards the region of maximum risk tension. This not only improves search efficiency by utilizing parallel exploration of the group, but also enhances the ability to escape local extrema through random velocity perturbation, achieving fast and robust convergence in high-risk regions.

[0016] Preferably, the discrete mapping mechanism includes: when the intelligent agent moves to a new location, calculating the Euclidean distance between the current coordinates of the intelligent agent and the coordinate vectors of each node in the weighted heterogeneous risk topology map, and attaching the intelligent agent to the node with the closest Euclidean distance.

[0017] Preferably, the output retrieval results include: extracting high-risk aggregation nodes and their neighboring nodes, forming a risk subgraph containing violators, related work orders, and affected equipment, and outputting it.

[0018] This invention counts the number of times each node is attracted by the intelligent agent, and when the risk exceeds the high-risk aggregation threshold, it extracts the node and its neighboring nodes to form a risk subgraph containing the violator, related work orders and affected equipment. It not only focuses on the core risk source, but also automatically associates upstream and downstream entities and outputs a risk event view with a complete business context, which is convenient for accurate handling.

[0019] Preferably, the power operation risk factor data association retrieval method based on graph topology is characterized in that the method further includes: pushing the risk subgraph to the data dashboard of the intelligent analysis platform for highlighted display; judging the risk level of the risk subgraph; if the risk level exceeds the threshold, sending an instruction to the two-ticket system to trigger violation blocking.

[0020] The beneficial effects of this invention are as follows: This invention establishes a multi-level diagnostic framework from data fusion and risk assessment to nonlinear correlation retrieval. It can not only uniformly represent multi-source heterogeneous factors such as personnel qualifications, equipment status, work tasks and on-site events, but also effectively identify implicit superimposed risks that are closely coupled in time, space and logic but whose single-point indicators have not exceeded limits through gravitational field modeling and intelligent agent aggregation behavior in vector space. It can clearly distinguish between isolated low-risk events and high-risk combination scenarios, significantly improve the earlyness, accuracy and interpretability of power operation risk identification, and reduce safety hazards and operation and maintenance costs caused by missed and false alarms. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the power operation risk factor data association retrieval method based on graph topology in this invention; Figure 2 This is a schematic diagram illustrating the distribution of node-initiated risk quality in this invention; Figure 3 This is a schematic diagram illustrating the risk tension field and swarm intelligence retrieval results in this invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0024] This invention discloses a method for data association and retrieval of power operation risk factors based on graph topology, referring to... Figure 1 This includes steps S1 to S4: S1: Connect to the power plant business system, the intelligent two-ticket system, and the intelligent analysis platform to obtain structured business data and unstructured perception data, and construct an empowered heterogeneous risk topology map.

[0025] It should be noted that in power operation safety management, the original business data and perception data are collected from independently operating power plant business systems, intelligent two-ticket systems, and intelligent analysis platforms, respectively. These business data and perception data have significant heterogeneity in structure, semantics, and temporal sequence. If they are directly used for risk assessment, it will lead to problems such as misalignment between personnel qualification status and on-site violations, disconnect between equipment attributes and work task logic, and separation between dynamic events and static risks, which will seriously affect the accuracy and completeness of risk association retrieval. Therefore, this step first integrates multi-source heterogeneous data such as business data and perception data to construct a weighted heterogeneous risk topology map under a unified semantic framework, and mathematizes it into a computable vector representation, providing a structured foundation for subsequent nonlinear risk transmission modeling.

[0026] First, standardized interfaces are used to connect with the power plant's business system, the intelligent two-ticket system, and the intelligent analysis platform to obtain structured business data and unstructured perception data. Based on the acquired multi-source heterogeneous data, five core entities are constructed: personnel entities are constructed by extracting personnel ID, job type, affiliated unit, and qualification status from the outsourcing management module of the intelligent two-ticket system; equipment entities are constructed by obtaining KKS codes, equipment names, rated voltages, and location information from the equipment ledger of the power plant's business system; task entities are constructed by obtaining work order numbers, work content, and risk levels from the intelligent two-ticket system; visual event entities are constructed by using YOLOv5 and ResNet-50 algorithms deployed on edge servers to analyze video streams in real time and extract events such as not wearing safety helmets and area intrusion; and environmental data entities are constructed by synchronously accessing micro-meteorological or sulfur hexafluoride sensor readings.

[0027] Furthermore, establishing edges between entities based on business logic includes: Logical bridge construction: Establish holding relationship edges between personnel entities and task entities, and association relationship edges between task entities and equipment entities. It should be noted that at this time, the task entity acts as an intermediary node connecting the physical world and the management world, connecting personnel entities and equipment entities in the logical topology. This ensures that when calculating semantic relevance in the future, the shortest path hop count between personnel entities and equipment entities is limited to within 2 hops, thereby establishing a close and traceable business logic association.

[0028] Spatial background construction: Establish spatial ownership edges for equipment entities, camera-pointed areas, and environmental data entities pointing to areas; when environmental sensor data is abnormal, it serves as a regional background risk field, which will raise the risk attributes of all equipment, operations, and personnel in the region, realizing the global impact model of environmental anomalies on local operation scenarios, thereby constructing a complete weighted heterogeneous risk topology map.

[0029] Finally, the weighted heterogeneous risk topology graph is trained using TransE or Node2Vec graph embedding algorithms. While preserving the structural proximity and semantic relevance between nodes, each node is mapped to a coordinate vector in a high-dimensional continuous vector space. The dimension of this coordinate vector can be adjusted according to the scale of the weighted heterogeneous risk topology graph and the computing resources. In this embodiment, it is set to 128 dimensions. In other embodiments, implementers can set the vector dimension to any reasonable value between 64 and 256 according to the actual deployment scenario or performance requirements, as a mathematical benchmark for subsequent hybrid semantic distance calculation and intelligent agent discrete mapping.

[0030] S2: Utilize static business levels and dynamic perception features to assess the proactive risk quality of nodes.

[0031] It should be noted that in the risk retrieval framework based on the gravitational field model, the mass of different entities determines the magnitude of their risk attraction. However, existing static risk levels cannot reflect violations occurring in real time, while simple video alarm events lack business background support such as personnel qualifications and equipment attributes, leading to risk assessments lagging behind actual working conditions or false alarms caused by isolated false alarms. Therefore, this step defines an active risk quality that integrates static business levels and dynamic perception features, comprehensively reflecting the true degree of danger of an entity at the current moment. The expression for active risk quality is: ; In the formula, For nodes exist Proactive risk quality at all times; This is the static-dynamic balance coefficient, used to adjust the contribution ratio of static and dynamic risks. If the value is too high, the system will be slow to respond to sudden violations; if If the value is too small, it will be easily affected by false alarms. Therefore, it is set to 0.6 in this embodiment. In other embodiments, it can be adjusted to a reasonable value between 0.4 and 0.8 according to the false alarm rate and business sensitivity. It is a natural exponential function; For nodes The static benchmark value is derived from the risk level of the work ticket in the intelligent two-ticket system or the equipment importance in the power plant business system. The high-risk task entity itself is a huge source of quality, amplifying the risk weight of the personnel and equipment connected to it. Let k be the natural logarithm function; k is Index of the set of perceived events; To connect directly to the node The set of perceived events includes not only visual violations identified by AI analysis, but also sensor anomalies triggered by environmental data entities. For time t, the first The abnormal feature values ​​of each perceived event are taken from the classification confidence of the AI ​​intelligent analysis platform; For the first The model credibility coefficient for each perceived event is preset based on the algorithm type and historical accuracy.

[0032] Among them, the exponent term Used to ensure that high-risk entities maintain a high basic quality, possessing significant risk attraction even without real-time events; for several items This causes the quality of proactive risk to increase non-linearly with the number and intensity of perceived events. When a single person commits a minor violation, the increase is gradual, but when multiple people commit intensive violations or high-confidence events overlap, the quality jumps significantly. This accurately portrays the cumulative effect and saturation characteristics of risk, avoiding excessive amplification caused by linear superposition.

[0033] For example, such as Figure 2 This is a schematic diagram illustrating the distribution of active risk quality at nodes in this invention. The scattered points represent physical nodes in the power scenario, and the size of the node area represents the level of active risk quality. The larger the area, the higher the active risk quality. The nodes ID:0 in the lower left corner and ID:1 in the upper right corner of the figure show significantly larger areas, indicating that these two nodes have extremely high active risk quality due to the superposition of high static risk benchmark and high dynamic violation events, providing core quality source points for the subsequent construction of risk tension field.

[0034] S3: Calculate the hybrid semantic distance in the weighted heterogeneous risk topology graph; calculate the total risk tension vector of the node based on the node's active risk quality and hybrid semantic distance.

[0035] It should be noted that in power operation scenarios, the correlation between entities is not determined solely by physical distance. Physically adjacent equipment may be completely isolated in electrical logic, while personnel in the central control room and equipment in the substation, which are far apart, may be directly linked by a work order. If risk gravity is calculated solely based on Euclidean distance in vector space, the actual risk transmission path will be seriously misjudged. Conversely, if only the graph hop count is relied upon, the fine-tuning effect brought about by semantic similarity is ignored. Therefore, this step defines a hybrid semantic distance that integrates the active risk quality of the weighted heterogeneous risk topology graph and the hybrid semantic distance to accurately characterize the real risk coupling strength between nodes. Based on this, the total risk tension vector of each node is calculated to model the nonlinear transmission and aggregation effects of risk in multi-source heterogeneous networks.

[0036] Specifically, firstly, based on the weighted heterogeneous risk topology graph and node coordinate vectors, the coordinates of any two nodes are calculated. and Mixed semantic distance between: ; In the formula, For nodes With nodes Mixed semantic distance between them; To assign weight to the shortest path hop count between two nodes in the heterogeneous risk topology graph, reflecting the degree of correlation in business logic; Let be the coordinate vector of node i in the high-dimensional continuous vector space; Let be the coordinate vector of node j in the high-dimensional continuous vector space; This represents the Euclidean distance.

[0037] This design ensures that logically strongly related nodes maintain a small mixed semantic distance even if the vector distance is slightly large, thereby maintaining high-risk propagation strength.

[0038] Furthermore, combining proactive risk quality with hybrid semantic distance calculation nodes The expression for the total risk tension vector at time t is: ; In the formula, For nodes exist The total risk tension vector at any given moment, whose direction is only used to guide the trend of the intelligent agent's movement in the embedded space, does not represent physical displacement; The set of all nodes in the weighted heterogeneous risk topology graph; for The gravitational coefficient at time t; For node i in Proactive risk quality at all times; For node j in Proactive risk quality at all times; For nodes With nodes Mixed semantic distance between them; The distance decay index is used to control the local intensity of risk impact decay with distance. The larger the value, the faster the risk tension decays with the increase of mixed semantic distance, and the more concentrated the scope of action is on nearby high-risk nodes. Therefore, it is set to 2.0 in this embodiment. In other embodiments, the value can be adjusted in the range [1.5, 3.0] according to the complexity of the operation scenario and the requirements of the risk transmission range. To prevent tiny constants with a denominator of zero; Let be the coordinate vector of node i in the high-dimensional continuous vector space; Let be the coordinate vector of node j in the high-dimensional continuous vector space; This represents the Euclidean distance.

[0039] The expression constructs a risk gravitational field with physical significance: when two nodes are closely related in their business, have high risk quality, and have high event novelty, they will generate a strong mutual attraction tension, driving intelligent agents to gather in such areas, thereby effectively identifying the implicit superimposed risks formed by the spatiotemporal coupling of qualification thresholds, high-risk operations, and on-site violations.

[0040] It is important to note that power operation risks are highly time-sensitive. Newly triggered violations or sudden anomalies should receive the highest priority, while the impact of older events should naturally diminish over time to avoid historical data interfering with current judgments. Therefore, this invention does not set the gravity coefficient as a constant, but rather defines it as a function that dynamically decays with the most recent risk update time of the target node. The gravity coefficient satisfies the following expression: ; In the formula, for The gravitational coefficient at time t; The maximum initial gravitational constant represents the maximum gravitational intensity that can be generated when a risk event just occurs. If the value is too small, the system will not be able to respond adequately to sudden high risks. If the value is too large, it will be prone to oscillation due to instantaneous disturbances. Therefore, it is set to 20 in this embodiment. In other embodiments, it can be reasonably adjusted within the range [10,50] according to the on-site risk sensitivity and system stability requirements. It is a natural constant; As a time decay factor, it controls the rate at which the impact of risk decays over time. If the value is too large, the risk attraction will disappear quickly, which may cause the intelligent agent to fail to capture transient high-risk events in time; if the value is too small, historical risks will remain for a long time, which may easily cause false aggregation. Therefore, it is set to 0.05 in this embodiment. In other embodiments, it can be flexibly set in the range [0.01, 0.1] according to the operation rhythm and the characteristics of risk persistence. The current system time; For the target node The timestamp of the most recent update to proactive risk quality.

[0041] Under this mechanism, when the difference between the current system time and the target node's most recent risk update time approaches zero, At time 1, the gravitational coefficient approaches the maximum initial gravitational constant, at which point gravity is strongest and can rapidly attract intelligent agents to gather at that node; as this time difference gradually increases, The gravitational coefficient at any given time decays rapidly according to an exponential law, causing the influence of expired risks to naturally diminish over time and gradually fade from the current risk focus. This enables the system to adaptively focus on high-risk events while effectively suppressing the interference of historical noise, ensuring that the risk tension field always reflects the most pressing security threats.

[0042] S4: Driven by the total risk tension vector, intelligent agent search is performed and a discrete mapping mechanism is executed. The number of adsorptions is counted. When the number of adsorptions exceeds the high-risk aggregation threshold, the node is identified as a high-risk aggregation node and the result is output.

[0043] It should be noted that the total risk tension vector is a vector field in a continuous vector space, which itself represents the strength and direction of risk attraction. However, the ultimate goal of power operation risk management is to locate specific business entities. To this end, this step introduces intelligent agent search, transforming the total risk tension vector into an operable search engine: multiple intelligent agents move autonomously in the vector space based on local tension fields, naturally tending towards the region with the strongest attraction. Subsequently, through a discrete mapping mechanism, the position of the intelligent agents in the continuous space is precisely anchored to the nearest business node in the graph. This preserves the global guidance capability of the risk field while achieving a connection from mathematical representation to actual objects, thus achieving efficient risk retrieval. Finally, a high-risk aggregation threshold is set as the decision boundary to ensure that a high-risk judgment is triggered only when a risk event forms a significant cumulative effect in the business logic, effectively filtering out instantaneous noise interference and achieving a reliable mapping from the mathematical model to the actual business object.

[0044] Specifically, random initialization in a high-dimensional continuous vector space One intelligent agent; the number of intelligent agents It is a key hyperparameter that affects retrieval coverage and computational efficiency, and needs to be set empirically in combination with the size of the map.

[0045] Regarding the initial number of smart agents Settings: If If the proxy is too small, the sparse proxy cannot cover the vast high-dimensional vector space and is prone to getting trapped in a local optimum gravitational trap, causing it to miss globally high-risk aggregation centers that are far away; if While an excessively large number of agents can improve coverage, it leads to a linear increase in computational overhead, and oversaturated agents will repeatedly stack at the same high-risk nodes, wasting computing resources and defeating the purpose of low-overhead retrieval. Therefore, the number of intelligent agents should be limited. The recommended value range is 50% to 150% of the total number of nodes in the current heterogeneous risk topology graph; in this embodiment, considering the real-time requirements and hardware computing power limitations of power operation scenarios, the number of intelligent agents is... Set to 100% of the total number of nodes in the graph to achieve the best balance between coverage and computational efficiency.

[0046] Each intelligent agent updates its motion state based on the local risk tension vector of its location; An agent at any time velocity vector Update using the following expression: ; In the formula, For the first A smart agent in The velocity vector at any given moment; For the first A smart agent in The velocity vector at any given moment; for The random coefficients within the interval are used to introduce exploratory perturbations to prevent the population from falling into local extrema too early; For the first A smart agent in The acceleration vector at time t, interpolated from that point to obtain the total risk tension vector. Decide.

[0047] It should be noted that the total risk tension vector calculated in step S3 While existing only at discrete graph node coordinates, the intelligent agent roams in a continuous vector space, and its location may not perfectly coincide with any node coordinate. Therefore, this invention adopts a nearest neighbor field strong mapping strategy, assuming that the force experienced by the agent is dominated by the entity node with the nearest spatial location. The specific calculation logic is as follows: at time... Calculate the first The Euclidean distance between the current position coordinates of each intelligent agent and the coordinate vectors of all nodes in the graph is used to identify the node with the smallest Euclidean distance. The total risk tension vector calculated for this node in step S3 is then directly assigned to the intelligent agent at time [time value missing]. acceleration vector This mechanism ensures that agents can sense the dominant risk attraction within their neighborhood and are thus drawn to high-risk areas; Through this update mechanism, intelligent agents naturally and rapidly converge towards areas of highest risk tension.

[0048] Furthermore, a discrete mapping mechanism is executed: when the... After a smart agent completes a move, it calculates the Euclidean distance between its current coordinates and the coordinate vectors of the weighted heterogeneous risk topology graph node, and then attaches the smart agent to the nearest weighted heterogeneous risk topology graph node, thereby completing the mapping from the continuous search space to discrete business entities.

[0049] Based on this, the cumulative number of times each weighted heterogeneous risk topology node is absorbed by the intelligent agent is counted; let a certain node... The number of hits is ,when When a node is identified as a high-risk aggregation node, the system considers the risk retrieval to have converged. The preset high-risk aggregation threshold is set to 5% of the total number of all intelligent agents.

[0050] Finally, all high-risk aggregation nodes that meet the conditions and their neighboring nodes within 1 to 2 hops are extracted to construct a risk subgraph containing violators, related work tickets, and affected equipment. This risk subgraph is then pushed to the data dashboard of the AI ​​intelligent analysis platform for highlighting. If the overall risk level of the risk subgraph exceeds a preset threshold, an instruction is automatically sent to the intelligent two-ticket system to trigger the violation operation lockout, blocking the execution of high-risk processes and achieving closed-loop management from risk identification to intervention and handling.

[0051] For example, such as Figure 3 This diagram schematically illustrates the risk tension field and swarm intelligence retrieval results in this invention. The arrows represent the calculated risk tension field, and the arrow direction and length represent the gravitational direction and magnitude, respectively. A distinct gravitational trap region is formed near ID:0 and ID:1. The search trajectory of the intelligent agent shows that it accelerates and gathers towards the high-risk region from its initial random position, eventually stabilizing at the center of the gravitational trap. The high-risk aggregation nodes ID:0 and ID:1 in the diagram are connected through implicit business association, verifying that this scheme overcomes the physical spatial distance barrier by using hybrid semantic distance, effectively identifying the superimposed risks formed by the coupling of personnel qualifications, work tasks, and on-site violations.

Claims

1. A method for data association and retrieval of power operation risk factors based on graph topology, characterized in that, include: Access the power plant business system, the intelligent two-ticket system, and the intelligent analysis platform to obtain multi-source heterogeneous data; Based on structured business data and unstructured perception data from multi-source heterogeneous data, a weighted heterogeneous risk topology map is constructed, and a graph embedding algorithm is used to map the nodes in the weighted heterogeneous risk topology map into coordinate vectors in a high-dimensional continuous vector space. The exponential function is used to process static business levels and the logarithmic function is used to process dynamic perception features, thereby assessing the proactive risk quality of nodes. Calculate the mixed semantic distance between the shortest path hop count and the Euclidean distance between the coordinate vectors of the two nodes in a weighted heterogeneous risk topology graph; calculate the total risk tension vector of the node based on the node's active risk quality and the mixed semantic distance. Multiple intelligent agents are randomly initialized in the vector space. The intelligent agent heuristic search is driven by the total risk tension vector. The discrete mapping mechanism is performed by calculating the Euclidean distance between the coordinate vectors of the intelligent agents and the nodes in the weighted heterogeneous risk topology graph. The number of times each node is attracted by the intelligent agents is counted. When the number of times exceeds the high-risk aggregation threshold, it is located as a high-risk aggregation node and the search results are output.

2. The power operation risk factor data association retrieval method based on graph topology structure according to claim 1, characterized in that, The steps for constructing the weighted heterogeneous risk topology map include: Based on structured business data, construct personnel entities, equipment entities, and task entities; based on unstructured perception data, construct visual event entities and environmental data entities, establish holding relationship edges from personnel entities to task entities, and association relationship edges from task entities to equipment entities, thereby forming a logical bridge connecting personnel and equipment; establish spatial affiliation edges from environmental data entities to area or equipment entities, and dynamic status edges from visual event entities to personnel entities; thereby constructing an empowered heterogeneous risk topology map.

3. The power operation risk factor data association retrieval method based on graph topology structure according to claim 1, characterized in that, The active risk quality expression for the node is: ； In the formula, For nodes exist Proactive risk quality at all times; This is the dynamic-static balance coefficient; It is a natural exponential function; For nodes The static reference value; Let k be the natural logarithm function; k is Index of perceived events; To connect directly to the node A set of perceived events; For time t, the first Abnormal characteristic values ​​of a perceived event; For the first The model credibility coefficient for a perceived event.

4. The power operation risk factor data association retrieval method based on graph topology structure according to claim 1, characterized in that, The expression for the hybrid semantic distance: ； In the formula, For nodes With nodes Mixed semantic distance between them; To assign the shortest path hop count between two nodes in a heterogeneous risk topology graph; Let be the coordinate vector of node i in the high-dimensional continuous vector space; Let be the coordinate vector of node j in the high-dimensional continuous vector space; This represents the Euclidean distance.

5. The power operation risk factor data association retrieval method based on graph topology according to claim 1, characterized in that, The total risk tension vector experienced by the node satisfies the expression: ； In the formula, For nodes exist The total risk tension vector at any given moment; The set of all nodes in the weighted heterogeneous risk topology graph; for The gravitational coefficient at time t; For node i in Proactive risk quality at all times; Node j in Proactive risk quality at all times; For nodes With nodes Mixed semantic distance between them; The distance decay exponent; To prevent tiny constants with a denominator of zero; Let be the coordinate vector of node i in the high-dimensional continuous vector space; Let be the coordinate vector of node j in the high-dimensional continuous vector space; This represents the Euclidean distance.

6. The power operation risk factor data association retrieval method based on graph topology according to claim 5, characterized in that, The gravitational coefficient satisfies the expression: ； In the formula, for The gravitational coefficient at time t; The maximum initial gravitational constant; It is a natural constant; This is the time decay factor; The current system time; For the target node The timestamp of the most recent update to proactive risk quality.

7. The power operation risk factor data association retrieval method based on graph topology according to claim 1, characterized in that, The steps of driving the intelligent agent heuristic search include: randomly initializing multiple intelligent agents in the vector space, multiplying the velocity vector of the previous moment by a random coefficient and superimposing it with the acceleration vector of the current position, and driving the intelligent agents to accelerate and gather towards the region with the greatest risk tension.

8. The power operation risk factor data association retrieval method based on graph topology according to claim 1, characterized in that, The discrete mapping mechanism includes: When the intelligent agent moves to a new location, the Euclidean distance between the intelligent agent's current coordinates and the coordinate vectors of each node in the weighted heterogeneous risk topology graph is calculated, and the intelligent agent is attracted to the node with the closest Euclidean distance.

9. The power operation risk factor data association retrieval method based on graph topology according to claim 1, characterized in that, The output search results include: Extract high-risk aggregation nodes and their neighboring nodes to form a risk subgraph containing violators, related work orders, and affected equipment, and output it.

10. The power operation risk factor data association retrieval method based on graph topology according to claim 9, characterized in that, The method further includes: The risk subgraph is pushed to the data dashboard of the intelligent analysis platform for highlighting; the risk level of the risk subgraph is determined, and if the risk level exceeds the threshold, an instruction is sent to the two-ticket system to trigger the violation blocking.

Images