A vulnerability closed-loop treatment method and system based on intelligent collaboration
By constructing a multi-dimensional vulnerability semantic model and state perception diagram, combined with a time-series closed-loop prediction model and multi-source verification, the problems of dynamic adaptability and full-process control in existing vulnerability handling methods are solved, and efficient and accurate vulnerability closed-loop management is achieved.
Patent Information
- Application Number
- CN202511079445.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-04
- Publication Date
- 2025-10-10
- Estimated Expiration
- 2045-08-04
AI Technical Summary
Existing vulnerability handling methods cannot dynamically adapt to changes in the asset environment and lack high-precision perception and control of the entire process, resulting in false repairs, missed repairs, and repeated repairs. They also lack unified semantics and time-sensitivity modeling, affecting vulnerability management efficiency and security situation awareness.
Build a multi-dimensional vulnerability semantic model, generate a state perception diagram, predict the probability of vulnerability closure through a time-series closed-loop prediction model, dynamically divide responsibilities and optimize the disposal process, combine incremental learning and multi-source verification, and realize intelligent management of the entire process from vulnerability discovery to closure.
It achieves high-precision perception and control of the entire vulnerability process, improves vulnerability repair efficiency and response accuracy, reduces multiple transfer rates, improves task hit rates and inter-system state consistency, and is suitable for large-scale distributed security environments.
Smart Images

Figure CN120579194B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of information security management technology, and in particular to a closed-loop vulnerability handling method and system based on intelligent collaboration. Background Art
[0002] As information systems become increasingly complex, vulnerability management within enterprises faces challenges such as widespread asset distribution, a wide variety of systems, and fragmented management responsibilities. Traditional vulnerability management methods rely primarily on static rules and targeted dispatch mechanisms. These methods are unable to dynamically adapt to real-time changes in the asset environment, nor can they predict and evaluate vulnerability management effectiveness. This often leads to issues such as "false fixes," "missed fixes," and "duplicate fixes," impacting an organization's overall security situational awareness.
[0003] In addition, most existing technologies focus on vulnerability scanning and asset identification, ignoring the dynamic propagation path of vulnerability information, intelligent attribution of responsibility boundaries, and modeling of logical dependencies between disposal behaviors, making it difficult to achieve high-precision perception and effective control of the status of each link in the entire process of vulnerability "from discovery to closure."
[0004] In cross-system scenarios, vulnerability data is often distributed and lacks unified semantics and structure, resulting in inconsistent information and severely restricting the efficiency of coordinated vulnerability resolution. Most existing vulnerability resolution mechanisms lack the ability to model time-sensitive data and are unable to identify key transition points and causal chains in vulnerability state transitions, leading to delays or even interruptions in closed-loop operations.
[0005] Therefore, there is an urgent need to propose an intelligent vulnerability handling method that integrates state-driven modeling, adaptive responsibility assignment, dynamic priority adjustment, and closed-loop state verification mechanism to achieve full-link automatic response, predict repair effects, and optimize collaborative strategies, fundamentally solving the problems of vulnerability handling fragmentation, slow response, and lack of adaptive capabilities in existing technologies. Summary of the Invention
[0006] The purpose of the present invention is to provide a closed-loop vulnerability handling method and system based on intelligent collaboration to address the shortcomings of the background technology.
[0007] To achieve the above objectives, the present invention provides the following technical solutions: a closed-loop vulnerability handling method based on intelligent collaboration, comprising:
[0008] Collect vulnerability data, asset information, and disposal records from multiple independent security systems to build a multi-dimensional vulnerability semantic model;
[0009] Build a state perception graph based on the vulnerability semantic model;
[0010] Extract candidate state sequence paths from the state perception graph. Build a time-series closed-loop prediction model based on the matching degree between the shortest transition path required for the target state to be closed and the historical state trajectory. For each vulnerability to be addressed, predict its closed-loop probability and the required key behavior chain.
[0011] Adaptively divide vulnerability responsibility based on vulnerability impact scope, key business dependency chains, and key nodes in the predicted closed-loop path, generate a dynamic handling responsibility map, and assign vulnerabilities to the optimal responsible individuals.
[0012] Obtain the expected response time, availability status, and task saturation of each node in the responsibility graph, dynamically calculate vulnerability priority weights, and automatically sort and schedule vulnerability handling processes based on priority;
[0013] Monitor the actual behavior nodes and state transitions during the execution of vulnerability remediation tasks, modify the edge weights and transition probabilities in the state perception graph in real time, and update the time series closed-loop prediction model;
[0014] After the vulnerability is fixed and passes multi-source verification, it is automatically marked as closed-loop and the status is synchronized to the original system;
[0015] Output the complete vulnerability handling behavior chain and closed-loop cycle evaluation report for reverse training of the time series closed-loop prediction model.
[0016] Preferably, starting from the state perception graph, all reachable paths from the "current state" to the "closed loop" are enumerated to form a candidate path set P;
[0017] For the vulnerability v to be handled, its historical state trajectory is recorded as T_v, and each path in the candidate path set The matching degree is calculated by weighted summing of the path prefix overlap rate and the time feature matching degree.
[0018] A temporal closed-loop prediction model is established using the occurrence frequency of candidate paths, the historical closed-loop success rate, and the matching degree with the current vulnerability trajectory as input features;
[0019] The output of the model includes the predicted probability of the current vulnerability reaching the "closed loop" status.
[0020] Preferably, obtaining the expected response time, availability status, and task saturation of each node in the responsibility graph and dynamically calculating the vulnerability priority weight includes:
[0021] Obtain the expected response time Te, availability status A, and task saturation L from each responsible node to form the capacity state vector S of the responsible node;
[0022] Calculate the base priority score for the vulnerability itself , the expression is: Where CCVSS represents the CVSS score of the vulnerability; Iasset represents the business importance score of the affected asset; Rexploit represents the possibility of the vulnerability being exploited or the current status of its exploitation; λ1 to λ3 represent the weight coefficients of each factor;
[0023] The node capability state vector S and the basic priority score Vbase are integrated to calculate the comprehensive priority weight W for each vulnerability. The expression is: ;in is the normalized value of Te, and α, β, γ, and δ are system scheduling parameters.
[0024] Preferably, the updating of the time series closed-loop prediction model includes:
[0025] During the vulnerability handling process, we continuously collect task execution logs, status change timestamps, and work order processing behavior chains to generate vulnerability behavior trajectory sequences.
[0026] Map the trajectory sequence to the path nodes in the state perception graph, calculate the actual jump frequency and jump time between each state, and construct a state transition behavior tensor with timing constraints;
[0027] The Bayesian incremental learning method is used to fuse the new behavior tensor and the historical transition probability matrix, dynamically modify the edge weights and transition probabilities in the state graph, and generate a time-varying state graph.
[0028] The path weight and closed-loop reachability function of the temporal closed-loop prediction model are updated based on the time-varying state diagram.
[0029] Preferably, outputting the vulnerability handling behavior chain and closed-loop cycle evaluation report and reversely training the prediction model includes:
[0030] After the vulnerability is closed, the complete handling behavior chain is extracted, recording all state transition nodes, timestamps, and sequences of responsible persons from the initial state to the closed-loop state;
[0031] Counting the dwell time of key nodes and the total cycle of the paths in the behavior chain to construct a closed-loop cycle structure feature set;
[0032] Compare the closed-loop cycle characteristics with the corresponding path estimates in the original prediction model and calculate the error residual distribution;
[0033] Based on error feedback, a structure-preserving transfer learning method is used to reversely train and update the original time series closed-loop prediction model, dynamically optimizing the path estimation accuracy and the importance weights of behavior nodes.
[0034] Preferably, the security system includes a scanning platform, a configuration management database, a security operation platform and a work order system.
[0035] Preferably, the nodes in the state perception graph represent vulnerability state nodes, the edges represent the transition relationship between vulnerability states, and the weight of each edge is adaptively calculated based on the historical state transition probability to support vulnerability state prediction.
[0036] The present invention also provides a closed-loop vulnerability handling system based on intelligent collaboration, comprising:
[0037] The semantic modeling module collects vulnerability data, asset information, and disposal records from multiple independent security systems to build a multi-dimensional vulnerability semantic model;
[0038] The state modeling module builds a state perception diagram based on the vulnerability semantic model;
[0039] The prediction analysis module extracts a set of candidate state sequence paths from the state perception graph, builds a time series closed-loop prediction model based on the matching degree between the shortest transfer path required for the target state to be closed and the historical state trajectory, and predicts the closed-loop probability and the required key behavior chain for each vulnerability to be addressed;
[0040] The responsibility division module adaptively divides vulnerability responsibilities based on the vulnerability impact scope, key business dependency chains, and key nodes in the predicted closed-loop path, generates a dynamic handling responsibility map, and assigns vulnerabilities to the optimal responsible persons;
[0041] The scheduling engine module obtains the expected response time, availability status, and task saturation of each node in the responsibility graph, dynamically calculates vulnerability priority weights, and automatically sorts and schedules vulnerability handling processes based on priority.
[0042] The model evolution module monitors the actual behavior nodes and state transitions during the execution of vulnerability remediation tasks, corrects the edge weights and transition probabilities in the state perception graph in real time, and updates the time series closed-loop prediction model;
[0043] The closed-loop confirmation module automatically marks the vulnerability as closed after it is fixed and passes multi-source verification, and synchronizes the status to the original system;
[0044] Feedback optimization module, output vulnerability handling complete behavior chain and closed-loop cycle evaluation report, which is used for reverse training of time series closed-loop prediction model.
[0045] In the above technical solution, the technical effects and advantages provided by the present invention are:
[0046] 1. This invention, by constructing a multidimensional vulnerability semantic model and state perception graph, integrates path prediction, dynamic responsibility partitioning, and scheduling priority calculation, achieving for the first time structured modeling and intelligent prediction of the entire vulnerability process, from discovery, assignment, resolution, to closure. Compared to existing vulnerability management approaches that rely on manual rules or static mechanisms, this invention is highly scalable and adaptive, automatically identifying key closed-loop behavior chains, predicting vulnerability closure probabilities, and dynamically assigning responsibilities to the optimal individuals based on business impact and resolution capabilities, significantly improving vulnerability remediation efficiency and response accuracy.
[0047] 2. This invention incorporates closed-loop behavioral chain feedback and state graph self-evolution mechanisms. By continuously collecting actual handling behaviors and verification evidence, it dynamically modifies model structure and parameters, thereby constructing a continuously optimized intelligent prediction and scheduling system. This method effectively reduces the rate of multiple dispatches, improves task hit rates, and enhances inter-system state consistency. It provides an integrated vulnerability management system with global awareness, autonomous decision-making, and evolvable capabilities, making it suitable for intelligent security operations and maintenance environments in large-scale distributed scenarios. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.
[0049] Figure 1 This is a mind map of the method of the present invention.
[0050] Figure 2 This is a mind map of the system modules of the present invention. DETAILED DESCRIPTION
[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.
[0052] Example 1, please refer to Figure 1 As shown, the closed-loop vulnerability handling method based on intelligent collaboration described in this embodiment includes:
[0053] Collect vulnerability data, asset information, and disposal records from multiple independent security systems to build a multi-dimensional vulnerability semantic model;
[0054] Build a state perception graph based on the vulnerability semantic model;
[0055] Extract candidate state sequence paths from the state perception graph. Build a time-series closed-loop prediction model based on the matching degree between the shortest transition path required for the target state to be closed and the historical state trajectory. For each vulnerability to be addressed, predict its closed-loop probability and the required key behavior chain.
[0056] Adaptively divide vulnerability responsibility based on vulnerability impact scope, key business dependency chains, and key nodes in the predicted closed-loop path, generate a dynamic handling responsibility map, and assign vulnerabilities to the optimal responsible individuals.
[0057] Obtain the expected response time, availability status, and task saturation of each node in the responsibility graph, dynamically calculate vulnerability priority weights, and automatically sort and schedule vulnerability handling processes based on priority;
[0058] Monitor the actual behavior nodes and state transitions during the execution of vulnerability remediation tasks, modify the edge weights and transition probabilities in the state perception graph in real time, and update the time series closed-loop prediction model;
[0059] After the vulnerability is fixed and passes multi-source verification, it is automatically marked as closed-loop and the status is synchronized to the original system;
[0060] Output the complete vulnerability handling behavior chain and closed-loop cycle evaluation report for reverse training of the time series closed-loop prediction model.
[0061] In this embodiment, a vulnerability information modeling method based on multi-system data fusion is provided to support cross-platform and cross-departmental information security collaborative management.
[0062] First, to address the heterogeneous multi-source data problem prevalent in existing information systems, the system collects vulnerability management-related data from four typical security systems, including:
[0063] Scanning platform: Extract vulnerability scanning results, including vulnerability unique identifier, vulnerability type (such as remote execution, privilege escalation, etc.), CVE number, risk level, first discovery time, current status, verification results and other fields;
[0064] Configuration Management Database (CMDB): Obtain asset information associated with the vulnerability, including the asset's unique number, asset-owning department, asset owner, asset importance level, operating environment (test / production), deployment area, etc.
[0065] Security Operations Platform (SOC): Extracts information related to the vulnerability, including security event alarm logs, associated attack paths, threat behavior tags, and indicators of compromise (IOCs), to enrich vulnerability context.
[0066] Work Order System: Obtain fields such as the creation time, handler, processing status, operation action (such as ignore, repair, verify), and processing time of historical vulnerability handling work orders for modeling the vulnerability lifecycle status.
[0067] To achieve cross-system semantic consistency, after data collection is completed, the system performs standardized cleaning on all types of raw data to eliminate inconsistent naming, missing fields, or redundant information. By constructing a multi-dimensional feature vector space, each vulnerability instance is mapped into a unified semantic entity, including the following core dimensions:
[0068] Vulnerability attribute dimensions: such as vulnerability type, vulnerability severity, exploit complexity, impact scope, etc.
[0069] Asset association dimensions: including business line, asset criticality, equipment type, deployment location, etc.
[0070] Historical behavior dimension: includes the historical repair frequency, number of recurrences, average resolution cycle, etc. of the vulnerability;
[0071] Threat perception dimensions: including whether it is associated with known attack incidents, whether there are publicly available exploit tools, and whether it is marked as an active risk by the intrusion detection system;
[0072] Organizational collaboration dimension: such as which team handled it in the past, whether the person handling it responded promptly, the frequency of task transfer, etc.
[0073] Finally, the above dimensions are constructed into a multi-dimensional vulnerability semantic model with the following characteristics:
[0074] Entity unity: Information about the same vulnerability in different systems is automatically aggregated into a complete instance;
[0075] Time series traceability: retain the timestamp of each field to make the evolution of vulnerability status traceable;
[0076] Semantic scalability: New semantic dimensions can be automatically expanded based on future new access systems;
[0077] Risk correlation: It can realize the graphical correlation modeling between vulnerabilities, threat behaviors, and asset vulnerabilities.
[0078] For example, an "influence" relationship is established between the vulnerability node "Vuln_20250701_001" and the asset node "Asset_XYZ123", an "attribution" relationship is established with the responsible person node "Zhang Mou", and a "tracking" relationship is established with the work order node "Ticket_20250701123", thus forming a closed-loop tracking semantic structure.
[0079] The multi-dimensional vulnerability semantic model realized in the foregoing manner not only breaks through the static data silos among multiple systems, but also supports vulnerability context behavior analysis and subsequent closed-loop path prediction, thereby providing a unified data basis and modeling entry for subsequent intelligent collaborative disposal.
[0080] In this embodiment, to realize modeling and prediction of the vulnerability life cycle state transition process, a state-aware graph structure (hereinafter referred to as a “state graph”) is further constructed based on the foregoing multi-dimensional vulnerability semantic model, to express state nodes in the vulnerability disposal process and the transition relationship therebetween, and to realize intelligent prediction of future state trends through learning of historical state evolution behaviors.
[0081] First, core state nodes in the vulnerability life cycle are defined as basic constituent units of the state graph. According to actual vulnerability management processes and industry specifications, the state nodes can include, but are not limited to, the following types:
[0082] “Detected”, “Assigned”, “In Progress”, “Verified”, “Closed”, “Rejected”, and “On Hold”;
[0083] Each node represents a stage of the vulnerability in the life cycle, and the state attribute can be extracted from the “current state” and “state change record” fields in the vulnerability semantic model.
[0084] Subsequently, historical path data of the vulnerability migrating from one state to another state is analyzed. Time series data collected in multiple systems (such as a ticket system and a scanning platform) can provide the following information:
[0085] State transition order (for example, “Detected” → “Assigned” → “In Progress” → “Verified” → “Closed”);
[0086] Time interval of each migration;
[0087] Whether there is an abnormal jump in the transition (for example, “Assigned” directly changes to “Closed”).
[0088] Based on the foregoing historical path samples, edge relationships in the state graph are constructed, where each edge represents a transition path between two state nodes, and a transition probability is assigned as a weight of the edge. The weight is calculated by counting the frequency of occurrence of a certain state transition in historical data. For example, among 1000 vulnerabilities, 800 vulnerabilities transition from “In Progress” to “Verified”, and thus the initial transition probability of the edge is 0.8.
[0089] To enhance the model's adaptability, this embodiment introduces an adaptive learning mechanism. As new vulnerability status change records are continuously added, the system dynamically adjusts the transition weights of each edge in the state graph. For example, if a large number of vulnerabilities recently transition directly from "In Progress" to "Closed," the transition probability of that path will be increased accordingly, reflecting the actual trend of the current handling process.
[0090] A core advantage of statecharts is their ability to predict the most likely future state paths for a vulnerability. For example, if a vulnerability is currently in the "In Progress" state, the system can predict its next possible state (such as "Verified" or "Pending") and the average time to reach that state based on the multiple edges and their weights originating from the "In Progress" node in the statechart, thereby supporting subsequent priority adjustments and task scheduling.
[0091] In addition, the state diagram constructed in this embodiment also supports the following advanced functions:
[0092] State anomaly identification: When the vulnerability state transition path deviates significantly from the regular path in the state diagram, the system can prompt an anomaly, such as "bypassing verification and directly closing the loop".
[0093] Strategy evaluation: By performing cluster analysis on the transfer probability of each path, the handling efficiency and execution standardization of different responsible units or system modules can be evaluated.
[0094] In summary, by constructing a state-aware graph structure, we have achieved the formal expression and statistical modeling of the state change logic during the vulnerability handling process, providing a solid data foundation and intelligent support for the subsequent prediction engine and collaborative scheduling mechanism.
[0095] This embodiment aims to analyze the historical vulnerability state evolution behavior, extract typical closed-loop paths from the state perception diagram, and build a time-series closed-loop prediction model to achieve closed-loop trend prediction, disposal cycle estimation, and key behavior prompts for newly discovered vulnerabilities, thereby improving the intelligence and precision of vulnerability repair.
[0096] The vulnerability closure process is essentially a state evolution path consisting of a series of state nodes arranged in chronological order, ultimately leading to the target state of "closed loop." The core purpose of the "Time Series Closed Loop Prediction Model" is to:
[0097] Based on the historical trajectory of actual vulnerability handling, common and efficient closed-loop path sequences are summarized;
[0098] Compare the similarity between the current vulnerability status evolution process and the historical closed-loop path;
[0099] Output the predicted probability of the current vulnerability reaching the "closed loop" state, as well as its most likely key behavior chain.
[0100] This model not only considers the static state transition frequency, but also the time dependence of the behavior sequence and the structural shortest state path. It is a time-aware path evaluation mechanism.
[0101] Extract candidate state path set: Starting from the state perception graph, use breadth-first search (BFS) or depth-first search (DFS) algorithm to enumerate all reachable paths from "current state" to "closed loop" to form a candidate path set Each path represents an ordered sequence of state nodes, and n is the total number of paths; for example:
[0102] : "Discovered" → "Assigned" → "Processing" → "Verified" → "Closed loop";
[0103] : "Discovered" → "Processing" → "Pending" → "Verified" → "Closed";
[0104] For each path, record its length, state jump sequence, historical frequency and other attributes.
[0105] Define the path structure matching index: For the vulnerability v to be addressed, its historical state trajectory (such as the most recent state transition sequence) is recorded as T_v, and each path in the candidate path set is matched with the path structure matching index. Perform matching calculation.
[0106] The following elements can be used to calculate the matching degree:
[0107] Path prefix overlap rate: the state sequence in T_v and Prefix matching length;
[0108] Time feature consistency: the degree of deviation between the average time of state jump and the historical mean.
[0109] The result is the matching score , which ranges from [0, 1] and is calculated by weighted summing of the path prefix overlap rate and the time feature matching degree.
[0110] Constructing a closed-loop prediction model: Using the frequency of candidate paths, the historical closed-loop success rate, and the degree of match with the current vulnerability trajectory as input features, we leverage models such as Bayesian classification, Markov chains, or graph neural networks (GNNs) to build a temporal closed-loop prediction model P(Close | T_v). Close represents a terminal state node in the state perception graph, equivalent to "closed" or "resolved."
[0111] The output of the model includes: the predicted probability of the current vulnerability reaching the "closed loop" state; the path sequence that is most likely to achieve a closed loop and the key behavior nodes it contains (key behavior nodes such as those that require manual verification, rely on a certain security module to perform isolation operations, etc.).
[0112] In the specific implementation, the system can set a probability threshold T. When P<T, the system will automatically mark the vulnerability as "high closed-loop risk" and trigger the auxiliary intervention process; when P≥T, the system will recommend the optimal path and disposal action suggestions to improve the decision-making efficiency of the responsible person.
[0113] The closed-loop time-series prediction model constructed in this embodiment enables highly accurate predictions of future vulnerability state evolution trends, overcoming the limitations of traditional vulnerability management mechanisms that rely solely on current static states for decision-making. This model provides forward-looking insights into the entire vulnerability management process, enhancing the precision and intelligence of vulnerability management. It is particularly suitable for large-scale information system environments with large volumes of vulnerabilities, widespread distribution, and complex collaboration.
[0114] This embodiment addresses the problems of static rules, one-sided assignments, and unclear responsibilities in traditional vulnerability responsibility division methods. It proposes a responsibility attribution determination method that integrates the vulnerability impact scope, key business dependency chain, and key nodes of the closed-loop path. By constructing a dynamic handling responsibility graph, it achieves adaptive optimization of vulnerability work order assignment.
[0115] Specifically, the following steps are used to complete the establishment and implementation of the mechanism:
[0116] First, the system builds a business dependency graph of the vulnerability's impact based on the associated asset information extracted from the vulnerability semantic model and the business system topology in the configuration management database (CMDB). This graph, based on asset nodes, annotates their upstream and downstream relationships within the business chain and their critical path locations, identifying high-priority business-impacting paths.
[0117] Secondly, using the optimal closed-loop path structure output by the aforementioned predictive closed-loop model, the system extracts key behavioral nodes in the path. These nodes usually correspond to key roles or system modules that need to actually perform operations, such as patch distribution modules, manual review links, security baseline control points, etc.
[0118] By matching the above key node information with the organizational responsibility matrix, the system identifies a set of potential responsible entities, including technicians directly responsible for asset maintenance, business system teams, operation and maintenance platform interface modules, etc.
[0119] Next, the system introduces a dynamic capability evaluation mechanism to calculate the responsibility suitability of each candidate responsible entity. The suitability is based on the following three elements:
[0120] Historical vulnerability resolution time (i.e., average resolution cycle);
[0121] Current task load (task saturation);
[0122] Real-time availability status (whether online or in a maintenance window);
[0123] By normalizing and weighting the above factors, a "node capability vector" is formed to characterize the adaptability of the node to specific vulnerability handling tasks at the current moment.
[0124] Finally, based on the degree of matching between key nodes in the closed-loop path structure and candidate responsible entities, the system generates a dynamic handling responsibility graph with the vulnerability as the root node. Each node in the graph is a responsible person or a responsible system module, the edge represents the task transfer order and collaborative relationship, and the edge weight reflects the responsibility adaptability.
[0125] The responsible node with the highest adaptability in the closed-loop path is used as the main responsible unit to distribute vulnerability work orders, and the remaining key nodes simultaneously generate collaborative notification tasks to achieve collaborative closed-loop tracking during the handling process.
[0126] Through the above method, not only can the refined mapping of "responsibility to behavior" and "path to person" in the vulnerability handling process be achieved, but it can also dynamically adapt to changes in system operating status, personnel scheduling, and task load in different scenarios, and realize the "optimal path decoupling + precise responsibility binding" of vulnerability responsibility division.
[0127] The introduction of this mechanism effectively overcomes the limitations of traditional reliance on manual assignment or static asset attribution logic, improves the accuracy of vulnerability handling responsibility matching, reduces the rate of erroneous dispatch and the probability of task interruption midway, and provides structured guarantees for the efficiency and rate of vulnerability closure.
[0128] This embodiment proposes a dynamic calculation method for vulnerability priority weights for multi-source task capability assessment. It combines the task response capabilities and system availability of each responsible node in the vulnerability handling path to achieve the prioritization of vulnerability remediation tasks and optimal resource scheduling, effectively improving the vulnerability closure efficiency and the overall security response performance of the system.
[0129] First, based on the generated dynamic disposal responsibility graph, all responsibility nodes in the current disposal path are extracted. Each node represents a unit entity, which can be the asset responsible person, system platform module, or technical support team.
[0130] The system obtains the following three key operating parameters from each responsible node in real time:
[0131] Expected response time (Te): based on the average response cycle statistics of historical vulnerability tasks of this node;
[0132] Availability status (A): Comprehensively judge the current handling capacity of the node through indicators such as personnel online status and system interface smoothness;
[0133] Task saturation (L): The ratio of the number of tasks currently received but not yet completed to the maximum processing capacity of the node.
[0134] The above data are standardized to form the capability state vector S = {Te, A, L} of the responsible node, which serves as one of the inputs of the dynamic scheduling mechanism.
[0135] The base priority score of the vulnerability itself is calculated by combining the following information recorded in the vulnerability semantic model: CCVSS represents the CVSS score of the vulnerability (ranging from 0 to 10), indicating the technical severity of the vulnerability. Iasset represents the business importance of the affected assets, quantified by a hierarchical standard (e.g., 0 to 5, corresponding to "non-core systems" to "critical infrastructure systems"). Rexploit indicates the likelihood of the vulnerability being exploited or its current status, based on whether there is a PoC and whether it appears in attack intelligence (e.g., 0: no exploit, 1: PoC exists, 2: already exploited). λ1 to λ3 represent the weight coefficients of each factor, which can be set based on industry practices or model training (e.g., default values: 0.4, 0.3, 0.3).
[0136] Next, the node capability state vector S and the basic priority score Vbase are integrated to calculate the comprehensive priority weight W for each vulnerability using the following formula: ;in is the normalized value of Te, and α, β, γ, and δ are system scheduling parameters used to control the influence weights of different dimensions.
[0137] All vulnerabilities to be addressed are sorted according to their priority weight W to form a vulnerability scheduling queue. The scheduling system completes the work order task push according to the following rules:
[0138] Prioritize assigning high W-value vulnerabilities to responsible nodes with high current availability and low task load;
[0139] Merge similar vulnerability tasks under the same responsibility node to reduce duplication of operations;
[0140] For vulnerabilities with low priority but long waiting time, the system dynamically triggers the privilege escalation mechanism for redistribution evaluation.
[0141] When the status of a responsibility graph node changes (such as improved response capability, task release, or personnel coming online), the scheduling mechanism can recalculate vulnerability priorities in real time and automatically adjust the scheduling results to ensure the optimal match between tasks and resources.
[0142] Through this mechanism, the "value priority + resource matching + adaptive adjustment" triple fusion of vulnerability scheduling is achieved, breaking through the traditional single way of scheduling based on static rules or fixed vulnerability levels, effectively improving the efficiency and accuracy of the vulnerability handling closed loop, and avoiding the risk of resource overload and repair delay.
[0143] To solve the problem of traditional state transition graph static modeling and difficulty in reflecting the dynamic changes of state transition in the vulnerability handling process, an adaptive evolution method of state-aware graph based on behavior trajectory driving is proposed, and combined with the incremental learning mechanism, the dynamic optimization and update of the time series closed loop prediction model are realized.
[0144] In the vulnerability handling process, the system continuously collects data from the work order system, asset change log and scanning platform, including the occurrence time of each state change event, the operation type executed by the responsible node, the response time and the final state change result. After standardization processing, the above data generates a structured "vulnerability behavior trajectory sequence". This sequence takes the time axis as the benchmark, concatenates the jump logic of each state node and the behavior execution node, forming a "state-behavior-time" triple set.
[0145] The behavior trajectory sequence is mapped to the original state-aware graph structure, and by calculating the corresponding relationship between each state jump path in the current trajectory and the existing edge in the graph, the "actual jump frequency" and "jump time interval" are extracted to construct a new "state jump behavior tensor". This tensor is a three-dimensional structure that records state pairs, jump time distribution and behavior execution node characteristics, and has dynamic evolution capability.
[0146] In order to integrate new behavior data without reconstructing the original graph structure, this embodiment introduces the Bayesian incremental learning method, taking the edge weight (transition probability) of the original state-aware graph as the prior distribution and the new behavior tensor data as the observation sample, performing posterior inference, dynamically correcting the transition probability values of each edge in the state graph, and updating the edge weight function in the state transition path to form a time-varying state-aware graph with time sensitivity.
[0147] The updated state graph is input into the time series closed loop prediction model to replace the original static transition structure, which is used to re-evaluate the accessibility, risk value and key node importance of each candidate closed loop path. At the same time, based on the path entropy change, average state residence time and path density distribution in the new graph structure, the system dynamically adjusts the path priority scoring function in the model, thereby improving the recognition ability of irregular or abnormal state jumps.
[0148] Through this mechanism, the system has the ability of continuous learning and model self-evolution of vulnerability handling behavior, and can still maintain high prediction accuracy and scheduling adaptability in complex and nonlinear handling environment, providing accurate support for subsequent priority dynamic adjustment and resource reallocation strategies.
[0149] This embodiment addresses the problem that traditional closed-loop vulnerability judgment methods rely on a single verification source and are prone to misjudgment or missed judgments. It proposes a closed-loop confirmation method that integrates multi-source verification evidence, and uses a trusted scoring mechanism to determine whether the vulnerability status can be marked as "closed-loop". At the same time, it completes cross-system status synchronization to ensure that the vulnerability closed-loop logic chain is complete and reliable.
[0150] After the vulnerability remediation task is completed, the system starts an automatic closed-loop verification process. This process collects closed-loop evidence from the following three data sources in parallel:
[0151] Rescan the platform to verify that the vulnerability has been eliminated from the affected assets;
[0152] Configuration item comparison tool: confirms whether the key configuration after patch application is consistent with the expected status;
[0153] Manual feedback module: collects the final confirmation of the operation completion by the person responsible for the disposal, as well as verification screenshots, operation logs and other auxiliary materials.
[0154] To improve the objectivity and accuracy of judgments, the system designed a closed-loop credibility fusion model. This model is based on a weighted evidence fusion mechanism, which weights each verification source and introduces the following three evaluation indicators:
[0155] Weight of evidence: For example, scan verification is superior to manual subjective confirmation;
[0156] Timeliness of evidence: the time interval between the time when the verification data is generated and the time when the task ends;
[0157] Evidence consistency: whether the verification results from multiple sources corroborate each other or conflict with each other.
[0158] The model inputs the above factors into the scoring function and outputs a closed-loop credibility score C_score, which ranges from 0 to 1. The system sets a closed-loop threshold Q and automatically makes a judgment based on the scoring result: when C_score ≥ Q, the system marks the vulnerability as "closed-loop."
[0159] After the status change is completed, the system automatically synchronizes the status to the original source system through the interface decoupling mechanism, including but not limited to the vulnerability scanning system, work order platform, configuration management database (CMDB), etc., to ensure the consistency of vulnerability status and closed-loop traceability among multiple systems.
[0160] This method breaks through the traditional closed-loop logic that relies on a single verification method and has the following technical advantages:
[0161] Multi-source verification fusion enhances the accuracy of closed-loop judgment;
[0162] The trusted scoring mechanism improves the automation level of closed-loop decision-making;
[0163] The state synchronization mechanism ensures logical consistency and closed-loop response between systems;
[0164] Reduce the burden of manual verification and improve the efficiency and controllability of large-scale vulnerability handling.
[0165] This example proposes a model reverse training method based on a closed-loop vulnerability remediation behavior chain, aiming to continuously refine and enhance the existing time-series closed-loop prediction model. By extracting actual closed-loop path and cycle data, constructing a structural feature set, and combining it with prediction deviations for differential learning, the model's generalization performance and path prediction accuracy are effectively improved.
[0166] When a vulnerability is marked as "closed," the system automatically records its entire lifecycle from discovery to closure, forming a complete action chain. This action chain includes the following information: a list of state sequence nodes (e.g., discovered → assigned → in progress → verified → closed); entry and exit times for each state node; the responsible execution node and operation type; task transfer method (automatic system / manual intervention); and exception flags (e.g., pending, rejected).
[0167] Based on this behavior chain, the system calculates metrics such as dwell time, path hop count, and total path time for key states, constructing a structured closed-loop cycle feature set. This feature set is then used as the actual observation value and compared item by item with the output of the original prediction model to generate error distribution maps at the path and node levels.
[0168] To enhance the model's ability to learn about changes in different temporal paths, this embodiment adopts a structure-preserving transfer learning method to incrementally adjust the transition probability weights, state priority estimation factors, and behavior node importance in the model without destroying the original path structure expression capability.
[0169] This migration strategy is driven by error residuals and focuses on fine-tuning path areas where deviations are concentrated, while avoiding inadvertent modifications to paths with high confidence and stable performance, maintaining model prediction stability and convergence efficiency.
[0170] Finally, the system automatically generates a vulnerability closed-loop cycle assessment report, which includes:
[0171] Actual closed-loop behavior chain structure diagram;
[0172] Path cycle comparison analysis;
[0173] Forecast deviations and correction suggestions;
[0174] Performance comparison indicators before and after model optimization (such as improved prediction accuracy and reduced path error rate).
[0175] Through this mechanism, the system implements a dual-path evolution method of "self-supervised feedback + precise transfer learning" for the time series prediction model, overcoming the limitations of traditional static modeling methods that cannot adapt to actual path changes and disposal strategy adjustments, and effectively improving the accuracy, adaptability and long-term maintainability of vulnerability closed-loop prediction.
[0176] Example 2: To enhance the technical credibility and practicality of this application, the following provides a specific example that demonstrates the beneficial effects of the closed-loop intelligent vulnerability handling method in terms of efficiency, accuracy, and automation capabilities in a data-driven manner, specifically including:
[0177] In this embodiment, a prototype system is deployed in the real production environment of a financial industry customer, connected to its existing vulnerability scanning platform (Nessus), configuration management database (CMDB), security operation center (SOC) and work order system, and the full amount of data during the actual vulnerability handling process is collected to verify the closed-loop efficiency improvement capability and automatic dispatch accuracy of the method described in this application.
[0178] During its three-month operational cycle, the system processed 38,562 valid vulnerability tasks, impacting 2,196 asset nodes, 124 responsible individuals, and 41 business systems. A control group approach was used for pre- and post-test evaluations, with the original manual dispatch mechanism serving as the baseline and the intelligent collaborative approach serving as the experimental group.
[0179] Table 1: Closed-loop efficiency comparison
[0180]
[0181] For randomly selected vulnerability samples, we constructed a state perception diagram and predicted their closed-loop paths and key behavior chains. The results are as follows:
[0182] Closed-loop path prediction accuracy: 84.9%;
[0183] Closed-loop probability prediction error (RMSE): ±6.3%;
[0184] Key node hit rate (matching with real key responsibility nodes): 91.2%.
[0185] By dynamically disposing responsibility graphs to assign tasks, the system performs as follows in terms of resource scheduling:
[0186] The average task saturation of high-load nodes decreased by 37.8%;
[0187] Average response latency was shortened by 41.5%;
[0188] The node response failure rate dropped from 6.3% to 1.1%.
[0189] By collecting closed-loop behavior chains for model reverse training, the accuracy of the closed-loop prediction model increased by 8.6% and the path misjudgment rate decreased by 11.2%, demonstrating the sustainable evolution capability of the structure-preserving transfer learning mechanism.
[0190] Example 3, please refer to Figure 2 As shown, the closed-loop vulnerability handling system based on intelligent collaboration described in this embodiment includes:
[0191] The semantic modeling module collects vulnerability data, asset information, and disposal records from multiple independent security systems to build a multi-dimensional vulnerability semantic model;
[0192] The state modeling module builds a state perception diagram based on the vulnerability semantic model;
[0193] The prediction analysis module extracts a set of candidate state sequence paths from the state perception graph, builds a time series closed-loop prediction model based on the matching degree between the shortest transfer path required for the target state to be closed and the historical state trajectory, and predicts the closed-loop probability and the required key behavior chain for each vulnerability to be addressed;
[0194] The responsibility division module adaptively divides vulnerability responsibilities based on the vulnerability impact scope, key business dependency chains, and key nodes in the predicted closed-loop path, generates a dynamic handling responsibility map, and assigns vulnerabilities to the optimal responsible persons;
[0195] The scheduling engine module obtains the expected response time, availability status, and task saturation of each node in the responsibility graph, dynamically calculates vulnerability priority weights, and automatically sorts and schedules vulnerability handling processes based on priority.
[0196] The model evolution module monitors the actual behavior nodes and state transitions during the execution of vulnerability remediation tasks, corrects the edge weights and transition probabilities in the state perception graph in real time, and updates the time series closed-loop prediction model;
[0197] The closed-loop confirmation module automatically marks the vulnerability as closed after it is fixed and passes multi-source verification, and synchronizes the status to the original system;
[0198] Feedback optimization module, output vulnerability handling complete behavior chain and closed-loop cycle evaluation report, which is used for reverse training of time series closed-loop prediction model.
[0199] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.
Claims
1. A closed-loop vulnerability handling method based on intelligent collaboration, characterized by: include: Collect vulnerability data, asset information, and disposal records from multiple independent security systems to build a multi-dimensional vulnerability semantic model; Specifically, the method involves mapping each vulnerability instance into a unified semantic entity by constructing a multidimensional feature vector space, encompassing vulnerability attribute dimensions, asset association dimensions, historical behavior dimensions, threat perception dimensions, and organizational collaboration dimensions; and constructing each dimension into a multidimensional vulnerability semantic model. Build a state perception graph based on the vulnerability semantic model; Specifically, the method includes: defining the core state nodes in the vulnerability lifecycle as the basic building blocks of the state graph; analyzing the historical path data of the vulnerability migrating from one state to another; and constructing edge relationships in the state graph based on historical path samples, where each edge represents the transition path between two state nodes and assigns a transition probability as the edge weight. Extract candidate state sequence paths from the state perception graph. Build a time-series closed-loop prediction model based on the matching degree between the shortest transition path required for the target state to be closed and the historical state trajectory. For each vulnerability to be addressed, predict its closed-loop probability and the required key behavior chain. The construction of the time series closed-loop prediction model includes: starting from the state perception graph, enumerating all reachable paths from the "current state" to the "closed loop" to form a candidate path set P; For the vulnerability v to be handled, its historical state trajectory is recorded as T_v, and each path in the candidate path set The matching degree is calculated by taking the weighted sum of the path prefix overlap rate and the temporal feature matching degree. A temporal closed-loop prediction model is established using the candidate path's frequency of occurrence, historical loop closure success rate, and matching degree with the vulnerability's current trajectory as input features. The model output is the predicted probability of the current vulnerability reaching the "closed-loop" state. Adaptively divide vulnerability responsibility based on vulnerability impact scope, key business dependency chains, and key nodes in the predicted closed-loop path, generate a dynamic handling responsibility map, and assign vulnerabilities to the optimal responsible individuals. Each node in the dynamic handling responsibility graph is a responsible person or a responsible system module, the edge represents the task transfer sequence and coordination relationship, and the edge weight reflects the responsibility adaptability; Obtain the expected response time, availability status, and task saturation of each node in the responsibility graph, dynamically calculate vulnerability priority weights, and automatically sort and schedule vulnerability handling processes based on priority; The dynamic calculation of vulnerability priority weights includes: obtaining the expected response time Te, availability status A, and task saturation L from each responsible node to form the capability state vector S of the responsible node; calculating the basic priority score of the vulnerability itself , the expression is: Where CCVSS represents the CVSS score of the vulnerability; Iasset represents the business importance score of the affected asset; Rexploit represents the possibility of the vulnerability being exploited or the current status of its exploitation; λ1 to λ3 represent the weight coefficients of each factor; The node capability state vector S and the basic priority score Vbase are integrated to calculate the comprehensive priority weight W for each vulnerability. The expression is: ;in is the normalized value of Te, α, β, γ, and δ are system scheduling parameters; Monitor the actual behavior nodes and state transitions during the execution of vulnerability remediation tasks, modify the edge weights and transition probabilities in the state perception graph in real time, and update the time series closed-loop prediction model; After the vulnerability is fixed and passes multi-source verification, it is automatically marked as closed-loop and the status is synchronized to the original system; Output the complete vulnerability handling behavior chain and closed-loop cycle evaluation report for reverse training of the time series closed-loop prediction model.
2. The closed-loop vulnerability handling method based on intelligent collaboration according to claim 1, characterized in that: The updating of the time series closed-loop prediction model includes: During the vulnerability handling process, we continuously collect task execution logs, status change timestamps, and work order processing behavior chains to generate vulnerability behavior trajectory sequences. Map the trajectory sequence to the path nodes in the state perception graph, calculate the actual jump frequency and jump time between each state, and construct a state transition behavior tensor with timing constraints; The Bayesian incremental learning method is used to fuse the new behavior tensor and the historical transition probability matrix, dynamically modify the edge weights and transition probabilities in the state graph, and generate a time-varying state graph. The path weight and closed-loop reachability function of the temporal closed-loop prediction model are updated based on the time-varying state diagram.
3. The closed-loop vulnerability handling method based on intelligent collaboration according to claim 1, characterized in that: Output vulnerability handling behavior chain and closed-loop cycle evaluation report and reverse training prediction model include: After the vulnerability is closed, the complete handling behavior chain is extracted, recording all state transition nodes, timestamps, and sequences of responsible persons from the initial state to the closed-loop state; Counting the dwell time of key nodes and the total cycle of the paths in the behavior chain to construct a closed-loop cycle structure feature set; Compare the closed-loop cycle characteristics with the corresponding path estimates in the original prediction model and calculate the error residual distribution; Based on error feedback, a structure-preserving transfer learning method is used to reversely train and update the original time series closed-loop prediction model, dynamically optimizing the path estimation accuracy and the importance weights of behavior nodes.
4. The closed-loop vulnerability handling method based on intelligent collaboration according to claim 1 is characterized by: The security system includes a scanning platform, a configuration management database, a security operation platform and a work order system.
5. The closed-loop vulnerability handling method based on intelligent collaboration according to claim 1 is characterized by: The nodes in the state perception graph represent vulnerability state nodes, the edges represent the transition relationship between vulnerability states, and the weight of each edge is adaptively calculated based on the historical state transition probability to support vulnerability state prediction.
6. A closed-loop vulnerability handling system based on intelligent collaboration, used to implement the closed-loop vulnerability handling method based on intelligent collaboration according to any one of claims 1 to 5, characterized in that: include: The semantic modeling module collects vulnerability data, asset information, and disposal records from multiple independent security systems to build a multi-dimensional vulnerability semantic model; The state modeling module builds a state perception diagram based on the vulnerability semantic model; The prediction and analysis module extracts a set of candidate state sequence paths from the state perception graph, builds a time series closed-loop prediction model based on the matching degree between the shortest transfer path required for the target state to be closed and the historical state trajectory, and predicts the closed-loop probability and the required key behavior chain for each vulnerability to be addressed; The responsibility division module adaptively divides vulnerability responsibilities based on the vulnerability impact scope, key business dependency chains, and key nodes in the predicted closed-loop path, generates a dynamic handling responsibility map, and assigns vulnerabilities to the optimal responsible persons; The scheduling engine module obtains the expected response time, availability status, and task saturation of each node in the responsibility graph, dynamically calculates vulnerability priority weights, and automatically sorts and schedules vulnerability handling processes based on priority. The model evolution module monitors the actual behavior nodes and state transitions during the execution of vulnerability remediation tasks, corrects the edge weights and transition probabilities in the state perception graph in real time, and updates the time series closed-loop prediction model; The closed-loop confirmation module automatically marks the vulnerability as closed after it is fixed and passes multi-source verification, and synchronizes the status to the original system; Feedback optimization module, output vulnerability handling complete behavior chain and closed-loop cycle evaluation report, which is used for reverse training of time series closed-loop prediction model.
Citation Information
Patent Citations
Vulnerability risk assessment method and device, storage medium, and server
CN109617910A
Industrial control network vulnerability scanning method
CN117061178A
