Optimal tracing timing dynamic selection method, system and medium for APT attack

By dynamically selecting the timing for tracing APT attacks and optimizing the timing using multi-source data and game theory models, the problem of inappropriate timing for tracing is solved, improving the effectiveness and security of tracing. This approach is applicable to critical information infrastructure such as power systems.

CN123122953APending Publication Date: 2026-11-17CHINA ELECTRIC POWER RESEARCH INSTITUTE CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611393720.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-09
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing methods for attribution of APT attacks have failed to effectively address the issue of timing for attribution, leading to risks such as attacker escape, evidence decay, and increased impact of attacks due to insufficient evidence or late attribution.

Method used

This paper proposes a dynamic selection method for optimal source tracing timing in response to APT attacks. By collecting attack data from multiple sources, a unified attack event sequence is formed, the attack stage is identified, candidate source tracing times are dynamically generated, and the comprehensive benefits are calculated based on a game theory model to select the optimal source tracing start time.

Benefits of technology

It enables dynamic optimization of evidence accumulation and risk in APT attacks, avoids premature or late attribution, enhances attribution capabilities in complex scenarios, provides interpretable decision-making basis, and is suitable for security protection of critical information infrastructure.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123122953A_ABST
    Figure CN123122953A_ABST
Patent Text Reader

Abstract

The application provides an optimal tracing opportunity dynamic selection method for APT attack, and belongs to the technical field of network security. The method comprises the following steps: collecting multi-source security data and converting the multi-source security data into attack event sequences in a unified format; dynamically generating candidate tracing time points, and counting attack states at each time point; based on a dynamic game model of attackers and defenders, calculating comprehensive benefits of each candidate time point, wherein the comprehensive benefits are obtained by weighting evidence gain, tracing confidence, evidence integrity, strategic phase benefit and analysis cost, exposure risk, and delay risk, and selecting a time point with the maximum comprehensive benefit as an optimal tracing start time; and outputting a high-benefit tracing window, a strategy comparison result, and a regret value. The application converts the selection of tracing time into a quantifiable benefit optimization problem, can balance evidence accumulation and waiting risk, avoids early or late tracing, and provides interpretable and reviewable tracing decision basis for security operation personnel.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of network security attack tracing and dynamic defense decision-making technology, and in particular to a method, system and medium for dynamically selecting the optimal tracing timing for APT attacks. Background Technology

[0002] Advanced Persistent Threats (APTs) are characterized by long attack cycles, highly targeted attacks, covert behavior, complex attack phases, and dynamic changes in the attack infrastructure. APT attacks targeting power systems typically begin at the office network or remote maintenance entry point, progressing through stages such as initial access, malicious code execution, persistence, privilege escalation, internal asset discovery, lateral movement, command and control, and attack impact. They may further approach dispatch automation systems, engineering workstations, front-end communication servers, and other critical production control assets. Because attack activities are dispersed across different hosts, accounts, and time periods, defenders need to synthesize multi-source evidence, including endpoint logs, honeypot interactions, threat intelligence, and security alerts, to reconstruct the attack path and generate credible attribution results.

[0003] As the duration and complexity of APT attacks increase, existing research employs dynamic decision-making defense by deploying adaptive interactive honeypots. Wang et al. proposed an active deception defense model based on Bayesian attack graphs and Stackelberg game theory. This model uses honeypots to disguise legitimate services to detect covert attackers and leverages a game theory model to optimize the deployment strategy for deception resources. [1] Related research indicates that defenders do not need to immediately terminate the interaction after detecting an attack; appropriately maintaining the honeypot session can further obtain information such as attack commands, malicious tools, samples, accounts, and target selection.

[0004] Network attack attribution is a crucial step in transforming behavioral data collected by honeypots into proactive defense capabilities. In recent years, APT attack attribution has evolved from analyzing single IP addresses, domain names, or malicious samples to multi-source evidence fusion, behavioral sequence analysis, and system attribution graph analysis. Currently, MITRE ATT&CK provides a knowledge framework for the standardized representation of honeypot data and multi-source security events. Among these, the Indicator of Compromise (IOC) can represent technical clues such as IP addresses, domain names, file hashes, malicious files, accounts, and registry entries; Tactics, Techniques, and Procedures (TTP) are used to describe the attacker's targets, technical methods, and specific implementation methods. Mapping events such as honeypot interactions and decoy triggers to IOCs and TTPs helps determine the attack phase, correlate behaviors on different hosts, and reconstruct cross-regional attack paths.

[0005] In existing research, attack attribution techniques mainly revolve around attack evidence extraction, attack chain reconstruction, attack behavior correlation, and attack source inference. The main approaches are as follows: (1) Utilizing multi-source evidence fusion, by integrating data such as terminal logs, network traffic, alarm information, threat intelligence, and manual annotation, to extract clues such as the IP addresses, domain names, samples, accounts, commands, and tools used by attackers, thereby improving the credibility of attribution conclusions. [2-3] (2) Using behavioral sequence analysis, events such as process creation, file access, network connection, account login and command execution are organized into attack behavior sequences in chronological order, thereby identifying the attacker's behavioral evolution process from initial access to lateral movement and data collection. [4,6] (3) Based on the system source graph or cause-effect graph, the host, process, file, network connection, account and security event are represented as nodes and edges in the graph. The attack path is traced through the dependency and cause-effect relationship, thereby realizing the attack chain reconstruction and intrusion investigation. [4-7] .

[0006] The aforementioned attribution studies demonstrate that multi-source evidence, behavioral sequences, and attribution graphs can effectively enhance the completeness and credibility of APT attack analysis. Especially in honeypot and decoy scenarios, the attacker's interaction with the honeypot, decoy resource triggering records, and terminal and network-side evidence can corroborate each other, thus supporting attack phase identification, attack chain reconstruction, and attack intent assessment. However, existing attribution methods typically assume that the attribution task has already begun, with limited further research into when the defender should initiate attribution analysis as the attack continues and evidence accumulates. In other words, existing methods can effectively utilize the obtained evidence for attribution, but they have not yet fully addressed the timing issue of "at what level of evidence accumulation is it appropriate to initiate attribution analysis." Summary of the Invention

[0007] This invention aims to address at least one of the technical problems existing in the prior art. To this end, this invention proposes a dynamic selection method for optimal attribution timing in APT attacks, to solve the problems of fixed attribution timing, difficulty in quantifying benefits, and lack of comparable strategies in existing APT attribution methods.

[0008] This invention also proposes a system and medium with the above-mentioned dynamic selection method for optimal source tracing timing against APT attacks.

[0009] The method for dynamically selecting the optimal timing for attribution tracing against APT attacks according to a first aspect of the present invention is characterized by comprising the following steps: Attack data from different security systems is collected and converted into attack events with a unified format. The attack events are sorted based on a unified time base to form an attack event sequence. The attack events are also aggregated according to a preset aggregation rule to form a cross-host attack session. Based on the attack event sequence and the attack session, attack clues are extracted and the current attack stage is identified. Multiple candidate tracing moments are dynamically generated, and at each candidate tracing moment, the attack evidence observed up to that moment is statistically analyzed to update and form the attack state. Based on the dynamic game model between the attacker and the defender, and combined with the attack state, a comprehensive benefit is calculated for each candidate tracing moment. The comprehensive benefit is jointly determined by the tracing benefit item and the risk cost item, and the candidate moment with the largest comprehensive benefit is selected as the optimal tracing start time.

[0010] The dynamic selection method for optimal attribution timing against APT attacks according to embodiments of the present invention has at least the following beneficial effects: This invention addresses the issue of choosing the right time to initiate attribution for APT attacks, proposing a dynamic optimization method that balances evidence accumulation with the risk of waiting. This method calculates attribution benefit indicators such as evidence gain, attribution confidence, and evidence completeness in real time, while comprehensively evaluating negative costs such as analysis costs, exposure risks, and delay risks. It constructs a complete comprehensive benefit quantification model, transforming the traditionally subjective decision-making problem of "when to initiate attribution"—relying on human experience—into a calculable, comparable, and interpretable mathematical problem of maximizing benefits. This method avoids the problems of insufficient evidence and incomplete attack paths caused by premature attribution, while also avoiding the risks of attacker escape, evidence decay, and expanded attack impact caused by delayed attribution. Furthermore, this invention outputs information such as high-benefit attribution windows, strategy comparison tables, and regret values, providing security operations personnel with interpretable and verifiable auxiliary decision-making basis. This enhances practical attribution capabilities in complex scenarios such as highly concealed, long-term, and cross-host APT attacks, and is applicable to the security protection of critical information infrastructure such as power monitoring systems.

[0011] According to some embodiments of the present invention, in the steps of collecting attack data from different security systems, converting the attack data into attack events with a unified format, sorting the attack events based on a unified time base to form an attack event sequence, and aggregating the attack events according to preset aggregation rules to form a cross-host attack session: The different security systems that collect attack data include at least one of the following: honeypot platform, decoy triggering platform, endpoint detection system, network detection system, log auditing system, security alerting platform, and threat intelligence platform; The attack event includes at least the timestamp, source of evidence, host, account, process, file, source address, destination address, event type, behavior description, attack stage, compromise index (IOC), tactics and procedures (TTP), and decoy identification field; The preset aggregation rule aggregates attack events based on source address, host, account, process, session identifier, and time window.

[0012] According to some embodiments of the present invention, in the step of extracting attack clues and identifying the current attack stage based on the attack event sequence and the attack session, dynamically generating multiple candidate tracing moments, and statistically analyzing the attack evidence observed up to that moment at each candidate tracing moment to update and form the attack state, the step of dynamically generating multiple candidate tracing moments specifically includes: Candidate points with a fixed time step are generated according to a fixed time step, and event trigger candidate points are inserted when a preset key attack event occurs. The candidate source tracing time set is composed of the fixed step candidate points and the event trigger candidate points. The fixed time step is set according to the business scenario, and the preset key attack events include at least one of the following: successful login, command execution, sample download, credential access, lateral movement, decoy triggering, and access to sensitive assets.

[0013] According to some embodiments of the present invention, the steps of extracting attack clues and identifying the current attack stage based on the attack event sequence and the attack session, dynamically generating multiple candidate tracing moments, and statistically analyzing the attack evidence observed up to that moment at each candidate tracing moment to update and form the attack state, specifically include: Based on the commands, files, processes, network connections, alarm types, and ATT&CK technology tags in the attack event, the attack event is divided into the initial access, execution, persistence, privilege escalation, credential access, insider discovery, lateral movement, data collection, command control, cleanup, and general activity phases. The attack clues extracted from the attack incident include Indicators of Compromise (IOCs) and Tactical Techniques and Procedures (TTPs). The IOCs include IP addresses, domain names, and file hashes, while the TTPs represent the tactics, techniques, and procedures used by the attacker. If the current attack state involves credential access, lateral movement, data collection, use of decoy credentials, access to sensitive files, or proximity to critical assets, then that candidate moment is marked as a high-value candidate moment.

[0014] According to some embodiments of the present invention, in the step of calculating a comprehensive benefit for each candidate tracing moment based on a dynamic game model between the attacker and the defender, combined with the attack state, wherein the comprehensive benefit is jointly determined by a tracing benefit item and a risk cost item, and selecting the candidate moment with the largest comprehensive benefit as the optimal tracing start time, the tracing benefit item includes evidence gain, tracing confidence, evidence completeness, and strategic phase benefit. The evidence gain is used to describe the growth of the obtained attack evidence relative to the potential evidence in the attack chain up to the candidate time. It is obtained by setting stage weights according to different attack stages, accumulating the event weights up to that time, and then normalizing it with the global maximum evidence weight. The source tracing confidence is used to describe whether the current evidence is sufficient to support a reliable source tracing judgment. It is calculated by weighted summation of attack chain host coverage, IOC coverage, TTP coverage, consistency of multi-source evidence, and consistency of labeled evidence. The evidence completeness is used to describe the coverage of the current evidence to key components of the attack chain, and it is calculated based on host coverage, attack phase coverage, TTP coverage, and IOC coverage. The strategic phase benefits are used to increase the benefit value based on the number of high-value behaviors when covering credential access, lateral movement, data collection, decoy triggering, and key asset approach behaviors at candidate moments.

[0015] According to some embodiments of the present invention, the risk cost item includes analysis cost, exposure risk, and delay risk: The analysis cost is used to characterize the log size, number of events, computing resources, storage resources, and manual judgment costs required to initiate source tracing. It is calculated based on the number of events, log volume, and analysis task complexity. The exposure risk is used to characterize the likelihood that the honeypot, decoy, monitoring rules, or defense strategy will be identified by the attacker while the defender continues to wait. It is calculated based on the attacker's dwell time, the depth of decoy interaction, and the coverage of high-risk phases. The delay risk is used to characterize the risk that the defender's continued waiting will lead to the attacker's escape, evidence decay, infrastructure migration, log cleanup, or the expansion of the attack's impact. It is calculated based on the relative position of the candidate moment in the attack lifecycle, the degree of progress of the attack phase, and the proximity of key assets.

[0016] According to some embodiments of the present invention, the comprehensive benefit is calculated using the following formula: in, , ), and These represent evidence gain, source tracing confidence, evidence completeness, and strategic phase gains, respectively. , and These represent the analysis cost, exposure risk, and delay risk, respectively. , , , , and The weighting parameter can be set based on expert experience, historical attack data, red team exercise results, or business risk preferences. Positive terms are used to represent the benefits of tracing the source, while negative terms are used to represent costs and risks.

[0017] According to some embodiments of the present invention, the evidence gain Calculated as follows: , , in, This represents the overall evidence weight for the i-th attack event. The cumulative effective evidence weight for candidate time t. To prevent extremely small positive numbers with a denominator of 0.

[0018] The analysis cost Calculated as follows: in, This represents the number of telemetry events up to time t. Indicates storage and retrieval costs, This indicates the cost of manual assessment. , , These are preset weight parameters.

[0019] According to some embodiments of the present invention, the exposure risk Calculated as follows: in, This represents the normalized value of the attacker's dwell time. Indicates the depth of the decoy interaction. Indicates coverage of high-risk phases. This indicates the likelihood that the honeypot or bait has been identified. , , , These are preset weight parameters.

[0020] According to some embodiments of the present invention, the step of calculating a comprehensive benefit for each candidate tracing time based on a dynamic game model between the attacker and the defender, combined with the attack state, wherein the comprehensive benefit is jointly determined by a tracing benefit item and a risk cost item, and selecting the candidate time with the largest comprehensive benefit as the optimal tracing start time, further includes: If the overall returns of multiple candidate times are similar, then the earlier or more robust time to start the tracing process is selected by combining the business risk level, the proximity of key assets, and the preference of human judgment.

[0021] According to some embodiments of the present invention, in the steps of generating a high-yield tracing window based on the optimal tracing start time, calculating and outputting the benefit comparison results and regret values ​​of different tracing strategies, and outputting tracing start suggestions including the optimal tracing start time: The high-yield tracing window uses the optimal tracing start time. Centered on a fixed time range or benefit threshold, it is used to indicate that source tracing analysis can be initiated within a time interval where the evidence benefit is high and the risk cost is acceptable. The regret value represents the difference in returns between a certain strategy and the optimal strategy. The smaller the regret value, the closer the strategy is to the optimal tracing timing. The proposed source tracing initiation recommendations include optimal source tracing initiation time, high-yield source tracing window, attack phase, attack chain host, IOC, TTP, evidence source, strategy comparison table, and regret value.

[0022] According to some embodiments of the present invention, the method further includes: Based on the optimal source tracing start time, a high-yield source tracing window is generated, and the profit comparison results and regret values ​​of different source tracing strategies are calculated and output. Based on the high-yield source tracing window and the profit comparison results and regret values ​​of different source tracing strategies, a source tracing start suggestion containing the optimal source tracing start time is output.

[0023] A dynamic selection system for optimal attribution timing against APT attacks according to a second aspect embodiment of the present invention, characterized in that it comprises: The multi-source attack event access module is used to collect attack data from different security systems, convert the attack data into attack events with a unified format, form an attack event sequence, and aggregate the attack events to form a cross-host attack session. The attack status identification module is used to extract attack clues and identify the current attack stage based on the attack event sequence and the attack session, dynamically generate multiple candidate tracing moments, and statistically analyze the observed attack evidence at each candidate tracing moment to update the attack status. The dynamic game decision-making module is used to calculate the comprehensive benefit for each candidate tracing moment based on the dynamic game model between the attacker and the defender, combined with the attack state, and select the candidate moment with the largest comprehensive benefit as the optimal tracing start time. The modules interact with each other through data flow, with the output of the previous module serving as the input of the next module, in order to achieve dynamic selection of the optimal tracing timing.

[0024] According to some embodiments of the present invention, the multi-source attack event access module includes: The different security systems that collect attack data include at least one of the following: honeypot platform, decoy triggering platform, endpoint detection system, network detection system, log auditing system, security alerting platform, and threat intelligence platform; The attack event includes at least the timestamp, source of evidence, host, account, process, file, source address, destination address, event type, behavior description, attack stage, compromise index (IOC), tactics and procedures (TTP), and decoy identification field; The preset aggregation rule aggregates attack events based on source address, host, account, process, session identifier, and time window.

[0025] According to some embodiments of the present invention, in the attack state identification module, the dynamic generation of multiple candidate tracing moments specifically includes: Candidate points with a fixed time step are generated according to a fixed time step, and event trigger candidate points are inserted when a preset key attack event occurs. The candidate source tracing time set is composed of the fixed step candidate points and the event trigger candidate points. The fixed time step is set according to the business scenario, and the preset key attack events include at least one of the following: successful login, command execution, sample download, credential access, lateral movement, decoy triggering, and access to sensitive assets.

[0026] According to some embodiments of the present invention, the attack state identification module specifically includes: Based on the commands, files, processes, network connections, alarm types, and ATT&CK technology tags in the attack event, the attack event is divided into the initial access, execution, persistence, privilege escalation, credential access, insider discovery, lateral movement, data collection, command control, cleanup, and general activity phases. The attack clues extracted from the attack incident include Indicators of Compromise (IOCs) and Tactical Techniques and Procedures (TTPs). The IOCs include IP addresses, domain names, and file hashes, while the TTPs represent the tactics, techniques, and procedures used by the attacker. If the current attack state involves credential access, lateral movement, data collection, use of decoy credentials, access to sensitive files, or proximity to critical assets, then that candidate moment is marked as a high-value candidate moment.

[0027] According to some embodiments of the present invention, in the dynamic game decision-making module, the source tracing payoff includes evidence gain, source tracing confidence, evidence completeness, and strategic phase payoff: The evidence gain is used to describe the growth of the obtained attack evidence relative to the potential evidence in the attack chain up to the candidate time. It is obtained by setting stage weights according to different attack stages, accumulating the event weights up to that time, and then normalizing it with the global maximum evidence weight. The source tracing confidence is used to describe whether the current evidence is sufficient to support a reliable source tracing judgment. It is calculated by weighted summation of attack chain host coverage, IOC coverage, TTP coverage, consistency of multi-source evidence, and consistency of labeled evidence. The evidence completeness is used to describe the coverage of the current evidence to key components of the attack chain, and it is calculated based on host coverage, attack phase coverage, TTP coverage, and IOC coverage. The strategic phase benefits are used to increase the benefit value based on the number of high-value behaviors when covering credential access, lateral movement, data collection, decoy triggering, and key asset approach behaviors at candidate moments.

[0028] According to some embodiments of the present invention, the risk cost item includes analysis cost, exposure risk, and delay risk: The analysis cost is used to characterize the log size, number of events, computing resources, storage resources, and manual judgment costs required to initiate source tracing. It is calculated based on the number of events, log volume, and analysis task complexity. The exposure risk is used to characterize the likelihood that the honeypot, decoy, monitoring rules, or defense strategy will be identified by the attacker while the defender continues to wait. It is calculated based on the attacker's dwell time, the depth of decoy interaction, and the coverage of high-risk phases. The delay risk is used to characterize the risk that the defender's continued waiting will lead to the attacker's escape, evidence decay, infrastructure migration, log cleanup, or the expansion of the attack's impact. It is calculated based on the relative position of the candidate moment in the attack lifecycle, the degree of progress of the attack phase, and the proximity of key assets.

[0029] According to some embodiments of the present invention, the system further includes: a tracing strategy output module, configured to generate a high-yield tracing window based on the optimal tracing start time, calculate and output the profit comparison results and regret values ​​of different tracing strategies, and output a tracing start suggestion including the optimal tracing start time.

[0030] According to a third aspect of the present invention, a computer-readable storage medium stores computer-executable instructions for performing the above-described method for dynamically selecting the optimal timing for tracing APT attacks.

[0031] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0032] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a schematic diagram illustrating the steps of a method for dynamically selecting the optimal source tracing timing for APT attacks according to an embodiment of the present invention. Figure 2 This is a structural block diagram of an optimal tracing timing dynamic selection system for APT attacks, provided in an embodiment of the present invention. Detailed Implementation

[0033] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0034] In the description of this invention, "several" means one or more, "more than" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0035] Example 1 To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should all fall within the scope of protection of the present invention.

[0036] This invention provides a dynamic method for selecting the optimal timing for attribution tracing in response to APT attacks. This method takes continuously arriving multi-source attack events as input, candidate attribution times as the decision-making object, and aims to maximize the overall benefit for the defender, dynamically selecting the optimal attribution initiation time. Figure 1 As shown, the method includes at least the following steps. Step S1: Collect attack data from different security systems, convert the attack data into attack events with a unified format, sort the attack events based on a unified time base to form an attack event sequence, and aggregate the attack events according to a preset aggregation rule to form a cross-host attack session.

[0037] Step S1 involves collecting attack data from different system platforms and uniformly converting security data from different sources into attack event sequences. Specifically, this includes: Step S11: Multi-source data acquisition.

[0038] The system collects attack-related data from honeypot platforms, decoy triggering platforms, endpoint detection systems (such as EDR), network detection systems (such as IDS / IPS), log auditing systems, security alerting platforms (such as SIEM), and threat intelligence platforms. This data includes, but is not limited to, attacker login attempts, command execution, file access, sample downloads, network connections, account usage, decoy access, malicious domains, suspicious IPs, file hashes, security alerts, and manually labeled information. In this embodiment, the system preferably obtains the above data through standardized interfaces or log collection proxies.

[0039] Step S12: Unify event format conversion.

[0040] The system converts data from different sources into a unified attack event e. This attack event includes at least the following fields: timestamp, evidence source, host, account, process, file, source address, destination address, event type, behavior description, attack stage, Indicator of Compromise (IOC), Tactics, Techniques and Procedures (TP), and decoy identifier. Through this unified event format, data generated by different security devices can enter the same attribution timing calculation process. For example, login logs generated by a honeypot platform and process creation logs generated by an endpoint system, after conversion, have the same field structure, facilitating subsequent processing.

[0041] Step S13: Generate an attack event sequence.

[0042] Attack events are sorted according to a unified time base to form an attack event sequence. .in, This represents the latest observed attack event up to time t. The attack event sequence describes the gradual unfolding of attacker behavior over time. The subsequence of events observable before time t is represented as follows: (1) Step S14: Attack session aggregation.

[0043] The system aggregates attack events based on source address, host, account, process, session identifier, and time window. It groups original security events scattered across different data sources, hosts, and time points but belonging to the same attack activity into a single attack session, forming a cross-host attack chain. For honeypot and decoy trigger events, the system retains the decoy identifier and trigger time to subsequently determine whether the attacker has entered a high-value interaction phase.

[0044] Attack session aggregation can be represented as: (2) in, It is an association key generated by combining the source address, host, account, process, and session identifier. This represents the time window for session aggregation. In this embodiment, It can be set to 30 minutes, 1 hour or 24 hours depending on the network environment.

[0045] After processing in step S1, the system outputs a sorted and aggregated sequence of attack events and attack sessions, providing a data foundation for subsequent attack status identification.

[0046] Step S2: Based on the attack event sequence and the attack session, extract attack clues and identify the current attack stage, dynamically generate multiple candidate tracing moments, and at each candidate tracing moment, count the attack evidence observed up to that moment to update and form the attack state.

[0047] Step S2 is used to extract attack clues from the attack event, generate a candidate attribution time set based on key attack phases, and statistically analyze the observed attack evidence and update the attack status at each candidate time. Specifically, it includes: Step S21: Attack phase division and clue extraction.

[0048] The system categorizes attack events into the following phases based on commands, files, processes, network connections, alarm types, and ATT&CK technical tags: initial access, execution, persistence, privilege escalation, credential access, insider discovery, lateral movement, data collection, command and control, cleanup, and general activities. The specific criteria for these phase divisions can be found in the tactical classifications within the MITREATT&CK framework.

[0049] Simultaneously, the system extracts Indicators of Compromise (IOCs) and Tactical Techniques and Procedures (TTPs) from attack events. IOCs include, but are not limited to, technical clues such as IP addresses, domain names, file hashes, malicious files, accounts, and registry entries; TTPs represent the tactics, techniques, and procedures used by the attacker, such as PowerShell execution, network scanning, and sample downloads.

[0050] Step S22: Dynamically generate candidate source tracing moments.

[0051] The system generates candidate moments according to a fixed time step, or inserts candidate moments when critical attack events occur. Critical events include successful login, command execution, sample download, credential access, lateral movement, decoy triggering, and access to sensitive assets. The candidate tracing moment set T consists of candidate points with a fixed time step. and event trigger candidate points Together they form the following: (3) (4) (5) in, This indicates a fixed time step, which can be set to 1 minute, 5 minutes, 10 minutes, or 30 minutes depending on the business scenario. This function represents a key event trigger. When the event is a preset key event such as successful login, command execution, sample download, credential access, lateral movement, decoy triggering, or access to sensitive assets, this function returns a true value.

[0052] Step S23: Statistical analysis of the attack status.

[0053] At each candidate tracing time t, the attack phase set, attack chain host set, IOC set, TTP set, evidence source set, number of decoy trigger events, and number of telemetry events observed up to that time are counted to form the attack state. This attack status is used to represent the attack progress and the quality of evidence that the defender can grasp at time t.

[0054] Step S24: Mark high-value candidate moments.

[0055] If the current attack state involves actions such as credential access, lateral movement, data collection, use of decoy credentials, access to sensitive files, or proximity to critical assets, the system will mark the candidate moment as a high-value candidate moment and add strategic phase benefits to the subsequent benefit calculation.

[0056] After processing in step S2, the system generates a set of candidate tracing moments and associates each moment with a corresponding attack state, providing complete input for dynamic game decision-making.

[0057] Step S3: Based on the dynamic game model between the attacker and the defender, and combined with the attack state, calculate the comprehensive benefit for each candidate tracing time. The comprehensive benefit is determined by the tracing benefit item and the risk cost item, and select the candidate time with the largest comprehensive benefit as the optimal tracing start time.

[0058] Step S3 is used to calculate the comprehensive payoff of candidate tracing times based on the game model between the attacker and the defender after each attack state update, according to the dynamic game payoff function, and select the candidate time with the highest payoff as the optimal tracing start time. Specifically, it includes: Step S31, Game Theory Model Framework.

[0059] The offensive-defense game model comprises two main parts: the benefits of tracing the source and the risks and costs. The benefits of tracing the source include evidence gain, confidence in tracing the source, completeness of evidence, and strategic phase gains; the risks and costs include analysis costs, exposure risks, and delay risks. The defender at the candidate moment... t The overall benefits of initiating traceability U D (t) is obtained by weighting the above terms.

[0060] Step S32: Calculate the various indicators of traceability benefits.

[0061] 1. Gain of Evidence This describes the growth of attack evidence obtained by the defender relative to the potential evidence in the attack chain up to candidate time t. The system sets stage weights according to different attack stages, accumulates the event weights up to time t, and then normalizes them with the global maximum evidence weight to obtain the final result. .

[0062] The evidence gain is calculated as follows: (6) (7) in, This represents the weight of the aggregate evidence observed for the i-th attack event t. The cumulative effective evidence weight for candidate time t. This represents any candidate time in the set of candidate source times. This indicates the prevention of extremely small positive numbers with a denominator of 0. In this embodiment, the event weights for different attack phases can be set as follows: 0.2 for the initial access phase, 0.3 for the command execution phase, 0.6 for the credential access phase, 0.8 for the lateral movement phase, and 0.9 for the access to critical assets phase.

[0063] 2. Source tracing confidence(t) describes whether the current evidence is sufficient to support a reliable source tracing judgment. The system calculates confidence(t) by weighting and accumulating the attack chain host coverage, IOC coverage, TTP coverage, consistency of multi-source evidence, and consistency of labeled evidence. The confidence increases accordingly when different sources of evidence can corroborate each other.

[0064] 3. Evidence completeness, which describes the coverage of current evidence to key components of the attack chain. The system obtains completeness(t) based on host coverage, attack phase coverage, TTP coverage, and IOC coverage to avoid misjudging the current moment as a high-yield moment based solely on a large number of low-value events.

[0065] 4. The strategic phase bonus(t) is used to highlight high-value behaviors such as credential access, lateral movement, data collection, decoy triggering, and proximity to key assets. When a candidate moment covers the above behaviors, the system adds a certain amount of bonus(t) to that moment based on the number of high-value behaviors, making the model focus more on attack phases that contribute significantly to tracing value.

[0066] Step S33: Calculate the various risk cost indicators. Specifically, this includes: 1. Cost Analysis This is used to characterize the log size, number of events, computing resources, storage resources, and manual analysis costs required when initiating source tracing. Typically, as attack observation time increases, the number of telemetry events continues to grow, and the analysis cost rises accordingly. The system can calculate `analysis_cost(t)` based on the number of events, log volume, and analysis task complexity, as follows: (8) in, This represents the number of telemetry events up to time t. Indicates storage and retrieval costs, This indicates the cost of manual assessment; , , These are preset weight parameters.

[0067] 2. Exposure Risk This value represents the likelihood that the attacker will identify the honeypot, decoy, monitoring rules, or defense strategy while the defender continues to wait. In a honeypot scenario, the longer the attacker stays and the deeper their interaction, the more likely they are to discover environmental anomalies. In real-world business scenarios, attackers may also identify monitoring mechanisms or cleanup traces. The system calculates `exposure_risk(t)` based on the attacker's stay time, the depth of decoy interaction, and the coverage of high-risk phases. The exposure risk is calculated as follows: (9) in, This represents the normalized value of the attacker's dwell time. This indicates the depth of the decoy interaction (e.g., the number of commands executed and files accessed by the attacker within the honeypot). This indicates coverage of high-risk phases (higher values ​​are used if the attack phase involves lateral movement or access to critical assets). This indicates the likelihood that the honeypot or bait will be identified (this can be quantified based on the degree of honeypot camouflage). , , , These are preset weight parameters.

[0068] 3. Delay risk (delay_risk(t)) characterizes the risk that continued waiting by the defender could lead to attacker escape, evidence decay, infrastructure migration, log cleanup, or escalation of the attack's impact. The system calculates delay_risk(t) based on the relative position of the candidate moment in the attack lifecycle, the progress of the attack phase, and the proximity of critical assets.

[0069] Step S34: Calculate the overall benefit and select the optimal timing.

[0070] The system models the attacker's attack progression and the defender's attribution initiation behavior as a dynamic game relationship. The comprehensive payoff for the defender to initiate attribution at candidate time t is expressed as: (10) in, , , , , and These are weighting parameters that can be set based on expert experience, historical attack data, red team drill results, or business risk preferences. Positive terms represent the benefits of tracing the source, while negative terms represent costs and risks.

[0071] The system considers the candidate set T={ , ,..., Calculation at each moment in} And select the candidate moment with the greatest overall benefit as the optimal source tracing start time, that is: If the returns from multiple candidate times are similar, an earlier or more stable time can be selected by combining the business risk level, the proximity of key assets, and human judgment preferences.

[0072] After processing in step S3, the system calculates the comprehensive benefit of all candidate time points and determines the optimal source tracing start time. .

[0073] Step S4: Based on the optimal source tracing start time, generate a high-yield source tracing window, calculate and output the profit comparison results and regret values ​​of different source tracing strategies, and output source tracing start suggestions including the optimal source tracing start time.

[0074] Step S4 is used to calculate the high-yield tracing window based on the optimal tracing start time, compare and evaluate different tracing strategies, and finally output tracing start recommendations. Specifically, it includes: Step S41: Generate a high-yield traceability window.

[0075] The system starts with the optimal source tracing time. Centered on a fixed time frame or benefit threshold, a high-yield attribution window is generated. This window indicates that the defender can initiate attribution analysis within a time interval where the evidence benefit is high and the risk cost is acceptable. The high-yield attribution window is as follows: (11) in, This represents the window revenue tolerance threshold, which can be set to 0.05, 0.1, or 0.15 depending on business needs.

[0076] If a fixed window is used, it can also be represented as: (12) in, and These represent the time range for the left and right expansion of the window, respectively, and can be set to 10 minutes, 30 minutes, or 1 hour.

[0077] Step S42: Calculate the strategy comparison results.

[0078] The system compares the method of this invention with various baseline strategies, including but not limited to: immediate attribution (started immediately upon detection of an attack), fixed-delay attribution (started after a fixed delay of 10 minutes, 30 minutes, or 1 hour), credential-triggered attribution (started only upon detection of credential access), and lateral movement-triggered attribution (started only upon detection of lateral movement). The system outputs the benefit value, evidence gain, confidence level, completeness, and risk cost of each strategy, presented in tabular or graphical form.

[0079] Step S43: Calculate the regret value.

[0080] The regret value is used to represent the difference in payoff between a given strategy and the optimal strategy. If the payoff of the optimal strategy is... The payoff of a certain candidate strategy is The regret value is The smaller the regret value, the closer the strategy is to the optimal time for tracing the source.

[0081] Step S44: Output a suggestion for starting the source tracing.

[0082] The system ultimately outputs the optimal tracing start time, high-yield tracing window, attack stage, attack chain host, IOC, TTP, evidence source, policy comparison table, and regret value, providing a basis for security operations personnel to carry out attack chain reconstruction, evidence solidification, alarm escalation, and manual analysis.

[0083] Example 2 This invention also provides a dynamic selection system for optimal attribution timing in response to APT attacks, such as... Figure 2 As shown, the system 20 includes: The multi-source attack event access module 201 is used to perform all the functions of step A above, namely, to collect attack data from different security systems, convert the attack data into attack events with a unified format, form an attack event sequence, and aggregate the attack events to form a cross-host attack session. The attack status identification module 202 is used to perform all the functions of step B above, namely, extracting attack clues and identifying the current attack stage based on the attack event sequence and the attack session, dynamically generating multiple candidate tracing moments, and statistically analyzing the observed attack evidence at each candidate tracing moment to update the attack status. The dynamic game decision module 203 is used to perform all the functions of step C above, namely, based on the dynamic game model between the attacker and the defender, combined with the attack state, to calculate the comprehensive benefit for each candidate tracing time, and select the candidate time with the largest comprehensive benefit as the optimal tracing start time. The source tracing strategy output module 204 is used to perform all the functions of step D above, namely, generating a high-yield source tracing window based on the optimal source tracing start time, calculating and outputting the benefit comparison results and regret values ​​of different source tracing strategies, and outputting source tracing start suggestions including the optimal source tracing start time.

[0084] The modules mentioned above interact via data flow, with the output of one module serving as the input for the next, enabling dynamic selection of the optimal time for source tracing. These modules can be deployed on the same security analysis platform or distributed across different platforms, communicating via message queues or API interfaces.

[0085] Example 3 This embodiment uses an APT attack in a power company's network as an example. The system collects terminal logs, network traffic, honeypot interaction records, decoy trigger records, and security alerts, and maps the attack behavior into stages such as login attempts, command execution, credential access, lateral movement, and access to critical assets. Candidate source tracing moments are dynamically generated according to a fixed 10-minute time interval and the triggering method of key attack events. to All indicators were normalized to The formula for calculating the comprehensive return over the given period is as follows: (13) Based on the calculation of the benefits and risks of tracing the source, the indicator calculation results for each candidate time point are as follows:

[0086] The calculation results show that the candidate time... The overall benefit was 0.518, higher than other candidate times, therefore the moment when the attacker initiated lateral movement was determined as the optimal time to initiate attribution. At this point, the attack evidence and attack path were already highly complete, while the exposure risk and latency risk had not yet increased excessively. The system then fixed relevant logs and attack evidence, extracted clues such as IP addresses, accounts, malicious processes, and attack tools, and generated the attack path from the attack entry point to the engineer's station. The above indicator values ​​and weights can be adjusted according to the actual network environment, asset level, and security policy.

[0087] Example 4 Another embodiment of the present invention provides a computer-readable storage medium storing computer-executable instructions for performing the above-described... Figure 1 The diagram shows the steps of the dynamic selection method for optimal source tracing of APT attacks.

[0088] The method provided by this invention calculates source tracing benefit indicators such as evidence gain, source tracing confidence, and evidence completeness in real time, while comprehensively evaluating negative costs such as analysis costs, exposure risks, and delay risks. It constructs a complete comprehensive benefit quantification model, transforming the traditional subjective decision-making problem of "when to initiate source tracing"—relying on human experience—into a calculable, comparable, and interpretable mathematical problem of maximizing benefits. This method avoids the problems of insufficient evidence and incomplete attack paths caused by premature source tracing, while also avoiding the risks of attacker escape, evidence decay, and expanded attack impact caused by delayed source tracing. Furthermore, this invention outputs information such as high-benefit source tracing windows, strategy comparison tables, and regret values, providing security operations personnel with interpretable and verifiable auxiliary decision-making basis. This enhances practical source tracing capabilities in complex scenarios such as highly concealed, long-term, and cross-host APT attacks, and is applicable to the security protection of critical information infrastructure such as power monitoring systems.

[0089] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0090] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0091] The above is a detailed description of the preferred embodiments of this application. However, this application is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. An optimal tracing timing dynamic selection method for APT attacks, characterized in that, Includes the following steps: Attack data from different security systems is collected and converted into attack events with a unified format. The attack events are sorted based on a unified time base to form an attack event sequence. The attack events are also aggregated according to a preset aggregation rule to form a cross-host attack session. Based on the attack event sequence and the attack session, attack clues are extracted and the current attack stage is identified. Multiple candidate tracing moments are dynamically generated, and at each candidate tracing moment, the attack evidence observed up to that moment is statistically analyzed to update and form the attack state. Based on the dynamic game model between the attacker and the defender, and combined with the attack state, a comprehensive benefit is calculated for each candidate tracing moment. The comprehensive benefit is jointly determined by the tracing benefit item and the risk cost item, and the candidate moment with the largest comprehensive benefit is selected as the optimal tracing start time.

2. The method of claim 1, wherein, In the steps of collecting attack data from different security systems, converting the attack data into attack events with a unified format, sorting the attack events based on a unified time base to form an attack event sequence, and aggregating the attack events according to preset aggregation rules to form a cross-host attack session: The different security systems that collect attack data include at least one of the following: honeypot platform, decoy triggering platform, endpoint detection system, network detection system, log auditing system, security alerting platform, and threat intelligence platform; The attack event includes at least the timestamp, source of evidence, host, account, process, file, source address, destination address, event type, behavior description, attack stage, compromise index (IOC), tactics and procedures (TTP), and decoy identification field; The preset aggregation rule aggregates attack events based on source address, host, account, process, session identifier, and time window.

3. The method of claim 1, wherein, The step of extracting attack clues and identifying the current attack stage based on the attack event sequence and the attack session, dynamically generating multiple candidate tracing moments, and statistically analyzing the attack evidence observed up to that moment at each candidate tracing moment to update and form the attack state, specifically includes the following steps: Candidate points with a fixed time step are generated according to a fixed time step, and event trigger candidate points are inserted when a preset key attack event occurs. The candidate source tracing time set is composed of the fixed step candidate points and the event trigger candidate points. The fixed time step is set according to the business scenario, and the preset key attack events include at least one of the following: successful login, command execution, sample download, credential access, lateral movement, decoy triggering, and access to sensitive assets.

4. The method of claim 1, wherein, The steps of extracting attack clues and identifying the current attack stage based on the attack event sequence and the attack session, dynamically generating multiple candidate tracing moments, and statistically analyzing the attack evidence observed up to that moment at each candidate tracing moment to update and form the attack state, specifically include: Based on the commands, files, processes, network connections, alarm types, and ATT&CK technology tags in the attack event, the attack event is divided into the initial access, execution, persistence, privilege escalation, credential access, insider discovery, lateral movement, data collection, command control, cleanup, and general activity phases. The attack clues extracted from the attack incident include Indicators of Compromise (IOCs) and Tactical Techniques and Procedures (TTPs). The IOCs include IP addresses, domain names, and file hashes, while the TTPs represent the tactics, techniques, and procedures used by the attacker. If the current attack state involves credential access, lateral movement, data collection, use of decoy credentials, access to sensitive files, or proximity to critical assets, then that candidate moment is marked as a high-value candidate moment.

5. The method of claim 1, wherein, The step of calculating the comprehensive benefit for each candidate tracing moment based on the dynamic game model between the attacker and the defender, combined with the attack state, where the comprehensive benefit is jointly determined by the tracing benefit item and the risk cost item, and selecting the candidate moment with the largest comprehensive benefit as the optimal tracing start time, includes the following tracing benefit items: evidence gain, tracing confidence, evidence completeness, and strategic phase benefit. The evidence gain is used to describe the growth of the obtained attack evidence relative to the potential evidence in the attack chain up to the candidate time. It is obtained by setting stage weights according to different attack stages, accumulating the event weights up to that time, and then normalizing it with the global maximum evidence weight. The source tracing confidence is used to describe whether the current evidence is sufficient to support a reliable source tracing judgment. It is calculated by weighted summation of attack chain host coverage, IOC coverage, TTP coverage, consistency of multi-source evidence, and consistency of labeled evidence. The evidence completeness is used to describe the coverage of the current evidence to key components of the attack chain, and it is calculated based on host coverage, attack phase coverage, TTP coverage, and IOC coverage. The strategic phase benefits are used to increase the benefit value based on the number of high-value behaviors when covering credential access, lateral movement, data collection, decoy triggering, and key asset approach behaviors at candidate moments.

6. The method of claim 1, wherein, The risk cost item includes analysis costs, exposure risk, and delay risk: The analysis cost is used to characterize the log size, number of events, computing resources, storage resources, and manual judgment costs required to initiate source tracing. It is calculated based on the number of events, log volume, and analysis task complexity. The exposure risk is used to characterize the likelihood that the honeypot, decoy, monitoring rules, or defense strategy will be identified by the attacker while the defender continues to wait. It is calculated based on the attacker's dwell time, the depth of decoy interaction, and the coverage of high-risk phases. The delay risk is used to characterize the risk that the defender's continued waiting will lead to the attacker's escape, evidence decay, infrastructure migration, log cleanup, or the expansion of the attack's impact. It is calculated based on the relative position of the candidate moment in the attack lifecycle, the degree of progress of the attack phase, and the proximity of key assets.

7. The method of claim 1, wherein, The overall return is calculated using the following formula: in, , ), and These represent evidence gain, source tracing confidence, evidence completeness, and strategic phase gains, respectively. , and These represent the analysis cost, exposure risk, and delay risk, respectively. , , , , and The weighting parameter can be set based on expert experience, historical attack data, red team exercise results, or business risk preferences. Positive terms are used to represent the benefits of tracing the source, while negative terms are used to represent costs and risks.

8. The method of claim 7, wherein, The evidence gain is calculated by , , in, This represents the overall evidence weight for the i-th attack event. The cumulative effective evidence weight for candidate time t. To prevent extremely small positive numbers with a denominator of 0.

9. The method of claim 7, wherein, The analysis cost is calculated by in, This represents the number of telemetry events up to time t. Indicates storage and retrieval costs, This indicates the cost of manual assessment. , , These are preset weight parameters.

10. The method of claim 1, wherein, The exposure risk By calculating: in, This represents the normalized value of the attacker's dwell time. Indicates the depth of the decoy interaction. Indicates coverage of high-risk phases. This indicates the likelihood that the honeypot or bait has been identified. , , , These are the preset weight parameters.

11. The method of claim 1, wherein, The step of calculating the comprehensive benefit for each candidate tracing time based on the dynamic game model between the attacker and the defender, combined with the attack state, where the comprehensive benefit is determined by both the tracing benefit and risk cost, and selecting the candidate time with the largest comprehensive benefit as the optimal tracing start time, further includes: If the overall returns of multiple candidate times are similar, then the earlier or more robust time to start the tracing process is selected by combining the business risk level, the proximity of key assets, and the preference of human judgment.

12. The method of claim 1, wherein, In the steps of generating a high-yield tracing window based on the optimal tracing start time, calculating and outputting the profit comparison results and regret values ​​of different tracing strategies, and outputting tracing start suggestions including the optimal tracing start time: the high-reward provenance window is generated at the optimal provenance launch time centered, generated according to a fixed time range or a reward threshold, for indicating that a provenance analysis can be launched within a time interval where the reward of the evidence is high and the risk cost is acceptable; The regret value represents the difference in returns between a certain strategy and the optimal strategy. The smaller the regret value, the closer the strategy is to the optimal tracing timing. The proposed source tracing initiation recommendations include optimal source tracing initiation time, high-yield source tracing window, attack phase, attack chain host, IOC, TTP, evidence source, strategy comparison table, and regret value.

13. The method of claim 1, wherein, The method further includes: Based on the optimal source tracing start time, a high-yield source tracing window is generated, and the profit comparison results and regret values ​​of different source tracing strategies are calculated and output. Based on the high-yield source tracing window and the profit comparison results and regret values ​​of different source tracing strategies, a source tracing start suggestion containing the optimal source tracing start time is output.

14. A dynamic selection system for optimal attribution timing in response to APT attacks, characterized in that, include: The multi-source attack event access module is used to collect attack data from different security systems, convert the attack data into attack events with a unified format, form an attack event sequence, and aggregate the attack events to form a cross-host attack session. The attack status identification module is used to extract attack clues and identify the current attack stage based on the attack event sequence and the attack session, dynamically generate multiple candidate tracing moments, and statistically analyze the observed attack evidence at each candidate tracing moment to update the attack status. The dynamic game decision-making module is used to calculate the comprehensive benefit for each candidate tracing moment based on the dynamic game model between the attacker and the defender, combined with the attack state, and select the candidate moment with the largest comprehensive benefit as the optimal tracing start time. The modules interact with each other through data flow, with the output of the previous module serving as the input of the next module, in order to achieve dynamic selection of the optimal tracing timing.

15. The system of claim 14, wherein, The multi-source attack event access module includes: The different security systems that collect attack data include at least one of the following: honeypot platform, decoy triggering platform, endpoint detection system, network detection system, log auditing system, security alerting platform, and threat intelligence platform; The attack event includes at least the timestamp, source of evidence, host, account, process, file, source address, destination address, event type, behavior description, attack stage, compromise index (IOC), tactics and procedures (TTP), and decoy identification field; The preset aggregation rule aggregates attack events based on source address, host, account, process, session identifier, and time window.

16. The system of claim 14, wherein, In the attack state identification module, the dynamic generation of multiple candidate tracing moments specifically includes: Candidate points with a fixed time step are generated according to a fixed time step, and event trigger candidate points are inserted when a preset key attack event occurs. The candidate source tracing time set is composed of the fixed step candidate points and the event trigger candidate points. The fixed time step is set according to the business scenario, and the preset key attack events include at least one of the following: successful login, command execution, sample download, credential access, lateral movement, decoy triggering, and access to sensitive assets.

17. The system of claim 14, wherein, The attack status identification module specifically includes: Based on the commands, files, processes, network connections, alarm types, and ATT&CK technology tags in the attack event, the attack event is divided into the initial access, execution, persistence, privilege escalation, credential access, insider discovery, lateral movement, data collection, command control, cleanup, and general activity phases. The attack clues extracted from the attack incident include Indicators of Compromise (IOCs) and Tactical Techniques and Procedures (TTPs). The IOCs include IP addresses, domain names, and file hashes, while the TTPs represent the tactics, techniques, and procedures used by the attacker. If the current attack state involves credential access, lateral movement, data collection, use of decoy credentials, access to sensitive files, or proximity to critical assets, then that candidate moment is marked as a high-value candidate moment.

18. The system of claim 14, wherein, In the dynamic game decision-making module, the source tracing payoff includes evidence gain, source tracing confidence, evidence completeness, and strategic phase payoff: The evidence gain is used to describe the growth of the obtained attack evidence relative to the potential evidence in the attack chain up to the candidate time. It is obtained by setting stage weights according to different attack stages, accumulating the event weights up to that time, and then normalizing it with the global maximum evidence weight. The source tracing confidence is used to describe whether the current evidence is sufficient to support a reliable source tracing judgment. It is calculated by weighted summation of attack chain host coverage, IOC coverage, TTP coverage, consistency of multi-source evidence, and consistency of labeled evidence. The evidence completeness is used to describe the coverage of the current evidence to key components of the attack chain, and it is calculated based on host coverage, attack phase coverage, TTP coverage, and IOC coverage. The strategic phase benefits are used to increase the benefit value based on the number of high-value behaviors when covering credential access, lateral movement, data collection, decoy triggering, and key asset approach behaviors at candidate moments.

19. The system of claim 14, wherein, The risk cost item includes analysis costs, exposure risk, and delay risk: The analysis cost is used to characterize the log size, number of events, computing resources, storage resources, and manual judgment costs required to initiate source tracing. It is calculated based on the number of events, log volume, and analysis task complexity. The exposure risk is used to characterize the likelihood that the honeypot, decoy, monitoring rules, or defense strategy will be identified by the attacker while the defender continues to wait. It is calculated based on the attacker's dwell time, the depth of decoy interaction, and the coverage of high-risk phases. The delay risk is used to characterize the risk that the defender's continued waiting will lead to the attacker's escape, evidence decay, infrastructure migration, log cleanup, or the expansion of the attack's impact. It is calculated based on the relative position of the candidate moment in the attack lifecycle, the degree of progress of the attack phase, and the proximity of key assets.

20. The system according to claim 14, characterized in that, The system further includes: a tracing strategy output module, used to generate a high-yield tracing window based on the optimal tracing start time, calculate and output the profit comparison results and regret values ​​of different tracing strategies, and output a tracing start suggestion including the optimal tracing start time.

21. A computer-readable storage medium storing computer-executable instructions for performing the method of any one of claims 1 to 12.