Mimicry arbitration method, device and equipment and storage medium

Abstract

The application provides a mimic decision method, device, equipment and storage medium, which issues a request to each online execution body in a heterogeneous execution body pool, collects and normalizes response data to generate an initial normalized data set, inputs a multi-round progressive decision process, and sequentially executes a judgment strategy of relative credibility, absolute credibility and global forced reset. If the output results of each online execution body are inconsistent, the abnormal execution body is replaced, the normalized data is re-collected and input into the process cycle iteration after replacement, and the output of each execution body converges until the output of each execution body converges and the decision result is output. The application provides a step-by-step error correction path through the heterogeneous strategy of rounds, isolates abnormal nodes by inconsistent triggering replacement and re-sampling iteration to block error accumulation. A convergence closed loop is constructed to provide an explicit process exit condition, overcome the majority misjudgment defect of single-round static voting, avoid decision process deadlock, and realize accurate convergence and business continuous output of the decision result under complex interference.

Description

Technical Field

[0001] This invention relates to the field of network security mimicry defense technology, and in particular to a mimicry adjudication method, apparatus, device, and storage medium. Background Technology

[0002] In the control plane of Software-Defined Networking (SDN) and distributed high-availability systems, heterogeneous execution pools are typically used to redundancy-determine critical control commands or status query requests. Existing technologies mostly rely on a single-round static majority voting mechanism, which involves synchronously sending pending requests to each online execution in the pool, collecting the response data generated by each execution in response to the request, performing consistency comparison after normalization, and using the output result of the majority of executions as the final decision.

[0003] However, existing single-round adjudication architectures have revealed significant flaws in real-world network warfare and long-term operation. For example, static majority voting mechanisms lack dynamic fault tolerance against common-cause attacks and slow, insidious infiltration. When external interference or malicious control causes the output of the majority of executors to deviate, the system is highly susceptible to majority misjudgment, leading to distorted adjudication results. On the other hand, existing mechanisms lack effective intervention methods for abnormal executors, allowing erroneous nodes to continue participating in subsequent business, resulting in error accumulation and reputation imbalance. Furthermore, in cases where the output results of multiple rounds of adjudication remain inconsistent, the system lacks reset capabilities, which can easily lead to deadlock in the adjudication process or interruption of control plane business, making it difficult to guarantee the long-term reliability and high availability of heterogeneous adjudication processes. Summary of the Invention

[0004] This invention provides a mimicry adjudication method, apparatus, device, and storage medium to solve the technical problems of existing single-round adjudication mechanisms being susceptible to interference leading to numerous misjudgments, continuous participation of abnormal nodes causing error accumulation, and process deadlock easily occurring when multiple rounds of judgments are inconsistent.

[0005] This invention provides a mimicry adjudication method comprising: A request to be processed is sent to each online executor in the heterogeneous executor pool. The response data generated by each online executor in response to the request to be processed is collected. The response data is normalized to generate an initial normalized dataset. The initial normalized dataset is input into a multi-round progressive adjudication process to execute the judgment strategy progressively according to the adjudication round. The first round executes a relatively reliable judgment strategy, the second round executes an absolutely reliable judgment strategy, and the third and subsequent rounds execute a global forced reset judgment strategy. After each execution of the corresponding judgment strategy, if the output results of each online executor are inconsistent, an abnormal executor is triggered for replacement. After replacing the abnormal executor, the response data is collected again, normalized, and then the next round of judgment dataset is generated. The next round of judgment dataset is used as the current input of the multi-round progressive adjudication process for the next round of judgment. The process is iterated until the output results of each online executor converge and are consistent. The convergence result is then output as the adjudication result.

[0006] According to a mimicry adjudication method provided by the present invention, after each execution of the judgment strategy, if the output results of each online executor converge to the same value, the multi-round progressive adjudication process is terminated, and the convergence result is output as the adjudication result.

[0007] According to a mimicry adjudication method provided by the present invention, if the output results of the online executors are consistent, the method further includes: Obtain the current adjudication round when the output results are consistent, and update the weights of each executor in the first round according to the differences in the current adjudication round; Wherein, if the output results after executing the determination strategy in the first round are consistent, the weights of each executor in the first round remain unchanged; If the output results are consistent after the second round of execution of the judgment strategy, then the first preset score is deducted from the abnormal execution body that was replaced in the first round of judgment, and the first preset score is added to the remaining execution bodies in the first round. If the output results are consistent after the third round and subsequent rounds of executing the determination strategy, then the execution body whose output result is consistent with the convergence result in the first round of determination is increased by a second preset score, and the other execution bodies in the first round are deducted from the second preset score, and the second preset score is greater than the first preset score.

[0008] According to a mimicry adjudication method provided by the present invention, after updating the weights of each executor in the first round based on the current adjudication round, when the weight of any executor in the first round reaches a preset global upper limit or a preset global lower limit, the method further includes: Obtain the maximum and minimum weights of each executor in the first round; The current weights of each executor in the first round are linearly mapped to a preset safe interval, such that the original maximum value corresponds to the upper limit of the preset safe interval and the original minimum value corresponds to the lower limit of the preset safe interval. The weights of other executors in the first round are synchronously mapped to the preset safe interval according to their original weights relative to the maximum and minimum values. Update the mapped weights to the current weights of each executor in the first round; Wherein, the upper limit of the preset safe interval is less than the preset global upper limit, and the lower limit of the preset safe interval is greater than the preset global lower limit.

[0009] According to a mimicry adjudication method provided by the present invention, the replacement of the triggering abnormal executor includes: If the current adjudication round is the first round, mark the executors that are not assigned to the relatively trustworthy group among the online executors as abnormal executors and replace them; If the current adjudication round is the second round, among the online executors, the executor whose output result is inconsistent with the output result of the benchmark executor is marked as the abnormal executor and replaced. The benchmark executor is the executor that was re-entered online in the first round of replacement operation. If the current adjudication round is the third round or a subsequent round, all online executors will be marked as abnormal executors and replaced.

[0010] According to a mimicry adjudication method provided by the present invention, before issuing pending requests to each online executor in the heterogeneous executor pool, the method further includes: Verify whether the number of each online executor is greater than or equal to a preset security threshold; If so, and the interval since the last time the abnormal executor was replaced is greater than the update duration threshold, then at least one executor is selected from the online executors to sequentially perform abnormal offline, cleaning and reconstruction, and re-online operations, and the timer for the interval duration is reset.

[0011] According to a mimicry adjudication method provided by the present invention, before issuing the pending request to each online executor in the heterogeneous executor pool, the method further includes: caching the pending request to a request queue; After outputting the convergence result as the adjudication result, the method further includes: sequentially releasing the cached pending requests from the request queue, and sending the adjudication result to downstream business nodes.

[0012] According to a mimicry adjudication method provided by the present invention, the normalization of each response data includes: If the request to be processed is a flow table query instruction, then the flow table rules in each response data are sorted according to priority and matching field, and redundant fields are removed. If the request to be processed is a network configuration instruction, then the non-critical fields in each response data are filtered out, and the filtered response data are recombined according to a preset sequence; If the request to be processed is a topology status query instruction, then the topology node and link information in each response data will be converted into a unified structured graph model.

[0013] The present invention also provides a mimicry adjudication device, comprising: The request distribution and normalization module is used to distribute pending requests to each online executor in the heterogeneous executor pool, collect the response data generated by each online executor in response to the pending request, normalize each response data, and generate an initial normalized dataset. The first adjudication module is used to input the initial normalized dataset into a multi-round progressive adjudication process to execute the judgment strategy progressively according to the adjudication round, wherein the first round executes a relatively reliable judgment strategy, the second round executes an absolutely reliable judgment strategy, and the third and subsequent rounds execute a global forced reset judgment strategy. The second adjudication module is used to trigger the replacement of the abnormal executor if the output results of each online executor are inconsistent after each execution of the corresponding judgment strategy. After replacing the abnormal executor, the response data is collected again and normalized to generate the next round of judgment dataset. The convergence and output module is used to take the next round of judgment dataset as the current input of the multi-round progressive adjudication process for the next round of judgment, and iterate in a loop until the output results of each online executor converge to be consistent, and output the convergence result as the adjudication result.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the mimicry adjudication method as described above.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the mimicry adjudication method as described above.

[0016] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the mimicry adjudication method as described above.

[0017] This invention provides a mimicry adjudication method, apparatus, device, and storage medium. It issues pending requests to each online executor in a heterogeneous executor pool and collects response data generated in response to these requests. After normalizing the response data, an initial normalized dataset is obtained and input into a multi-round progressive adjudication process. The dataset is then subjected to relative reliability, absolute reliability, and global forced reset judgment strategies sequentially according to the adjudication round. After each execution of the corresponding judgment strategy, if the outputs of the online executors are inconsistent, an abnormal executor is triggered for replacement. After replacement, normalized data is re-collected and input into the process for iterative iteration until the outputs of all executors converge and a judgment result is output. This invention provides a step-by-step intervention path through a progressive heterogeneous judgment strategy, achieving state error correction when most outputs deviate. It isolates abnormal nodes through inconsistency-triggered replacement and re-collection iteration, preventing error accumulation. Furthermore, it provides a clear process exit condition through a closed-loop mechanism that iterates until convergence, thereby overcoming the majority misjudgment defects of single-round static voting, avoiding deadlock in the adjudication process, and achieving accurate convergence of adjudication results and continuous business output under complex interference. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 A schematic diagram of the system architecture of the mimicry adjudication method provided by the present invention.

[0020] Figure 2 This is a flowchart illustrating the mimicry adjudication method provided by the present invention.

[0021] Figure 3 This is a schematic diagram of a multi-round progressive adjudication process in a common-cause attack scenario provided by the present invention.

[0022] Figure 4 This is one of the structural schematic diagrams of the mimicry adjudication device provided by the present invention.

[0023] Figure 5 This is the second schematic diagram of the mimicry adjudication device provided by the present invention.

[0024] Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0026] In heterogeneous executor collaborative adjudication scenarios, existing systems mostly employ a single-round static majority voting mechanism. This mechanism synchronously sends requests to each online executor, collects responses, and directly performs consistency judgment and outputs results based on the principle that the majority output is correct. This mode relies on a single static comparison and lacks dynamic verification and state intervention capabilities. However, under actual complex interference or long-term operation, this mechanism in existing technologies is susceptible to common-cause attacks, leading to deviations in the coordination of the majority nodes and causing majority errors. Simultaneously, the lack of means to identify and isolate abnormal nodes causes errors to accumulate continuously. Furthermore, when the judgment results remain inconsistent, the system, lacking gradual intervention and a clear convergence exit point, is highly prone to deadlock in the adjudication process or business interruption.

[0027] To address the aforementioned problems in existing technologies, this invention provides a mimicry adjudication method, apparatus, device, and storage medium. The inventive concept of this invention lies in proposing a mimicry adjudication approach based on progressive strategy switching and dynamic anomaly isolation iteration. This reconstructs a single-round static judgment into a multi-level progressive intervention process. Different judgment strategies are executed progressively in each round to provide a step-by-step error correction path. When the output deviates, anomaly node replacement and response resampling are triggered immediately to block error propagation. A closed-loop mechanism is constructed that iterates cyclically until the result converges. This transforms the adjudication logic from passively accepting the majority to actively progressively correcting deviations, ensuring reliable convergence of the adjudication process and accurate result output even under complex interference.

[0028] The mimicry adjudication method, apparatus, device, and storage medium provided by the present invention are described below with reference to the accompanying drawings and embodiments.

[0029] Figure 1 This diagram illustrates the system architecture of the mimicry adjudication method provided by the present invention. In the mimicry defense architecture, the heterogeneous execution pool serves as the carrier for realizing dynamic heterogeneous redundancy. To enhance the credibility and convergence capability of mimicry adjudication in complex adversarial environments, this diagram schematically illustrates the deployment environment using an SDN controller platform such as ONOS. Figure 1 As shown, this system architecture can include a heterogeneous execution pool, a southbound proxy service, and an intrinsic security component. The heterogeneous execution pool is deployed at the bottom layer of the architecture and can contain multiple executions built based on different hardware and software environments. For example, the execution pool may be configured with five heterogeneous executions (e.g., ...). Figure 1The system comprises P1 to P5, with three running online and two in standby mode. The online executables P1, P2, and P3 can be deployed on environments such as Ubuntu 20.04 + ONOS 2.7.0, CentOS 8 + ONOS 2.5.0, and Debian 11 + ONOS 2.6.0, respectively. The standby executables P4 and P5 can be deployed on environments such as Ubuntu 22.04 + ONOS 2.5.2 and Debian 12 + ONOS 2.7.1.

[0030] At the data interaction level, southbound communication messages generated by each online executor are uniformly forwarded and protocol adapted via the southbound proxy service to ensure control plane functional interaction and to direct the response data generated by each executor for the pending request to the intrinsic security component. The intrinsic security component integrates a scheduling service and a multi-round adjudication module. The scheduling service monitors and maintains the online or standby state switching of each executor in real time through the executor state synchronization link. The multi-round adjudication module is used to obtain the adjudication information returned by each executor, perform normalization processing of the information, and drive a multi-round progressive adjudication process. The multi-round adjudication module's process can be as follows: the normalized response data first enters the first round of adjudication for relatively reliable judgment. If the output results are not converged and consistent, an abnormal executor is triggered to replace and the dataset is refreshed, entering the second round of absolutely reliable judgment for baseline error correction. If still inconsistent, a third round of global forced reset adjudication is performed. When the judgment results of any round converge and consistent, the process is marked as adjudicated successfully and the adjudication result is output.

[0031] Furthermore, to support the relatively reliable grouping determination and subsequent differentiated update mechanism in the multi-round progressive adjudication of the mimicry adjudication method provided by this invention, a reputation weight value needs to be configured for each executor. For example, the initial weight of each executor is configured to 50 points, responsible for processing flow table information, network configuration (netcfg), device status information, and topology information. To drive the orderly execution and state management of each step of the mimicry adjudication method provided by this invention, corresponding operating parameters also need to be configured. For example, the operating parameters can be configured as follows: a periodic adjudication cycle of 20 seconds, a forced update duration threshold of 1200 seconds, global upper and lower limits of weight of 1 point and 100 points respectively, and a maximum request queue length of 1000, etc.

[0032] It should be noted that in actual operating conditions, the number of execution entities, hardware / software combinations, and operating parameters can be adaptively adjusted according to the actual business load and security level. Furthermore, the mimicry adjudication method provided by this invention is also applicable to other SDN controllers or general intrinsic security architecture systems. Figure 1 This is merely an illustrative example.

[0033] Figure 2 This is a flowchart illustrating the mimicry adjudication method provided by the present invention, as shown below. Figure 2 As shown, the method includes: S101. Verify whether the number of each online executor is greater than or equal to the preset safety threshold.

[0034] The adjudication service periodically calls the scheduling service interface in the intrinsic security component to obtain information about online executors, thereby determining the set of executors participating in the adjudication. It then calls the corresponding interface of each executor to obtain the relevant data required for the adjudication. Each time a periodic adjudication is triggered, a pre-verification check is performed. For example, it first checks whether the number of online executors is greater than or equal to a preset security threshold. The preset security threshold refers to the minimum number of online nodes required to maintain the validity of multiple rounds of adjudication, preventing the judgment from losing statistical significance due to insufficient heterogeneous redundancy. For example, the preset security threshold can be configured to 3, and the check might verify whether the current number of online executors is greater than or equal to 3. If the number of online executors is less than the preset security threshold (i.e., the judgment result is negative), it indicates that the number is insufficient to guarantee the statistical reliability of the adjudication. The current adjudication process is then terminated, and step S102a can be executed to generate a warning message, waiting for a backup executor to be added before re-executing step S101. Conversely, if the number of online executors is greater than or equal to the preset security threshold (i.e., the judgment result is positive), step S102b is executed, which checks whether the interval since the last abnormal executor replacement is greater than the update time threshold.

[0035] S102a. Generate a warning message to instruct the standby executor to supplement.

[0036] S102b: Determine whether the interval since the last time the abnormal execution body was replaced is greater than the update time threshold.

[0037] The update duration threshold is used to prevent the execution entity state from becoming fixed due to long-term latent attacks; for example, it can be configured to be 1200 seconds. After judgment, if the interval duration is greater than the update duration threshold, step S103 is executed, which involves selecting at least one execution entity from the online execution entities to sequentially perform abnormal offline, cleanup and reconstruction, and re-online operations, and resetting the interval timer. Conversely, if the interval duration is less than or equal to the update duration threshold, it is determined that the current execution entity pool is within a safe update cycle, and the update operation in step S103 can be skipped, maintaining the current node state and directly proceeding to the subsequent pending request distribution process, i.e., step S104.

[0038] S103. Select at least one execution unit from each online execution unit to sequentially perform abnormal offline, cleaning and reconstruction and re-online operations, and reset the timer for the interval.

[0039] For example, strategies such as random selection, polling based on online time, or evaluation based on operational metrics can be used to select at least one executor and sequentially perform abnormal offline, cleanup and reconstruction, and re-online operations to reset the executor's state. Abnormal offline immediately cuts off the executor's business traffic and communication links, achieving node isolation to prevent the spread of potential contamination. Cleanup and reconstruction clears runtime memory, resets system configuration, and reloads the baseline security image to delete any potential latent control chains or traces of common-cause attacks. Re-online involves reinjecting the updated executor into the online pool, allowing it to participate in subsequent adjudication processes. Furthermore, resetting the timer interval can redetermine the security cycle for the next forced update, ensuring the dynamic trustworthiness of the executor pool while avoiding frequent state switching that could interfere with the continuity of adjudication.

[0040] S104. Send pending requests to each online executor in the heterogeneous executor pool, collect the response data generated by each online executor for the pending requests, normalize each response data, and generate an initial normalized dataset.

[0041] For example, the SDN controller synchronously sends pending requests to each online executor in the heterogeneous executor pool, triggering parallel computation within the same business context. Subsequently, it collects response data independently generated by each executor in real time to obtain the original output state and potential discrepancies. Then, addressing the format heterogeneity and field differences in the original responses, the SDN controller standardizes, aligns, and cleans the collected response data, eliminating incomparable structural noise and generating an initial normalized dataset with unified structure and aligned fields.

[0042] In some embodiments, the normalization processing of each response data can be implemented using differentiated adaptation rules based on the instruction type of the request to be processed. For example, when the request to be processed is a flow table query instruction, the flow table rules in each response data can be sorted according to priority and matching fields, and redundant fields can be removed to eliminate the impact of differences in rule arrangement order on consistency comparison. When the request to be processed is a network configuration instruction, non-critical fields are filtered and the filtered response data is recombined according to a preset sequence to ensure the standardization of configuration parameter structure. When the request to be processed is a topology status query instruction, the topology node and link information in each response data is converted into a unified structured graph model to achieve lossless mapping of heterogeneous topology representation. It should be noted that this normalization rule is only an illustrative example. In actual working conditions, equivalent substitutions can be made according to actual business protocols and comparison requirements, and this invention does not limit this.

[0043] It's important to note that before issuing pending requests to each online executor in the heterogeneous executor pool, these requests are temporarily stored in a request queue for isolation and caching. During the multi-round progressive adjudication process, the request queue remains frozen, preventing new requests from interfering with the real-time state of the executors and avoiding timing misalignment or state drift caused by concurrent requests. Once the adjudication process iterates until the output results converge and are consistent, and the convergence result is output as the adjudication result, the SDN controller releases the cached pending requests from the request queue sequentially, completing the final issuance of the business instruction. This caching isolation ensures that multiple rounds of decision-making are always based on the same transaction, effectively maintaining baseline consistency in the adjudication process and continuous output of control plane services.

[0044] S105. Input the initial normalized dataset into the multi-round progressive adjudication process to execute the judgment strategy progressively according to the adjudication round.

[0045] In this round, a relatively reliable determination strategy is executed in the first round, an absolutely reliable determination strategy is executed in the second round, and a global forced reset determination strategy is executed in the third round and subsequent rounds.

[0046] The initial normalized dataset is input into a multi-round progressive adjudication process. Judgment strategies are executed sequentially according to a preset round. For example, the first round uses a relatively reliable judgment strategy to perform preliminary grouping and reliability assessment based on local anomalies. If the output results do not converge, the second round executes an absolutely reliable judgment strategy, using a high-reliability benchmark node as a reference for forced error correction. If the output results are still inconsistent, a global forced reset judgment strategy is executed in the third and subsequent rounds, performing a full state reset of the execution pool.

[0047] S106a. After each execution of the decision strategy, if the output results of each online executor converge to the same value, the multi-round progressive decision process is terminated, and the convergence result is output as the decision result.

[0048] S106b. After each execution of the corresponding judgment strategy, if the output results of each online executor are inconsistent, an abnormal executor is triggered for replacement. After replacing the abnormal executor, the response data is collected again and normalized to generate the judgment dataset for the next round.

[0049] After each execution of the corresponding decision-making strategy, for example, the SDN controller compares the output results of each online executor and performs subsequent control based on the comparison results. Specifically, if the output results of each online executor converge to the same value, it is determined that the current executor pool has reached a valid consensus. The SDN controller terminates the multi-round progressive decision-making process and outputs the convergence result as the decision result. This branch proposes a clear process exit mechanism to avoid unnecessary iterations consuming computing resources and ensure low-latency response and deterministic release of control plane services. If the output results of each online executor are inconsistent, it is determined that there is a disagreement or abnormal interference in the current decision. The SDN controller triggers the replacement of the abnormal executor. After replacing the abnormal executor, the response data is re-collected and normalized to generate the decision dataset for the next round. This branch can cut off the error propagation link by isolating anomalies and updating the baseline, providing new data input for the next round of policy decision-making and driving the system state towards the convergence point.

[0050] S107. Use the next round of judgment dataset as the current input for the multi-round progressive adjudication process, and iterate in a loop until the output results of each online executor converge and are consistent, and output the convergence result as the adjudication result.

[0051] The normalized dataset for the next round of decision is fed into the multi-round progressive decision-making process as input for the new round of decision-making, allowing the execution pool to re-determine the outcome under the progressive strategy. If no consensus is reached in this round of decision-making, the process automatically returns to the exception replacement and data update branch, continuing the iterative loop until the outputs of all online execution entities converge completely. Once convergence is detected, the consistent output can be locked and used as the final decision result. This effectively eliminates the risks of state drift and computational deadlock during multi-round decision-making through deterministic iterative convergence conditions and clear process exit boundaries, ensuring that the system can still obtain a high-confidence deterministic output under complex disturbances.

[0052] The mimicry adjudication method provided by this invention issues pending requests to each online executor in a heterogeneous executor pool and collects response data generated in response to the pending requests. After normalizing the response data, an initial normalized dataset is obtained and input into a multi-round progressive adjudication process, which sequentially performs judgment strategies based on relative reliability, absolute reliability, and global forced reset according to the adjudication round. After each execution of the corresponding judgment strategy, if the output results of each online executor are inconsistent, an abnormal executor is triggered for replacement. After replacement, normalized data is re-collected and input into the process for iterative looping until the outputs of all executors converge to a consistent level, and an adjudication result is output. This invention provides a step-by-step intervention path through a progressive heterogeneous adjudication strategy, achieving state error correction when most outputs deviate. It also isolates abnormal nodes through inconsistency-triggered replacement and re-collection iteration, preventing error accumulation. Furthermore, it provides a clear process exit condition through a closed-loop mechanism that iterates until convergence, thereby overcoming the majority misjudgment defects of single-round static voting, avoiding deadlock in the adjudication process, and achieving accurate convergence of adjudication results and continuous business output under complex interference.

[0053] In the iterative loop of the above steps, if the output results of each online executor are inconsistent, an abnormal executor is triggered for replacement to block error propagation. Specifically, if the current adjudication round is the first round, that is, after executing the relatively trustworthy determination strategy in the first round, the output results of each online executor are inconsistent, the SDN controller, based on the grouping results of the relatively trustworthy determination strategy, marks the executors that are not classified into the relatively trustworthy group as abnormal executors and replaces them. This design logic follows the principle of local fault tolerance, prioritizing the isolation of nodes with low initial trustworthiness and retaining high-weight nodes as benchmark references to avoid global cleanup caused by local deviations in the first round. If the current adjudication round is the second round, meaning the outputs of the online executors are inconsistent after the absolute trust determination strategy of the second round, the SDN controller will use the executor that was re-brought online in the first round of replacement operations as the baseline executor. Since this executor has completed standardization and cleaning and is in a standard initialization state that removes historical operational remnants, the SDN controller uses its output as the judgment benchmark. Online executors whose outputs are inconsistent with the baseline executor's output are marked as abnormal executors and replaced, achieving an upgrade from a relatively fault-tolerant strategy to an absolute alignment strategy. This forcibly removes interfering nodes that deviate from the security benchmark and blocks collaborative attack chains. However, if the current adjudication round is the third round or later, it indicates that the system faces complex common-cause attacks or deep state contamination. In this case, the SDN controller executes a global forced reset judgment strategy, marking all current online executors as abnormal executors and triggering a full replacement. Healthy nodes in the backup pool simultaneously take over online slots, cutting off potential control links and hidden paths.

[0054] Furthermore, after the replacement action is completed, the abnormal node is taken offline and isolated and the backup node is brought online and injected. The pending request is resent to the updated execution pool, the newly generated response data is collected and normalized, and the next round of judgment dataset is generated. This dataset is then used as the current input to reconnect to the multi-round progressive adjudication process and continue to participate in the next round of strategy judgment.

[0055] The mimicry adjudication method provided by this invention employs a multi-level isolation strategy, including local initial screening, benchmark error correction, and global reset, to prevent abnormal nodes from continuously participating in business interactions. Combined with the baseline update operation after replacement, it ensures that each iteration runs on a clean and comparable decision-making environment, eliminating the risk of error accumulation and process deadlock, and providing a reliable convergence driving force for multi-round adjudication.

[0056] In some embodiments, when an online executor is determined to be abnormal and is taken offline, a healthy candidate executor can be prioritized from the standby pool to take over the online slot and migrate the business load. The abnormal executor that is replaced and taken offline then enters a standardized cleaning process, which sequentially performs memory state erasure, storage data reset, and security baseline image reload to thoroughly remove potential attack traces and state contamination. After the cleaning process is completed, the executor is restored to its initial trusted state and rejoined the standby pool for standby. When the standby pool resources are exhausted, an emergency takeover strategy can be enabled to directly reactivate the executor that has just completed the cleaning process and bring it online to the online pool. Through this dynamic replacement and cleaning return mechanism, the number of online executors is always dynamically maintained above a preset security baseline, such as 3 or 5, thereby ensuring the minimum redundancy and majority decision validity required for heterogeneous adjudication and achieving sustainable rotation and high availability of the executor pool.

[0057] In some embodiments, after the multi-round progressive adjudication process converges and outputs the adjudication result, the mimicry adjudication method provided by the present invention further includes a reputation feedback and weight maintenance step, which aims to establish a dynamic reputation evaluation model that is positively correlated with the difficulty of adjudication convergence. Through a round-bound differentiated update strategy and an anti-polarization remapping algorithm, the long-term accurate characterization of the executor's credibility can be achieved.

[0058] The weight update tied to the round is to obtain the current adjudication round at which convergence is achieved, and to update the initial weights of each executor in the first round accordingly. For example, if convergence is achieved in the first round, it indicates that the output of each node is highly reliable and there is no abnormal interference. In this case, the weights of each executor in the first round remain unchanged, maintaining the initial reputation baseline. If convergence is achieved after the absolute trust determination in the second round, it indicates that there is a local anomaly or slight interference. In this case, the first preset score is deducted from the abnormal executor that was marked and replaced in the first round, and the first preset score is increased for the remaining executors in the first round, realizing the reputation penalty and compensation of the secondary error correction process. If convergence is achieved only after a global forced reset in the third round and subsequent rounds, it indicates that the system is facing complex common-cause attacks or deep state pollution. In this case, the second preset score is increased for the executors whose output results are consistent with the final convergence result in the first round, and the second preset score is deducted for the remaining executors in the first round. The second preset score is strictly greater than the first preset score. This differentiated update allows the weight adjustment magnitude to be coupled with the difficulty of adjudication convergence, ensuring that the reputation score can objectively reflect the true credibility of the enforcer in an adversarial environment.

[0059] Furthermore, after updating the weights of each executor in the first round according to the current adjudication round, weight boundary verification can be performed simultaneously. For example, when it is detected that the current weight of any executor in the first round reaches the preset global upper limit or the preset global lower limit, an anti-polarization remapping mechanism can be triggered. Specifically, the maximum and minimum values ​​of the weights of each executor in the current first round are obtained, and the current weights of each executor are linearly mapped to the preset safe interval according to the proportion. During the mapping process, the original maximum weight corresponds to the upper limit of the preset safe interval, the original minimum weight corresponds to the lower limit of the preset safe interval, and the weights of other executors in the first round are synchronously mapped according to the relative proportion of their original weights between the maximum and minimum values. After the mapping is completed, the updated weight values ​​are used as the current weights of each executor. The upper and lower limits of the preset safe interval are both within the internal threshold range of the preset global upper and lower limits, that is, the upper limit of the preset safe interval is smaller than the preset global upper limit, and the lower limit of the preset safe interval is larger than the preset global lower limit. By using this weight boundary verification, the weight distribution scale can be dynamically reset while keeping the relative reputation ranking of each node unchanged. This effectively avoids the monopoly of a single node's weight or the depletion of its reputation, and reserves sufficient dynamic adjustment space for subsequent long-term adjudication.

[0060] It should be noted that the above weight update and anti-polarization mapping mechanism are only performed when convergence is consistent. The specific values ​​of the first preset score, the second preset score, and the preset safety interval can be set according to the actual working conditions. For example, the first preset score can be configured as 5 points, the second preset score can be configured as 10 points, and the preset safety interval can be set as [20, 80], etc. Those skilled in the art can make adaptive adjustments to the above score parameters and safety intervals according to the actual business security level and computing resources. This invention does not limit this.

[0061] Figure 3 This is a schematic diagram of a multi-round progressive adjudication process in a common-cause attack scenario provided by the present invention, as shown below. Figure 3 As shown, the following example of a common cause attack scenario involving SDN controller flow table tampering illustrates the execution logic of the mimicry adjudication method provided by this invention in an adversarial environment. In the initial state, P1, P2, and P3 are running online with a weight of 50 points each. P2 and P3 are subjected to a common cause attack and output an incorrect flow table key value Ip2, while P1 outputs the correct value Ip1.

[0062] Step S1: Pre-verification and request control.

[0063] The SDN controller triggers an adjudication task periodically, verifies that the number of online nodes meets the safety baseline, and detects that the interval since the last refresh has exceeded the update duration threshold. It then randomly selects P1 to perform offline cleaning and re-online operations. Simultaneously, pending business requests are cached and frozen in the request queue, and released uniformly after the adjudication converges.

[0064] Step S2: First round of relatively reliable judgment and anomaly replacement.

[0065] The controller sends a request to the online nodes and normalizes the collected responses. P1 outputs Ip1, and P2 and P3 output Ip2. Relative trust grouping is performed based on the initial weights. The maximum combined weight of P2 and P3 is tentatively designated as the trust group, and P1 is marked as an anomaly. Because the output results do not converge, the controller triggers P1 to replace and schedules P4 to come online. After re-collecting and normalizing the data, the next round begins.

[0066] Step S3: Second round of absolutely reliable judgment and benchmark error correction.

[0067] The system adopts an absolutely trusted strategy, using the recently re-launched P4 as the absolutely trusted benchmark. Data collection shows that P4 outputs Ip1, while P2 and P3 output Ip2. The controller determines that P2 and P3 deviate from the benchmark, marks them as anomalies, and performs replacement. After re-sampling and normalization, a third-round dataset is generated.

[0068] Step S4: Third round of global reset judgment and convergence output.

[0069] Entering the global forced reset strategy, the collected data shows that P4, P5, and P1 all output Ip1, and the results of each node are completely converged and consistent. The controller terminates the multi-round iteration process and locks Ip1 as the final decision result.

[0070] Step S5: Round-binding weight update and queue release.

[0071] After the decision converges, the controller backtracks to the first round of output states based on the final result and applies differentiated weighted rewards and penalties. If P1's output matches the final state, it gains 2 points, bringing its total to 52 points. If P2 and P3's outputs do not match, they each lose 2 points, bringing their total to 48 points. P4 and P5 remain at 50 points. After the weight maintenance is complete, the controller unfreezes the queue and releases cached requests sequentially according to the final state Ip1.

[0072] Step S6: Weighted anti-polarization remapping.

[0073] To address potential weight polarization scenarios during long-term operation, if the weight distribution is detected to reach the global boundary (e.g., P1=1 point, P2=100 points, P3=55 points, P4=60 points, P5=45 points), the controller triggers linear order-preserving mapping. The extreme values ​​are compressed to a safe range according to the preset formula shown below. The calculated mapping values ​​are 20 points for P1, 80 points for P2, 53 points for P3, 56 points for P4, and 47 points for P5. While maintaining the relative reputation ranking, the distribution scale is reset, restoring the flexibility of dynamic adjustment.

[0074] Wnew=20+(Wold-Wmin)×60 / (Wmax-Wmin) Wnew represents the updated weights, Wold represents the original weights, and Wmax and Wmin represent the maximum and minimum values, respectively.

[0075] The above embodiments illustrate the full lifecycle execution chain of the mimicry adjudication method under common-cause attack interference. Those skilled in the art can make equivalent substitutions for the threshold parameters, score rules, and mapping ranges according to actual business load, security level, or data protocol; this invention does not limit this. Furthermore, it is understood that the descriptions of the above embodiments using an SDN controller as the deployment platform are merely illustrative examples. The mimicry adjudication logic provided in this application can also be adapted to other mimicry defense scenarios such as Web application firewalls, industrial control systems, and cloud-native gateway decision engines. Moreover, without departing from the technical concept of this invention, equivalent substitutions and adaptive modifications to the execution pool size, data interaction protocol, or business normalization rules all fall within the protection scope of this invention.

[0076] As can be seen from the description of the above embodiments, the mimicry adjudication method provided by the present invention has at least the following beneficial effects: By constructing a multi-round progressive adjudication process and a heterogeneous strategy switching mechanism, the misjudgment defects caused by the default assumption of majority rule under common-cause attacks in traditional redundant architectures can be effectively overcome. In the first round of adjudication, potential anomalies are identified through relatively reliable initial screening based on weighted grouping. In the second round, the system proactively switches to the absolute reliability principle to implement mandatory baseline error correction. If convergence is still not achieved, the system advances to a global mandatory reset strategy. This progressive logical switching can break the rigid paradigm of static majority voting, preventing attackers from continuously bypassing the defense based on a single strategy, and significantly improving the system's dynamic adaptive defense capability in complex adversarial environments.

[0077] In addition, by combining a differentiated weight update mechanism that is deeply tied to the convergence rounds of the adjudication, the reputation of the executor can be accurately and dynamically characterized. The reward and punishment range is automatically matched according to the specific rounds required to achieve convergence. The more times the adjudication is iterated, the more complex the interference or the higher the degree of node abnormality. The corresponding weight adjustment step size is also increased accordingly, so that the reputation score can highly sensitively map the true trustworthiness of the executor. This effectively accelerates the identification, isolation and reconstruction of the reputation system of abnormal nodes and avoids the spread of local errors to the global system.

[0078] Furthermore, regarding system robustness and process consistency assurance, this application incorporates a timed forced update mechanism in the pre-adjudication stage. This mechanism automatically triggers random node cleanup, reconstruction, and re-launch operations when the execution pool experiences prolonged periods without state changes, effectively cutting off potential attack links for attackers to exploit long-term latency for consistency compromise. Simultaneously, through a request queue caching isolation mechanism, all pending instructions are frozen and temporarily stored during the adjudication cycle, only being released sequentially after multiple rounds of iterative convergence. This eliminates execution state drift caused by a surge of new requests during the adjudication process, ensuring absolute consistency of the baseline comparisons across multiple rounds and the continuous and reliable execution of control plane services.

[0079] Furthermore, for the complex data interaction scenarios unique to the control plane of software-defined networks, this invention incorporates multi-dimensional normalization adaptation rules at the data acquisition and processing end. These rules can effectively eliminate format noise and expression differences in heterogeneous execution environments by sorting and reorganizing flow table rules according to priority and matching domain, filtering non-critical fields in network configuration instructions, and performing structured graph model transformation on topology states.

[0080] Through the aforementioned synergistic effect, the multi-round heterogeneous adjudication logic can be accurately adapted to high-concurrency and high-dynamic control plane business scenarios, ultimately achieving accurate convergence of adjudication results, effective blocking of abnormal interference, and deterministic issuance of core control commands under complex interference and long-cycle operation environments.

[0081] The mimicry adjudication device provided by the present invention is described below. The mimicry adjudication device described below can be referred to in correspondence with the mimicry adjudication method described above.

[0082] Figure 4 This is one of the structural schematic diagrams of the mimicry adjudication device provided by the present invention, such as... Figure 4 As shown, the mimicry adjudication device 400 includes: The request distribution and normalization module 401 is used to distribute pending requests to each online executor in the heterogeneous executor pool, collect the response data generated by each online executor for the pending requests, normalize each response data, and generate an initial normalized dataset. The first adjudication module 402 is used to input the initial normalized dataset into the multi-round progressive adjudication process to execute the judgment strategy progressively according to the adjudication round. The first round executes the relatively reliable judgment strategy, the second round executes the absolutely reliable judgment strategy, and the third and subsequent rounds execute the global forced reset judgment strategy. The second adjudication module 403 is used to trigger the replacement of abnormal executors if the output results of each online executor are inconsistent after each execution of the corresponding judgment strategy. After replacing the abnormal executor, the response data is collected again and normalized to generate the next round of judgment dataset. The convergence and output module 404 is used to take the current input of the next round of decision dataset as the current input of the multi-round progressive decision process, and iterate in a loop until the output results of each online executor converge to be consistent, and output the convergence result as the decision result.

[0083] In one possible design, the convergence and output module 404 is also used for: After each execution of the decision strategy, if the output results of each online executor converge to the same value, the multi-round progressive decision process is terminated, and the convergence result is output as the decision result.

[0084] exist Figure 4 On this basis, Figure 5 This is a second schematic diagram of the mimicry adjudication device provided by the present invention, as shown below. Figure 5 As shown, the mimicry adjudication device 400 further includes: a weight update module 405, used for: Obtain the current adjudication round when the output results are consistent, and update the weights of each executor in the first round based on the differences in the current adjudication round; If the output results after the first round of executing the judgment strategy are consistent, then the weights of each executor in the first round remain unchanged. If the output results are consistent after the second round of execution of the judgment strategy, the first preset score is deducted from the abnormal execution body that was replaced in the first round of judgment, and the first preset score is added to the remaining execution bodies in the first round. If the output results after executing the judgment strategy in the third round and subsequent rounds are consistent, then the execution body whose output result is consistent with the convergence result in the first round of judgment will have a second preset score added, and the other execution bodies in the first round will have a second preset score deducted, and the second preset score is greater than the first preset score.

[0085] In one possible design, when the weight of any executor in the first round reaches a preset global upper limit or a preset global lower limit, the weight update module 405 is further used for: Obtain the maximum and minimum weights of each executor in the first round; The current weights of each executor in the first round are linearly mapped to the preset safe interval, so that the original maximum value corresponds to the upper limit of the preset safe interval, and the original minimum value corresponds to the lower limit of the preset safe interval. The weights of other executors in the first round are synchronously mapped to the preset safe interval according to the relative proportion of their original weights between the maximum and minimum values. Update the mapped weights to the current weights of each executor in the first round; Among them, the upper limit of the preset safe zone is smaller than the preset global upper limit, and the lower limit of the preset safe zone is larger than the preset global lower limit.

[0086] In one possible design, the second adjudication module 403 is also used for: If the current adjudication round is the first round, the executors in each online executor that are not classified into the relatively trustworthy group will be marked as abnormal executors and replaced; If the current adjudication round is the second round, among the online executors, the executor whose output result is inconsistent with the output result of the benchmark executor is marked as an abnormal executor and replaced. The benchmark executor is the executor that was re-brought online in the first round of replacement operation. If the current adjudication round is the third round or subsequent rounds, all online executors will be marked as abnormal executors and replaced.

[0087] In one possible design, the request distribution and normalization module 401 is also used for: Verify whether the number of each online executor is greater than or equal to the preset safety threshold; If so, and the interval since the last abnormal execution body replacement is longer than the update time threshold, then at least one execution body is selected from each online execution body to sequentially perform abnormal offline, cleaning and reconstruction, and re-online operations, with the timer for the interval being the same.

[0088] In one possible design, the request distribution and normalization module 401 is also used to cache the pending requests to the request queue before distributing the pending requests to each online executor in the heterogeneous executor pool. After outputting the convergence result as the adjudication result, it is also used to: release the cached pending requests from the request queue in sequence, and send the adjudication result to the downstream business nodes.

[0089] In one possible design, the request distribution and normalization module 401 is specifically used for: If the request to be processed is a flow table query command, then the flow table rules in each response data are sorted according to priority and matching field, and redundant fields are removed. If the request to be processed is a network configuration command, then the non-critical fields in each response data are filtered, and the filtered response data is reassembled according to a preset sequence; If the request to be processed is a topology status query command, then the topology node and link information in each response data will be converted into a unified structured graph model.

[0090] Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 6 As shown, the electronic device may include: a processor 610, a communications interface 620, a memory 630, and a communications bus 840, wherein the processor 610, the communications interface 620, and the memory 630 communicate with each other through the communications bus 640. The processor 610 can call logic instructions in the memory 630 to execute a mimicry adjudication method. This method includes: issuing pending requests to each online executor in the heterogeneous executor pool; collecting response data generated by each online executor for the pending request; normalizing the response data to generate an initial normalized dataset; inputting the initial normalized dataset into a multi-round progressive adjudication process to progressively execute a judgment strategy according to the adjudication round, wherein the first round executes a relatively reliable judgment strategy, the second round executes an absolutely reliable judgment strategy, and the third and subsequent rounds execute a global forced reset judgment strategy; if the output results of each online executor are inconsistent after each execution of the corresponding judgment strategy, an abnormal executor is triggered for replacement; after replacing the abnormal executor, response data is re-collected and normalized to generate the next round's judgment dataset; using the next round's judgment dataset as the current input to the multi-round progressive adjudication process for the next round of adjudication, iterating cyclically until the output results of each online executor converge to a consistent level, and outputting the convergence result as the adjudication result.

[0091] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0092] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the mimicry adjudication method provided by the above methods. The method includes: sending pending requests to each online executor in a heterogeneous executor pool; collecting response data generated by each online executor in response to the pending requests; normalizing each response data to generate an initial normalized dataset; and inputting the initial normalized dataset into a multi-round progressive adjudication process to progressively execute the adjudication according to the adjudication rounds. The decision-making strategy is implemented as follows: the first round executes a relatively reliable decision-making strategy, the second round executes an absolutely reliable decision-making strategy, and the third and subsequent rounds execute a global forced reset decision-making strategy. If the output results of each online executor are inconsistent after each execution of the corresponding decision-making strategy, an abnormal executor is triggered for replacement. After replacing the abnormal executor, the response data is re-collected, normalized, and used to generate the decision dataset for the next round. The decision dataset for the next round is used as the current input for the multi-round progressive adjudication process, and the process is iterated until the output results of each online executor converge and become consistent. The convergence result is then output as the adjudication result.

[0093] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the mimicry adjudication method provided by the above-described methods. The method includes: issuing pending requests to each online executor in a heterogeneous executor pool; collecting response data generated by each online executor in response to the pending requests; normalizing the response data to generate an initial normalized dataset; inputting the initial normalized dataset into a multi-round progressive adjudication process to progressively execute a judgment strategy according to the adjudication round, wherein the first round executes a relatively reliable judgment strategy, the second round executes an absolutely reliable judgment strategy, and the third and subsequent rounds execute a global forced reset judgment strategy; if the output results of each online executor are inconsistent after each execution of the corresponding judgment strategy, an abnormal executor is triggered for replacement; after replacing the abnormal executor, the response data is re-collected and normalized to generate the next round judgment dataset; using the next round judgment dataset as the current input of the multi-round progressive adjudication process for the next round of judgment, iterating cyclically until the output results of each online executor converge to a consistent level, and outputting the convergence result as the adjudication result.

[0094] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; 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. Those skilled in the art can understand and implement this without any creative effort.

[0095] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A mimicry adjudication method, characterized in that, include: A request to be processed is sent to each online executor in the heterogeneous executor pool. The response data generated by each online executor in response to the request to be processed is collected. The response data is normalized to generate an initial normalized dataset. The initial normalized dataset is input into a multi-round progressive adjudication process to execute the judgment strategy progressively according to the adjudication round. The first round executes a relatively reliable judgment strategy, the second round executes an absolutely reliable judgment strategy, and the third and subsequent rounds execute a global forced reset judgment strategy. After each execution of the corresponding judgment strategy, if the output results of each online executor are inconsistent, an abnormal executor is triggered for replacement. After replacing the abnormal executor, the response data is collected again, normalized, and then the next round of judgment dataset is generated. The next round of judgment dataset is used as the current input of the multi-round progressive adjudication process for the next round of judgment. The process is iterated until the output results of each online executor converge and are consistent. The convergence result is then output as the adjudication result.

2. The method according to claim 1, characterized in that, After each execution of the determination strategy, if the output results of each online executor converge to the same value, the multi-round progressive adjudication process is terminated, and the convergence result is output as the adjudication result.

3. The method according to claim 2, characterized in that, If the outputs of the various online executors are consistent, the following also applies: Obtain the current adjudication round when the output results are consistent, and update the weights of each executor in the first round according to the differences in the current adjudication round; Wherein, if the output results after executing the determination strategy in the first round are consistent, the weights of each executor in the first round remain unchanged; If the output results are consistent after the second round of execution of the judgment strategy, then the first preset score is deducted from the abnormal execution body that was replaced in the first round of judgment, and the first preset score is added to the remaining execution bodies in the first round. If the output results are consistent after the third round and subsequent rounds of executing the determination strategy, then the execution body whose output result is consistent with the convergence result in the first round of determination is increased by a second preset score, and the other execution bodies in the first round are deducted from the second preset score, and the second preset score is greater than the first preset score.

4. The method according to claim 3, characterized in that, After updating the weights of each executor in the first round according to the current adjudication round, when the weight of any executor in the first round reaches a preset global upper limit or a preset global lower limit, the method further includes: Obtain the maximum and minimum weights of each executor in the first round; The current weights of each executor in the first round are linearly mapped to a preset safe interval, such that the original maximum value corresponds to the upper limit of the preset safe interval and the original minimum value corresponds to the lower limit of the preset safe interval. The weights of other executors in the first round are synchronously mapped to the preset safe interval according to their original weights relative to the maximum and minimum values. Update the mapped weights to the current weights of each executor in the first round; Wherein, the upper limit of the preset safe interval is less than the preset global upper limit, and the lower limit of the preset safe interval is greater than the preset global lower limit.

5. The method according to claim 1, characterized in that, The replacement of the trigger exception execution body includes: If the current adjudication round is the first round, mark the executors that are not assigned to the relatively trustworthy group among the online executors as abnormal executors and replace them; If the current adjudication round is the second round, among the online executors, the executor whose output result is inconsistent with the output result of the benchmark executor is marked as the abnormal executor and replaced. The benchmark executor is the executor that was re-entered online in the first round of replacement operation. If the current adjudication round is the third round or a subsequent round, all online executors will be marked as abnormal executors and replaced.

6. The method according to claim 1, characterized in that, Before issuing pending requests to each online executor in the heterogeneous executor pool, the method further includes: Verify whether the number of each online executor is greater than or equal to a preset security threshold; If so, and the interval since the last time the abnormal executor was replaced is greater than the update duration threshold, then at least one executor is selected from the online executors to sequentially perform abnormal offline, cleaning and reconstruction, and re-online operations, and the timer for the interval duration is reset.

7. The method according to claim 1, characterized in that, Before sending the pending request to each online executor in the heterogeneous executor pool, the method further includes: caching the pending request to a request queue; After outputting the convergence result as the adjudication result, the method further includes: sequentially releasing the cached pending requests from the request queue, and sending the adjudication result to downstream business nodes.

8. The method according to any one of claims 1 to 7, characterized in that, The normalization of each response data includes: If the request to be processed is a flow table query instruction, then the flow table rules in each response data are sorted according to priority and matching field, and redundant fields are removed. If the request to be processed is a network configuration instruction, then the non-critical fields in each response data are filtered out, and the filtered response data are recombined according to a preset sequence; If the request to be processed is a topology status query instruction, then the topology node and link information in each response data will be converted into a unified structured graph model.

9. A mimicry adjudication device, characterized in that, include: The request distribution and normalization module is used to distribute pending requests to each online executor in the heterogeneous executor pool, collect the response data generated by each online executor in response to the pending request, normalize each response data, and generate an initial normalized dataset. The first adjudication module is used to input the initial normalized dataset into a multi-round progressive adjudication process to execute the judgment strategy progressively according to the adjudication round, wherein the first round executes a relatively reliable judgment strategy, the second round executes an absolutely reliable judgment strategy, and the third and subsequent rounds execute a global forced reset judgment strategy. The second adjudication module is used to trigger the replacement of the abnormal executor if the output results of each online executor are inconsistent after each execution of the corresponding judgment strategy. After replacing the abnormal executor, the response data is collected again and normalized to generate the next round of judgment dataset. The convergence and output module is used to take the next round of judgment dataset as the current input of the multi-round progressive adjudication process for the next round of judgment, and iterate in a loop until the output results of each online executor converge to be consistent, and output the convergence result as the adjudication result.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the mimicry adjudication method as described in any one of claims 1 to 8.

11. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the mimicry ruling as described in any one of claims 1 to 8.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the mimicry ruling as described in any one of claims 1 to 8.

Images