Information alignment system based on AI proxy network and dynamic consensus mechanism

Abstract

The invention relates to the crossing field of artificial intelligence and enterprise computing, in particular to an information alignment system based on an AI proxy network and a dynamic consensus mechanism, which comprises an AI proxy network module, a consensus verification system and an arbitration module, the AI proxy network module comprises one or more sites; information is transmitted between the sites through a structured communication protocol; the AI proxy network module and the consensus verification system transmit response information through a data link; the consensus verification system and the arbitration module transmit a consistency check result through a preset communication channel; the arbitration module is connected with the AI proxy network module and communicates with an external entity through a man-machine interaction interface. According to the invention, a distributed AI proxy network and a dynamic consensus mechanism are adopted, so that decentralized reliable operation is realized, and a single-point fault is avoided; conflicts are accurately detected through semantic alignment, deduction alignment and echo technologies, and the AI error rate is reduced; site roles are flexibly adjusted through modular design, multi-scene tasks are adapted, cost is reduced, efficiency is improved, and decision-making precision is improved.

Description

Technical Field

[0001] The present invention relates to the cross - field of artificial intelligence and enterprise computing, and particularly to an information alignment system based on an AI agent network and a dynamic consensus mechanism. Background Art

[0002] The evolution of artificial intelligence technology has experienced a development stage from single - point intelligence to initial collaboration. Early technologies focused on automation in specific scenarios, such as basic functions like automatic email reply and document classification through rule engines, providing users with initial efficiency improvements. With the breakthrough of natural language processing technology, AI voice assistants centered on text interaction have gradually become popular, capable of assisting individual users in tasks such as information retrieval, text generation, and cross - platform communication, significantly enhancing the convenience of human - machine collaboration. At the same time, robotic process automation technology has achieved the automation of repetitive processes such as data entry and form filling by simulating manual operations, further promoting the digital transformation of business processes. However, these technologies have gradually revealed limitations when dealing with complex organizational environments: single - point intelligence lacks cross - scenario semantic understanding capabilities, AI voice assistants are difficult to support information alignment among multiple users, and RPA systems are restricted by fixed - rule process choreography and cannot adapt to dynamically changing business requirements.

[0003] There are three core problems in the existing technologies in the field of organizational - level information processing and collaboration. First, the semantic understanding ability of single - point intelligence technology is limited to specific scenarios. For example, an automatic email reply system cannot parse complex term differences in cross - departmental collaboration, resulting in semantic breaks during information transmission. Second, although mainstream AI voice assistants can process the text input of a single user, they lack an information alignment mechanism among multiple agents and are difficult to achieve knowledge sharing and consensus building in team collaboration. For example, in the generation of multi - person meeting records or cross - departmental requirement analysis, the output results of different AI agents may be contradictory or redundant. Third, process automation technology relies on a predefined rule engine and often leads to process interruptions or errors due to the lack of dynamic decision - making ability when dealing with business processes that require multi - node collaboration. In addition, existing solutions generally rely on manual review to verify information consistency. For example, in scenarios such as contract review and data analysis, manual intervention not only increases the time cost but may also cause decision - making delays due to subjective judgment differences. Summary of the Invention

[0004] The purpose of the present invention is to overcome the above - mentioned problems and provide an information alignment system based on an AI agent network and a dynamic consensus mechanism. To achieve the above purpose, the present invention adopts the following technical solutions:

[0005] An information alignment system based on an AI agent network and a dynamic consensus mechanism, including an AI agent network module, a consensus verification system, and an arbitration module; the AI agent network module includes a single or multiple sites; information is transmitted between the sites through a structured communication protocol; the AI agent network module and the consensus verification system transmit response information through a data link; the consensus verification system and the arbitration module transmit the consistency check result through a preset communication channel; the arbitration module is connected to the AI agent network module and communicates with external entities through a human-computer interaction interface.

[0006] Further, each site in the AI agent network module includes an entity port and a virtual port; the entity port corresponds to a real individual or organization; the virtual port is a local AI agent, and the virtual port is connected to the entity port through a bidirectional communication interface and is connected to the virtual ports of other sites and the consensus verification system through a network protocol.

[0007] Further, the communication method between the virtual port and the entity port uses a predefined template and professional terms as a structured format. Each communication includes at least one echo operation. The echo operation is that the receiving virtual port generates feedback information and transmits it back to the sending virtual port. Each communication generates a log file and stores it in the log database of the AI agent network module.

[0008] Further, the operation mode of the AI agent network module includes the following steps:

[0009] Step S11: The entity port of the source site publishes a task including a task objective, a list of related sites, and task acceptance requirements.

[0010] Step S12: The source site automatically connects to the related sites to form an AI agent network, and the virtual port of the source site distributes the task information to the virtual ports of the related sites.

[0011] Step S13: The virtual port of the related site gives a response using the model calculation ability and inputs the response information into the consensus verification system.

[0012] Step S14: The consensus verification system performs a consistency check on the outputs of all related sites and feeds back the information that violates the consistency to the relevant sites and the source site.

[0013] Step S15: If the consistency is not achieved, return to Step S13. If the number of times of non - achievement reaches the set threshold, enter the arbitration module.

[0014] Step S16: If the consistency is achieved, the consensus verification system sends the confirmed information to the virtual ports and entity ports of each site, and the entity port makes the final confirmation.

[0015] Further, the consensus verification system performs semantic alignment checks and deductive alignment checks on the response information; the semantic alignment aims to find conflicts in semantics and logic among information between sites; the deductive alignment aims to find conflicts in the implementation of information at each site.

[0016] Further, the above-mentioned deductive alignment is completed by the association method or the call simulation method. The association method regenerates new information from the original information obtained by the virtual port from other ports through AI association, and performs semantic alignment on the new information; the simulation method extracts relevant information from the information through the association algorithm, calls the existing emulator to quantify the degree of association of the relevant information, substitutes the emulator result into the relevant information to generate new relevant information and performs semantic alignment.

[0017] Further, the operation mode of the consensus verification system includes the following steps:

[0018] Step S21: Task reception and consistency check startup. After the relevant site receives the task information from the source site, it completes echo communication with the source site to confirm information consistency; after confirmation, based on the polling check principle, the semantic alignment algorithm is started to initiate consistency checks on other relevant sites separately or simultaneously. If a conflict with the source site information is detected, the conflict result is fed back to the arbitration module.

[0019] Step S22: Semantic alignment result processing. If the semantic alignment algorithm passes the check, the deductive alignment algorithm is triggered to execute; if the check fails, the conflict information is output to the arbitration module and the virtual ports of relevant sites, and the current task process is terminated.

[0020] Step S23: Deductive alignment result processing. If the deductive alignment algorithm passes the check, the consistency results of all sites are summarized and sent to the waiting entity ports of each site for confirmation; if the check fails, the conflict information is output to the arbitration module and the virtual ports of relevant sites, and the current task process is terminated.

[0021] Step S24: Final confirmation and task termination. If both the semantic alignment and the deductive alignment checks pass, and all site entity ports confirm no conflict, the task is determined to be completed; if any entity port confirmation fails, the conflict information is sent to the arbitration module and the virtual ports of relevant sites, and the task process is terminated.

[0022] Further, the arbitration module uses algorithms or manual intervention to report, analyze, and arbitrate the response results of the consensus verification system that fail the consistency check; the arbitration module sends the content that fails the consistency check of each module to the relevant sites and the entity ports of the source site in a user-defined format.

[0023] The advantages of the present invention are:

[0024] 1. The present invention realizes decentralized collaboration and highly reliable operation by constructing a distributed AI agent network architecture and a dynamic consensus mechanism. Each site is equipped with an independent AI agent, which completes information interaction based on a local model and a structured communication protocol, avoiding the complex global algorithm design of traditional centralized AI systems; through polling checks and multi-node redundancy mechanisms, it ensures that even if some sites fail, the system can still maintain its core functions, significantly enhancing the continuity guarantee ability of the enterprise's key business.

[0025] 2. The present invention effectively reduces the error rate of generative AI and optimizes the cross-departmental collaboration efficiency through the synergistic effect of semantic alignment, deductive alignment, and echo technology. The AI agent network replaces manual operations to complete the detection of term definition conflicts, time logic verification, and policy target consistency verification, significantly reducing information transmission deviation; by combining the association method to expand parameter boundaries and the simulation method to quantify physical limitations, it accurately identifies semantics and implementation conflicts in scenarios such as hardware design and tourism planning, significantly improving the decision-making accuracy and response speed of complex tasks.

[0026] 3. The present invention realizes flexible adaptation to multi-domain task scenarios through a modular site architecture and a dynamic network reconstruction mechanism. The system supports dynamically adjusting site roles and communication rules according to business requirements, calling a professional simulator to verify the extreme value of signal attenuation in chip development, and generating a compliance verification process based on historical templates in logistics planning, enabling a single framework to cover diverse requirements from engineering development to business planning, significantly enhancing the enterprise's adaptability to complex and changing business scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The drawings constituting a part of this application are used to provide a further understanding of this application, making other features, objectives, and advantages of this application more obvious. The schematic embodiments and descriptions of the drawings of this application are used to explain this application and do not constitute an improper limitation of this application.

[0028] In the drawings:

[0029] Figure 1 It is a schematic diagram of the site structure of the information alignment system based on the AI agent network and the dynamic consensus mechanism in Embodiment 1.

[0030] Figure 2 It is a schematic diagram of the communication process of the information alignment system based on the AI agent network and the dynamic consensus mechanism in Embodiment 1.

[0031] Figure 3 It is a logical diagram of the operation of the AI agent network of the information alignment system based on the AI agent network and the dynamic consensus mechanism in Embodiment 1.

[0032] Figure 4Schematic diagram of the consensus verification algorithm of the information alignment system based on the AI agent network and the dynamic consensus mechanism in Embodiment 1. Detailed implementation manners

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Usually, the components of the embodiments of the present invention described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations.

[0034] The present invention will be introduced in detail and specifically below through specific embodiments to better understand the present invention. However, the following embodiments do not limit the protection scope of the present invention.

[0035] Embodiment 1

[0036] As Figures 1-4 shown, the information alignment system based on the AI agent network and the dynamic consensus mechanism includes an AI agent network module, a consensus verification system, and an arbitration module; the AI agent network module includes one or more sites; information is transmitted between the sites through a structured communication protocol; the AI agent network module and the consensus verification system transmit response information through a data link; the consensus verification system and the arbitration module transmit the consistency check result through a preset communication channel; the arbitration module is connected to the AI agent network module and communicates with external entities through a human-computer interaction interface.

[0037] Different from the traditional unified large model AI system, the AI agent network in this embodiment consists of one or more sites, and each site consists of an entity port and a virtual port. The entity port corresponds to real individuals, such as employees, organizations, institutions, etc.; the virtual port is a localized AI agent, and the original data of its AI model can be from existing models or only from the company's internal database, which is an AI model customized for the entity port. For example, users can download the open-source deepseek model for basic training and then give it to each employee for offline operation. At this time, each employee himself becomes an entity port, and the deepseek AI in their hands is their respective virtual ports. Each employee and their own deepseek AI form a site.

[0038] Furthermore, each site in the AI agent network module includes an entity port and a virtual port; the entity port corresponds to real individuals or organizations; the virtual port is a localized AI agent, and the virtual port is connected to the entity port through a bidirectional communication interface and is connected to the virtual ports of other sites and the consensus verification system through a network protocol.

[0039] The virtual port has the ability to communicate with the physical ports of this site and the virtual ports of other sites. The input and output of each physical port are only connected to the virtual ports of this site; the input of each virtual port is the physical port of this site, the consensus verification system, and the virtual ports of all other sites. The virtual port is by default the information agent of the physical port. In theory, the physical port only needs to process information decisions and does not need to align and communicate information with the physical ports of other sites.

[0040] Furthermore, the communication method between the virtual port and the physical port uses predefined templates and professional terms as structured formats. Each communication includes at least one echo operation. The echo operation is that the receiving virtual port generates feedback information and transmits it back to the sending virtual port. Each communication generates a log file and stores it in the log database of the AI agent network module.

[0041] The communication methods between sites and between the physical ports and virtual ports within a site need to meet the following characteristics: The communication must be in a structured format, such as fixed templates and professional terms, etc. Each communication needs to have at least one echo. The echo means that the information receiver of the communication feeds back the information it understands to the information sender for confirmation. The number of echo times for each communication can be set according to the maturity of the model, and each communication needs to leave a log for the sites involved in the communication to consult. A single communication can be divided into two types: sending and receiving. The sending and receiving mentioned later in the text must meet the communication definition in this section.

[0042] For example: The virtual port A1 of site A sends a piece of information TEXT1 to the virtual port B1 of site B. After B1, as an AI model, understands the content of TEXT1, it sends TEXT2 regenerated after its own understanding to B1. If B1 confirms that TEXT1 and TEXT2 are consistent, it is considered that the echo is successful and a success message is sent to A1 and B1. If it is considered that TEXT1 and TEXT2 are inconsistent, it is considered that the echo fails and a failure message is sent to A1 and B1.

[0043] The virtual port and the physical port have the same database call rights. Example: An employee is a physical port, and their AI agent is a virtual port. The company data that the employee can access, their AI agent can also access. The database here is not within the scope of the present invention and belongs to common knowledge, so it will not be elaborated.

[0044] After the virtual port is generated, any acquisition of training data requires the consent of the physical port of this site or to be directly given by the physical port of this site. For example: Employees are physical ports, and their AI models are virtual ports. If it is stipulated that each employee's AI model must run offline or only within the company's internal local area network and the model training window is only open to employees, then the training data of the model can only be given by the employees.

[0045] The sites in the proxy network are divided into source sites and related sites. The source site refers to the task publishing site, and the related site refers to the task-related site. For example: A travel agency needs to develop a product for a one-week trip to Japan with a per capita cost of less than 10,000 yuan. The product manager and their AI agent constitute the source site, and the market researchers, flight and hotel reservation staff, visa processing staff, and their AI agents constitute the related sites. There will be only one source site in each task, and there can be multiple related sites.

[0046] After a site publishes a task, a new related network will be formed, and the site that publishes the task will become the source site. The related sites can be specified by the entity port of the source site or generated by the virtual port of the source site with the help of the experience template of historical projects. Any site can serve as the source site for its own task and also as a related site for the tasks of other sites.

[0047] Furthermore, the operation mode of the AI agent network module includes the following steps:

[0048] Step S11: The entity port of the source site publishes a task including the task objective, the list of related sites, and the task acceptance requirements;

[0049] Step S12: The source site automatically connects to the related sites to form an AI agent network, and the virtual port of the source site sends the task information to the virtual ports of the related sites;

[0050] Step S13: The virtual port of the related site gives a response using the model computing ability and inputs the response information into the consensus verification system;

[0051] Step S14: The consensus verification system checks the consistency of the outputs of all related sites and feeds back the information that violates the consistency to the relevant sites and the source site;

[0052] Step S15: If the consistency is not achieved, return to Step S13. If the number of times the consistency is not achieved reaches the set threshold, enter the arbitration module;

[0053] Step S16: If the consistency is achieved, the consensus verification system sends the confirmed information to the virtual ports and entity ports of each site, and the entity port makes the final confirmation.

[0054] The consensus verification system is responsible for the computing and information processing work by the virtual ports of each site in the AI agent network, and no additional new AI computing units need to be added. Its working mode is called polling check, that is, each site will check the consistency of other related sites in the AI agent network and send the check results to the arbitration module 3-6. The arbitration module will wait until all sites have sent the consistency check results.

[0055] The response may include task decomposition, decomposing the current task information into subtasks, implementation plans, progress schedules, etc. for the current site according to the division of labor of the current site in the task. The conditions for triggering the response may be task release, internal information update of the site, the consensus verification system finding conflicts in relevant information between sites, and the arbitration module reinitiating arbitration.

[0056] Furthermore, the consensus verification system performs semantic alignment checks and deductive alignment checks on the response information; the semantic alignment aims to find conflicts in semantics and logic between information of sites; the deductive alignment aims to find conflicts in the implementation of information of each site.

[0057] Furthermore, the above-mentioned deductive alignment is completed by the association method or the simulation method. The association method regenerates new information from the original information obtained by the virtual port from other ports through AI association, and performs semantic alignment on the new information; the simulation method extracts relevant information from the information through the association algorithm, calls the existing emulator to quantify the degree of association of the relevant information, substitutes the emulator result into the relevant information to generate new relevant information and performs semantic alignment.

[0058] The input of the consensus verification system is the response results of the virtual ports of each site in the AI agent network, and the output is connected to the arbitration module and the virtual ports of each site. The consensus verification system is used to find implicit conflicts between input information, and its essence is to replace the communication and alignment actions between entity ports. It includes at least two algorithms: semantic alignment and deductive alignment. The goal of semantic alignment is to find conflicts in semantics and logic between information of sites. For example: for a Japanese travel project, from the perspective of optimal cost, the flight route site plans the itinerary to depart from November to January of the following year, but from the perspective of optimal experience, the route planning site believes the itinerary should be from March to May of each year. Then the optimal cost strategy of the flight route site and the optimal experience strategy of the route planning site are semantic conflicts in the project information, and the different specific itinerary arrangements at the two places are logical conflicts.

[0059] The method of semantic alignment is to perform an "echo" operation between the output information of each virtual port and all other virtual ports. For any two nodes multiple times, if the echo fails, a failure message is sent to the corresponding node and the arbitration module. The semantic alignment of each node can be carried out simultaneously or batch by batch.

[0060] The deductive alignment goal is to find conflicts in the implementation of site information. For example, in a single-board design project, at the initial stage of target decomposition, the hardware node sets the maximum length of the single board to 5 inches for customer experience requirements. The chip interface node requires that the link signal attenuation be less than 30 dB for the stability of chip docking performance. There is no conflict between the requirements of these two nodes semantically and logically. However, when the single-board length is set to 5 inches, in the face of extreme conditions such as cables, it may lead to the inability to meet the signal attenuation requirements of the chip interface. This belongs to the implicit conflict in the implementation process.

[0061] Furthermore, the operation mode of the consensus verification system includes the following steps:

[0062] Step S21: Task reception and consistency check startup. After the stakeholder site receives the task information from the source site, it completes echo communication with the source site to confirm information consistency. After confirmation, based on the polling check principle, the semantic alignment algorithm is started, and consistency checks are initiated for other stakeholder sites separately or simultaneously. If a conflict with the source site information is detected, the conflict result is fed back to the arbitration module.

[0063] Step S22: Processing of semantic alignment results. If the semantic alignment algorithm passes the check, the deductive alignment algorithm is triggered to execute. If the check fails, the conflict information is output to the arbitration module and the virtual ports of relevant sites, and the current task process is terminated.

[0064] Step S23: Processing of deductive alignment results. If the deductive alignment algorithm passes the check, the consistency results of all sites are summarized and sent to the waiting entity ports of each site for confirmation. If the check fails, the conflict information is output to the arbitration module and the virtual ports of relevant sites, and the current task process is terminated.

[0065] Step S24: Final confirmation and task termination. If both the semantic alignment and deductive alignment checks pass, and all site entity ports confirm no conflict, the task is determined to be completed. If any entity port confirmation fails, the conflict information is sent to the arbitration module and the virtual ports of relevant sites, and the task process is terminated.

[0066] Deductive alignment can be accomplished through the association method. Its core principle is to regenerate Information 1-1 to Information 1-N, Information N-1 to Information N-N from the information obtained by the virtual port from other ports through AI association, and then perform semantic alignment on the regenerated information. Regarding the single-board length issue, using general association algorithms such as keyword retrieval, it can be known that the length is related to the attenuation of the signal link. Then, based on the given single-board length and the historical single-board attenuation coefficient, the current single-board attenuation amount can be calculated, thereby calculating the maximum value X of the link signal attenuation in the current application scenario. Then, the requirement of "the single-board length is set to a maximum of 5 inches" proposed by the hardware node can be extended to "the single-board length is set to a maximum of 5 inches, and the maximum value of the link signal attenuation in the application scenario is X". Then, by making the hardware node and the chip interface contact point have an echo once, a semantic conflict can be discovered.

[0067] Deductive alignment can be completed by calling the simulation method 4-2. Its core principle is to extract relevant information from Information 1 to Information N through an association algorithm, and call an existing simulator, which can be a professional field simulator such as the cadence simulation in the chip field or a generative AI such as chatgpt to quantify the degree of association of the relevant information in the real situation. Then, substitute the simulator results into the relevant information to generate new relevant information 1 to relevant information N, and then perform semantic alignment on relevant information 1 to relevant information N. For the single-board length issue, after obtaining the information related to the single-board length and the signal link attenuation through the association algorithm, the "maximum value X of the link signal attenuation in the current application scenario" can be obtained by calling the simulator. Then, the requirement of "the single-board length is set to a maximum of 5 inches" proposed by the hardware node can be extended to "the single-board length is set to a maximum of 5 inches, and the maximum value of the link signal attenuation in the application scenario is X". Then, by making the hardware node and the chip interface contact point have a semantic alignment once, a semantic conflict can be discovered.

[0068] Furthermore, the arbitration module uses an algorithm or manual intervention to report, analyze, and arbitrate the response results of the consensus verification system's consistency check failure; the arbitration module sends the content of the consistency check failure of each module to the relevant sites and the source site entity ports in a user-defined format.

[0069] The specific embodiments of the present invention have been described in detail above, but they are only examples, and the present invention is not equivalent to the specific embodiments described above. For those skilled in the art, any equivalent modifications and substitutions made to the present invention are also within the scope of the present invention. Therefore, all equivalent transformations and modifications made without departing from the spirit and scope of the present invention should be covered within the scope of the present invention.

Claims

1. An information alignment system based on an AI agent network and a dynamic consensus mechanism, characterized in that: It includes an AI agent network module, a consensus verification system, and an arbitration module; the AI agent network module includes one or more sites; information is transmitted between the sites through a structured communication protocol; the AI agent network module and the consensus verification system transmit response information through a data link; the consensus verification system and the arbitration module transmit the consistency check result through a preset communication channel; the arbitration module is connected to the AI agent network module and communicates with external entities through a human-computer interaction interface.

2. The information alignment system based on the AI agent network and the dynamic consensus mechanism according to claim 1, wherein: Each site in the AI agent network module includes an entity port and a virtual port; the entity port corresponds to a real individual or organization; the virtual port is a local AI agent, and the virtual port is connected to the entity port through a bidirectional communication interface and is connected to the virtual ports of other sites and the consensus verification system through a network protocol.

3. The information alignment system based on the AI agent network and the dynamic consensus mechanism according to claim 2, wherein: The communication method between the virtual port and the entity port uses a predefined template and professional terms as a structured format. Each communication includes at least one echo operation. The echo operation is that the receiving virtual port generates feedback information and transmits it back to the sending virtual port. Each communication generates a log file and stores it in the log database of the AI agent network module.

4. The information alignment system based on the AI agent network and the dynamic consensus mechanism according to claim 3, characterized in that: The operation mode of the AI agent network module includes the following steps: Step S11: The entity port of the source site publishes a task including a task target, a list of related sites, and task acceptance requirements. Step S12: The source site automatically connects to related sites to form an AI agent network, and the virtual port of the source site sends the task information to the virtual ports of related sites. Step S13: The virtual port of the related site gives a response using the model calculation ability and inputs the response information into the consensus verification system. Step S14: The consensus verification system performs a consistency check on the outputs of all related sites and feeds back the information that violates the consistency to the relevant sites and the source site. Step S15: If the consistency is not achieved, return to Step S13. If the number of times of non-achievement reaches the set threshold, enter the arbitration module. Step S16: If the consistency is achieved, the consensus verification system sends the confirmed information to the virtual ports and entity ports of each site, and the entity port makes the final confirmation.

5. The information alignment system based on the AI agent network and the dynamic consensus mechanism according to claim 4, wherein: The consensus verification system performs semantic alignment check and deductive alignment check on the response information; the semantic alignment target is to find the conflicts in semantics and logic between the information of sites. The deductive alignment target is to find the conflicts in the implementation of the information of each site.

6. The information alignment system based on the AI agent network and the dynamic consensus mechanism according to claim 5, wherein: The deductive alignment is completed by the association method or the simulation method. The association method regenerates new information from the original information obtained by the virtual port from other ports through AI association and performs semantic alignment on the new information. The simulation method extracts relevant information from the information through an association algorithm, calls an existing emulator to quantify the degree of association of the relevant information, substitutes the emulator result into the relevant information to generate new relevant information and performs semantic alignment.

7. The information alignment system based on the AI agent network and the dynamic consensus mechanism according to claim 6, characterized in that: The operation mode of the consensus verification system includes the following steps: Step S21: Task reception and consistency check start. After the relevant site receives the task information from the source site, it completes echo communication with the source site to confirm information consistency. After confirmation, based on the polling check principle, the semantic alignment algorithm is started to initiate consistency checks for other relevant sites separately or simultaneously. If a conflict with the source site information is detected, the conflict result is fed back to the arbitration module. Step S22: Processing of semantic alignment results. If the semantic alignment algorithm passes the check, the deductive alignment algorithm is triggered to execute. If the check fails, the conflict information is output to the arbitration module and the virtual ports of relevant sites, and the current task process is terminated. Step S23: Processing of deductive alignment results. If the deductive alignment algorithm passes the check, the consistency results of all sites are aggregated and sent to the waiting entity ports of each site for confirmation. If the check fails, the conflict information is output to the arbitration module and the virtual ports of relevant sites, and the current task process is terminated. Step S24: Final confirmation and task termination. If both the semantic alignment and deductive alignment checks pass, and all site entity ports confirm no conflict, the task is determined to be completed. If any entity port confirmation fails, the conflict information is sent to the arbitration module and the virtual ports of relevant sites, and the task process is terminated.

8. The information alignment system based on the AI agent network and the dynamic consensus mechanism according to claim 7, wherein: The arbitration module uses an algorithm or manual intervention to report, analyze, and arbitrate the response results of the consensus verification system's failed consistency checks. The arbitration module sends the content of the failed consistency checks of each module to the relevant sites and the entity ports of the source site in a user-defined format.

Images