A remote digital conference management system combining blockchain and AI large models

Abstract

The present invention discloses a remote digital conference management system combining blockchain and AI large models. The present invention relates to the field of information technology and includes a hierarchical dynamic verification chain module, which is composed of a real-time layer verification chain and an evidence storage layer audit chain connected through a data synchronization interface. The real-time layer verification chain adopts a lightweight consortium chain architecture, and the evidence storage layer audit chain is built based on a public chain. This remote digital conference management system combining blockchain and AI large models improves the security and intelligence level of remote conference management through the deep integration of blockchain and AI large models. The blockchain technology constructs a hierarchical dynamic verification chain to ensure the immutability of the full-process data of the conference and traceability throughout the life cycle, effectively solving the defects of traditional centralized systems being vulnerable to attacks and data being easily forged.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and specifically to a remote digital conference management system combining blockchain and AI large models. Background Art

[0002] The core challenge faced by traditional remote conference management systems in large international conferences or high-profile events is that it is difficult to synergistically achieve the security, trustworthiness of the entire conference process data and intelligent decision-making capabilities. Existing systems usually adopt a centralized architecture to store conference data, including key contents such as participant information, agenda records, resource allocation, and post-meeting analysis results. However, the centralized storage mode is vulnerable to network attacks or internal tampering. Especially in scenarios involving multinational multi-party collaboration, the authenticity and integrity of data are difficult to verify, resulting in frequent problems such as forged conference registration information, disputes over voting results, and loss of resource allocation records. Although some systems introduce basic encryption technology or third-party audits, they cannot achieve non-tamperable traceability throughout the data life cycle, and lack an automated trust mechanism when dynamically adjusting the conference process. At the same time, the intelligent level of each link in conference management is limited to rule engines or simple algorithms. For example, resource scheduling such as room allocation and dining seating depends on manually preset parameters, and it is difficult to cope with temporary agenda changes or sudden participant needs; complex tasks such as cross-language communication and topic hot-spot mining still rely on manual processing, resulting in low efficiency and easy to produce errors. Existing technologies attempt to apply blockchain or artificial intelligence separately to the conference scenario, such as using blockchain to store data in a single link, or optimizing local processes through traditional algorithms, but fail to solve two key contradictions: First, there is a response delay between the on-chain data verification mechanism of blockchain and the real-time decision-making requirements, making it difficult to support in-meeting dynamic scheduling; second, the data quality and source credibility on which the artificial intelligence model training depends lack guarantee, which may lead to algorithmic bias or decision-making distortion. This technical fragmentation makes the conference management system unable to meet the dual requirements of high security standards and high intelligence at the same time, severely restricting the management efficiency and user experience of large-scale digital conferences. Summary of the Invention

[0003] (1) Technical Problems to be Solved

[0004] In view of the deficiencies of the prior art, the present invention provides a remote digital conference management system combining blockchain and AI large models, and solves the problem of how to construct a secure, trustworthy and self-decision-making full-process conference management architecture through the deep integration of blockchain and artificial intelligence.

[0005] (2) Technical Solutions

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: A remote digital conference management system combining blockchain and AI large models, including:

[0007] Hierarchical Dynamic Verification Chain Module, which is composed of a real-time layer verification chain and an evidence storage layer audit chain connected through a data synchronization interface. The real-time layer verification chain adopts a lightweight consortium chain architecture, and the evidence storage layer audit chain is built based on a public chain;

[0008] Federated Multi-modal Decision Network Module, whose input end is communicatively connected to the output end of the hierarchical dynamic verification chain module, and is used to generate a conference resource scheduling plan and risk response instructions;

[0009] Double-chain Heterogeneous Consensus Engine, embedded in the hierarchical dynamic verification chain module, including an asynchronous sharding consensus unit and a global anchoring unit. The asynchronous sharding consensus unit is used for parallel verification of real-time layer data, and the global anchoring unit is used to periodically write the hash fingerprint of real-time layer data into the evidence storage layer audit chain;

[0010] Dynamic Defense Network Module, whose attack feature library is data-connected to the evidence storage layer audit chain of the hierarchical dynamic verification chain module, and the defense strategy output end is connected to the control end of the federated multi-modal decision network module.

[0011] During the operation of the remote digital conference management system, the system realizes the trusted storage and real-time verification of the whole process data of the conference through the hierarchical dynamic verification chain module. When participants submit registration information, room allocation requests or agenda adjustment applications, the data first enters the AI pre-screening gateway of the real-time layer verification chain for preliminary screening. The anomaly detection model embedded in the AI pre-screening gateway constructs a logical relationship graph based on the historical trusted data set, and analyzes potential contradiction points in the input data, such as repeated registration behaviors of the same identity information in different conferences, operations that violate the regional isolation rules in the room allocation request, or abnormal requests for permission applications that exceed the role level.

[0012] For the detected high-risk operations, the system directly intercepts and generates a risk log and stores it in the evidence storage layer audit chain; for the data that passes the preliminary screening, the on-chain smart contract further performs rule engine verification, including the consistency comparison of the participant's identity certificate and the biometric hash value, agenda time conflict detection, and real-time verification of resource capacity.

[0013] If the data passes the double verification, it triggers the federated multi-modal decision network module to start the dynamic decision-making process; if any link fails, the system returns a specific error type description to the user, and writes the complete verification path and failure reason into the blockchain for evidence storage.

[0014] Preferably, the real-time layer verification chain includes an AI pre-screening gateway, and the AI pre-screening gateway is composed of an anomaly detection model and an on-chain smart contract:

[0015] The anomaly detection model is constructed based on a graph neural network, and the training data comes from the registration information, room allocation records and permission change logs in the historical conference data set;

[0016] The on-chain smart contract includes rules for verifying the identities of participants, rules for verifying the compliance of the agenda, and rules for detecting resource allocation conflicts;

[0017] Among them, the output end of the anomaly detection model is connected in series with the input end of the on-chain smart contract. Only when the data passes through the screening of the anomaly detection model and the on-chain smart contract passes the verification, the federated multi-modal decision network module is triggered to start.

[0018] Preferably, the federated multi-modal decision network module includes:

[0019] A resource scheduling agent, whose data input end is connected to the resource allocation interface of the real-time layer verification chain, generates a room allocation or seat allocation plan based on a reinforcement learning model, and dynamically optimizes the plan weight through the Monte Carlo tree search algorithm;

[0020] A semantic collaboration agent, with a multilingual large model built-in, whose corpus is stored in the audit chain of the evidence storage layer. The translation result is signed by the private key of the participant's terminal and then sent back to the audit chain of the evidence storage layer;

[0021] A risk prediction agent, which uses a temporal graph convolutional network to analyze the historical meeting data in the audit chain of the evidence storage layer, predicts the probability of agenda conflicts and outputs it to the on-chain smart contract.

[0022] The federated multi-modal decision network module consists of multiple intelligent agents working together. The resource scheduling agent generates an initial resource allocation plan based on a reinforcement learning model and dynamically optimizes it through the Monte Carlo tree search algorithm, including: when a new or cancelled participation request is received, the algorithm combines the real-time venue status, the identity weights of participants, and the physical layout constraints, calculates the priorities of different allocation combinations, gives priority to ensuring the adjacent needs of associated participants, and simultaneously reserves emergency buffer resources dynamically.

[0023] During the translation process, the semantic collaboration agent calls the cross-cultural corpus in the audit chain of the evidence storage layer to constrain the output of the multilingual large model, ensuring the accuracy of terms and cultural adaptability. The translation result needs to be signed by the private key of the participant's terminal, and the validity of the signature is verified by the fast consensus of the nodes in the real-time layer verification chain before it can be written into the multilingual evidence chain partition of the audit chain of the evidence storage layer.

[0024] The risk prediction agent analyzes the historical meeting data stored on the blockchain, identifies risk patterns such as agenda conflicts and equipment failures. When the detected risk probability exceeds the threshold, it triggers the on-chain smart contract to automatically execute the emergency plan, including switching to a backup venue, adjusting the agenda time, or diverting participants to a virtual sub-venue. All operation instructions need to be verified by the blockchain and the complete execution path is recorded.

[0025] Preferably, after the resource scheduling agent generates a room allocation or seat allocation plan, it synchronously writes two types of on-chain records to the hierarchical dynamic verification chain module:

[0026] Execution chain record, including a solution identifier, an execution timestamp, and resource utilization metrics;

[0027] Logical chain record, encapsulating the parameter update path of the decision model and the fairness constraint verification result in the form of zero-knowledge proof;

[0028] Among them, the execution chain record and the logical chain record are associated by hash and input into the feedback learning unit of the federated multi-modal decision network module.

[0029] Preferably, the hierarchical dynamic verification chain module performs triple verification on VIP permission change operations:

[0030] The first verification generates a preliminary proposal by the risk prediction agent of the federated multi-modal decision network module, and the preliminary proposal includes the permission change scope, timeliness, and risk score;

[0031] The second verification performs fast consensus of the node group on the proposal by the asynchronous sharding consensus unit of the real-time layer verification chain, and the verification content includes user identity hash matching, permission timeliness conflict detection, and role permission upper limit verification;

[0032] The third verification randomly spot-checks the logical consistency between the proposal and the historical permission records by the final audit node of the evidence storage layer audit chain, including the compliance of the permission upgrade path and the rationality of the risk score distribution;

[0033] Only when the results of the triple verification are consistent, the proposal is marked as valid and written into the permission change partition of the evidence storage layer audit chain.

[0034] For permission change operations with high security requirements, the system implements a triple verification mechanism. The risk prediction agent first generates a change proposal and evaluates its risk level; the node group of the real-time layer verification chain conducts distributed consensus verification on the proposal, checking identity consistency, time compliance, and permission scope legality; the final audit node of the evidence storage layer audit chain verifies the historical logical consistency through random sampling. Only when the results of the triple verification are consistent, the operation instruction party is marked as finally valid. For detected abnormal operations, the system automatically freezes the process and triggers a multi-level response: pushing warning information to the administrator, updating the AI model training parameters, generating a blockchain audit report, and simultaneously activating the collaborative protection of the defense network module.

[0035] Preferably, the dynamic defense network module includes:

[0036] An attack feature extraction unit, extracting the data pattern of historical attack events from the evidence storage layer audit chain;

[0037] An AI adversarial model training unit, generating a defense strategy based on the extracted attack features and transmitting it back to the federated multi-modal decision network module;

[0038] Among them, the defense strategies include data poisoning attack interception rules and model parameter encryption rules, and the policy version numbers are synchronously updated to the audit chain of the evidence storage layer.

[0039] The dynamic defense network module continuously analyzes the security event logs stored in the blockchain to build an attack feature library and train an AI countermeasure model. When a data poisoning attack is detected, the system implants a dynamic filter at the data entry point to reduce the influence weight of abnormal data in the AI model; in the face of a model stealing attack, a parameter sharding encryption storage strategy is adopted, and the key parameters are dynamically assembled only after legal verification. The defense strategies are dynamically adjusted according to the attack level, including: traffic restriction and logging for low-risk attacks, and triggering a deep protection protocol for high-risk attacks, including interface access restriction, enhanced multi-factor authentication, and emergency resource switching. The version iteration of all defense strategies forms a traceable knowledge base through blockchain evidence storage to ensure the continuous evolution of the protection system.

[0040] Preferably, after the translation result of the semantic collaboration agent is generated, it needs to be signed with the private key of the participant's terminal, and the signature validity is verified by the asynchronous sharding consensus unit of the real-time layer verification chain. After the verification passes, it is written into the multi-language evidence chain partition of the audit chain of the evidence storage layer.

[0041] Preferably, when the risk prediction agent detects that the probability of agenda conflict exceeds the preset threshold, it triggers the on-chain smart contract to perform the following operations:

[0042] Call the venue status interface of the real-time layer verification chain to retrieve the availability of the alternate venue;

[0043] If the alternate venue is available, send a switching instruction to the federated multi-modal decision network module and update the agenda record of the audit chain of the evidence storage layer.

[0044] Preferably, the node grouping strategy of the asynchronous sharding consensus unit is dynamically adjusted according to the network load of the real-time layer verification chain. The adjustment parameters take effect after being verified by the global anchoring unit of the audit chain of the evidence storage layer, and the historical grouping strategy versions are stored in the audit chain of the evidence storage layer.

[0045] Preferably, the agenda compliance verification rules of the on-chain smart contract include:

[0046] Detection of time conflicts between parallel agendas of the same participant;

[0047] Verification of regional isolation between VIPs and ordinary participants in the room allocation plan;

[0048] Verification of permission level matching for ad-hoc agenda adjustments;

[0049] After the verification result passes the node consensus of the real-time layer verification chain, it triggers the resource scheduling agent to regenerate the plan.

[0050] In the specific implementation process, in a multi-time zone scenario across countries, the system introduces a time zone adaptation mechanism. When adjusting the agenda time, it automatically associates with the registered time zone data of participants. When it detects that a large number of participants are in inactive periods, it triggers a hierarchical notification and multi-version agenda generation process. The translation service adds a time zone consistency check during the signature verification phase and initiates biometric secondary authentication for requests with time zone conflicts. When allocating virtual sub-conference rooms, the resource scheduling agent gives priority to the geographical location of participants and network latency metrics, and generates multilingual guidance information through a semantic collaboration agent to ensure the consistency of the cross-regional participation experience.

[0051] The system realizes end-to-end auditability through a double-chain recording mechanism. The execution chain records the operation results and real-time metrics of storage, and the logic chain records the decision-making basis and rule matching paths encapsulated with zero-knowledge proofs. The two are associated through hashing to form an inseparable pair of audit evidences. Any data tampering or logical contradiction can be quickly located through two-way verification. When a dispute occurs, the auditor can completely trace back the decision-making life cycle based on the blockchain evidence, including the AI model version, contract rule iteration records, and network environment snapshots, to ensure the transparency and credibility of the system behavior.

[0052] Through the deep collaboration between blockchain and AI, a secure, trustworthy, and self-optimizing conference management architecture is constructed. By realizing real-time intelligent decision-making on the data of the entire conference process without tampering for the first time, it solves the fundamental contradiction that it is difficult to reconcile the security mechanism and efficiency requirements in traditional systems, and forms a self-reinforcing technical ecosystem through the integrated design of double-chain verification, federated decision-making network, and dynamic defense system.

[0053] (III) Beneficial effects

[0054] The present invention provides a remote digital conference management system combining blockchain and AI large models. It has the following beneficial effects:

[0055] (I) The remote digital conference management system combining blockchain and AI large models improves the security and intelligence level of remote conference management through the deep integration of blockchain and AI large models. The blockchain technology constructs a hierarchical dynamic verification chain to ensure the immutability and full life cycle traceability of the data of the entire conference process, effectively solving the defects of traditional centralized systems being vulnerable to attacks and data being easily forged; the federated multi-modal decision-making network driven by the AI large model realizes autonomous optimization and real-time response in links such as resource scheduling, risk prediction, and cross-language collaboration, reducing the cost of manual intervention and improving management efficiency.

[0056] (2) The remote digital conference management system combining blockchain and AI large models solves the problem of coordinating security mechanisms and decision-making efficiency, and balances the strong audit requirements of blockchain and the real-time response capabilities of AI through a dual-chain architecture; the dynamic defense network and self-optimization mechanism further enhance the system's robustness, can actively identify and intercept new types of attacks such as data poisoning and permission tampering, and continuously optimize the AI model and resource allocation strategy based on federated learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a schematic diagram of the overall framework of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0058] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0059] Please refer to Figure 1 , the present invention provides a technical solution: a remote digital conference management system combining blockchain and AI large models, including:

[0060] The hierarchical dynamic verification chain module is composed of a real-time layer verification chain and an evidence storage layer audit chain connected through a data synchronization interface. The real-time layer verification chain adopts a lightweight consortium chain architecture, and the evidence storage layer audit chain is built based on a public chain;

[0061] The federated multimodal decision network module, whose input end is communicatively connected to the output end of the hierarchical dynamic verification chain module, is used to generate a conference resource scheduling plan and risk response instructions;

[0062] The dual-chain heterogeneous consensus engine is embedded in the hierarchical dynamic verification chain module and includes an asynchronous sharding consensus unit and a global anchoring unit. The asynchronous sharding consensus unit is used for parallel verification of real-time layer data, and the global anchoring unit is used to periodically write the hash fingerprint of real-time layer data into the evidence storage layer audit chain;

[0063] The dynamic defense network module, whose attack feature library is data-connected to the evidence storage layer audit chain of the hierarchical dynamic verification chain module, and the defense strategy output end is connected to the control end of the federated multimodal decision network module.

[0064] The real-time layer verification chain includes an AI pre-screening gateway, which is composed of an anomaly detection model and an on-chain smart contract:

[0065] The anomaly detection model is constructed based on a graph neural network, and the training data comes from the registration information, room allocation records, and permission change logs in the historical conference dataset;

[0066] The on-chain smart contract includes rules for verifying the identities of participants, rules for verifying the compliance of the agenda, and rules for detecting resource allocation conflicts;

[0067] Among them, the output end of the anomaly detection model is connected in series with the input end of the on-chain smart contract. Only when the data passes the screening of the anomaly detection model and the on-chain smart contract verification, the federated multi-modal decision network module is triggered to start.

[0068] It should be further noted that in the specific implementation process, in the real-time layer verification chain of the conference management system, the AI pre-screening gateway realizes trusted data filtering through the series verification mechanism of the anomaly detection model and the on-chain smart contract. When a participant submits a registration information, a room allocation request, or a permission change application, the anomaly detection model first analyzes the logical relevance between the input data and the historical conference dataset based on the graph neural network. For example: detecting conflicting registration behaviors of the same identity information in different conferences, abnormal mixing of VIPs and ordinary participants in the room allocation request, or matching contradictions between the permission change application and the user role level.

[0069] If high-risk operations are detected, such as duplicate registration, illegal cross-regional room allocation requests, the model will directly intercept the data and generate a risk log for storage in the audit chain of the evidence storage layer; for the data that passes the preliminary screening, the on-chain smart contract further performs rule engine verification: for the rule of verifying the identity of participants, the contract automatically compares the pre-stored digital identity certificate in the blockchain with the current submitted biometric hash value; for the agenda compliance rule, the contract verifies whether the agenda time conflicts with other agendas that have been certified; for the rule of detecting resource allocation conflicts, the contract retrieves the real-time occupancy status of the venue and accommodation in real time and rejects over-capacity allocation requests.

[0070] If the data passes the dual verification, the system will trigger the federated multi-modal decision network to generate a dynamic plan; if any link verification fails, a specific error code will be returned to the user, such as "identity verification failed - biometric mismatch" or "room allocation request conflict - target area full", and at the same time, the failure record and the hash value of the verification path will be written into the audit chain of the evidence storage layer for traceability.

[0071] For sudden requirements such as temporary agenda adjustments, the AI pre-screening gateway starts the fast-track mode, preferentially matches the decision mode cache of historical similar scenarios in the anomaly detection stage, shortens the verification response time, and at the same time, the on-chain smart contract dynamically relaxes some non-core rules, such as the seating priority of non-VIP participants, to ensure that the system maintains high response efficiency while ensuring security.

[0072] The federated multi-modal decision network module includes:

[0073] A resource scheduling agent, whose data input end is connected to the resource allocation interface of the real-time layer verification chain, generates a room or seat allocation plan based on a reinforcement learning model, and dynamically optimizes the plan weight through the Monte Carlo tree search algorithm;

[0074] A semantic collaboration agent, with a multilingual large model built in, whose corpus is stored in the audit chain of the evidence storage layer. The translation result is signed with the private key of the participant's terminal and then transmitted back to the audit chain of the evidence storage layer;

[0075] A risk prediction agent, which uses a temporal graph convolutional network to analyze the historical meeting data in the audit chain of the evidence storage layer, predicts the probability of agenda conflicts and outputs it to the on-chain smart contract.

[0076] It should be further noted that in the specific implementation process, during the operation of the federated multi-modal decision network module, the resource scheduling agent generates an initial room or seat allocation plan based on the verified participant data and venue resource status provided by the real-time layer verification chain, and uses the Monte Carlo tree search algorithm to dynamically optimize the plan: when there is a new participant or a request to cancel participation temporarily, the algorithm calculates the priority scores of different allocation combinations according to the real-time updated seat occupancy rate, participant identity weight, and venue physical layout constraints, and preferentially retains adjacent seats for participants with strong relevance, that is, preferentially retains seats for members of the same organization, and at the same time dynamically adjusts the buffer capacity of unallocated seats to cope with sudden demands; among them, the participant identity weight includes VIP level and historical participation records, and the venue physical layout constraints include epidemic prevention safety distances.

[0077] During the real-time translation process, the semantic collaboration agent extracts the cross-cultural corpus of the target language from the audit chain of the evidence storage layer as context constraints, inputs it into the multilingual large model to generate a preliminary translation, and then performs semantic consistency verification on the translation. Among them, the cross-cultural corpus includes an industry glossary and a cultural taboo word library.

[0078] The process of subsequent semantic consistency verification of the translation includes: if it is detected that the deviation of the key terms from the official definition stored in the blockchain exceeds the threshold, such as the translation error rate of technical terms is greater than 5%, the manual review process is triggered and the write permission of the translation in the evidence storage chain is frozen; if the verification passes, the participant needs to use the private key of the terminal to digitally sign the translation, and the signed translation is written into an independent partition of the audit chain of the evidence storage layer after the consensus of the nodes of the real-time layer verification chain, forming an immutable multi-language communication evidence chain.

[0079] The risk prediction agent analyzes historical meeting conflict events stored in the audit chain of the evidence storage layer through a temporal graph convolutional network, such as the time distribution of equipment failures and the frequency of agenda timeouts, to construct a dynamic risk probability map. When it detects that the current agenda conflict probability exceeds a preset threshold, such as the interval between two agendas in the same venue is less than 10 minutes and the attendee overlap rate is greater than 30%, it sends the risk level and response plan to the on-chain smart contract, and the contract automatically retrieves the availability status of the backup venue.

[0080] The process by which the contract automatically retrieves the availability status of the backup venue includes: if there are sufficient backup venue resources, it immediately sends a switching instruction to the resource scheduling agent and updates the agenda record in the audit chain of the evidence storage layer; if the backup resources are insufficient, it activates a downgrading plan, such as extending the agenda interval, splitting the attendees into virtual sub-venues, and pushing notifications to the relevant attendees through the semantic collaboration agent. The decision data flow between the agents is deeply coupled with the blockchain verification mechanism to ensure that all decision-making bases are traceable and the decision results are auditable.

[0081] After the resource scheduling agent generates a room allocation or seating arrangement plan, it synchronously writes two types of on-chain records to the hierarchical dynamic verification chain module:

[0082] Execution chain record, including the plan identifier, execution timestamp, and resource utilization metrics;

[0083] Logical chain record, which encapsulates the parameter update path of the decision model and the fairness constraint verification result in the form of zero-knowledge proof;

[0084] Among them, the execution chain record and the logical chain record are associated through a hash and input into the feedback learning unit of the federated multi-modal decision network module.

[0085] It should be further noted that in the specific implementation process, after the resource scheduling agent generates a room allocation or seating arrangement plan, the system synchronously writes two types of data, namely the execution chain record and the logical chain record, to the hierarchical dynamic verification chain module. The execution chain record includes the plan identifier, execution timestamp, and resource utilization metrics, and is written to the resource allocation partition of the real-time layer verification chain, triggering the resource scheduling agent to execute the room allocation or seating arrangement plan; at the same time, the system monitors the execution effect of the plan in real time, such as the actual seat occupancy rate and the attendee satisfaction feedback. If it detects that the resource utilization rate is lower than the preset threshold, such as the vacancy rate in the VIP area exceeds 20% or the general area is overcrowded by 10%, it sends a dynamic adjustment request to the federated multi-modal decision network module.

[0086] The logical chain record encapsulates the parameter update path of the decision-making model and the fairness constraint verification result in the form of zero-knowledge proof. The generation process includes: calculating the regional isolation degree score between VIPs and ordinary participants based on the identity weight distribution of the participants in the current room allocation plan; verifying whether the adjacent rate of associated participants in the seating plan meets the preset requirements, where the associated participants include those from the same institution or cooperation party; verifying whether the permission level of the temporary adjustment request matches the operation scope. The logical chain record passing the verification is written into the decision logic partition of the audit chain in the evidence storage layer after hash operation, and is two-way bound with the hash value of the execution chain record to form an inseparable audit evidence pair.

[0087] When the feedback learning unit of the federated multi-modal decision-making network module receives a new room allocation request, it first retrieves the resource utilization rate index of the historical execution chain record and the fairness verification result of the logical chain record from the audit chain in the evidence storage layer, and analyzes their relevance: if the historical data shows high resource utilization rate but the fairness score is lower than the threshold, such as the VIP area being over-occupied, resulting in a decline in the experience of ordinary participants, then adjust the weight allocation strategy of the reinforcement learning model, reduce the identity level weight and increase the spatial balance weight; if it is detected that there are unpassed fairness verification items in the logical chain record, such as insufficient adjacent rate of associated participants, then trigger the constraint condition strengthening module of the Monte Carlo tree search algorithm to limit the allocation combinations with low correlation in the search branches.

[0088] For abnormal scenarios, such as a sudden large-scale cancellation of participants, the system reversely deduces the optimal recovery strategy based on the time stamp and the time sequence of resource status changes in the execution chain record, and verifies the compliance of the strategy through the parameter path of the logical chain record. If it is detected that the strategy conflicts with the historical fairness rules, such as the recovered seats being preferentially allocated to non-VIP participants violating the existing agreement, then freeze the execution of the strategy and push an artificial review instruction to the administrator terminal.

[0089] The above interactive mechanism for double-chain records extends to the exception handling scenario, including: when the hash correlation between the execution chain record and the logic chain record is detected to be broken, such as the deviation between the actual resource utilization rate and the predicted value of the decision-making model exceeds 30%, the system automatically freezes all subsequent operations of the room allocation plan, extracts the associated AI decision parameters and verification paths from the audit chain of the evidence storage layer, and starts the root cause analysis process, including: if the problem stems from the data synchronization delay of the real-time layer verification chain, such as the venue status update not being uploaded to the chain in a timely manner, then roll back the room allocation plan and re-trigger the resource scheduling agent; if the problem is caused by a fairness rule vulnerability in the logic chain record, such as the failure to identify a new type of associated participant relationship, then update the verification rules of the smart contract on the chain and generate a patched version to be synchronized to all nodes. For the performance bottleneck in the high-concurrency scenario, the system starts a dynamic caching mechanism for the frequently accessed execution chain records, caches the hash value mapping relationship between the recently active plans and the logic chain in the edge nodes, reduces the cross-chain retrieval latency, and at the same time sets the validity period of the cached data to be synchronized with the global anchoring period of the audit chain of the evidence storage layer to ensure strong data consistency.

[0090] The hierarchical dynamic verification chain module implements triple verification for VIP permission change operations:

[0091] The first verification generates a preliminary proposal by the risk prediction agent of the federated multi-modal decision-making network module. The preliminary proposal includes the permission change scope, timeliness, and risk score.

[0092] The second verification conducts a fast consensus among node groups on the proposal by the asynchronous sharding consensus unit of the real-time layer verification chain. The verification content includes the user identity hash matching, permission timeliness conflict detection, and role permission upper limit verification.

[0093] The third verification randomly spot-checks the logical consistency between the proposal and the historical permission records by the final audit node of the evidence storage layer audit chain, including the compliance of the permission upgrade path and the rationality of the risk score distribution.

[0094] Only when the results of the triple verification are consistent, the proposal is marked as valid and written into the permission change partition of the evidence storage layer audit chain.

[0095] It should be further noted that in the specific implementation process, in the triple verification process of VIP permission change operations, when the system receives a permission change request, first, the risk prediction agent of the federated multi-modal decision network module generates a preliminary proposal. The risk prediction agent analyzes the rationality of the permission change based on the user's historical permission records, current meeting role, and associated agenda sensitivity stored in the audit chain of the evidence storage layer: If the request involves cross-security level operations, such as an ordinary attendee applying for VIP background access permission, the agent will extract the user's past behavior data and calculate the risk score through a temporal graph convolutional network. If the score exceeds the preset threshold, such as the risk value is greater than 0.7, the request will be directly rejected and a high-risk log will be generated; if the risk score is within an acceptable range, the agent generates a proposal containing the reason for the change, the scope of permissions, and the timeliness, and attaches a risk score explanation; among them, the user's past behavior data includes the historical agenda participation rate and the permission usage frequency.

[0096] Subsequently, the proposal enters the asynchronous sharding consensus unit of the real-time layer verification chain for the second verification. The node group dynamically divides the verification shards according to the proposal type, and the verification nodes within each shard independently execute the checks; among them, the proposal types include permission level elevation and access scope expansion.

[0097] The process of each verification node within each shard independently executing the checks includes: comparing whether the user identity hash value in the proposal is consistent with the identity certificate stored in the blockchain evidence; verifying whether the permission change timestamp conflicts with the associated agenda, such as applying for post-meeting permissions but taking effect during the meeting; checking whether the permission scope exceeds the preset upper limit of this role, such as an ordinary attendee applying for speaker-exclusive permissions. If more than 2 / 3 of the node group verifies and passes, the proposal is marked as "temporarily valid" and temporarily stored in the real-time layer buffer; if the number of opposing nodes in any verification shard exceeds 1 / 3, the cross-shard arbitration mechanism is triggered, and the main shard node rechecks the disputed items and generates a final conclusion.

[0098] The third verification is randomly sampled by the final audit node of the evidence storage layer audit chain, including: retrieving all historical permission change records of this user from the evidence storage layer, verifying whether the current proposal conforms to the permission upgrade logic, such as: progressing step by step from ordinary to VIP to administrator; checking whether the risk score in the proposal is consistent with the score distribution of historical operations of the same type, such as being considered suspicious if it deviates abnormally from the mean by plus or minus 3σ; verifying whether the timestamp is synchronized with the global meeting clock to prevent maliciously tampering with the local time to fabricate the validity period.

[0099] After the random sampling passes, the final audit node anchors the proposal and the verification path hash value to the immutable area of the evidence storage layer, and synchronously releases the temporary mark in the real-time layer buffer as "finally effective".

[0100] If logical inconsistencies are found during random inspection, such as a user skipping the VIP level and directly applying for administrator privileges, the proposal will be immediately frozen and the following processing will be triggered, including: sending a model correction instruction to the federal multi-modal decision network module to require retraining of the permission upgrade logic of the risk scoring model; pushing an alarm message containing details of the abnormal path to the administrator terminal; generating an audit report with a digital signature in the audit chain of the evidence storage layer for subsequent judicial forensics.

[0101] For high-frequency permission change scenarios, such as batch VIP guest permission granting in a large conference, the system enables a batch proposal optimization channel, including: the risk prediction agent clusters requests of the same type into group proposals and uses group behavior pattern analysis to replace single-user risk assessment; the real-time layer verification node group parallelly verifies the common conditions within the group according to a preset rule set, such as the consistency between the guest list and the evidence storage list of the organizer, and only initiates independent verification for abnormal individuals; the sampling ratio of the final audit node is dynamically reduced to 10%-30%, and the threshold is automatically adjusted according to the historical sampling accuracy. If a high-risk individual request is mixed in the group proposal, such as a person not on the list disguising as a guest, the system automatically strips the abnormal item and generates a sub-case processing queue during the second verification stage to ensure the quick execution of compliant requests and the independent interception of risk requests.

[0102] The dynamic defense network module includes:

[0103] An attack feature extraction unit that extracts the data patterns of historical attack events from the audit chain of the evidence storage layer;

[0104] An AI countermeasure model training unit that generates defense strategies based on the extracted attack features and sends them back to the federal multi-modal decision network module;

[0105] Among them, the defense strategies include data poisoning attack interception rules and model parameter encryption rules, and the strategy version number is synchronously updated to the audit chain of the evidence storage layer.

[0106] It should be further noted that during the specific implementation process, during the operation of the dynamic defense network module, the attack feature extraction unit continuously scans the security event logs in the audit chain of the evidence storage layer to identify potential attack patterns: when detecting a high-frequency abnormal data write request, such as more than 500 room allocation plan queries initiated by the same IP address within a short period of time, the unit first extracts the context features of this operation, including the request time distribution, the depth of the data access path, and the associated user behavior portrait, and performs a similarity match with the historical attack pattern library stored in the blockchain. If a known attack type is matched, the defense strategy generation process is triggered; if it is identified as a new type of attack, such as a timing confusion attack using AI decision delay, the emergency mode is activated, and the original attack data is sliced and encrypted and then broadcast to the blockchain nodes for collaborative analysis.

[0107] The AI adversarial model training unit generates targeted defense rules based on attack characteristics: For data poisoning attacks, the model constructs a dynamic filter, implants a feature weight detection mechanism at the data entry of the real-time layer verification chain, and verifies the distribution rationality of the participant information fields. If it is detected that the concentration of a certain type of feature deviates abnormally from the historical distribution, such as the proportion of registered people from the same institution suddenly increasing to 80%, the sampling weight of this type of data in the AI decision model will be automatically reduced, and an artificial review process will be triggered; among them, the participant information fields include the institution name and position level. For model stealing attacks, the model generates a parameter sharding encryption strategy, splits the key parameters of the federated multi-modal decision network into multiple ciphertext segments, stores them in different blockchain nodes respectively, and dynamically assembles available parameters only when the verification requester has a legitimate identity certificate and the operation context matches.

[0108] In the defense strategy implementation stage, the system dynamically adjusts the protection intensity according to the attack type, including: when a low-risk exploratory attack is detected, such as a single abnormal permission detection, a lightweight interception mode is adopted, only the attack characteristics are recorded and the request frequency of this IP is restricted; for high-risk persistent attacks, such as data pollution by distributed node collaboration, a deep defense protocol is started, including temporarily closing the public writing interface of the real-time layer verification chain, forcibly switching to a multi-signature approval mode, and pushing a security verification challenge to the associated participant terminal, such as two-factor authentication based on blockchain certificates. The version number and effective timestamp of all defense strategies are written into the defense log partition of the evidence storage layer audit chain to form a traceable protection knowledge base.

[0109] For the iterative optimization of defense strategies, the system sets up a dual-channel verification mechanism, including: after the new strategy is deployed, historical attack data is automatically replayed in the sandbox environment to monitor the effectiveness of the strategy and the system performance loss. If it is detected that the key indicators deteriorate beyond the tolerance threshold, such as the interception rate increases by less than 10% and the delay increases by more than 200ms, roll back to the previous stable version and mark this strategy as "to be optimized"; among them, monitoring the effectiveness of the strategy includes the interception rate of poisoned data, and the system performance loss includes the increase in AI decision delay. For the strategies that pass the verification, the system extracts their core features to generate a simplified knowledge graph, and selects the optimal version through the blockchain node voting mechanism and synchronizes it to the whole network; among them, the core features include the filter rule set and the number of encryption shards.

[0110] In extreme attack scenarios, such as suffering from multiple types of composite attacks at the same time, the defense network starts a cross-module collaboration mechanism: calls the risk prediction agent to evaluate the overall risk level of the system. If it reaches a critical state, such as the availability of the core AI model is lower than 70%, the degradation mode of the federated multi-modal decision network is triggered, giving priority to ensuring the basic resource allocation function, and closing non-critical modules, such as multi-language real-time translation, until the defense network restores the system to the security baseline.

[0111] After the translation result of the semantic collaboration agent is generated, it needs to be signed with the private key of the participant's terminal, and the asynchronous sharding consensus unit of the real-time layer verification chain verifies the signature validity. After passing the verification, it is written into the multi-language evidence chain partition of the archival storage layer audit chain.

[0112] It should be further noted that in the specific implementation process, after the semantic collaboration agent generates the translation result, the system starts the trusted archival storage process: when the participant's terminal receives the real-time translation manuscript, the terminal automatically calls the locally stored private key to digitally sign the translated text content, generates a signature data packet containing the translation hash value, timestamp, and signer identity identifier, and transmits it to the temporary buffer of the real-time layer verification chain through an encrypted channel. The asynchronous sharding consensus unit of the real-time layer verification chain verifies the signature validity: first, it parses the identity identifier in the signature data packet and matches it with the digital identity certificate pre-stored in the archival storage layer audit chain; second, it checks whether the translation hash value is consistent with the original translation text; finally, it verifies whether the timestamp is within the meeting validity period, such as 1 hour before the agenda starts to 2 hours after the end. If any verification item fails, such as the identity certificate expires, the hash value is tampered with, or the submission times out, the system immediately freezes the translation and triggers the following processing: sends a re-translation instruction to the semantic collaboration agent, requiring it to regenerate the translation based on the original corpus stored on the blockchain; pushes a signature exception warning to the participant's terminal, prompting to initiate the signature process again; marks the abnormal event characteristics to the attack feature library of the dynamic defense network module. Among them, the abnormal event characteristics include the same IP address with frequent signature failures.

[0113] For the translation that passes the verification, the system writes it into the multi-language evidence chain partition of the archival storage layer audit chain: it is stored according to the three-level index structure of "meeting number - agenda number - language pair", and each translation is associated with the complete verification path data, including the signature certificate chain, the list of consensus nodes, and the timestamp sequence, forming a traceable translation evidence chain.

[0114] When a participant raises an objection to the accuracy of the translation, the auditor can trace back to the original text and signature context through the hash value in the evidence chain. If it is detected that the translated text content is inconsistent with the version at the time of signature, such as the hash value recorded in the archival storage chain cannot be reproduced, it is determined that the data in the archival storage layer is abnormal, and the blockchain node consistency verification and data repair process is triggered.

[0115] For the simultaneous interpretation scenario with high real-time requirements, the system enables the fast channel mode: the semantic collaboration agent preloads the high-frequency term library during the translation generation stage, extracts the standard term list of the current meeting field from the archival storage chain, and gives priority to ensuring the accuracy of key information; in the signature verification link, a streaming processing mechanism is adopted, and the long translation is sliced into paragraph-level data blocks for parallel signature, and the verification results of each data block are fed back to the AI model in real time for context coherence correction.

[0116] If a logical break is detected between paragraphs, such as more than 3 inconsistent terms before and after, the local retranslation mechanism is automatically triggered, and only the problematic paragraph is replaced while preserving the signature status of the verified part.

[0117] In the scenario of multinational multi-time zone meetings, the system introduces time zone adaptive verification rules, including: when the timestamp of the translation signature deviates from the standard time of the main venue by more than a preset threshold, such as when the timestamp of the translation signature deviates from the standard time of the main venue by more than plus or minus 4 hours, the geographical location data of the associated participants is automatically extracted from the registration information. If the location information conflicts with the signature time zone logically, the translation archiving process is frozen and secondary biometric authentication is required. It can be written into the blockchain only after passing. For the signature time drift caused by network latency, such as the deviation between the actual generation time and the system reception time is less than 5 minutes, the system starts the time window compensation mechanism, and records the double timestamps of the original time and the calibrated time in the audit chain of the archiving layer for cross-verification during subsequent audits.

[0118] When the risk prediction agent detects that the probability of agenda conflict exceeds the preset threshold, it triggers the on-chain smart contract to perform the following operations:

[0119] Call the venue status interface of the real-time layer verification chain to retrieve the availability of the alternative venue;

[0120] If the alternative venue is available, send a switching instruction to the federated multi-modal decision network module and update the agenda record in the audit chain of the archiving layer.

[0121] It should be further noted that in the specific implementation process, when the risk prediction agent detects that the probability of agenda conflict exceeds the preset threshold, the system triggers the on-chain smart contract to execute the automated process of venue switching and agenda adjustment: the on-chain smart contract first calls the venue status interface of the real-time layer verification chain to retrieve the device status, the number of online participants and the network load metrics of the current main venue, and at the same time extracts the historical usage records and real-time availability data of the alternative venue from the audit chain of the archiving layer.

[0122] If the resources of the alternative venue are sufficient, such as the online capacity is greater than or equal to 80% of the main venue's demand and the device ready time is less than or equal to 5 minutes, the contract immediately sends a switching instruction to the resource scheduling agent. The instruction includes the switching time window, the participant diversion ratio and the access key of the new venue; the resource scheduling agent generates the optimal migration path based on the reinforcement learning model, such as preferentially migrating the low-priority participants in the overlapping agenda to the alternative venue, and pre-writes the hash value of the migration plan into the temporary decision area of the real-time layer verification chain. After the participant terminal confirms the reception, the agenda record in the audit chain of the archiving layer is officially updated.

[0123] If the resources of the backup venue are insufficient or the equipment is not ready, the contract activates the dynamic adjustment mode of the agenda time: according to the urgency of the conflicting agenda and the overlapping rate of the participants, calculate the deferrable time interval and push the time adjustment plan to the associated speaker terminal; among them, the agenda involving the signing ceremony is marked as the highest priority in the urgency of the conflicting agenda; if the speaker accepts the adjustment, the contract locks the new agenda time and updates the audit chain of the deposit layer; if the speaker refuses or fails to respond within the time limit, the participant diversion mechanism is triggered, and some participants are guided to the virtual sub-venue, such as grouping by region or topic preference. After being translated by the semantic collaboration agent, the access link of the virtual sub-venue is pushed to the target participant terminal through the encrypted channel verified by the blockchain.

[0124] For high-frequency small-scale conflicts, such as the interval between three consecutive agendas being less than 5 minutes, the system enables a progressive optimization strategy, including: the risk prediction agent no longer triggers a global venue switch, but fine-tunes the parameters of a single agenda through an on-chain smart contract, such as compressing the speech duration by 5% and extending the interval of the Q&A session. Each adjustment requires the private keys of the participants and speakers for signature confirmation. After being signed, the adjustment record is hashed and associated with the original agenda version and then written into the audit chain of the deposit layer. For conflicts caused by sudden equipment failures, such as the network interruption in the main venue, the contract bypasses the regular verification process and directly calls the emergency plan template pre-stored in the audit chain of the deposit layer, such as switching to a dedicated disaster recovery venue, and completes the blockchain verification record within 24 hours after execution, including: the resource scheduling agent traces back all operation logs during the failure period, recalculates the compliance score of the resource allocation plan, and if unauthorized temporary occupation of backup resources is detected, a default report is automatically generated and the relevant account permissions are frozen.

[0125] In the scenario of multiple time zones across countries, time zone synchronization verification needs to be added to the agenda time adjustment: when the contract modifies the agenda timestamp, it automatically associates the registered time zone data of the participants. If it is detected that more than 20% of the participants are in an inactive period due to the time modification, such as local time 23:00 - 6:00, a hierarchical notification mechanism is triggered, that is: a mandatory confirmation request is pushed to the core participants, an optional participation time questionnaire is sent to the non-core participants, and multiple agenda copies are dynamically generated according to the feedback results. The participation records of each copy of the agenda are independently stored and finally merged into the main agenda hash tree.

[0126] For the ambiguity of the agenda description caused by time zone conversion, such as "9:00 the next day" without specifying the time zone, the semantic collaboration agent automatically inserts the time zone identifier and triggers a secondary signature confirmation to avoid subsequent audit disputes.

[0127] Anti-oscillation mechanisms are set for all automated decisions in the above process: If the same agenda is adjusted more than 3 times continuously within 1 hour, the system will freeze the automatic switching function and transfer to the federated multi-modal decision-making network to initiate multi-factor evaluations, such as participant satisfaction prediction and equipment loss cost models, generate a final manual review plan, and mark the decision logic chain as a high-priority audit item to ensure the stability and interpretability of the system in emergency scenarios.

[0128] The node grouping strategy of the asynchronous sharding consensus unit is dynamically adjusted according to the network load of the real-time layer verification chain. The adjusted parameters take effect after being verified by the global anchoring unit of the deposit layer audit chain, and the historical grouping strategy versions are stored in the deposit layer audit chain.

[0129] It should be further noted that during the dynamic adjustment of the node grouping strategy of the asynchronous sharding consensus unit in the specific implementation process, the system continuously collects key metrics through the network load monitoring module of the real-time layer verification chain, including: data transfer latency between nodes, transaction verification throughput within a shard, and node computing resource occupancy rate.

[0130] When the latency of any shard is detected to exceed the preset threshold, such as the main shard latency being greater than 500ms or the edge shard latency being greater than 1200ms, the grouping strategy optimization algorithm is triggered, including: First, freeze the new transaction verification requests of this shard, and count the effective verification rate of the nodes within the shard in the past 1 hour. The effective verification rate is the number of successfully verified transactions / the total number of transactions. If the effective verification rate is lower than 80%, it is determined that the node performance is insufficient or there is interference from malicious nodes; if the effective verification rate meets the standard but the latency exceeds the standard, it is determined that the network topology structure is unreasonable.

[0131] For the scenario of insufficient performance, the algorithm retrieves a list of standby nodes from the node reputation library of the deposit layer audit chain, with a reputation score greater than 90 points and an online rate greater than 95%, and replaces the inefficient nodes according to the load balancing principle; for network topology problems, the algorithm re-divides the geographical distribution boundaries of the shards, such as clustering cross-border nodes by time zone, and generates a new shard communication routing table.

[0132] The adjusted grouping strategy parameters need to be submitted to the global anchoring unit of the deposit layer audit chain for compliance verification, including: The verification content includes whether the number of shards exceeds the system's maximum carrying threshold, whether the node allocation conforms to the principle of decentralization, and whether the routing table contains high-risk paths; among them, the grouping strategy parameters include the number of shards, node allocation rules, and routing table version; the system's maximum carrying threshold: the lower limit of the number of nodes per shard is 10, the principle of decentralization: the proportion of nodes of a single institution does not exceed 30%, and high-risk paths: passing through nodes in known low-reputation areas.

[0133] If the verification passes, the global anchoring unit writes the hash value of the new policy and the effective timestamp into the configuration history area of the evidence storage layer audit chain, and synchronizes it to all real-time layer nodes for enforcement; if the verification fails, it rolls back to the previous valid policy version, marks this adjustment attempt as a "high-risk operation" and triggers the following processing: sending node anomaly feature data to the dynamic defense network module to update the attack pattern library; freezing the policy adjustment permission of this shard for 24 hours, during which only manual intervention is allowed.

[0134] In response to abnormal load fluctuations caused by network attacks, the system enables attack scenario-specific policies, including: switching the packet policy optimization algorithm to the conservative mode, suspending the automatic node replacement function, and instead calling the emergency shard template pre-stored in the evidence storage layer audit chain, such as the minimum available shard set, and forcibly enabling the multi-signature verification mechanism, that is: at least 3 high-reputation nodes are required to jointly sign and confirm the shard operation; at the same time, the traffic scheduling module of the real-time layer verification chain starts attack traffic cleaning, routes suspected malicious requests to the sandbox shard for behavior analysis, and releases them to the main shard after confirming security.

[0135] During low-load periods, such as during meeting breaks, the system performs preventive policy optimization: based on the historical load data prediction model in the evidence storage layer audit chain, it merges redundant shards in advance, such as merging 4 low-active shards into 2, releasing node resources for other high-priority tasks, such as AI model training; during the merging process, the system migrates the unfinished transactions within the original shards across shards, and the migration path needs to be verified by the global anchoring unit for compatibility with the target shard, such as the protocol version being the same and the node certificate not being expired.

[0136] The historical grouping policy version backtracking mechanism is automatically activated when detecting anomalies in the current policy: when the shard verification error rate rises by more than 50% in 3 consecutive detection cycles, the system retrieves the last 5 valid policy versions from the evidence storage layer audit chain, replays the load data and verification results for the corresponding time periods one by one, and selects the version with the lowest error rate as the temporary rollback target; the rollback operation needs to be secondarily verified by the global anchoring unit to ensure that the historical policy parameters are compatible with the current network environment, such as the node certificate expiration date not being terminated, and evaluates the impact of the rollback on the overall system performance through the federated multi-modal decision network, such as predicting the change range of the delay after rollback. If the evaluation result meets the security threshold, such as the delay increase not exceeding 15%, then the version rollback is executed and a root cause analysis report is generated and stored in the blockchain.

[0137] This dynamic adjustment mechanism extends to cross-border, multi-node collaboration scenarios: When a collective offline state of nodes in a particular region is detected, such as a regional network failure, the system automatically activates cross-chain mirroring, temporarily transferring the validation responsibilities of the affected shard to a backup chain node group in another region. This transfer process must meet the following conditions: the average node reputation score of the mirrored shard is no lower than that of the original shard; the data synchronization delay after the transfer is no more than 1 second; and a complete mirror relationship mapping table is retained in the audit chain of the evidence layer. After the failure is recovered, the system synchronizes the data status of the mirrored shard with the original shard using a difference comparison algorithm, ensuring data consistency and then releasing the mirror binding.

[0138] The agenda compliance verification rules for on-chain smart contracts include:

[0139] Time conflict detection of parallel agendas of the same participant;

[0140] Verification of regional isolation between VIPs and ordinary participants in the room allocation plan;

[0141] Verify the authority level matching of temporary agenda adjustments;

[0142] After the verification result passes the node consensus of the real-time layer verification chain, the resource scheduling agent is triggered to regenerate the plan.

[0143] It should be further explained that in the specific implementation process, when the on-chain smart contract performs agenda compliance verification, the system implements differentiated processing procedures for different types of rules. When a participant submits a new agenda registration request, the contract first retrieves the timestamp of the user's already stored agenda in the evidence layer audit chain. If it is detected that the time overlap exceeds the preset tolerance value, such as the interval between parallel agendas is less than 15 minutes and the distance between physical venues is greater than 500 meters, a conflict report will be automatically generated and the following processing will be triggered: a priority assessment request for the conflicting agenda will be sent to the federated multimodal decision network module. Based on the agenda type, participant identity weight and historical participation data, if the assessment result is optimizable, such as a low-priority agenda for non-core participants, the resource scheduling agent will be called to generate a time fine-tuning plan, such as a 30-minute delay or migration to a virtual branch venue; if the assessment is not adjustable, such as the organizer's mandatory agenda, the user's new agenda request will be frozen and the conflict details will be pushed to the administrator terminal.

[0144] To verify the VIP area isolation of the room allocation plan, the contract extracts the distribution of participant identity tags in the current room allocation plan from the real-time layer verification chain. If it is detected that the proportion of ordinary participants in the VIP area exceeds 5% or there is physical mixing of VIP and ordinary areas, such as shared entrances and exits, the isolation is determined to have failed.

[0145] At this time, the contract sends a regional reset instruction to the resource scheduling agent, requiring it to regenerate the room allocation plan based on the reinforcement learning model and increase the isolation weight coefficient in the new plan. For example, the boundary buffer zone of the VIP area is expanded to 3 meters. The reset plan needs to pass a double verification, including: first, the AI pre-screening gateway verifies whether it complies with the isolation rules, and then the on-chain smart contract reviews the historical violation records, such as whether the agent has triggered a reset more than 3 times due to the same type of problem in the past 24 hours. If both verifications pass, the hash value of the new plan is anchored to the audit chain of the evidence storage layer; otherwise, the manual review process is initiated.

[0146] When processing a request for adjusting the temporary agenda, the contract performs a three-level verification that matches the permission level: First, it parses the digital identity certificate of the request initiator and compares it with the pre-stored role permission table in the evidence storage chain. For example, only the agenda organizer has the permission to modify the time. Second, it verifies whether the adjusted agenda parameters exceed the authorized scope of the original agenda. For example, the extended duration shall not exceed 50% of the initial value. Finally, it checks whether the operation timestamp is within the legal window period. For example, adjustments are only allowed within 2 hours before the start of the agenda.

[0147] If all three-level verifications pass, the contract marks the adjustment record as the "to be executed" status and broadcasts it to the verification chain nodes in the real-time layer for quick consensus. If any stage fails, a differentiated response is triggered according to the failure type, that is: for the problem of insufficient permissions, an encrypted error code and a permission upgrade guidance link are returned to the requester; for the problem of parameter out-of-bounds, a compliance parameter plan is automatically generated and an AI optimization path description is attached, such as recommended adjustment ranges and impact predictions.

[0148] In a high-concurrency scenario, the system enables an asynchronous queue mechanism for agenda verification, including: sorting the requests to be processed according to priority, where VIP user requests take precedence over ordinary users. For non-urgent operations, they are temporarily stored in the delayed processing area of the evidence storage layer audit chain, and the risk prediction agent evaluates the risk of queue backlog. Among them, non-urgent operations include requests for post-meeting data analysis. If it is detected that the backlog exceeds the system's bearing threshold, such as the number of unprocessed requests is greater than 1000, the verification rules are dynamically adjusted: temporarily relax the conflict tolerance value for non-core agendas, such as shortening the parallel interval to 10 minutes, and record the temporary rule version and the effective period in the evidence storage chain. After the load drops, the standard rules are automatically restored.

[0149] For a multi-regulatory environment across countries, the contract incorporates a dynamic compliance rule engine: when it detects a change in the region of the participant's registration information, it automatically associates the data protection regulations corresponding to the region in the evidence storage chain and reconstructs the verification rule set. The rule switch needs to pass cross-chain verification: the compliance node group of the evidence storage layer audit chain verifies the authenticity of the region label, such as the consistency between the IP geolocation and the registered address, and generates a multilingual compliance statement through the semantic collaboration agent, which can only take effect after the participant's secondary confirmation.

[0150] The results and process data of all verification operations form a double-chain evidence storage system: the agenda status change records at the execution level are written into the operation log partition of the real-time layer verification chain, and the rule matching paths and verification bases at the logical level are encapsulated in the form of zero-knowledge proofs and written into the compliance evidence library of the storage layer audit chain; among them, the verification bases include the conflict detection algorithm version and the source of the isolation weight coefficient. When an audit dispute occurs, based on the hash correlation of the double-chain data, the full life cycle path of the verification decision can be completely traced back, including the AI model version, the contract rule set iteration record, and the network load snapshot, ensuring that the system meets the judicial-level audit requirements while ensuring efficient operation.

[0151] A remote digital conference management method combining blockchain and AI large models includes the following steps:

[0152] Step S1: Participants submit registration information, resource allocation requests, or agenda adjustment applications through terminals. The system receives the data and starts the preprocessing process, encapsulating the original data into a standardized transaction request packet.

[0153] Step S2: The AI pre-screening gateway calls the anomaly detection model to conduct the first logical screening of the requests, analyzes potential conflicts based on the historical trusted data set, intercepts high-risk operations, and generates risk logs to be stored in the blockchain; the data that passes the screening is forwarded to the on-chain smart contract; among them, high-risk operations include duplicate registration and over-authorization requests.

[0154] Step S3: The on-chain smart contract performs secondary rule verification, including identity certificate verification, agenda time conflict detection, and resource capacity compliance check. The transactions that pass the verification trigger the real-time layer node consensus verification of the hierarchical dynamic verification chain.

[0155] Step S4: The real-time layer verification chain verifies the transactions at the millisecond level through the asynchronous sharding consensus mechanism, and the verification results are temporarily stored in the buffer; at the same time, the global anchoring unit periodically writes the hash fingerprint of the real-time layer data into the storage layer audit chain to complete the full life cycle evidence storage.

[0156] Step S5: The federated multi-modal decision network starts dynamic decision-making based on the verification results: the resource scheduling agent generates a room or seat allocation plan, the semantic collaboration agent performs multilingual translation and requests signatures, and the risk prediction agent evaluates the conflict probability and generates response instructions.

[0157] Step S6: For operations involving permission changes or high risks, including VIP permission upgrades and alternate venue switches, the system performs triple verification: AI proposal generation, real-time layer node consensus, and storage layer final audit, which takes effect only after a unanimous vote.

[0158] Step S7: The dynamic defense network continuously monitors the security event logs of the evidence storage layer, generates defense strategies after identifying attack patterns, dynamically adjusts data filtering rules, encryption policies, and access control permissions, and synchronously writes the defense records to the blockchain.

[0159] Step S8: After the multilingual translation results are signed with the private keys of the participants, the real-time layer verification chain verifies the validity of the signature. After passing, they are written to the evidence storage layer according to the three-level index of "conference - agenda - language" to form a traceable evidence chain.

[0160] Step S9: When agenda conflicts or resource anomalies are detected, the on-chain smart contract automatically triggers an adjustment process: retrieves the status of alternative resources, generates migration plans or modifies agenda parameters, and updates the blockchain records after the signatures of the participants or administrators are confirmed.

[0161] Step S10: The system periodically performs self-optimization operations: dynamically adjusts the node grouping strategy, updates the federated learning of the defense knowledge base, and checks the consistency of the data in the dual chains. All optimization parameters take effect after passing the compliance verification of the audit chain in the evidence storage layer.

[0162] Through the deep integration of the blockchain and the AI large model, the security and intelligence level of remote meeting management have been improved. The blockchain technology constructs a hierarchical dynamic verification chain to ensure the immutability and full-life-cycle traceability of the data in the entire meeting process, effectively solving the defects of traditional centralized systems being vulnerable to attacks and data being easily altered. The federated multi-modal decision-making network driven by the AI large model realizes autonomous optimization and real-time response in links such as resource scheduling, risk prediction, and cross-language collaboration, reducing the cost of manual intervention and improving management efficiency.

[0163] It solves the coordination problem between the security mechanism and decision-making efficiency, and balances the strong audit requirements of the blockchain and the real-time response ability of the AI through the dual-chain architecture. The dynamic defense network and the self-optimization mechanism further strengthen the system robustness, can actively identify and intercept new types of attacks such as data poisoning and permission tampering, and continuously optimize the AI model and resource allocation strategy based on federated learning.

[0164] It should be noted that, in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0165] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A remote digital conference management system combining blockchain and AI large models, characterized in that, It includes: A hierarchical dynamic verification chain module, which is composed of a real-time layer verification chain and an evidence storage layer audit chain connected through a data synchronization interface. The real-time layer verification chain adopts a lightweight consortium chain architecture, and the evidence storage layer audit chain is built based on a public chain; A federated multi-modal decision-making network module, whose input end is communicatively connected to the output end of the hierarchical dynamic verification chain module, and is used to generate a conference resource scheduling plan and risk response instructions; A dual-chain heterogeneous consensus engine, embedded in the hierarchical dynamic verification chain module, including an asynchronous sharding consensus unit and a global anchoring unit. The asynchronous sharding consensus unit is used for parallel verification of real-time layer data, and the global anchoring unit is used to periodically write the hash fingerprint of real-time layer data into the evidence storage layer audit chain; A dynamic defense network module, whose attack feature library is data-connected to the evidence storage layer audit chain of the hierarchical dynamic verification chain module, and the defense strategy output end is connected to the control end of the federated multi-modal decision-making network module.

2. The remote digital conference management system combining blockchain and AI large models according to claim 1, characterized in that: The real-time layer verification chain includes an AI pre-screening gateway, which is composed of an anomaly detection model and an on-chain smart contract: The anomaly detection model is built based on a graph neural network, and the training data comes from the registration information, room allocation records, and permission change logs in the historical conference dataset; The on-chain smart contract contains rules for verifying the identities of participants, rules for verifying the compliance of the agenda, and rules for detecting resource allocation conflicts; Among them, the output end of the anomaly detection model is connected in series with the input end of the on-chain smart contract. Only when the data passes the screening of the anomaly detection model and the verification of the on-chain smart contract, the federated multi-modal decision-making network module is triggered to start.

3. The remote digital conference management system combining blockchain and AI large model according to claim 1, characterized in that: The federated multi-modal decision-making network module includes: A resource scheduling agent, whose data input end is connected to the resource allocation interface of the real-time layer verification chain, generates a room allocation or seating plan based on a reinforcement learning model, and dynamically optimizes the plan weight through the Monte Carlo tree search algorithm; A semantic collaboration agent, built-in with a multi-language large model, whose corpus is stored in the evidence storage layer audit chain, and the translation result is signed by the private key of the participant's terminal and then sent back to the evidence storage layer audit chain; A risk prediction agent, which uses a temporal graph convolutional network to analyze the historical conference data in the evidence storage layer audit chain, predicts the probability of agenda conflicts and outputs it to the on-chain smart contract.

4. The remote digital conference management system combining blockchain and AI large models according to claim 3, characterized in that: After the resource scheduling agent generates a room allocation or seating plan, it synchronously writes two types of on-chain records to the hierarchical dynamic verification chain module: An execution chain record, which includes a plan identifier, an execution timestamp, and a resource utilization rate indicator; A logic chain record, which encapsulates the parameter update path of the decision model and the fairness constraint verification result in the form of zero-knowledge proof; Among them, the execution chain record and the logic chain record are associated through a hash and input into the feedback learning unit of the federated multi-modal decision-making network module.

5. The remote digital conference management system combining blockchain and AI large models according to claim 1, characterized in that: The hierarchical dynamic verification chain module implements triple verification on VIP permission change operations: The first verification generates a preliminary proposal by the risk prediction agent of the federated multi-modal decision-making network module. The preliminary proposal includes the scope of permission change, timeliness, and risk score; The second verification is to perform node group fast consensus on the proposal by the asynchronous sharding consensus unit of the real-time layer verification chain. The verification content includes user identity hash matching, permission timeliness conflict detection, and role permission upper limit verification; The third - level verification randomly spot - checks the logical consistency between the proposed plan and historical permission records by the final - audit nodes of the audit chain in the evidence - depositing layer, including the compliance of the permission - escalation path and the rationality of the risk - score distribution; Only when the results of the three - level verifications are consistent, the proposed plan is marked as valid and written into the permission - change partition of the audit chain in the evidence - depositing layer.

6. The remote digital conference management system combining blockchain and AI large models according to claim 1, characterized in that: The dynamic defense network module includes: An attack - feature extraction unit that extracts the data patterns of historical attack events from the audit chain in the evidence - depositing layer; An AI counter - measure model training unit that generates defense strategies based on the extracted attack features and sends them back to the federated multi - modal decision - making network module; Among them, the defense strategies include data - poisoning attack interception rules and model - parameter encryption rules, and the strategy version numbers are synchronously updated to the audit chain in the evidence - depositing layer.

7. The remote digital conference management system combining blockchain and AI large models according to claim 3, characterized in that: After the translation result of the semantic collaboration agent is generated, it needs to be signed by the private key of the participant's terminal and the signature validity is verified by the asynchronous sharding consensus unit of the real - time layer verification chain. After passing the verification, it is written into the multi - language evidence - chain partition of the audit chain in the evidence - depositing layer.

8. The remote digital conference management system integrating blockchain and AI large models according to claim 3, characterized in that: When the risk - prediction agent detects that the probability of agenda conflict exceeds the preset threshold, it triggers the on - chain smart contract to perform the following operations: Call the venue - status interface of the real - time layer verification chain to retrieve the availability of the backup venue; If the backup venue is available, send a switching instruction to the federated multi - modal decision - making network module and update the agenda record in the audit chain of the evidence - depositing layer.

9. The remote digital conference management system integrating blockchain and AI large models according to claim 1, characterized in that: The node - grouping strategy of the asynchronous sharding consensus unit is dynamically adjusted according to the network load of the real - time layer verification chain. The adjusted parameters take effect after being verified by the global anchoring unit of the audit chain in the evidence - depositing layer, and the historical grouping - strategy versions are stored in the audit chain of the evidence - depositing layer.

10. The remote digital conference management system integrating blockchain and AI large model according to claim 2, characterized in that: The agenda - compliance verification rules of the on - chain smart contract include: Detecting the time conflict of parallel agendas of the same participant; Verifying the regional isolation between VIPs and ordinary participants in the room - allocation plan; Verifying the matching of permission levels for ad - hoc agenda adjustments; After the verification results pass the node consensus of the real - time layer verification chain, trigger the resource - scheduling agent to regenerate the plan.

Images