Blockchain-based industrial digitization real-time risk control and decision support system

Abstract

The present application relates to the technical field of blockchain and data processing, in particular to an industry digitization real-time risk control and decision support system based on blockchain, comprising a cause-effect entropy chain construction module, a business plan simulation module, a multi-party trusted cooperation module and a business credit evaluation module.The cause-effect entropy chain construction module uses dynamic graph neural network to infer the cause-effect relationship of enterprise operation data, quantifies the entropy value, and packs the result into a block to record on the blockchain to form a cause-effect relationship account book; based on the risk link with the entropy value in the cause-effect relationship account book exceeding the preset threshold, the business plan simulation module performs adversarial risk simulation to generate a business decision plan; the multi-party trusted cooperation module allows each enterprise to cooperatively train a global risk prediction model through federated learning and zero-knowledge proof without exposing private data; and the business credit evaluation module generates a dynamic trusted two-dimensional code that can represent the enterprise's operating capacity and risk level for various business scenarios.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of blockchain and data processing, in particular to an industrial digital real-time risk control and decision support system based on blockchain. BACKGROUND

[0002] In the era of industrial digitalization, the systematicness and linkage of enterprise operation risks are increasingly enhanced, which poses a severe challenge to the traditional lagging business risk management mode.

[0003] The existing business risk control system relies on historical data for correlation analysis, and lacks insight into the causal transmission of risks, resulting in lagging decision-making. Although blockchain is used for data authorization and evidence storage, the application is relatively shallow. Due to concerns about business data privacy and security, cross-subject collaborative risk modeling is difficult to carry out, and data silos are common. The credit certificates issued by enterprises to the outside are mostly static reports, which are poor in timeliness and difficult to verify. Therefore, there is an urgent need for a business risk decision support system that can realize forward-looking, trusted collaboration, and dynamic verification.

[0004] To this end, an industrial digital real-time risk control and decision support system based on blockchain is proposed. SUMMARY

[0005] The purpose of the present application is to provide an industrial digital real-time risk control and decision support system based on blockchain, which provides dynamic and verifiable business decision support through the integration of deep causal insight, forward-looking risk simulation, and multi-party trusted collaboration.

[0006] To achieve the above purpose, the present application provides the following technical solutions:

[0007] The industrial digital real-time risk control and decision support system based on blockchain comprises:

[0008] A causal entropy chain construction module: a dynamic graph neural network is used to model enterprise operation data as a dynamic graph, infer the causal relationships between enterprise operation data, and use the information entropy principle to quantify the entropy value of the causal relationships, and identify risk links with entropy values exceeding a preset threshold. The causal relationships, entropy values, and model parameter hash values are packaged into causal relationship blocks and recorded on the blockchain to form an unalterable causal relationship ledger.

[0009] A business contingency simulation module: based on the risk links in the causal relationship ledger with entropy values exceeding the preset threshold, an adversarial risk simulation is performed to proactively discover operational risks and generate business decision contingency plans.

[0010] Multi-party trusted collaboration module: allows enterprises to maintain local digital twins on their own private data, and periodically collaboratively train and update a shared global risk prediction model through federated learning without exposing private data. Each participant's local model update is accompanied by zero-knowledge proof and updated model parameter hash value.

[0011] Commercial credit assessment module: generates a dynamic trusted two-dimensional code that represents the operating capacity and risk level of an enterprise, which is used in commercial business scenarios.

[0012] Preferably, the enterprise operation data includes business and judicial data, tax declaration data, invoice data, intellectual property data, and enterprise internal operation data.

[0013] Preferably, the causal entropy chain construction module includes a dynamic graph modeling unit, an entropy value quantization unit, and a causal block generation unit. The dynamic graph modeling unit models entities in enterprise operations as nodes of a dynamic graph and models materials and information between entities as edges of the graph. It uses a dynamic graph neural network to learn the propagation pattern and conditional probability of risks between nodes to infer causal relationships. The entropy value quantization unit uses the information entropy principle to quantify the entropy values of the causal relationships and identify risk links with entropy values exceeding a preset threshold. The causal block generation unit packages the inferred causal relationships, their corresponding entropy values, and model parameter hash values used for inference into a causally signed causal relationship block to form an unalterable and auditable causal relationship ledger on the blockchain.

[0014] Preferably, the business contingency simulation module includes a risk synthesis unit and a strategy optimization unit. The risk synthesis unit uses a generative artificial intelligence model to automatically generate simulated risk scenarios by combining multiple low-probability events based on risk links in the causal relationship ledger with entropy values exceeding a preset threshold. The strategy optimization unit uses reinforcement learning to find and output a business decision plan that minimizes operating losses in the simulated risk scenarios through trial-and-error learning.

[0015] Preferably, the multi-party trusted cooperation module comprises a local training and verification unit, a homomorphic encryption and secure aggregation unit, and a multi-party collaborative decryption unit; the local training and verification unit is used for each enterprise to generate a local model update on the respective private data through a federated learning algorithm, and generate a zero-knowledge proof for the updated calculation process to prove the compliance of the calculation without exposing the private data; the homomorphic encryption and secure aggregation unit is used to receive the encrypted local model update and the zero-knowledge proof submitted by each enterprise, and after verifying that the zero-knowledge proof is valid, the homomorphic encryption technology is used to aggregate all valid local model updates in the ciphertext state to form an encrypted global model; and the multi-party collaborative decryption unit organizes multiple enterprises to use the private key fragments held by each enterprise to collaboratively decrypt the encrypted global model through a secure multi-party computation protocol.

[0016] Preferably, the dynamic trusted two-dimensional code is a dynamic verifiable credential, comprising an access address pointing to a smart contract on a block chain; the smart contract records an evaluation result generated by the commercial credit evaluation module, representing the current operating ability and risk level of the enterprise; a zero-knowledge proof for proving that the evaluation result is calculated by an authorized model and trusted data; and an index pointing to the cause-and-effect relationship account book and data authorization record.

[0017] Preferably, the dynamic updating mechanism of the dynamic trusted two-dimensional code comprises that when an instant evaluation request from a third-party service institution is received and / or a preset updating period is reached, the system will automatically trigger a new round of enterprise operation data collection and analysis process, and refresh and update the evaluation result and the corresponding zero-knowledge proof recorded in the smart contract.

[0018] Preferably, the commercial plan simulation module further comprises an expert decision upgrading mechanism, which automatically suspends the simulation when the strategy optimization unit cannot find a decision plan within a preset operating loss threshold, and packs the simulated risk scenario into an on-chain governance proposal and submits it to the decentralized expert unit for research and voting, and finally executes the commercial decision formed by the expert consensus according to the voting result.

[0019] Compared with the prior art, the present application has the following beneficial effects:

[0020] 1. By constructing a business cause-and-effect relationship account book, the present application deeply understands the internal logic of risk transmission, and combines with forward-looking adversarial simulation, so that the enterprise can actively find and develop optimized plans for potential and complex unknown risks, and fundamentally improves the forward-looking and systematic anti-fragility of commercial decisions.

[0021] 2. This invention utilizes technologies such as federated learning, zero-knowledge proof, and homomorphic encryption to enable multiple enterprises within an industry chain or region to collaborate on modeling securely and pool ecological wisdom while fully protecting their respective trade secrets, thereby achieving accurate predictions of macro-market and systemic risks.

[0022] 3. This invention transforms a company's complex, real-time operating status into a concise, reliable, and readily verifiable digital credential by generating a dynamically verifiable, trusted QR code. This replaces traditional, delayed, and easily forged static reports, significantly reducing trust costs and business friction in scenarios such as finance and supply chains. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Fig. 1 A flowchart of the method for the blockchain-based industrial digital real-time risk control and decision support system proposed in an embodiment of the present invention;

[0024] Fig. 2 This is a system structure diagram of the blockchain-based industrial digital real-time risk control and decision support system proposed in an embodiment of the present invention;

[0025] Fig. 3 This is a system flow chart of the blockchain-based industrial digital real-time risk control and decision support system proposed in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0027] Example 1

[0028] See also Figs. 1-3 The present invention provides a blockchain-based industrial digital real-time risk control and decision support system. The technical solution is as follows:

[0029] Industrial digital real-time risk control and decision support system based on blockchain, such as Figs. 1-3 ,include:

[0030] Causal Entropy Chain Building Module: Utilizing dynamic graph neural networks, enterprise operational data is modeled as dynamic graphs, inferring causal relationships between these data points. The information entropy principle is then used to quantify the entropy of these causal relationships, identifying risk links where the entropy exceeds a preset threshold. Causal relationships, entropy values, and model parameter hash values ​​are packaged into causal blocks and recorded on the blockchain, forming an immutable causal ledger.

[0031] Business plan simulation module: Based on the risk links in the causal relationship ledger whose entropy value exceeds the preset threshold, adversarial risk simulation is performed to actively discover business risks and generate business decision plans.

[0032] Multi-party trusted collaboration module: This module allows enterprises to maintain local digital twins based on their own private data. Without exposing private data, the module periodically collaborates on training and updating a shared global risk prediction model through federated learning. Each participant's submitted local model update is accompanied by a zero-knowledge proof and a hash of the updated model parameters.

[0033] Commercial credit assessment module: Generates a dynamic and credible QR code that can represent the business operation capability and risk level of the enterprise, for use in commercial business scenarios.

[0034] Furthermore, the enterprise operation data includes industrial and commercial and judicial data, financial and tax declaration data, invoice data, intellectual property data and internal enterprise operation data.

[0035] For enterprise operational data, we use the predefined JSON Schema to perform structured cleaning on data from different sources; we use entity recognition technology to perform alignment; and we define a feature mapping table to map the enterprise's multi-dimensional operating indicators to the corresponding numerical attributes of the enterprise nodes in the graph model after normalization.

[0036] By integrating and standardizing multi-source data, we can effectively avoid the one-sidedness and fraud risks of a single data source, and lay a high-quality data foundation for building a comprehensive and objective enterprise risk profile.

[0037] Furthermore, the causal entropy chain construction module includes a dynamic graph modeling unit, an entropy value quantification unit and a causal block generation unit. The dynamic graph modeling unit is used to model the entities in the enterprise operation as nodes of the dynamic graph, and model the materials and information between the entities as edges of the graph; and use the dynamic graph neural network to learn the risk propagation pattern and conditional probability between nodes to infer the causal relationship; the entropy value quantification unit is used to use the information entropy principle to quantify the entropy value of the causal relationship and identify the risk link whose entropy value exceeds a preset threshold; the causal block generation unit is used to package the inferred causal relationship, its corresponding entropy value and the hash value of the model parameter used for inference into a cryptographically signed causal relationship block to form an unalterable and auditable causal relationship ledger on the blockchain.

[0038] The dynamic graph modeling unit specifically performs modeling in the following manner:

[0039] Node definition: define the market subjects such as enterprises, banks, suppliers and customers as entity nodes, and define invoices, contracts, litigation cases and intellectual property rights as event nodes. Each node has different attributes according to its type.

[0040] Edge definition: if there is a direct business transaction or legal relationship between two nodes, an edge with direction and type is defined. Dynamic graph construction: all nodes and edges generated within a month are constructed as a graph snapshot.

[0041] Causal inference model: the dynamic graph neural network specifically selects a time-aware graph attention network. To achieve causal inference, the time-aware graph attention network is combined with Granger causality test. The specific steps are as follows: first, the time-aware graph attention network is used to learn the dynamic embedding representation of each node in the sequence graph. Then, for any two linked nodes i and j, the time series Finally, the vector autoregressive model is executed on the two time series, and Granger causality test is performed to screen out node pairs with statistically significant predictive association, identify potential risk transmission paths, and infer probabilistic causal relationships.

[0042] By modeling abstract operational data into specific dynamic causal graphs, a deep transition from "association" to "causality" is achieved, enabling the system to understand the fundamental logic of risk transmission and providing a solid basis for accurate decision-making and risk auditing.

[0043] The entropy quantification unit specifically adopts the following logical steps to quantify the entropy value of the causal relationship:

[0044] For each causal relationship inferred by the dynamic graph neural network, the unit obtains its corresponding probability score, which represents the possibility of the occurrence of the result event after the occurrence of the cause event.

[0045] The entropy quantification unit evaluates the probability score according to the information entropy principle, and the core logic is: if the probability score of a causal relationship tends to the certainty extreme, i.e. the result is highly predictable, then a low entropy value is assigned to it; on the contrary, if the probability score tends to the intermediate value, i.e. the result has high unpredictability, then a high entropy value is assigned to it.

[0046] ​​​​​​Meanwhile, a threshold of entropy value score is set, and when the entropy value score of a certain causal relationship exceeds the preset threshold, the causal link is marked as a risk link with an entropy value exceeding the preset threshold, and serves as a key input of a subsequent business plan simulation module.

[0047] Through the principle of information entropy, the entropy value is accurately quantified by mathematics, so that the system can automatically identify the core risk link that deserves the most attention from a large number of relationships, and provide a key input for subsequent simulation decision-making.

[0048] By modeling abstract operation data into a specific causal relationship graph, a deep transition from correlation analysis to causal logic inference is achieved. The quantification of risk entropy can help enterprises accurately identify and focus on core risk points. The final on-chain causal ledger provides an unalterable basis for decision-making audit and attribution.

[0049] Further, the business plan simulation module includes a risk synthesis unit and a strategy optimization unit; the risk synthesis unit uses a generative artificial intelligence model to automatically generate a simulated risk scenario by combining multiple low-probability events based on the risk links with entropy values exceeding the preset threshold in the causal relationship ledger; and the strategy optimization unit uses a reinforcement learning method to find and output a business decision plan that can minimize operating losses in the simulated risk scenario through trial and error learning.

[0050] The risk synthesis unit specifically uses a conditional variational autoencoder (CVAE) as a generative artificial intelligence model. The training data of CVAE is real risk events in history (for example, a certain enterprise's financial chain breaks down), and each event is composed of a group of precursor events and a label. The label is a risk link with an entropy value exceeding the preset threshold identified from the causal entropy chain, such as "upstream core supplier A stops production". In the simulation stage, the operator can select one or more causal links with high entropy values from the causal entropy chain as input conditions. CVAE will sample and decode in the latent space based on these conditions to generate a set of logically possible precursor event combinations that have not occurred in history. For example, the input condition is "upstream core supplier A stops production", and the model may generate a combination of "30% increase in raw material prices", "sea transportation disruption", and "competitor B obtains large financing", which together constitute a new and complex simulated risk scenario.

[0051] The strategy optimization unit specifically uses a proximal policy optimization algorithm as a reinforcement learning method. Its core elements are defined as follows:

[0052] State space: the current simulated risk scenario, represented by a vector containing the event combination generated by the risk synthesis unit and the current enterprise's core operating indicators (such as cash flow, inventory, and order volume).

[0053] Action Space: a discrete set of decisions representing the countermeasures the enterprise can take. For example: {“Emergency procurement of alternative raw materials” “Apply for a short-term loan of 1 million yuan” “Start up the backup production line” “Issue a notice of delayed delivery to customers”}.

[0054] Reward Function: defined as a weighted combination of changes in the enterprise’s key performance indicators (KPIs). At each simulation time step t, after performing action , the reward is calculated as follows = . Where, is the change in cash flow, is the change in inventory holding cost, is the amount of fines incurred due to delayed delivery, etc. , , are preset weight coefficients representing the enterprise’s emphasis on cash, cost, and reputation, respectively.

[0055] Through trial and error in thousands of simulations, the reinforcement learning model gradually learns which actions to take in which states to achieve the highest long-term cumulative reward. This learned optimal strategy is ultimately output as a “business decision plan”.

[0056] Through the adversarial simulation of generative AI and reinforcement learning, the invention realizes a strategic shift from passive risk response to active “immunity”. The system can automatically reorganize and deduce historical risk factors to generate complex risk combination scenarios that experts may overlook, and find the optimal emergency plan in the process, thereby completing stress testing and strategy reserves before a real crisis occurs.

[0057] Further, the multi-party trusted collaboration module includes a local training and verification unit, a homomorphic encryption and secure aggregation unit, and a multi-party collaborative decryption unit; the local training and verification unit is used for each enterprise to generate a local model update on its own private data through a federated learning algorithm, and generate a zero-knowledge proof for the updated calculation process, proving the compliance of the calculation without exposing the private data; the homomorphic encryption and secure aggregation unit is used to receive the encrypted local model updates and zero-knowledge proofs submitted by each enterprise, and after verifying the validity of the zero-knowledge proofs, it aggregates all valid local model updates in a ciphertext state using homomorphic encryption technology to form an encrypted global model; the multi-party collaborative decryption unit organizes multiple enterprises to use private key shards held by each enterprise to collaboratively decrypt the encrypted global model through a secure multi-party computation protocol.

[0058] The specific implementation process of the multi-party trusted collaboration module is as follows, and each step is designed to ensure data privacy and the trustworthiness of the computing process:

[0059] Each participating enterprise adopts a common federated averaging strategy. First, each enterprise independently trains a global risk prediction model on its own private data, calculating a local adjustment value for the global model parameters, also known as model updates. This model update is the only data that needs to leave the local environment of the enterprise. Before submitting the model update, each enterprise generates a zero-knowledge proof. This is an independent encrypted data package whose purpose is to prove to other parties in the system that the enterprise has generated this model update strictly in accordance with pre-established rules (e.g., using the correct data format and standard training algorithm). Most importantly, this proof process does not reveal any private data itself used for calculation. To protect business secrets, the enterprise encrypts the model update using a homomorphic encryption scheme that supports addition operations before uploading the model update and zero-knowledge proof. After receiving the encrypted model update and zero-knowledge proof, the central aggregation server first verifies the validity of the proof to eliminate non-compliant submissions. Then, the aggregator uses the magical properties of homomorphic encryption to directly sum all valid model updates in an encrypted state, ultimately obtaining an encrypted, aggregated global model update.

[0060] The encrypted global model update cannot be decrypted by any single party, including the aggregator. The complete key required for decryption is divided into multiple key shards, each of which is held by a participating enterprise. Only when a pre-set number of enterprises (e.g., more than half) participate in a secure multi-party computation protocol and simultaneously provide their respective key shards can the decryption operation be completed cooperatively, restoring the clear and usable global model update. This mechanism ensures that the final result is the result of collective consensus, reducing the risk of single-point failure or malicious manipulation.

[0061] The global risk prediction model is a three-layer feedforward neural network, with the input layer containing 128 neurons corresponding to 128 macroeconomic and industry risk factors defined jointly by the industry chain partners; the hidden layer contains 64 neurons using the ReLU activation function; and the output layer contains 1 neuron using the Sigmoid activation function, outputting the probability of systemic risk occurring in the next quarter.

[0062] The local training and verification unit generates zero-knowledge proofs using the zk-SNARKs protocol (such as the Groth16 algorithm); the homomorphic encryption and secure aggregation unit encrypts and aggregates the model updates using the Paillier cryptosystem; and the multi-party collaborative decryption unit implements secure multi-party computation decryption based on the Shamir secret sharing scheme.

[0063] By means of multi-layered cryptography design such as zero-knowledge proof and homomorphic encryption, the application fundamentally solves the trust paradox of data collaboration, guarantees the honesty of the calculation process of each party, completely removes the dependence on a centralized trusted third party, and enables entities that are competitors to each other to safely carry out deep data collaboration.

[0064] Further, the dynamic trusted two-dimensional code is a dynamic verifiable credential, which includes an access address pointing to a smart contract on a blockchain, the smart contract records an evaluation result generated by the commercial credit evaluation module, the evaluation result represents the current operating capacity and risk level of an enterprise, a zero-knowledge proof for proving that the evaluation result is calculated by an authorized model and trusted data, and an index pointing to the cause-and-effect relationship ledger and data authorization record.

[0065] The enterprise credit is upgraded to a dynamic and verifiable digital credential. By means of zero-knowledge proof, a third party can verify the authenticity and compliance of the evaluation result without accessing private data, which provides an audit basis for business decision-making and greatly reduces the cost of commercial mutual trust.

[0066] Further, the dynamic updating mechanism of the dynamic trusted two-dimensional code includes that when an instant evaluation request from a third-party service institution is received and / or a preset updating period is reached, the system will automatically trigger a new round of enterprise operation data collection and analysis process, refresh and update the evaluation result and the corresponding zero-knowledge proof recorded in the smart contract.

[0067] Through the on-demand or periodic automatic updating mechanism, it is ensured that the trusted credential can reflect the latest operating status of the enterprise in real time, so that the third party can make more solid and controllable risk when making business decisions such as credit approval and cooperation review.

[0068] Further, the business contingency simulation module further includes an expert decision upgrading mechanism, when the strategy optimization unit cannot find a decision plan within a preset operating loss threshold, the simulation is automatically suspended, and the simulated risk scenario is packaged into a chain governance proposal and submitted to the decentralized expert unit for judgment and voting, and finally the business decision formed by the expert consensus is executed according to the voting result.

[0069] Through the governance mode of man-machine combination, the rigid boundary problem of pure automatic decision is solved. When AI faces extremely complex risks, it can be seamlessly upgraded to human expert consensus, which takes into account the efficiency of machines and the wisdom of experts. The on-chain governance process ensures the transparency, fairness and traceability of crisis handling.

[0070] The system includes two core processes of real-time assessment and periodic collaborative modeling. The multi-party trusted collaboration module is periodically executed (e.g., once a week) to securely generate and update a global risk prediction model using federated learning and multiple cryptography techniques. The causal entropy chain construction module and the commercial credit assessment module of each enterprise call the latest version of the global risk prediction model stored locally as an environmental parameter input when performing real-time assessment, thereby achieving immediate response to risks.

[0071] The present application provides a novel commercial risk decision support system. It realizes the strategic shift from passive response to active prediction through causal insight and forward-looking risk simulation; breaks data silos using advanced cryptography to enable ecosystem-level collaborative intelligence while protecting commercial privacy; and finally converts complex analysis results into dynamically verifiable commercial credit certificates, significantly reducing trust costs in financial and supply chain scenarios and improving overall industry efficiency and security.

[0072] Embodiment two

[0073] This embodiment will take a high-end precision manufacturing industry chain as a specific scenario to further illustrate the implementation of the present application. The core enterprise of the industry chain is an automobile engine manufacturer A, which involves dozens of key component suppliers (such as pistons, turbochargers, electronic control units ECU, etc.) upstream and multiple large automobile manufacturers downstream.

[0074] First step: construction of causal entropy chain and risk identification

[0075] Automobile manufacturer A as the chain master first uses the system. The system accesses the enterprise internal operation data of A company (such as production plan, inventory data, order data in ERP system), and authorizedly obtains its invoice data, financial and tax data, and related public judicial proceedings and intellectual property data. At the same time, its core suppliers (for example, German turbocharger supplier B and domestic ECU supplier C) also join the system as participants, and authorize to share part of the desensitized production and logistics data.

[0076] The dynamic graph modeling unit in the system models enterprises A, suppliers B, C, logistics company D, etc. as "entity nodes". The purchase contract between A and B, the invoice issued by B to A, the payment of A to B, and the logistics documents of D company are modeled as "event nodes" and "transaction" type edges with time and amount attributes.

[0077] The system constructs a graph snapshot in a weekly unit. By analyzing the dynamic graph sequence of consecutive weeks, the system's time-aware graph attention network and Granger causality test model discover a potential causal relationship: "the degree of port congestion in the region where supplier B is located" and "the risk of engine production line downtime of manufacturer A" have a probabilistic causal relationship, with a confidence score of 60%. Since this score is close to 50%, the system's entropy quantification unit assigns it an entropy score of 0.7 based on the principle of information entropy. Since this score exceeds the system's pre-set risk threshold of 0.65 for this type of link, it is marked as a "risk link with entropy exceeding the pre-set threshold".

[0078] Step 2: Simulation and optimization of business scenarios

[0079] The business scenario simulation module is triggered, and the risk synthesis unit takes the high-entropy link "port congestion of supplier B" as the basis. Its built-in conditional variational autoencoder (CVAE) model, combined with other unrelated low-probability events in history, generates a complex simulated risk scenario: "port congestion of supplier B is severe, causing a delay in the delivery of turbochargers by sea for at least 3 weeks; at the same time, a domestic competitor releases a new engine and announces a 15% price reduction to seize the market".

[0080] In the face of this simulated scenario (state space), the proximal policy optimization algorithm of the strategy optimization unit begins trial-and-error learning.

[0081] Action 1: Execute the decision to "air transport part of the emergency materials". Simulation and deduction find that although this can solve part of the problem, the high air transportation cost will seriously erode product profits, resulting in a lower negative feedback score from the reward function.

[0082] Action 2: Execute the decision to "start the authentication and procurement process for backup domestic turbocharger supplier E". Simulation and deduction find that although the performance of supplier E's products is slightly lower than that of B, it can ensure supply chain continuity and is more cost-effective, and can still maintain profits in the face of competitors' price reductions. The reward function gives a higher positive feedback score based on the comprehensive evaluation of financial and operational impacts.

[0083] After thousands of rounds of simulation, the system finally outputs the optimal business decision plan: "immediately start the emergency authentication of supplier E, and inform major customers of potential delivery risks to manage expectations, and use part of the risk reserve to offset short-term profit fluctuations".

[0084] Step 3: Multi-party trusted collaboration and macro risk prediction

[0085] Manufacturer A, supplier C and dozens of other enterprises in the industry chain jointly participate in a multi-party trusted collaboration network to collaboratively train a "macro risk prediction model of the automobile industry chain".

[0086] Each enterprise independently calculates model updates using local data and uploads them after encryption through a homomorphic encryption scheme.

[0087] The central aggregator aggregates all encrypted updates without touching the plaintext data using the properties of homomorphic encryption.

[0088] The decryption process requires more than half of the enterprises to use their respective private key shards to collaboratively complete, ensuring the fairness of the results and the confidentiality of the process.

[0089] Finally, this global model can predict macro risks such as "overall chip price trend rising in the next six months", providing a broader decision-making perspective for all participating enterprises.

[0090] Step 4: Application of dynamic trusted QR code

[0091] A commercial bank F plans to provide supply chain financing for manufacturer A. Traditional due diligence takes weeks. Now, the credit officer of bank F only needs to scan the "dynamic trusted QR code" provided by A company.

[0092] The QR code contains an access address pointing to a smart contract on the blockchain.

[0093] The credit officer can instantly view the assessment results of A company's operating capacity and risk level generated by the commercial credit assessment module up to the current day by accessing the address. Bank F's system can also automatically verify the zero-knowledge proof attached to the assessment results to confirm that the results are indeed calculated by the authorized model based on A company's real and trusted data (such as the latest invoices, tax data), and that the entire calculation process is compliant. Since the QR code will automatically refresh when manufacturer A completes a new large order or reaches the preset update period (e.g. every week), bank F can make credit decisions based on near real-time and trusted data, greatly improving approval efficiency and reducing risk.

[0094] The invention provides comprehensive support for core enterprises in the industry chain and their partners from micro risk planning to macro risk insight, to trusted financing, demonstrating its great value in improving the stability and efficiency of the entire industry digital ecosystem.

[0095] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be effected without departing from the spirit and scope of the present application, which is not limited to the exact construction and arrangement described. It is intended, therefore, to cover all modifications and adaptions that fall within the scope of the claims and their equivalents.

Claims

1. The blockchain-based industrial digital real-time risk control and decision support system is characterized by: include: Causal entropy chain building module: Utilizes dynamic graph neural networks to model enterprise operational data into dynamic graphs, inferring causal relationships between enterprise operational data, and uses the information entropy principle to quantify the entropy value of causal relationships, identifying risk links whose entropy values ​​exceed a preset threshold. The causal relationship, entropy value and model parameter hash value are packaged into a causal relationship block and recorded on the blockchain to form an unalterable causal relationship ledger; Business plan simulation module: Based on the risk links in the causal relationship ledger whose entropy value exceeds the preset threshold, it conducts adversarial risk simulation to proactively identify operational risks and generate business decision plans; The business plan simulation module includes a risk synthesis unit and a strategy optimization unit; The risk synthesis unit uses a generative artificial intelligence model to automatically generate simulated risk scenarios by combining multiple low-probability events based on risk links in the causal relationship ledger whose entropy values ​​exceed a preset threshold; The strategy optimization unit uses a reinforcement learning method to find and output a business decision plan that can minimize operating losses through trial and error learning in the simulated risk scenario; The business plan simulation module also includes an expert decision upgrade mechanism: When the strategy optimization unit cannot find a decision plan within the preset operating loss threshold, the simulation is automatically suspended. The simulated risk scenario is packaged into an on-chain governance proposal and submitted to the decentralized expert unit for analysis and voting. Finally, the business decision formed by expert consensus is executed based on the voting results. Multi-party trusted collaboration module: This module allows enterprises to maintain local digital twins on their own private data. Without exposing private data, it periodically collaborates to train and update a shared global risk prediction model through federated learning. Each participant's submitted local model update is accompanied by a zero-knowledge proof and a hash of the updated model parameters. The multi-party trusted collaboration module includes a local training and verification unit, a homomorphic encryption and security aggregation unit, and a multi-party collaborative decryption unit; The local training and verification unit is used by each enterprise to generate local model updates using a federated learning algorithm on its own private data, and to generate a zero-knowledge proof for the updated calculation process, proving the compliance of the calculation without exposing private data; The homomorphic encryption and security aggregation unit is used to receive the encrypted local model updates and zero-knowledge proofs submitted by each enterprise. After verifying the validity of the zero-knowledge proofs, it uses homomorphic encryption technology to aggregate all valid local model updates in the ciphertext state to form an encrypted global model. The multi-party collaborative decryption unit organizes multiple enterprises to use their respective private key shards to collaboratively decrypt the encrypted global model through a secure multi-party computing protocol; Commercial credit assessment module: Generates a dynamic and credible QR code that can represent the business operation capability and risk level of the enterprise, for use in commercial business scenarios.

2. The blockchain-based real-time risk control and decision support system for industrial digitization according to claim 1 is characterized by: The enterprise operation data includes industrial and commercial and judicial data, financial and tax declaration data, invoice data, intellectual property data and internal enterprise operation data.

3. The blockchain-based industrial digital real-time risk control and decision support system according to claim 1 is characterized by: The causal entropy chain construction module includes a dynamic graph modeling unit, an entropy value quantification unit, and a causal block generation unit; The dynamic graph modeling unit is used to model entities in enterprise operations as nodes of a dynamic graph, and to model materials and information between entities as edges of the graph; and to use a dynamic graph neural network to learn the risk propagation pattern and conditional probability between nodes and infer causal relationships; The entropy value quantification unit is used to quantify the entropy value of the causal relationship using the information entropy principle and identify risk links whose entropy value exceeds a preset threshold; The causal block generation unit is used to package the inferred causal relationship, its corresponding entropy value, and the hash value of the model parameter used for inference into a cryptographically signed causal relationship block, forming an unalterable and auditable causal relationship ledger on the blockchain.

4. The blockchain-based industrial digital real-time risk control and decision support system according to claim 1 is characterized by: The dynamic trusted QR code is a dynamic and verifiable credential that contains an access address pointing to a smart contract on the blockchain; the smart contract records an assessment result generated by the business credit assessment module that represents the company's current operating capabilities and risk level; A zero-knowledge proof that the evaluation result is calculated using an authorized model and trusted data; A reference to the causal relationship ledger and data authorization records.

5. The blockchain-based real-time risk control and decision support system for industrial digitization according to claim 4 is characterized in that: The dynamic update mechanism of the dynamic trusted QR code includes: When an immediate evaluation request is received from a third-party service agency and / or the preset update cycle is reached, the system will automatically trigger a new round of enterprise operation data collection and analysis process, refresh and update the evaluation results and corresponding zero-knowledge proof recorded in the smart contract.

Images