Miniature data transaction device and intelligent transaction system thereof

Abstract

The invention relates to the field of edge computing and Internet of Things, and discloses a miniature data transaction device which comprises an edge computing unit used for executing a centralized or decentralized matchmaking algorithm and a transaction protocol; the multi-dimensional detection module comprises a compliance detection unit, a safety detection unit, a quality evaluation unit and a scene applicability evaluation unit; the hardware accelerator is used for accelerating a detection task and encryption operation; the communication module supports multi-protocol data transmission; the storage unit is used for caching transaction data and intelligent contract codes; the power management unit is used for controlling the total power consumption of the equipment, and the edge computing unit comprises a hierarchical trusted execution environment architecture consisting of a safe enclave and a general computing unit. According to the method, a hardware isolation architecture of the secure enclave and the general computing unit is adopted, and the dynamic key encryption shared memory technology is combined, so that the effects of privacy computing and efficient matching parallel execution are achieved.

Description

Technical Field

[0001] The present invention relates to the technical fields of edge computing and the Internet of Things, and specifically provides a micro data exchanger and its intelligent trading system. Background Art

[0002] A micro data exchanger is a dedicated device deployed in edge computing nodes or Internet of Things terminals, used to achieve decentralized trading and intelligent matching of data resources in a distributed environment. Its intelligent trading system supports data ownership verification, privacy-preserving transactions, and multi-chain contract collaborative execution by integrating modules such as secure computing, dynamic pricing, and cross-chain communication, and is applicable to scenarios such as energy trading and medical data sharing. The core features of this system lie in the design of a miniaturized hardware architecture (such as RISC-V SoC integration) and low-power optimization, enabling high-throughput transaction processing in resource-constrained environments.

[0003] Existing data trading technologies are mostly based on cloud computing platforms or general blockchain nodes, using a centralized matching engine and fixed security policies. For example: Trusted Execution Environment (TEE) technology relies on a single enclave (such as Intel SGX) to achieve data isolation, and software encryption is used to ensure transaction privacy; data pricing adopts a static double auction model, and the transaction price is determined by the historical average price or a fixed premium coefficient; the hardware platform uses a general-purpose processor (such as an x86 CPU) to run full-node software, and software-defined policies are used to manage power consumption.

[0004] The core problem of the existing technology is the mismatch between the trusted execution environment architecture and the computing resources of micro devices. Traditional solutions adopt a single TEE design (such as SGX), which requires exclusive computing resources to execute encryption operations, resulting in competition for hardware resources between transaction matching and privacy computing tasks. For example, zero-knowledge proof verification and transaction matching need to be serially executed within the enclave, causing the transaction delay to exceed 100 ms, which cannot meet the millisecond-level response requirements of scenarios such as smart grids. The root cause of the contradiction is that the hardware isolation mechanism is not optimized for the computing power characteristics of micro devices, resulting in mutual exclusion between security and efficiency. Summary of the Invention

[0005] In view of the deficiencies of the existing technology, the present invention provides a micro data exchanger and its intelligent trading system, which solves the problem that the single trusted execution environment architecture of the existing technology cannot adapt to the resource characteristics of micro devices, resulting in difficulty in achieving both security isolation and computing efficiency.

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: A micro data exchanger, comprising:

[0007] An edge computing unit, used to execute centralized or decentralized matching algorithms and transaction protocols;

[0008] The multi-dimensional detection module includes a compliance detection unit, a security detection unit, a quality assessment unit, and a scenario applicability assessment unit;

[0009] The hardware accelerator is used to accelerate detection tasks and encryption operations;

[0010] The communication module supports multi-protocol data transmission;

[0011] The storage unit is used to cache transaction data and smart contract codes;

[0012] The power management unit controls the total power consumption of the device.

[0013] Preferably, the edge computing unit includes:

[0014] The hierarchical trusted execution environment architecture is composed of a secure enclave and a general computing unit;

[0015] The secure enclave runs independently based on the RISC-V instruction set and is used to execute zero-knowledge proof verification, post-quantum key negotiation, and data privacy computing tasks;

[0016] The general computing unit runs a decentralized matching algorithm and the Gossip protocol, and interacts with the secure enclave for sensitive data through a memory isolation wall;

[0017] The decentralized matching algorithm module integrates an improved double auction model and a dynamic pricing engine, with a response time ≤ 500 ms;

[0018] The cross-chain smart contract conversion middleware converts the source chain contract logic into target chain executable code through a semantic adaptation layer;

[0019] The hardware acceleration interface is directly connected to the hardware accelerator through the PCIe4.0 bus to offload compliance detection, hash calculation, and encryption tasks.

[0020] Preferably, the dynamic pricing engine calculates the equilibrium price based on the improved double auction model, and its formula is:

[0021]

[0022] where α is the supply and demand elasticity coefficient, is the buyer's quote, is the seller's asking price;

[0023] The distributed broadcast unit uses the Gossip protocol to broadcast transaction demands in the P2P network, with a coverage radius ≥ 1 km, and the node selection strategy is:

[0024] N eighbor(u) ={v|distance(u,v)≤R∧deg(v)≤deg max};

[0025] where deg(v) is the number of neighbors of node v, and deg max is the maximum connection number threshold.

[0026] Preferably, the decentralized matching algorithm module further includes:

[0027] A cross-chain smart contract conversion middleware that realizes contract logic mapping through the following steps:

[0028] Parse the syntax structure of the source chain contract to generate an intermediate language;

[0029] Reconstruct the IL into equivalent contract logic according to the syntax rules of the target chain;

[0030] Optimize the contract execution order based on the Gas cost model of the target chain to minimize transaction fees;

[0031] A real-time status feedback unit that integrates a Q-learning reinforcement learning model to dynamically adjust the cross-chain routing strategy. Its scoring formula is:

[0032] Score = w1·T h + w2·(1 - C)+ w3·(1 - L);

[0033] where T h , C, and L are the throughput, transaction fee, and latency of the target chain respectively, and the weights w1, w2, and w3 are dynamically updated according to the transaction success rate.

[0034] Preferably, the multi-dimensional detection module includes:

[0035] A compliance detection unit for verifying whether a data transaction complies with the data privacy regulations and data sovereignty requirements of the target region. It includes:

[0036] A dynamic legal knowledge graph that integrates version tracking and conflict detection of data regulations in more than 50 countries and generates a compliance report through semantic matching;

[0037] A cross-language clause alignment module that realizes real-time mapping of multilingual legal clauses based on the LaBSE multilingual embedding model and the FAISS index library.

[0038] Preferably, the security detection unit further includes:

[0039] Integrate the national cryptographic algorithm acceleration module and the post-quantum encryption module, including:

[0040] The SM2 / SM4 national cryptographic algorithm hardware accelerator is used for data signature and encrypted transmission;

[0041] CRYSTALS-Kyber post-quantum key agreement module, achieving quantum attack protection for NIST PQC standard through ASIC chips.

[0042] Preferably, the quality assessment unit verifies data quality through the following steps:

[0043] Hash check: Calculate the BLAKE3 hash value of the data stream and compare it with the hash value recorded on the evidence storage platform;

[0044] Timestamp deviation detection: Verify that the timestamp of data generation deviates from the standard time of the time service center by ≤ 0.5 seconds.

[0045] Preferably, the scenario applicability assessment unit uses a natural language processing model to parse metadata tags, including:

[0046] BERT semantic matching engine, generating a vector representation of the data description text and calculating the cosine similarity with the buyer's scenario tags;

[0047] Authorization chain verification sub-module, querying the evidence storage of the buyer's data usage rights for the target scenario through the blockchain.

[0048] Preferably, the hardware accelerator includes:

[0049] Scenario assessment accelerator, integrating a low-precision quantization AI inference engine, supporting parallel computing in INT8 data format, and used for real-time processing of semantic matching tasks of data scenario tags;

[0050] Security compliance accelerator, designed based on ASIC chips, and supporting the following functions simultaneously:

[0051] Hardware-level acceleration of national cryptographic algorithms, including key generation, data signature, and encryption;

[0052] Key agreement and encapsulation of NIST post-quantum cryptographic algorithm CRYSTALS-Kyber;

[0053] Parallel computing of BLAKE3 hash algorithm, optimizing throughput through SIMD instruction set;

[0054] Hardware interface, directly connected to the edge computing unit through the PCIe4.0 bus, and implementing the following coordination mechanism:

[0055] Offloading the hash calculation and signature tasks in compliance detection to the security compliance accelerator;

[0056] Offloading the NLP inference task of scenario applicability assessment to the scenario assessment accelerator;

[0057] Dynamic power management unit, adopting voltage-frequency regulation technology, and dynamically switching the following modes according to the task load:

[0058] High-power consumption mode: Enable all acceleration modules, with a voltage of 1.2V and a frequency of 200MHz;

[0059] Low-power consumption mode: Only enable the SM4 encryption module, with a voltage of 0.8V and a frequency of 50MHz.

[0060] The present invention also provides an intelligent trading system for a micro data exchanger, including a plurality of micro data exchangers, and the following collaborative components:

[0061] National computing power network nodes, used for storing and verifying distributed transaction data;

[0062] Cross-chain routing center, dynamically selecting the optimal blockchain path through a reinforcement learning model;

[0063] Edge-cloud collaborative management platform, monitoring device status and dynamically updating the compliance rule library.

[0064] The present invention provides a micro data exchanger and its intelligent trading system. It has the following beneficial effects:

[0065] 1. The present invention adopts a hardware isolation architecture of a secure enclave and a general computing unit, combined with a dynamic key encryption shared memory technology, achieving the effect of parallel execution of privacy computing and efficient matching. Compared with the performance bottleneck caused by a single trusted execution environment in the prior art, it solves the problem that it cannot balance security isolation and high-throughput data processing, and is especially suitable for the resource-constrained scenario of a micro data exchanger.

[0066] 2. The present invention integrates four-dimensional detection modules of compliance, security, quality, and scenario applicability, combined with an improved double auction dynamic pricing formula, realizing full-dimensional trusted verification and real-time price adaptation of transaction data. Compared with the isolated detection or static pricing model in the traditional scheme, it solves the problems of single evaluation dimension and lagging market response, and meets the rapid decision-making needs of micro devices for complex transaction scenarios.

[0067] 3. The present invention significantly improves the data label matching, encryption operation, and cross-chain synchronization efficiency through a scenario evaluation accelerator, a security compliance accelerator, and a multi-link aggregation transmission protocol. Compared with the excessive latency caused by the existing software scheme relying on a general-purpose processor, it solves the defect that it is difficult to support high-concurrency real-time transactions of micro devices, and ensures the balance of transaction throughput and energy efficiency.

[0068] 4. The present invention realizes extended device battery life and reliable cross-chain data synchronization through DVFS load prediction, multi-power switching, and CRDT consistency protocol. Compared with the fixed power consumption strategy or single-chain storage architecture in the traditional scheme, it solves the pain points of uncontrollable energy consumption and frequent cross-chain data conflicts, and ensures the long-term stable operation of the micro data exchanger in a distributed environment. Brief Description of the Drawings

[0069] Figure 1 is the main framework diagram of the present invention;

[0070] Figure 2 is the flowchart for generating the dynamic data fingerprint of the present invention. Detailed Description of the Preferred Embodiments

[0071] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings 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.

[0072] Please refer to the attached Figure 1-2 , the embodiments of the present invention provide a micro data exchanger, including:

[0073] An edge computing unit for executing a centralized or decentralized matching algorithm and a trading protocol;

[0074] A multi-dimensional detection module, including a compliance detection unit, a security detection unit, a quality evaluation unit, and a scenario applicability evaluation unit;

[0075] A hardware accelerator for accelerating detection tasks and cryptographic operations;

[0076] A communication module that supports multi-protocol data transmission;

[0077] A storage unit for caching trading data and smart contract codes;

[0078] A power management unit for controlling the total power consumption of the device.

[0079] The edge computing unit includes:

[0080] A hierarchical trusted execution environment architecture, composed of a secure enclave and a general computing unit;

[0081] The secure enclave runs independently based on the RISC-V instruction set and is used to execute zero-knowledge proof verification, post-quantum key negotiation, and data privacy calculation tasks;

[0082] The general computing unit runs a decentralized matching algorithm and the Gossip protocol, and interacts with the secure enclave for sensitive data through a memory isolation wall;

[0083] A decentralized matching algorithm module, integrating an improved double auction model and a dynamic pricing engine, with a response time ≤ 500 ms;

[0084] Cross-chain smart contract conversion middleware that converts the source chain contract logic into executable code for the target chain through a semantic adaptation layer;

[0085] Hardware acceleration interface that is directly connected to the hardware accelerator through the PCIe4.0 bus to offload compliance detection, hash calculation, and encryption tasks.

[0086] Dynamic pricing engine that calculates the equilibrium price based on an improved double auction model, and its formula is:

[0087]

[0088] where α is the supply and demand elasticity coefficient, is the buyer's quote, is the seller's asking price;

[0089] Distributed broadcast unit that broadcasts transaction demands in the P2P network using the Gossip protocol, with a coverage radius ≥ 1 km, and the node selection strategy is:

[0090] N eighbor(u) ={v|distance(u,v)≤R∧deg(v)≤deg max};

[0091] where deg(v) is the number of neighbors of node v, and deg max is the maximum connection number threshold.

[0092] The decentralized matching algorithm module also includes:

[0093] Cross-chain smart contract conversion middleware that realizes contract logic mapping through the following steps:

[0094] Parse the syntax structure of the source chain contract to generate intermediate language;

[0095] Reconstruct the IL into equivalent contract logic according to the syntax rules of the target chain;

[0096] Optimize the contract execution order based on the Gas cost model of the target chain to minimize transaction fees;

[0097] Real-time status feedback unit that integrates the Q-learning reinforcement learning model to dynamically adjust the cross-chain routing strategy, and its scoring formula is:

[0098] Score=w1·T h +w2·(1-C)+w3·(1-L);

[0099] where T h , C, L are the throughput, transaction fees, and latency of the target chain respectively, and the weights w1, w2, w3 are dynamically updated according to the transaction success rate.

[0100] The multi-dimensional detection module includes:

[0101] A compliance detection unit for verifying whether a data transaction complies with the data privacy regulations and data sovereignty requirements of the target region, including:

[0102] A dynamic legal knowledge graph that integrates version tracking and conflict detection of data regulations in more than 50 countries, and generates a compliance report through semantic matching;

[0103] A cross-language clause alignment module that realizes real-time mapping of multi-language legal clauses based on the LaBSE multi-language embedding model and the FAISS index library.

[0104] The security detection unit also includes:

[0105] An integrated national cryptographic algorithm acceleration module and a post-quantum encryption module, including:

[0106] An SM2 / SM4 national cryptographic algorithm hardware accelerator for data signature and encrypted transmission;

[0107] A CRYSTALS-Kyber post-quantum key agreement module that realizes quantum attack protection of the NISTPQC standard through an ASIC chip.

[0108] The quality assessment unit verifies the data quality through the following steps:

[0109] Hash check: Calculate the BLAKE3 hash value of the data stream and compare it with the hash value recorded in the evidence storage platform;

[0110] Timestamp deviation detection: Verify that the deviation between the data generation timestamp and the standard time of the time service center is ≤ 0.5 seconds.

[0111] The scenario applicability evaluation unit uses a natural language processing model to parse metadata tags, including:

[0112] A BERT semantic matching engine that generates a vector representation of the data description text and calculates the cosine similarity with the buyer's scenario tags;

[0113] An authorization chain verification sub-module that queries the evidence storage of the buyer's data usage rights for the target scenario through the blockchain.

[0114] The hardware accelerator includes:

[0115] A scenario evaluation accelerator that integrates a low-precision quantization AI inference engine, supports parallel computing in INT8 data format, and is used for real-time processing of semantic matching tasks of data scenario tags;

[0116] A security compliance accelerator designed based on an ASIC chip, supporting the following functions simultaneously:

[0117] Hardware-level acceleration of national cryptographic algorithms, including key generation, data signature, and encryption;

[0118] Key agreement and encapsulation of the NIST post-quantum cryptographic algorithm CRYSTALS-Kyber;

[0119] Parallel computing of the BLAKE3 hash algorithm to optimize throughput through the SIMD instruction set;

[0120] Hardware interface, directly connected to the edge computing unit through the PCIe4.0 bus to implement the following cooperation mechanism:

[0121] Offload the hash calculation and signature tasks in compliance detection to the security compliance accelerator;

[0122] Offload the NLP inference task of scenario applicability assessment to the scenario assessment accelerator;

[0123] Dynamic power management unit, using voltage-frequency regulation technology to dynamically switch the following modes according to the task load:

[0124] High-power mode: Enable all acceleration modules, voltage 1.2V, frequency 200MHz;

[0125] Low-power mode: Only enable the SM4 encryption module, voltage 0.8V, frequency 50MHz.

[0126] In this embodiment, the edge computing unit adopts a hierarchical trusted execution environment architecture to achieve secure isolation and efficient computing of data transactions. Specifically, the edge computing unit includes a secure enclave and a general computing unit, and the two implement parallel processing of sensitive tasks and general tasks through a hardware-level isolation mechanism, ensuring that privacy computing tasks are executed in a protected environment while general computing tasks maintain high throughput.

[0127] As an option, the secure enclave is implemented based on the reduced instruction set architecture. Exemplarily, the secure enclave uses the RISC-V instruction set to build an independent operating environment, configured with a dedicated secure storage area and a cryptographic coprocessor. Among them, the secure storage area is used to store key materials and intermediate results of privacy computing, and the cryptographic coprocessor is used to execute tasks with high security requirements such as zero-knowledge proof verification and post-quantum key agreement. It should be noted that the secure enclave divides access permissions through the memory management unit. If an unauthorized process attempts to access the memory area of the secure enclave, a hardware exception interrupt will be triggered and sensitive data will be cleared, thus avoiding the risk of side-channel attacks.

[0128] In a possible implementation, the general computing unit runs a lightweight operating system kernel for processing decentralized matching algorithms, transaction broadcast protocols, and smart contract parsing tasks. It can be understood that the data interaction between the general computing unit and the secure enclave is achieved through a shared memory area. Preferably, the shared memory area adopts a dynamic encryption policy. When data is written, the secure enclave generates a temporary session key for encryption, and when data is read, it is decrypted after passing the permission verification of the secure enclave.

[0129] Specifically, the decentralized matching algorithm module integrates an improved double auction model and a dynamic pricing engine. The dynamic pricing engine calculates the equilibrium price based on the real-time supply and demand relationship, and its formula is defined as:

[0130]

[0131] Where: is the bid price of the jth buyer, representing the highest bid price of the buyer for the target data or service;

[0132] is the asking price of the kth seller, representing the lowest acceptable price for which the seller provides data or services;

[0133] D supply is the current total market supply, which is obtained by accumulating the total amount of data to be traded of all sellers;

[0134] D demand is the current total market demand, which is obtained by accumulating the total demand of all buyers;

[0135] α is the supply and demand elasticity coefficient, which is used to adjust the sensitivity of price to supply and demand imbalance. The value range is 0 < α ≤ 1. Preferably, it is dynamically adjusted according to the sliding window analysis of historical transaction data.

[0136] Formula operation process: Quote aggregation: Collect the set of buyer bids and the set of seller asking prices and sort them by price to generate the supply and demand curve;

[0137] Equilibrium calculation: Calculate the weighted average price base value and adjust the impact of supply and demand imbalance on price through the normalization factor max(D supply ,D demand );

[0138] Transaction matching: At the equilibrium price P match , cross-match the highest buyer bid with the lowest seller asking price to generate the final transaction pairs.

[0139] Normalization factor: Eliminate the dimension difference between supply and demand through max(D supply ,D demand ) to make the price fluctuation range proportional to the market imbalance degree;

[0140] Dynamic update of elasticity coefficient: When the market supply and demand fluctuate frequently, increase α to enhance price sensitivity; when the market is stable, decrease α to maintain price stability.

[0141] The distributed broadcast unit uses the Gossip protocol to broadcast transaction demands in the P2P network, and the node selection strategy is defined as:

[0142] N eighbor(u) ={v|distance(u,v)≤R∧deg(v)≤deg max};

[0143] distance(u,v): The physical distance between node u and node v, estimated by GPS coordinates or network latency;

[0144] R: The communication coverage radius. Exemplarily, it is set to 1km to meet the regional transaction requirements;

[0145] deg(v): The current number of connections of node v, indicating the number of its neighbor nodes;

[0146] deg max : The maximum connection number threshold, used to avoid network congestion.

[0147] Implementation process:

[0148] Neighbor screening: Only select nodes with distance ≤ R and connection number ≤ deg max as the broadcast target;

[0149] Load balancing: When the connection number of node v exceeds deg max x, automatically stop receiving new requests;

[0150] Priority broadcast: High-priority transaction requests (such as real-time energy data) are preferentially diffused through the Gossip protocol.

[0151] Physical distance limit: Limit the communication radius R to ensure that transaction demands are diffused within a reasonable geographical range, applicable to the regional energy trading scenario;

[0152] Connection number threshold: Limit the load of a single node through deg max to prevent network topology overload

[0153] The middleware realizes the compatibility conversion of multi-chain contract logic through the semantic adaptation layer, specifically including the following steps:

[0154] Syntax parsing: Parse the abstract syntax tree (AST) of the source chain smart contract, and extract function logic, state variables, and event listeners;

[0155] Intermediate language generation: Convert the parsing result into a chain-independent intermediate representation (IR), preferably in the LLVM IR format;

[0156] Target chain reconstruction: Map the IR to equivalent contract code according to the characteristics of the target chain virtual machine instruction set (such as EVM bytecode, WASM instruction set);

[0157] Gas optimization: Based on the Gas cost model of the target chain, adjust the contract execution order through static code analysis, for example, split loops into parallel batches to reduce Gas consumption.

[0158] Intermediate language layer: Achieve an abstract expression of multi-chain contract logic through IR, and solve the interaction barriers of heterogeneous chains such as Ethereum and Fabric;

[0159] Gas cost optimization: Optimize the contract code structure to adapt to the economic model of the target chain and reduce the user's transaction cost.

[0160] As an option, the compliance detection unit integrates a dynamic legal knowledge graph and a cross-language clause alignment module. Specifically, the dynamic legal knowledge graph extracts key entities of data privacy regulations (such as "personal data", "cross-border transmission restrictions") by crawling the official websites of multinational legislative bodies and regulatory documents, and constructs a graph structure with legal articles, regions, and data types as nodes.

[0161] It should be noted that the graph realizes the correlation analysis between clauses and transaction data through a semantic matching engine. Exemplarily, when it is detected that the data recipient is located in the European Union, the clause regarding the consent of the data subject in the GDPR is automatically matched, and a compliance report is generated.

[0162] In a possible implementation, the cross-language clause alignment module encodes multi-lingual legal clauses into a unified vector space based on the LaBSE (Language-agnostic BERT Sentence Embedding) model.

[0163] Preferably, the FAISS (Facebook AI Similarity Search) index library is used to achieve real-time semantic similarity retrieval. It can be understood that this module solves the problem of aligning Chinese and English legal articles, for example, semantically mapping Article 35 of the Data Security Law of the People's Republic of China to GDPR Article 44 to ensure that cross-border transactions comply with the requirements of both jurisdictions.

[0164] The security detection unit integrates a national cryptographic algorithm acceleration module and a post-quantum encryption module. Specifically, the national cryptographic algorithm acceleration module supports SM2 digital signature and SM4 symmetric encryption algorithms. Exemplarily, the SM2 signature private key is generated within a secure enclave, and the signature process is implemented through a hardware accelerator. Preferably, the signature speed can meet the requirements of high-concurrency transactions; SM4 encryption uses the CBC mode to perform block encryption on the data stream to ensure the confidentiality of data during transmission.

[0165] As an option, the post-quantum encryption module implements quantum-resistant key agreement based on the CRYSTALS-Kyber algorithm. It should be noted that the module accelerates polynomial multiplication and modular reduction operations through an ASIC chip. Preferably, the key encapsulation and decapsulation latency are controlled within milliseconds to meet the requirements of real-time transaction scenarios.

[0166] The quality assessment unit verifies the integrity and timeliness of data through hash verification and timestamp deviation detection. Specifically, the hash verification includes the following steps:

[0167] Data shard hash calculation: Shard the input data stream and calculate the BLAKE3 hash value of each shard;

[0168] Evidence comparison: Aggregate the shard hash values into a Merkle tree structure and compare it with the root hash value recorded on the blockchain evidence platform to ensure that the data has not been tampered with.

[0169] It can be understood that the timestamp deviation detection module synchronizes with the national time service center through the NTP protocol to verify whether the deviation between the data generation timestamp and the standard time is within the allowable range. Exemplarily, when the detected deviation exceeds the threshold, the data is automatically marked as invalid and an alarm is triggered.

[0170] The scenario applicability assessment unit uses a natural language processing model to parse metadata tags and match them with the buyer's scenario requirements. Specifically:

[0171] BERT semantic matching engine: Encode the data description text (such as "Lung cancer CT image dataset") and the buyer's scenario label (such as "AI drug research and development") into 768-dimensional vectors and calculate the cosine similarity;

[0172] Authorization chain verification: Query the blockchain for the evidence of the buyer's data usage rights for the target scenario. Exemplarily, verify whether a medical institution has an authorization certificate for the "cancer research" scenario.

[0173] In a possible implementation, the semantic matching engine integrates low-precision quantization technology to convert the model weights from FP32 to INT8 format. Preferably, the inference speed is improved through a hardware accelerator to meet the real-time requirements.

[0174] As an option, the scenario evaluation accelerator integrates a low-precision quantization AI inference engine for real-time processing of semantic matching tasks of data scenario tags. Specifically, the accelerator supports parallel computing in the INT8 data format, significantly reducing computational resource consumption by quantizing the weights of pre-trained models (such as BERT) from FP32 to INT8. It can be understood that the quantization process adopts dynamic range calibration technology. Preferably, by calibrating the activation value distribution of the calibration dataset, the quantization parameters are dynamically adjusted to minimize precision loss.

[0175] Exemplarily, in the medical data trading scenario, the scenario evaluation accelerator parses the metadata tags of CT image data (such as "lung cancer", "high resolution"), calculates its semantic similarity with the buyer's scenario requirements (such as "AI drug research and development") through the INT8 quantization model, and outputs a matching score to assist in trading decisions.

[0176] The security and compliance accelerator is based on ASIC chip design and supports hardware-level acceleration of national cryptographic algorithms, post-quantum cryptographic algorithms, and hash algorithms. Specifically:

[0177] National cryptographic algorithm acceleration module:

[0178] SM2 signature: Key generation and signature verification are implemented through a dedicated elliptic curve operation circuit. Preferably, the private key generation is completed within a secure enclave, and the signature process is parallel processed through a hardware pipeline;

[0179] SM4 encryption: The data stream is encrypted in blocks using the CBC (Cipher Block Chaining) mode, and high-throughput encrypted transmission is achieved through a Substitution-Permutation Network (SPN) structure hardware circuit.

[0180] Post-quantum cryptography module:

[0181] Key agreement and encapsulation against quantum attacks are implemented based on the CRYSTALS-Kyber algorithm. It should be noted that the module integrates a polynomial multiplier and a modular reduction circuit through an ASIC chip. Preferably, the key encapsulation process includes the following steps:

[0182] Generate the Kyber parameter set (such as Kyber-768) to construct a lattice-based difficult problem;

[0183] Accelerate polynomial multiplication operations through NTT (Fast Number Theory Transform);

[0184] Encapsulate the shared key and bind it to the recipient's public key to generate a ciphertext resistant to quantum attacks.

[0185] The BLAKE3 algorithm is adopted to implement data integrity verification, and the parallelism of hash calculation is optimized through the SIMD (Single Instruction Multiple Data) instruction set. Exemplarily, the module divides the input data stream into 512-bit data blocks and processes multiple data blocks in parallel through a multi-level pipeline, significantly improving the hash throughput.

[0186] The hardware accelerator is directly connected to the edge computing unit through the PCIe4.0 bus to achieve low-latency task offloading. Specifically:

[0187] Task offloading queue: The edge computing unit encapsulates tasks such as hash calculation and signature verification in the compliance detection into instruction packets and submits them to the task queue of the accelerator;

[0188] Interrupt notification mechanism: After the accelerator completes task processing, it sends an interrupt signal to the edge computing unit through MSI-X (Message Signaled Interrupt Extension) to trigger the result return;

[0189] Data consistency protocol: The DMA (Direct Memory Access) technology is adopted to achieve zero-copy data transfer between the accelerator and the main memory, avoiding performance loss caused by CPU intervention.

[0190] The dynamic power management unit dynamically adjusts the voltage and frequency of the accelerator according to the task load. Specifically:

[0191] High-power consumption mode: When the number of concurrent tasks ≥ 3 or hash calculation requests are intensive, all acceleration modules are activated, the voltage rises to 1.2V, and the frequency is adjusted to 200MHz;

[0192] Low-power consumption mode: When the device idle time exceeds the threshold or the power is lower than the preset value, only the SM4 encryption module is enabled, the voltage drops to 0.8V, the frequency is adjusted to 50MHz, and the rest of the modules enter the sleep state.

[0193] It can be understood that power management dynamically switches the working mode by real-time monitoring the task queue depth and the system energy consumption status to balance performance and energy efficiency.

[0194] As an option, the dynamic protocol switching engine dynamically selects the transmission protocol according to the network status, device power, and data priority. Specifically, the engine triggers protocol switching through the following parameters:

[0195] Network signal strength: When the received signal strength indication (RSSI) is lower than the threshold, switch to a high-penetration protocol (such as LoRa);

[0196] Data urgency: High-priority transaction data (such as real-time energy trading instructions) is switched to a low-latency protocol (such as 5GNR);

[0197] Remaining battery power of the device: When the power is lower than the preset value, switch to a low-power protocol (such as NB-IoT).

[0198] It can be understood that the switching logic is implemented through a state machine model. Exemplarily, the state transition function is defined as:

[0199] Protocol next =f(RSSI,Power remain ,Priority data );

[0200] where Protocol next is the next state protocol type, and f is the switching function based on the weighted decision tree.

[0201] The adaptive coding and modulation module dynamically adjusts the coding rate and modulation method according to the channel quality. Specifically, the module optimizes the transmission efficiency through the following steps:

[0202] Channel state estimation: The receiving end periodically feeds back the channel quality indication (CQI), including the signal-to-noise ratio (SNR) and the bit error rate (BER);

[0203] Modulation and coding scheme (MCS) selection: Based on the CQI, look up the table to select the optimal MCS level, and its mapping relationship is:

[0204]

[0205] where γ min is the lowest SNR threshold, and Δγ is the SNR interval;

[0206] Dynamic adjustment: If the number of consecutive transmission failures exceeds the threshold, lower the MCS level to improve robustness.

[0207] It should be noted that the module supports two coding methods, LDPC (low-density parity-check code) and Polar code. Preferably, the coding scheme is selected according to the protocol type (for example, 5GNR preferably uses Polar code).

[0208] The communication module integrates the Q-learning algorithm to optimize the multi-hop transmission path. Specifically:

[0209] State space: Defined as {remaining battery power of the node, link delay, node load};

[0210] Action space: Select the next-hop node or switch the transmission protocol;

[0211] Reward function:

[0212]

[0213] Among them, w1, w2, and w3 are weight coefficients, Delay is the end-to-end delay, Power saving is the energy-saving gain, and PacketLoss is the packet loss rate.

[0214] Exemplarily, in the energy data trading scenario, the Q-learning model selects a low-latency and high-reliability transmission path through the exploration-exploitation strategy while balancing the energy consumption of nodes.

[0215] The multi-link aggregation transmission module improves the throughput by binding multiple physical links. Specifically:

[0216] Link shunting: Shard the data stream and transmit it concurrently through different protocols (such as 5GNR and LoRa dual channels);

[0217] Load balancing: Allocate the data shard size based on the real-time bandwidth of the link, and its allocation weight is:

[0218]

[0219] Among them, B i is the available bandwidth of the i-th link;

[0220] Recombination verification: The receiving end recombines the data shards according to the sequence number and verifies the integrity through the BLAKE3 hash.

[0221] The low-power wake-up mechanism reduces the standby power consumption by combining periodic listening and event triggering. Specifically:

[0222] Periodic listening: The device wakes up every fixed time window (such as 60 seconds) to receive beacon frames and detects whether there is data to be received;

[0223] Event triggering: When high-priority data arrives, the sending end wakes up the receiving device through a long preamble. Preferably, the length of the preamble is positively correlated with the transmission distance.

[0224] As an option, the communication module uniquely identifies and verifies the integrity of the data packet through a dynamic data fingerprint generation algorithm before data transmission. Specifically, the algorithm process includes the following steps:

[0225] Feature extraction: Generate a feature vector based on the data content, and its dimension compression formula is:

[0226]

[0227] Among them, D raw is the original data, LSH is the locality-sensitive hashing function, k is the number of hash bits (preferably, k = 128), represents the exclusive OR operation;

[0228] Dynamic fingerprint generation: Based on the current timestamp t and the device unique identifier ID dev Generate dynamic fingerprint:

[0229] F fingerprint = SM3(V feature ||t||ID dev );

[0230] Among them, SM3 is the national cryptographic hash algorithm, and || represents data concatenation;

[0231] Fingerprint embedding and verification: Embed F fingerprint into the data packet header, and the receiving end verifies the data integrity and source legality through the same algorithm.

[0232] It should be noted that the algorithm adapts to the computing power of micro-devices through lightweight design, and the single fingerprint generation time is less than 5ms (tested based on the Cortex-M7 kernel).

[0233] In this embodiment, the storage unit is used to cache transaction data and smart contract code, and realizes efficient access and secure persistence through a hierarchical storage architecture and a dynamic data management strategy. Specifically, the storage unit includes a transaction buffer, a contract code library, and a distributed storage interface, supporting multi-mode data reading and writing and cross-chain data indexing.

[0234] As an option, the storage unit adopts the LRU-K (Least Recently Used - K times) algorithm to manage the transaction buffer. Specifically, the cache eviction policy dynamically adjusts the priority based on the access frequency and temporal locality of data blocks, and its weight calculation formula is defined as:

[0235]

[0236] Among them, f i is the access frequency of data block i, t last is the most recent access time, w1 and w2 are weight coefficients, and T max is the maximum storage duration threshold. It should be noted that when the cache space is insufficient, the data block with the lowest score is preferentially evicted.

[0237] Exemplarily, in the energy trading scenario, the real-time electricity price data with high-frequency access is retained in the buffer, and the low-frequency historical data is transferred to the distributed storage node.

[0238] The storage unit isolates sensitive data and non-sensitive data through hardware-level security partitioning. Specifically:

[0239] Sensitive data area: Stores key materials, user privacy data, etc., and adopts a dynamic key encryption strategy. Preferably, the encryption key is generated by a secure enclave and encrypted and stored through the SM4 algorithm;

[0240] Non-sensitive data area: Stores transaction logs, contract codes, etc., and uses Transparent Data Encryption (TDE) technology. The encryption key is managed by a Hardware Security Module (HSM).

[0241] It can be understood that the dynamic key generation process includes:

[0242] Securely generate a temporary session key K temp ;

[0243] Use K temp After encrypting the data, encapsulate the ciphertext and the key into a secure object;

[0244] Write the secure object into the storage unit and record the integrity check value through a hash chain.

[0245] The storage unit realizes data persistence and cross-node synchronization through a distributed storage protocol (such as IPFS, Arweave). Specifically:

[0246] Data sharding and redundant coding: Shard the transaction data into data blocks of a fixed size, and generate redundant check blocks through Erasure Coding. Preferably, set the ratio of data blocks to check blocks as k:n;

[0247] Cross-chain indexing service: Construct a B+ tree index structure with data hashes as the primary key to support fast retrieval of cross-chain transaction records. Exemplarily, the index metadata is stored on a lightweight blockchain (such as Merkle Patricia Trie) to ensure immutability.

[0248] It should be noted that the redundant coding parameters k and n are dynamically adjusted according to the availability of storage nodes and network latency. When the online rate of nodes is lower than the threshold, the proportion of check blocks is automatically increased to improve fault tolerance.

[0249] The storage unit optimizes storage resources through a hot and cold data classification strategy. Specifically:

[0250] Hot data: Frequently accessed transaction status data (such as the unconfirmed transaction queue) is retained in the NVMe SSD to support low-latency reading and writing;

[0251] Warm data: Periodically accessed contract codes and historical transaction logs are transferred to the SATA SSD;

[0252] Cold data: Infrequently accessed archived data (such as transaction records that have not been accessed for more than 6 months) is migrated to distributed storage nodes, and a compression algorithm (such as Zstandard) is triggered to reduce storage costs.

[0253] In a possible implementation, the data migration strategy is based on an access pattern prediction model, and its input features include data age, access frequency, and associated transaction scenarios. Exemplarily, an LSTM (Long Short-Term Memory) model is used to predict the access probability in the next 7 days, and the storage hierarchy is dynamically adjusted.

[0254] The storage unit achieves cross-chain data eventual consistency through an improved CRDT (Conflict-Free Replicated Data Type) protocol. Specifically:

[0255] Operation log synchronization: Encoding data update operations as incremental logs and broadcasting them to other chain nodes through the Gossip protocol;

[0256] Conflict resolution: When a version conflict is detected, merge operation records based on timestamps and transaction priorities. Preferably, a vector clock is used to mark the operation timing, and the resolution rule is defined as:

[0257]

[0258] where Priority(O) is the priority weight of transaction operation O, which is comprehensively calculated by factors such as transaction amount and participant credit score.

[0259] The storage unit dynamically switches storage media according to real-time I / O pressure. Specifically:

[0260] High-load mode: When it is detected that the number of concurrent read / write requests exceeds the threshold, enable the collaborative processing of NVMe SSD and memory cache, and achieve zero-copy data transfer by bypassing the CPU through DMA (Direct Memory Access) technology;

[0261] Low-load mode: When the system is idle, migrate data to high-density QLC SSD to reduce power consumption and extend the hardware lifespan.

[0262] It can be understood that the switching strategy dynamically adjusts the activation state of the storage media by real-time monitoring of the I / O queue depth, latency, and error rate.

[0263] In this embodiment, the power management unit is used to control the total power consumption of the device, and achieves the balance between energy consumption and performance through dynamic voltage and frequency scaling (DVFS), task scheduling strategy optimization, and multi-power input switching technology. Specifically, the unit dynamically adjusts the power supply strategy by real-time monitoring of the system load, remaining battery power, and external power supply status to ensure the reliable operation of the device under various working conditions.

[0264] As an option, the DVFS module dynamically adjusts the voltage and frequency of the processor core based on system load prediction. Specifically, the module is implemented through the following steps:

[0265] Load factor calculation:

[0266]

[0267] Among them, Q active is the current number of active tasks, and Q max is the maximum concurrent task capacity of the processor;

[0268] Voltage-frequency mapping: Select the optimal voltage-frequency pair (V, g) according to β by looking up the table, and its mapping relationship satisfies:

[0269] f = f base ·min(1, β + δ);

[0270] Among them, f base is the reference frequency, and δ is the load margin coefficient, which is used to prevent frequency mutation;

[0271] Smooth transition: Adopt the ramp control algorithm to gradually adjust the voltage and frequency to avoid instantaneous current impact. It should be noted that the load margin coefficient δ is dynamically calibrated through historical load fluctuation data. Preferably, the exponentially weighted moving average (EWMA) algorithm is used for updating.

[0272] It should be noted that the load margin coefficient δ is dynamically calibrated through historical load fluctuation data. Preferably, the exponentially weighted moving average (EWMA) algorithm is used for updating.

[0273] The intelligent task scheduling strategy allocates computing resources according to task priorities and energy consumption costs. Specifically, the task priority weight W i is calculated by the following formula:

[0274]

[0275] Among them, P task is the task urgency (determined by the transaction type), E saving is the estimated energy-saving benefit, and α is the weight adjustment factor. It can be understood that high-priority tasks (such as real-time transaction matching) are allocated to high-computing-power cores, and low-priority tasks (such as log archiving) are allocated to low-power coordination processors.

[0276] Exemplarily, in the medical data transaction scenario, the patient privacy computing task is given the highest priority and scheduled to be executed on a high-performance core; the non-real-time data backup task is delayed until the device is idle.

[0277] The multi-power input switching module supports seamless switching between an external DC power supply, a lithium battery, and an energy harvesting unit (such as solar energy). Specifically:

[0278] Priority strategy: Preferentially use the external power supply for power supply. When the external power supply is detected to be disconnected, automatically switch to the lithium battery;

[0279] Hybrid power supply mode: When the external power supply is insufficient, the maximum power point tracking (MPPT) technology is used to coordinate the parallel power supply of the external power supply and the lithium battery. The output current distribution formula is as follows:

[0280]

[0281] Among them, P req is the total power consumption demand of the system, P ext is the available power of the external power supply, η is the conversion efficiency, and V bat is the battery voltage;

[0282] Energy harvesting optimization: When a solar panel is connected, the MPPT algorithm dynamically adjusts the load impedance to make the energy harvesting unit work at the maximum power point.

[0283] It should be noted that during the power supply switching process, a voltage ramp-up / ramp-down circuit is adopted to avoid device reset caused by power supply interruption or voltage sudden change.

[0284] The energy prediction module predicts the future energy consumption trend of the device through an LSTM (Long Short-Term Memory) model. Specifically:

[0285] Input features: historical power consumption sequence, current task queue depth, ambient temperature;

[0286] Output result: power consumption curve E pred (t) within the future T time window;

[0287] State machine switching: According to

[0288] If then enter the power-saving mode and turn off non-critical peripherals (such as the display screen, redundant communication module);

[0289] If then restore the full-function mode.

[0290] It can be understood that the prediction model combines offline training and online fine-tuning. Preferably, the federated learning framework is used to aggregate the energy consumption pattern data of multiple devices to improve the prediction accuracy.

[0291] It can be understood that the prediction model combines offline training and online fine-tuning. Preferably, the federated learning framework is used to aggregate the energy consumption pattern data of multiple devices to improve the prediction accuracy.

[0292] The power consumption monitoring module collects the current, voltage, and temperature data of each functional unit in real time and realizes abnormal protection through the following mechanism:

[0293] Overcurrent protection: If the instantaneous current value is detected to exceed the threshold I max, immediately cut off the power supply and trigger the fusing mechanism;

[0294] Thermal management: When the chip temperature T junc > T safe , dynamically reduce the frequency and start the cooling fan;

[0295] Battery health management: Estimate the battery cycle life through coulomb counting. When the health H bat <80%, limit the maximum discharge current to 70% of the rated value.

[0296] The intelligent trading system of a micro data trader described below can be cross-referenced with a micro data trader described above.

[0297] An intelligent trading system of a micro data trader includes a plurality of micro data traders and the following collaborative components:

[0298] National computing power network nodes for storing and verifying distributed trading data;

[0299] Cross-chain routing center that dynamically selects the optimal blockchain path through a reinforcement learning model;

[0300] Edge-cloud collaborative management platform that monitors device status and dynamically updates the compliance rule library

[0301] The system of this embodiment can be used to execute the above device embodiment, and its principle and technical effects are similar, so they will not be elaborated here.

[0302] Embodiment 1:

[0303] In the scenario of cross-institutional medical data sharing, the micro data trader executes the following processes:

[0304] Privacy data verification: Patient electronic medical record data is processed through a hierarchical trusted execution environment. Sensitive fields (such as name, ID number) are encrypted within a secure enclave, and non-sensitive fields (such as diagnosis conclusions) are semantically matched by a general computing unit;

[0305] Dynamic fingerprint generation: The communication module generates a dynamic fingerprint F for the encrypted medical record data fingerprint , and transmits it to the cloud medical data platform through the 5G NR protocol;

[0306] Cross-chain deposit and proof: The distributed storage interface shards and stores the transaction records in the medical alliance chain and IPFS nodes, and synchronizes the on-chain indexes of each institution through the CRDT protocol;

[0307] Compliance detection: The security compliance accelerator continuously verifies the access rights of data users. If an unauthorized institution requests decryption, immediately trigger the termination of the transaction and record the violation log.

[0308] Example 2:

[0309] In the industrial Internet of Things device status data transaction, the micro data trader operates as follows:

[0310] Multi - protocol switching: The communication module automatically switches to the LoRa protocol according to the network signal strength (RSSI = - 90dBm) in the factory and uploads the device vibration sensor data to the edge server;

[0311] Low - power optimization: The power management unit detects that the device power is below 20%, triggers the DVFS module to reduce the processor frequency to 800MHz, and shuts down non - essential peripherals (such as redundant communication links);

[0312] Dynamic pricing: The intelligent trading system adjusts the data price based on the real - time device health score (predicted by the LSTM model). The data price of devices with a health score below 70% is increased by 30%;

[0313] Conflict resolution: The distributed storage interface detects version conflicts of device status data among multiple edge nodes and merges the data records based on timestamps and device priorities (such as key devices on the production line having priority).

[0314] 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 micro data trader, characterized in that, Including: An edge computing unit for executing centralized or decentralized matching algorithms and trading protocols; A multi-dimensional detection module including a compliance detection unit, a security detection unit, a quality assessment unit, and a scenario applicability assessment unit; A hardware accelerator for accelerating detection tasks and cryptographic operations; A communication module supporting multi-protocol data transmission; A storage unit for caching transaction data and smart contract codes; A power management unit for controlling the total power consumption of the device.

2. A micro data exchanger according to claim 1, characterized in that, The edge computing unit includes: A hierarchical trusted execution environment architecture composed of a secure enclave and a general computing unit; The secure enclave runs independently based on the RISC-V instruction set and is used to execute zero-knowledge proof verification, post-quantum key negotiation, and data privacy computing tasks; The general computing unit runs a decentralized matching algorithm and the Gossip protocol, and interacts with the secure enclave for sensitive data through a memory isolation wall; A decentralized matching algorithm module integrating an improved double auction model and a dynamic pricing engine, with a response time ≤ 500 ms; A cross-chain smart contract conversion middleware that converts the source chain contract logic into executable code for the target chain through a semantic adaptation layer; A hardware acceleration interface directly connected to the hardware accelerator through a PCIe4.0 bus to offload compliance detection, hash calculation, and encryption tasks.

3. A micro data exchanger according to claim 1, characterized in that, The decentralized matching algorithm module includes: The dynamic pricing engine calculates the equilibrium price based on the improved double auction model, and its formula is: where α is the supply-demand elasticity coefficient, is the buyer's quoted price, is the seller's asking price; A distributed broadcast unit that uses the Gossip protocol to broadcast transaction demands in the P2P network, with a coverage radius ≥ 1 km, and the node selection strategy is: N eighbor(u) = {v | distance(u, v) ≤ R ∧ deg(v) ≤ deg max}; where deg(v) is the number of neighbors of node v, and deg max is the maximum connection number threshold.

4. A micro data exchanger according to claim 1, characterized in that, The decentralized matching algorithm module further includes: A cross-chain smart contract conversion middleware that realizes contract logic mapping through the following steps: Parse the syntax structure of the source chain contract to generate an intermediate language; Reconstruct the IL into equivalent contract logic according to the syntax rules of the target chain; Optimize the contract execution order based on the Gas cost model of the target chain to minimize transaction fees; A real-time status feedback unit integrating a Q-learning reinforcement learning model to dynamically adjust the cross-chain routing strategy, and its scoring formula is: Score = w1·T h + w2·(1 - C)+ w3·(1 - L); Among which T h , C, L are the throughput, handling fee, and latency of the target chain respectively, and the weights w1, w2, w3 are dynamically updated according to the transaction success rate.

5. A micro data exchanger according to claim 1, characterized in that, The multi-dimensional detection module includes: A compliance detection unit for verifying whether data transactions comply with data privacy regulations and data sovereignty requirements of the target region, and it includes: A dynamic legal knowledge graph integrating version tracking and conflict detection of data regulations in more than 50 countries, and generating a compliance report through semantic matching; A cross-language clause alignment module that realizes real-time mapping of multilingual legal clauses based on the LaBSE multilingual embedding model and the FAISS index library.

6. A micro data exchanger according to claim 1, characterized in that, The security detection unit further includes: Integrating a national cryptographic algorithm acceleration module and a post-quantum encryption module, including: An SM2 / SM4 national cryptographic algorithm hardware accelerator for data signature and encrypted transmission; A CRYSTALS-Kyber post-quantum key negotiation module that realizes anti-quantum attack protection for the NISTPQC standard through an ASIC chip.

7. A micro data trader according to claim 1, characterized in that, The quality assessment unit verifies data quality through the following steps: Hash verification: Calculate the BLAKE3 hash value of the data stream and compare it with the hash value recorded on the evidence storage platform; Timestamp Deviation Detection: Verify that the timestamp deviation between the data generation timestamp and the standard time of the time service center ≤ 0.5 seconds.

8. A micro data exchanger according to claim 1, characterized in that, The scenario applicability evaluation unit uses a natural language processing model to parse metadata tags, including: A BERT semantic matching engine that generates a vector representation of the data description text and calculates the cosine similarity with the buyer's scenario tags; An authorization chain verification sub-module that queries the blockchain for the data usage permission deposit of the buyer for the target scenario.

9. A micro data exchanger according to claim 1, characterized in that, The hardware accelerator includes: A scenario evaluation accelerator that integrates a low-precision quantization AI inference engine, supports parallel computing in INT8 data format, and is used to real-time process the semantic matching tasks of data scenario tags; A security compliance accelerator, designed based on ASIC chips, and supports the following functions simultaneously: Hardware-level acceleration of national cryptographic algorithms, including key generation, data signature, and encryption; Key agreement and encapsulation of the NIST post-quantum cryptographic algorithm CRYSTALS-Kyber; Parallel computing of the BLAKE3 hash algorithm, with optimized throughput through the SIMD instruction set; A hardware interface that is directly connected to the edge computing unit through a PCIe4.0 bus to achieve the following coordination mechanism: Offload the hash calculation and signature tasks in compliance detection to the security compliance accelerator; Offload the NLP inference tasks for scenario applicability evaluation to the scenario evaluation accelerator; A dynamic power management unit that uses voltage-frequency regulation technology to dynamically switch between the following modes according to the task load: High-power mode: Enable all acceleration modules, voltage 1.2V, frequency 200MHz; Low-power mode: Only enable the SM4 encryption module, voltage 0.8V, frequency 50MHz.

10. An intelligent trading system for a micro data exchanger, based on a micro data exchanger according to any one of claims 1-9, characterized in that, It includes multiple micro data traders and the following coordination components: National-level computing power network nodes for storing and verifying distributed transaction data; A cross-chain routing center that dynamically selects the optimal blockchain path through a reinforcement learning model; An edge-cloud coordination management platform that monitors device status and dynamically updates the compliance rule library.

Images