Multi-source data trusted desensitization and storage and traceability method fusing privacy calculation and blockchain

Abstract

The application relates to the field of electric digital data processing and discloses a multi-source data trusted desensitization and storage evidence tracing method combining privacy calculation and a block chain, which comprises the following steps: extracting a real-time state feature vector of a current block header of a block chain side chain, obtaining a multi-source input bit stream, initializing a desensitization operator associated with the real-time state feature vector, operating the operator to generate a desensitization bit stream, sampling and transforming logical gate level flip data in a process to extract an execution witness composed of bit offset cumulative features, encapsulating the execution witness into a metadata domain of the desensitization bit stream and storing the execution witness into the block chain, realizing endogenous coupling of a desensitization operator bit level interaction and a ledger state vector, eliminating a state synchronization vacuum period generated by a traditional asynchronous architecture, reducing a calculation entropy increase of a storage evidence link by using an intermediate state feedback of a gate circuit, and ensuring the logical consistency of a whole path of multi-source data flow conversion.

Description

Technical Field

[0001] This invention relates to a method for trusted desensitization, evidence storage and traceability of multi-source data that integrates privacy computing and blockchain, belonging to the field of electronic digital data processing technology. Background Technology

[0002] In current cross-institutional, multi-source data collaboration scenarios, processing units call privacy-preserving computation operators to transform sensitive bitstreams and use distributed ledgers to solidify the processing trajectory. The logical operations of privacy operators and the ledger notarization process are usually on separate logical planes, resulting in a time misalignment between data transformation and notarization actions. This logical vacuum period caused by the asynchronous coupling of functional modules not only causes inaccuracies in the desensitized state and stub fingerprint state when facing large-scale, high-concurrency bitstream processing, but also generates significant computational cycle redundancy due to the secondary hash scan of the desensitized results. When the data processing pipeline maintains high load throughput, traditional architectures, in order to ensure the determinism of notarization results, usually rerun the hash algorithm on the desensitized dataset after the transformation action. This secondary feature extraction process on the processed data consumes a large amount of processor computing resources and increases the entropy of the instruction cycle. Furthermore, when the operator parameters shift due to complex operating conditions, the subsequent hash digest cannot capture in real time and enforce the physical integrity of the transformation process.

[0003] To address these challenges, increasing hardware computing nodes or improving processor frequency not only raises deployment costs but also fails to resolve the inherent logical security flaws of asynchronous architectures. Attempting to force ledger state alignment by blocking processing flows would trigger pipeline backpressure and cause a non-linear deterioration in overall system throughput. In fact, just as existing architectures face insurmountable physical bottlenecks due to limitations imposed by specific roller shapes and other physical load-bearing components, current data storage methods focused on software control also suffer from deep-seated mechanism deficiencies. For example, Chinese invention patent application CN120631855A discloses a privacy data sharing method based on blockchain and privacy-preserving computing, employing local data desensitization preprocessing and... The asynchronous control logic that generates data summaries after the event and registers them on the blockchain, upon in-depth analysis of its underlying mechanism, still relies on secondary feature extraction after the transformation action is completed. Essentially, it is a macroscopic state mapping that is lagging behind the separate logic plane. When facing large-scale, high-concurrency, multi-source heterogeneous bit streams, the post-exhibition strategy that is detached from the underlying digital signal flow process causes significant computational redundancy. There is an irreconcilable temporal and spatial misalignment between the evidence generation mechanism and the underlying bit-level transformation action of the desensitization operator. The ledger state vector is not deeply integrated into the internal physical-level operation link. Once the operation component experiences transient offset under extreme conditions, the macroscopic control method based on the software application layer cannot capture the physical integrity of the forced constraint processing action in real time, resulting in the loss of global state consistency.

[0004] Therefore, the technical problem to be solved by this invention is how to construct a technical path that cohesively couples the desensitization action with the evidence generation logic, and generates execution credentials by using the intermediate state feedback of logic gate circuits while realizing data transformation, thereby eliminating the logical vacuum of state synchronization and reducing the computational load. Summary of the Invention

[0005] To address the problems in the background technology, the technical solution of the present invention is as follows: A method for trusted de-identification, evidence storage, and traceability of multi-source data integrating privacy computing and blockchain, comprising the following steps:

[0006] Step S1: The processing unit extracts the real-time state feature vector representing the real-time running state of the distributed ledger from the current block header of the blockchain sidechain, and obtains a multi-source input bit stream containing multiple heterogeneous sensitive fields.

[0007] Step S2: The processing unit performs sensitivity classification on the multi-source input bitstream according to the preset privacy policy library, and initializes the desensitization operator that has a bit-level correlation mapping relationship with the real-time state feature vector;

[0008] Step S3: The processing unit operates the desensitization operator to perform nonlinear shift transformation on the specified sensitive fields in the multi-source input bitstream to generate a desensitized bitstream;

[0009] In step S4, during the nonlinear shift transformation operation, the processing unit synchronously samples the level-flipping data sequence of the internal logic gate circuit and extracts the execution witness that represents the dynamic characteristics of the accumulated bit offset; wherein, the execution witness is the deterministic mathematical residual generated by the forced constraint of the real-time state feature vector when the desensitized operator operates at the bit level, and the execution witness, the desensitized bit stream, and the real-time state feature vector form a unique monotonic mapping association.

[0010] In step S5, the processing unit encapsulates the execution witness into the metadata field of the desensitized bitstream and sends the execution witness to the blockchain node for evidence storage.

[0011] Preferably, step S2 includes: step S21, the processing unit generates a dynamic obfuscation operator seed based on the real-time state feature vector; step S22, the processing unit uses the dynamic obfuscation operator seed to construct the logic mapping matrix of the desensitization logic pipeline to constrain the bit distribution probability of the multi-source input bit stream, so that the global bit feature entropy of the desensitized bit stream undergoes nonlinear migration with the update of the real-time state feature vector.

[0012] Preferably, the method further includes the following steps: Step S6, the processing unit pushes the generated execution witness into the multi-level aggregation buffer; Step S7, when the number of execution witnesses in the multi-level aggregation buffer reaches a preset threshold, the processing unit performs a cascade hash operation on all execution witnesses in the multi-level aggregation buffer to construct a Merkle tree; Step S8, the processing unit sends the root hash value of the Merkle tree as a global evidence storage anchor to the blockchain node for storage.

[0013] Preferably, step S3 includes: step S31, the processing unit imports the multi-source input bit stream into the trusted execution environment; step S32, the processing unit shields external interrupt requests in the trusted execution environment and strengthens the instruction pipeline of the desensitized operator according to the real-time state feature vector; step S33, the processing unit performs nonlinear displacement transformation in the strengthened instruction pipeline to synchronize the desensitized bit stream with the generation clock of the execution witness.

[0014] Preferably, the method further includes the following steps: the processing unit monitors the re-identification risk entropy value of the desensitized bitstream in real time; when the re-identification risk entropy value exceeds a preset security threshold, the processing unit captures the new block height state of the blockchain sidechain and regenerates the real-time state feature vector, triggering the dynamic reconstruction of the internal parameters of the desensitization operator.

[0015] Preferably, before obtaining the multi-source input bitstream in step S1, the method further includes: the processing unit performs topological reorganization on the received multi-source heterogeneous raw data and encapsulates it into a self-verification data packet containing a load data field and a hash association field; wherein, the load data field carries the multi-source input bitstream, and the hash association field stores the association verification value used to characterize the integrity of the electronic-to-digital conversion in the subsequent tracing process.

[0016] Preferably, the method further includes the following steps: when the computational load of the processing unit exceeds a preset load threshold, the processing unit sends a computation outsourcing instruction through the blockchain sidechain; the processing unit receives the sub-desensitization results fed back by adjacent consensus nodes, and uses real-time state feature vectors to perform homomorphic verification on the sub-desensitization results to ensure the logical consistency of the outsourced computation results.

[0017] Preferably, step S4 includes: step S41, the processing unit acquires the level-flipping data sequence of the logic gate circuit during nonlinear displacement transformation; step S42, the processing unit discretizes the level-flipping data sequence, extracts the bit offset deviation corresponding to the real-time state feature vector, and maps it to the binary representation of the execution witness.

[0018] Preferably, the method further includes the following steps: receiving a traceability query request, which includes the de-identified bitstream to be verified and the target evidence identifier; the processing unit reads the execution witness corresponding to the target evidence identifier from the blockchain node; the processing unit compares the bit distribution characteristics of the de-identified bitstream to be verified with the mapping consistency of the execution witness to determine the original processing trajectory of the de-identified bitstream to be verified.

[0019] Compared with the prior art, the beneficial effects of the present invention are:

[0020] 1. In the trusted desensitization and evidence preservation traceability of multi-source data, by introducing the distributed ledger state vector into the initialization process of the privacy transformation operator, the data transformation action and the generation of evidence fingerprints are logically aligned within the processing unit. This eliminates the spatiotemporal misalignment between the data transformation state and the ledger state in the traditional asynchronous architecture, prevents the risk of non-real business data being replaced during the transformation interval, and ensures the global state consistency of the electronic digital processing link.

[0021] 2. Since the evidence storage image originates from the intermediate state feedback of the logic gate during the execution of the desensitization operator, rather than the secondary hash operation after desensitization, it effectively reduces the computational entropy increase in the electronic digital processing process. This correlation mechanism transforms the evidence storage cost from the traditional linear growth to a constant-level distribution, releasing the logical load of the processing unit while maintaining the same security strength, and improving the processing throughput of large-scale heterogeneous data streams.

[0022] 3. By constructing a deterministic state folding path, the distribution of the desensitized bitstream contains the mathematical residual of the corresponding block height. Based on the data processing trajectory, it has self-proving properties. This mechanism makes subsequent audits no longer dependent on a full backtracking of massive amounts of original data. It is only necessary to verify the mapping relationship between the bit offset and the state seed to determine the compliance of the processing logic, thereby improving the traceability accuracy and trust transmission efficiency in cross-industry multi-source data exchange scenarios. Attached Figure Description

[0023] Figure 1 This is a flowchart of the multi-source data trusted de-identification and execution witness storage method of the present invention;

[0024] Figure 2 This is a logic diagram for the execution of witness extraction and multi-level aggregated evidence storage in this invention.

[0025] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0026] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0027] A method for trusted de-identification, evidence storage, and traceability of multi-source data that integrates privacy computing and blockchain includes the following steps:

[0028] Step S1: The processing unit extracts the real-time state feature vector representing the real-time running state of the distributed ledger from the current block header of the blockchain sidechain, and obtains a multi-source input bit stream containing multiple heterogeneous sensitive fields.

[0029] Step S2: The processing unit performs sensitivity classification on the multi-source input bitstream according to the preset privacy policy library, and initializes the desensitization operator that has a bit-level correlation mapping relationship with the real-time state feature vector;

[0030] Step S3: The processing unit operates the desensitization operator to perform nonlinear shift transformation on the specified sensitive fields in the multi-source input bitstream to generate a desensitized bitstream;

[0031] In step S4, during the nonlinear shift transformation operation, the processing unit synchronously samples the level-flipping data sequence of the internal logic gate circuit and extracts the execution witness that represents the dynamic characteristics of the accumulated bit offset; wherein, the execution witness is the deterministic mathematical residual generated by the forced constraint of the real-time state feature vector when the desensitized operator operates at the bit level, and the execution witness, the desensitized bit stream, and the real-time state feature vector form a unique monotonic mapping association.

[0032] In step S5, the processing unit encapsulates the execution witness into the metadata field of the desensitized bitstream and sends the execution witness to the blockchain node for evidence storage.

[0033] Preferably, step S2 includes: step S21, the processing unit generates a dynamic obfuscation operator seed based on the real-time state feature vector; step S22, the processing unit uses the dynamic obfuscation operator seed to construct the logic mapping matrix of the desensitization logic pipeline to constrain the bit distribution probability of the multi-source input bit stream, so that the global bit feature entropy of the desensitized bit stream undergoes nonlinear migration with the update of the real-time state feature vector.

[0034] Preferably, the method further includes the following steps: Step S6, the processing unit pushes the generated execution witness into the multi-level aggregation buffer; Step S7, when the number of execution witnesses in the multi-level aggregation buffer reaches a preset threshold, the processing unit performs a cascade hash operation on all execution witnesses in the multi-level aggregation buffer to construct a Merkle tree; Step S8, the processing unit sends the root hash value of the Merkle tree as a global evidence storage anchor to the blockchain node for storage.

[0035] Preferably, step S3 includes: step S31, the processing unit imports the multi-source input bit stream into the trusted execution environment; step S32, the processing unit shields external interrupt requests in the trusted execution environment and strengthens the instruction pipeline of the desensitized operator according to the real-time state feature vector; step S33, the processing unit performs nonlinear displacement transformation in the strengthened instruction pipeline to synchronize the desensitized bit stream with the generation clock of the execution witness.

[0036] Preferably, the method further includes the following steps: the processing unit monitors the re-identification risk entropy value of the desensitized bitstream in real time; when the re-identification risk entropy value exceeds a preset security threshold, the processing unit captures the new block height state of the blockchain sidechain and regenerates the real-time state feature vector, triggering the dynamic reconstruction of the internal parameters of the desensitization operator.

[0037] Preferably, before obtaining the multi-source input bitstream in step S1, the method further includes: the processing unit performs topological reorganization on the received multi-source heterogeneous raw data and encapsulates it into a self-verification data packet containing a load data field and a hash association field; wherein, the load data field carries the multi-source input bitstream, and the hash association field stores the association verification value used to characterize the integrity of the electronic-to-digital conversion in the subsequent tracing process.

[0038] Preferably, the method further includes the following steps: when the computational load of the processing unit exceeds a preset load threshold, the processing unit sends a computation outsourcing instruction through the blockchain sidechain; the processing unit receives the sub-desensitization results fed back by adjacent consensus nodes, and uses real-time state feature vectors to perform homomorphic verification on the sub-desensitization results to ensure the logical consistency of the outsourced computation results.

[0039] Preferably, step S4 includes: step S41, the processing unit acquires the level-flipping data sequence of the logic gate circuit during nonlinear displacement transformation; step S42, the processing unit discretizes the level-flipping data sequence, extracts the bit offset deviation corresponding to the real-time state feature vector, and maps it to the binary representation of the execution witness.

[0040] Preferably, the method further includes the following steps: receiving a traceability query request, which includes the de-identified bitstream to be verified and the target evidence identifier; the processing unit reads the execution witness corresponding to the target evidence identifier from the blockchain node; the processing unit compares the bit distribution characteristics of the de-identified bitstream to be verified with the mapping consistency of the execution witness to determine the original processing trajectory of the de-identified bitstream to be verified.

[0041] Example 1: In a high-frequency data exchange scenario within a cross-institutional medical data research consortium, the processing unit receives heterogeneous sensitive bitstreams from multiple terminal nodes. When the data throughput reaches the level of 10,000,000 per second, due to the physical asynchrony between the data transformation logic and the ledger evidence storage link, logical inaccuracies occur between the desensitized data and the evidence fingerprint at the instruction cycle level. The processing unit extracts a real-time state feature vector representing the real-time operating state of the distributed ledger from the current block header of the blockchain sidechain. and will The data is loaded into a high-speed cache register as a logical reference for operator generation. The processing unit reads the random sequence output by the on-chip hardware physical noise generator with a sampling frequency of not less than 50MHz and converts the real-time state feature vector... The system performs a bitwise XOR operation with a random sequence, calls the SHA-256 algorithm to hash the XOR result, extracts the first 128 bits of the output digest to generate a dynamic obfuscation operator seed, and sets the initial bias state of each shift register in the instruction pipeline based on the seed value. For subsequent sensitivity grading logic, a pre-defined privacy policy library loads explicit conditional filtering rules into the processing unit. The system performs characterization and restoration on the input multi-source bitstream, uses a pre-defined regular expression engine to scan and extract identity recognition fields with structured features, and simultaneously performs a sliding step traversal on unstructured data segments to solve the Shannon information entropy distribution within each data window. If the extracted segment matches the feature dictionary or its local information entropy exceeds a specific security baseline, the system issues a grading command to mark it as a high-sensitivity level and matches it with a high-order obfuscation parameter; otherwise, it is classified as a low-sensitivity level, thus forming a complete grading mapping judgment basis.

[0042] Based on a preset privacy policy library, feature analysis and sensitivity classification are performed on the multi-source input bitstream, and a [presumably a specific privacy policy] is initialized. Desensitization operators with bit-level correlation mapping relationships feed multi-source input bitstreams into the instruction pipeline. The desensitization operators perform nonlinear shift transformations on specified sensitive fields, generating a desensitized bitstream. During the transformation operation, the processing unit synchronously samples the level-flipping data sequence of its internal logic gates, extracting execution witnesses that characterize the cumulative dynamic features of bit offsets. ,in For desensitization operator operations bit-level interaction The deterministic mathematical residuals generated by the forced constraints make With desensitized bitstream and To form a unique monotonic mapping relationship, the processing unit will Encapsulated into the metadata field of the desensitized bitstream and pushed into a multi-level aggregation buffer, when the number of witnesses executed in the buffer reaches a preset threshold. At the same time, the evidence storage data is sent to the blockchain node, realizing the coupling of the de-identification action and the evidence storage fingerprint at the processor instruction set level, reducing the instruction overhead of performing secondary hash calculations on the processed data.

[0043] Example 2: In a distributed electronic digital data processing platform, a medical remote sensing analog bitstream containing 1024 sampling points is selected as the multi-source input bitstream. The platform simulates a sidechain environment with 10 consensus nodes, achieving a clock synchronization accuracy better than 1ms and an input channel superposition bit error rate of [missing value]. Interference signals, preset threshold The value is determined based on the matching relationship between the processing unit's cache capacity and the sidechain consensus period. When the sampling period is 10ms, it will... The value is set to 500; in the test group, the processing unit reads the real-time state feature vector. and multi-source input bitstream The import instruction pipeline uses a desensitization operator to perform nonlinear shift transformations to generate a desensitized bitstream. Simultaneously, it samples the level-flipping data of internal logic gates during operation to generate execution witnesses. When processing 1GB of data, the total number of instruction cycles consumed by the experimental group was: Execution Witness With desensitized bitstream The Hamming distance residuals remained within a deterministic range of 0.05%. The control group used a hash calculation method after desensitization, and its total instruction cycles were... Of these, the secondary hash scan performed on the processed data accounts for 35% of the computation cycle.

[0044] Partially missing control group removes real-time state feature vector The constraints on the initialization of the de-identification operator cause the generated execution witness to lose its association with the current state of the distributed ledger, resulting in a 100% failure rate in its traceability verification. This proves that the coupling between the ledger state vector and bit-level transformation actions is a prerequisite for achieving reliable traceability. In boundary verification, a preset threshold is used. The out-of-range control group, set at 5000, showed that the system's end-to-end evidence storage latency increased from 15.2ms to 148.6ms. The preset threshold was then adjusted. The control group with a lower limit of 50 showed that the CPU utilization rate increased from 42.5% to 88.3%, and a backpressure phenomenon occurred in the processing flow. Through quantitative analysis of instruction execution efficiency and evidence consistency under different operating conditions, the multi-source data processing method that integrates privacy computing and blockchain maintains high throughput electro-digital signal transformation while achieving underlying interlocking of data processing trajectory and distributed ledger state based on output bit stream with self-proving attributes through state folding path.

[0045] Example 3: This example combines Figures 1 to 2 This section explains a method for trusted de-identification and evidence preservation and traceability of multi-source data that integrates privacy computing and blockchain. Figure 1 As shown, real-time state feature vectors are extracted to obtain multi-source input bitstreams. The flow direction is based on the sensitivity-level initialization of the privacy policy library's desensitization operator, and the desensitization operator's nonlinear shift transformation is further computed to generate a desensitized bitstream. Simultaneously, the level-flipped data sequence is sampled to extract the execution witness. The accompanying association indicates that the execution witness is constrained by the feature vector, generating a deterministic mathematical residual. Finally, the execution witness is encapsulated and sent to the metadata domain for blockchain node storage. Figure 2 As shown, the process originates from the logic gate circuit and flows to the acquisition level-flipped data sequence. Simultaneously, it combines with the real-time state feature vector to converge to the discretization sampling to extract bit offset deviation, proceeds to the mapping execution witness binary representation, and continues to be pushed into the multi-level aggregation buffer. After that, it proceeds to the multi-level aggregation buffer to reach the preset threshold, and then flows sequentially to the cascaded hash operation and the construction of the Merkle tree. It then flows to the global evidence storage anchor point based on the hash value, and finally leads to the sending to the blockchain node for storage.

[0046] Example 4: In the digital data processing pipeline of the transaction clearing system, the processing unit receives bit streams from multiple source interfaces. To align the states during the processing, the system employs a state folding method. The processing unit retrieves the hash value of the current block height from the sidechain cache and converts it into a 256-bit real-time state feature vector. Start the logical mapping matrix Initialization, select The first 128 bits are used as the row vector offset, selected The last 128 bits are used as column vector offsets, and a 256×256 logical mapping matrix is ​​constructed through a circular left shift operation. , where matrix elements The value of determines the initial toggling threshold of the pipelined logic gate circuit, in the bit stream After entering the instruction pipeline, the processing unit performs a nonlinear shift transformation on the sensitive fields using a desensitization operator to generate a desensitized bit stream. The sampling module monitors the level status of internal logic gate circuits at a frequency of no less than 10 GHz. Specifically, the sampling module is a customized on-chip hardware probe array independent of the central processing unit. Its physical probes are directly connected to the source and drain nodes of the core shift register in the arithmetic logic unit. The probe array digitizes the nanosecond-level analog level transient fluctuations in real time through a high-speed analog-to-digital converter integrated on-chip, and writes the quantized hardware low-level level characteristic data directly into a specific physical register range that can be seamlessly addressed by the processing unit through a dedicated high-level peripheral bus. This overcomes the scale limitation of software instruction cycles and constructs a physical direct link for collecting the low-level analog electrical state as the input feature of the upper-level digital algorithm.

[0047] When a logic gate undergoes a level flip, the sampling module records the voltage change within a clock cycle. and compare it with a preset reference voltage. If compared, Exceed of If the percentage is %, it is determined to be a valid bit offset point. The processing unit, based on the intrinsic noise Poisson distribution model of semiconductor devices, converts the physical flip event into a quantization residual. The processing unit then uses the formula... Calculate the deterministic mathematical residual of a single effective bit offset. ,in, The transient change in the actual trigger voltage of the calibrated logic gate. The expected value of the switching voltage of a specific logic gate circuit is obtained after 10,000 no-load tests at a standard reference temperature of 25°C. The standard deviation of the no-load test voltage dataset is calibrated, and the processing unit extracts dimensionless real numbers. The mathematical signature constituting a single flip event, within this residual generation mechanism, is logically self-consistent based on the following: the extraction operation does not directly appropriate the physical thermal noise itself, which possesses random characteristics, but rather systematically removes environmental thermal noise and random scattering interference conforming to a Poisson distribution using the aforementioned standardized formula; after removal, the remaining level characteristic fluctuations are solely induced by a fixed Boolean logic sequence executed by the desensitization operator under the control of a specific real-time state characteristic vector; since a deterministic instruction pipeline inevitably triggers the coordinated switching actions of transistors at the same microscopic level, the signal offset caused by this specific energy release possesses reproducible mathematical determinism after removing background noise, thus eliminating causal mutual exclusion in the physical mechanism; the processing unit accumulates the timing characteristics of 512 consecutive effective bit offset points, generating an execution witness characterizing the cumulative dynamic characteristics of the bit offset. Under operating conditions with an ambient noise signal-to-noise ratio of 20dB, the generated Maintaining 99.999% stability, the output desensitized bitstream Witnessing the execution Logically constitutes the subject The monotonic association of constraints ensures that the logical consistency determination accuracy of data packets reaches the field level when they are verified by the source query request.

[0048] Example 5: In a distributed sidechain node's electronic digital data processing hardware environment, the system receives multi-source input bit streams. Before starting the logical mapping matrix During the parameter calibration process, the processing unit activates the on-chip dynamic voltage regulation module and extracts the real-time state feature vector. A specific bit segment is input to a digital-to-analog converter (DAC) with a resolution of at least 12 bits, outputting a reference bias voltage. This reference bias voltage is directly applied to the substrate of the instruction pipeline transistor. Based on the bulk effect principle of metal-oxide-semiconductor field-effect transistors (MOSFETs), the physical switching threshold of the logic gate circuit is quantitatively controlled by changing the substrate bias potential. In actual operation, the dynamic adjustment of this substrate bias potential is achieved through a high-precision programmable power management chip on the motherboard. The digital control interface of this power management chip is directly connected to the system data bus carrying the real-time state feature vector. When it receives the reference bias voltage digital sequence from the DAC, its internal linear regulator circuit responds instantly, outputting a corresponding level of precise DC bias potential. This is then directly injected into the physical backplane of the processing core where the instruction pipeline is located through a dedicated, independently isolated power supply network. This establishes a hard-wired response mechanism from the software feature vector state to the hardware micro-potential. After locking the hardware physical state, the processing unit injects a preset calibration bit sequence and monitors the internal instruction pipeline state, calculating a compensation coefficient based on the deviation between the gate circuit level switching frequency and the expected value. The compensation coefficient Superimposed on the real-time state feature vector To adjust the initial toggling threshold of the logic gate circuit, when the adjusted real-time state feature vector After loading the cache register, the system uses a logical mapping matrix. Constrain the bit shift step size of the desensitization operator so that different physical execution cores can process desensitization intensity at the same level. When generating the bitstream, the execution witness is generated. With desensitized bitstream The statistical characteristic distribution converges to the preset residual interval.

[0049] When the system faces a situation where there are slight timing and physical differences between sidechain block height updates and high-speed bitstream processing, the processing unit uses a synchronous pulse buffer pool to align the state and acquire the real-time state feature vector. Then, the pipeline controller sends a state lock signal to the instruction execution stack, triggering the desensitization operator parameters to complete the process based on the real-time state feature vector within the next clock cycle. Dynamic reconfiguration, by adjusting the voltage change The sampling sensitivity is adjusted to the reference voltage. The feature entropy fluctuation caused by sidechain communication jitter is suppressed by setting a sliding sampling window length of 512 bits and a value of 15.5%; the associated verification value is encapsulated in the desensitized data packet header. Indexing to the ledger state at a specific height in the blockchain enables the system to maintain physical reproducibility of its processing trajectory under different operating conditions.

[0050] Example 6: In the initial deployment phase of the electronic digital clearing environment, the processing unit determines the desensitization strength by inputting a calibration bit sequence with statistical characteristics. The initial value, when the desensitization intensity Starting from 0.10 and increasing in steps of 0.05 up to 0.80, the processing unit samples and obtains the re-identification risk entropy value corresponding to each intensity gradient. , determine Reduced to a preset safety threshold The following minimum value is taken as the desensitization intensity. Based on the remaining capacity of the processor core's L2 cache Compared to the current sidechain consensus cycle The proportional relationship, according to the formula Calculate encryption weight ,in, At the moment of metering sampling, the processor's L2 cache did not occupy any actual available bytes. The time interval between the generation of two adjacent consensus blocks in the distributed ledger is measured. After the processing unit calculates the ratio, it truncates the floating-point number of the ratio and directly assigns it as a dimensionless scalar parameter to the encryption weight. The values ​​are limited to the real number range of 0 to 1. The logic for calculating the re-identification risk entropy mentioned in the above calculation is based on an assessment and analysis of the difficulty of reverse decoding of the feature fingerprint in the frequency domain. During the calculation, the hardware logic calculation unit first sorts out the discretized frequency of each feature data group in the desensitized output dataset falling into the preset feature subspace, and normalizes the frequency to obtain the objective probability distribution characteristics of the feature group. Then, the arithmetic unit accumulates the negative value of the product of the probability of each feature occurrence and the corresponding logarithm with base 2, and defines the pure scalar calculation result as the re-identification risk entropy index that reflects the resistance to reverse derivation. The resulting encryption weight is defined as follows. Used to initialize the logical mapping matrix It also constrains the bit shift step size of the desensitization operator, so that the desensitized bit streams generated by different physical kernels are... Witnessing the execution The distribution characteristics remain consistent.

[0051] During the instruction cycle when the processing unit is operating without load, the sampling module calculates the first derivative of the voltage change rate by continuously monitoring the level fluctuations of the logic gates, and then compares the peak-to-average voltage fluctuation with the safety margin. The product obtained by multiplication is used as the reference voltage. The correction offset is used to adjust the execution witness. The sampling sensitivity is adjusted to eliminate random bit-flipping interference caused by circuit thermal drift when multiple source input bit streams are used. After the imported instructions are processed by the desensitization operator in the pipeline, the processing unit determines the input based on the logical mapping matrix. Generate execution witness containing high imprint of the ledger Will execute witness Associated check value Encapsulate the metadata field into the self-verifying data packet, and at a height of [height missing] in the distributed ledger. Batch storage is completed within the cycle, ensuring that the data transformation trajectory is controlled by the real-time state feature vector. Induced logical constraints, among which, Desensitization intensity; To re-identify risk entropy values; Preset safety threshold; For encrypted weights; This represents the remaining capacity of the L2 cache. Consensus period; It is a logical mapping matrix; For desensitized bitstream; To serve as witness; For safety margin; The reference voltage; It is a multi-source input bitstream; For associated verification values; For distributed ledger height; This is a real-time state feature vector; when the computational load of the processing unit exceeds a preset load threshold... At that time, the processing unit sends computation outsourcing instructions via sidechain and receives sub-desensitization results from adjacent kernels, and calls the real-time state feature vector. Homomorphic verification is performed on the sub-results; specifically, it involves comparing whether the bit distribution probability of the sub-results conforms to the real-time state feature vector. The distribution constraints of the generated dynamic obfuscation operator seed determine its logical consistency. Upon receiving a traceability query request, the processing unit reads the execution witness corresponding to the target evidence identifier from the blockchain node. And compare the statistical distribution characteristics of the desensitized bitstream to be verified. When the mapping error between the two is lower than the preset deviation threshold, At that time, it is determined that the original processing trajectory of the bit stream to be verified conforms to the operating state of the distributed ledger at a specific block height.

[0052] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for trusted de-identification, evidence storage, and traceability of multi-source data integrating privacy computing and blockchain, characterized in that, Includes the following steps: Step S1: The processing unit extracts the real-time state feature vector representing the real-time running state of the distributed ledger from the current block header of the blockchain sidechain, and obtains a multi-source input bit stream containing multiple heterogeneous sensitive fields. Step S2: The processing unit performs sensitivity classification on the multi-source input bitstream according to the preset privacy policy library, and initializes the desensitization operator that has a bit-level correlation mapping relationship with the real-time state feature vector; Step S3: The processing unit operates the desensitization operator to perform nonlinear shift transformation on the specified sensitive fields in the multi-source input bitstream to generate a desensitized bitstream; In step S4, during the nonlinear shift transformation operation, the processing unit synchronously samples the level-flipping data sequence of the internal logic gate circuit and extracts the execution witness that represents the dynamic characteristics of the accumulated bit offset; wherein, the execution witness is the deterministic mathematical residual generated by the forced constraint of the real-time state feature vector when the desensitized operator operates at the bit level, and the execution witness, the desensitized bit stream, and the real-time state feature vector form a unique monotonic mapping association. In step S5, the processing unit encapsulates the execution witness into the metadata field of the desensitized bitstream and sends the execution witness to the blockchain node for evidence storage.

2. The method for trusted de-identification, evidence storage, and traceability of multi-source data integrating privacy computing and blockchain as described in claim 1, characterized in that, Step S2 includes: Step S21, the processing unit generates a dynamic obfuscation operator seed based on the real-time state feature vector; Step S22, the processing unit uses the dynamic obfuscation operator seed to construct the logic mapping matrix of the desensitization logic pipeline to constrain the bit distribution probability of the multi-source input bit stream, so that the global bit feature entropy of the desensitized bit stream undergoes nonlinear migration with the update of the real-time state feature vector.

3. The method for trusted de-identification, evidence storage, and traceability of multi-source data integrating privacy computing and blockchain as described in claim 1, characterized in that, It also includes the following steps: Step S6: The processing unit pushes the generated execution witness into the multi-level aggregation buffer; Step S7: When the number of execution witnesses in the multi-level aggregation buffer reaches a preset threshold, the processing unit performs a cascade hash operation on all execution witnesses in the multi-level aggregation buffer to construct a Merkle tree; Step S8: The processing unit sends the root hash value of the Merkle tree as the global evidence storage anchor to the blockchain node for storage.

4. The method for trusted de-identification, evidence storage, and traceability of multi-source data integrating privacy computing and blockchain as described in claim 1, characterized in that, Step S3 includes: Step S31, the processing unit imports the multi-source input bit stream into the trusted execution environment; Step S32, the processing unit shields external interrupt requests in the trusted execution environment and strengthens the instruction pipeline of the desensitized operator according to the real-time state feature vector; Step S33, the processing unit completes nonlinear shift transformation in the strengthened instruction pipeline to synchronize the desensitized bit stream with the generation clock of the execution witness.

5. The method for trusted de-identification and evidence preservation and traceability of multi-source data integrating privacy computing and blockchain as described in claim 1, characterized in that, It also includes the following steps: The processing unit monitors the re-identification risk entropy value of the desensitized bitstream in real time; When the re-identification risk entropy value exceeds the preset security threshold, the processing unit captures the new block height state of the blockchain sidechain and regenerates the real-time state feature vector, triggering the dynamic reconstruction of the internal parameters of the desensitization operator.

6. The method for trusted de-identification, evidence storage, and traceability of multi-source data integrating privacy computing and blockchain as described in claim 1, characterized in that, Before step S1 acquires the multi-source input bitstream, the processing unit further includes: performing topological reorganization on the received multi-source heterogeneous raw data and encapsulating it into a self-verification data packet containing a load data field and a hash association field; wherein, the load data field carries the multi-source input bitstream, and the hash association field stores the association verification value used to characterize the integrity of the electronic-to-digital conversion in the subsequent tracing process.

7. The method for trusted de-identification and evidence preservation and traceability of multi-source data integrating privacy computing and blockchain as described in claim 1, characterized in that, It also includes the following steps: When the computational load of the processing unit exceeds the preset load threshold, the processing unit sends a computation outsourcing instruction through the blockchain sidechain; the processing unit receives the sub-desensitization results fed back by the adjacent consensus nodes, and uses the real-time state feature vector to perform homomorphic verification on the sub-desensitization results to ensure the logical consistency of the outsourced computation results.

8. The method for trusted de-identification, evidence storage, and traceability of multi-source data integrating privacy computing and blockchain as described in claim 1, characterized in that, Step S4 includes: Step S41, the processing unit acquires the level-flipping data sequence of the logic gate circuit during nonlinear displacement transformation; Step S42, the processing unit discretizes the level-flipping data sequence, extracts the bit offset deviation corresponding to the real-time state feature vector, and maps it to the binary representation of the execution witness.

9. A method for trusted de-identification, evidence storage, and traceability of multi-source data integrating privacy computing and blockchain as described in claim 1, characterized in that, It also includes the following steps: The system receives a traceability query request, which includes the de-identified bitstream to be verified and the target evidence identifier; the processing unit reads the execution witness corresponding to the target evidence identifier from the blockchain node; the processing unit compares the bit distribution characteristics of the de-identified bitstream to be verified with the mapping consistency of the execution witness to determine the original processing trajectory of the de-identified bitstream to be verified.

Images