Homomorphic encryption method for sensitive data transmission

Abstract

This invention relates to the field of data encryption technology and discloses a homomorphic encryption method for sensitive data transmission, comprising: a lower-level node left-shifting and mapping sensitive state parameters according to the starting position and length of binary bit segments, and superimposing anti-overflow isolation bits to construct a combined data packet; calling the public key to encrypt and generate state ciphertext and sending it to a relay node; the relay node aggregating multiple state ciphertexts in the ciphertext field through homomorphic modular multiplication in place, converging to generate a single aggregated ciphertext and transmitting it to the core decryption node; the core decryption node calling the private key to globally decrypt the single aggregated ciphertext to recover the plaintext value, and extracting the global accumulated value corresponding to each binary bit segment through hierarchical masking. This invention aggregates sensitive data in place in the ciphertext state, transforming the decryption overhead from a concurrent linear growth state to a constant low-load state, and using anti-overflow isolation bits to absorb numerical overflow, eliminating the computing power congestion of the core decryption node.

Description

Technical Field

[0001] This invention relates to a homomorphic encryption method for sensitive data transmission, belonging to the field of data encryption technology. Background Technology

[0002] Current wide-area, multi-level linkage service networks typically employ cryptographic algorithms to establish end-to-end data privacy protection in the network channel when handling the reporting of sensitive status parameters from massive numbers of terminal nodes. However, due to the objective need for phased statistical analysis and aggregation of regional indicator data by intermediate relay nodes, a ciphertext computation principle based on homomorphic algebra structures is used to provide a decryption-free data flow path when solving long-link relay aggregation. This principle allows data to directly complete product or summation algebraic operations while maintaining ciphertext at relay nodes. This enables the central node to directly recover the algebraic summation result of the global data after receiving the aggregated ciphertext and decrypting it using its private key. This is particularly relevant in multi-level linkage scenarios. When service networks face the demand for large-scale, high-frequency concurrent reporting of multidimensional heterogeneous state parameters, existing homomorphic computation methods exhibit significant mechanistic constraints when handling multiple independent performance indicators or data statistics. If each independent state parameter is configured with an independent ciphertext space to maintain the isolation of algebraic operations, the communication bandwidth and the frequency of algebraic multiplication at relay nodes increase linearly with the increase of the state parameter dimension. As a result, the entire transmission link suffers from huge bandwidth loads and transmission delays. This inherent structural configuration, which simply superimposes the single-variable homomorphic process onto multidimensional complex conditions, becomes the cumulative loss in the industry when reporting multidimensional data in real time over long links.

[0003] To alleviate the narrowing of communication bandwidth and the loss of computing power, conventional improvement strategies usually attempt to directly carry fragmented transparent ciphertext by increasing the hardware configuration of the central decryption node or introducing a multi-party collaborative interaction mechanism at the routing node. However, simply increasing hardware overhead cannot eliminate the physical limitations of concurrent congestion in a wide area network environment. Introducing complex interaction flows can lead to unpredictable timing delays in response synchronization between nodes. Relying solely on physical layer topology optimization or hardware configuration stacking is no longer sufficient to cope with the large-scale, high-frequency concurrent demands of multi-dimensional heterogeneous parameters. In addition to the limitations of hardware resource allocation, software-level encryption strategies and aggregation logic also face systemic bottlenecks. For example, the publication number C Chinese invention patent application N118427892A discloses a method, apparatus, and device for constructing a federated decision tree model based on batch homomorphic encryption. It improves transmission efficiency by concatenating multiple gradient information into a packet and then performing homomorphic encryption. This scheme mainly targets the local working condition of gradient exchange in federated learning and lacks in-depth consideration of the numerical evolution characteristics in long-link transmission of multi-level relays in wide area networks. In actual multi-level linkage scenarios, the continuous accumulation of ciphertext fields initiated by multiple relay nodes will lead to rapid numerical growth. If only simple plaintext concatenation is performed, without strict bit segment security margin control, the bit data of each independent dimension is very likely to cause bit cross-contamination during the overflow process.

[0004] Therefore, the technical problem to be solved by this invention is how to achieve in-situ dynamic aggregation of multi-dimensional state parameters and concurrent traffic peak reduction in a multi-level interconnected service network while maintaining strong data privacy protection. Summary of the Invention

[0005] To address the problems in the background art, the technical solution of the present invention is as follows: A homomorphic encryption method for sensitive data transmission, comprising the following steps: Step S101: The lower-level node obtains the sensitive state parameters collected locally. Based on the starting position of the binary bit segment corresponding to the management level and the effective bit segment length determined by the historical maximum extreme value, it performs a left shift operation on the sensitive state parameters to map the sensitive state parameters into the corresponding binary bit segment. It also adds a preset bit width anti-overflow isolation bit to the high bit side of the binary bit segment and packages it to generate a combined plaintext data packet. In step S102, the lower-level node calls the public key of the additive homomorphic encryption algorithm to perform a high-dimensional hidden calculation on the combined plaintext data packet to generate the state ciphertext, and transmits the state ciphertext to the relay aggregation node of the receiving end through the configured multi-level network channel. In step S103, the relay aggregation node aggregates multiple state ciphertexts belonging to the same reporting cycle in the undecrypted ciphertext field using homomorphic modular multiplication, and generates a single aggregated ciphertext containing the global accumulation result of each binary bit segment. The single aggregated ciphertext is then transmitted upward to the core decryption node at the center. In step S104, the core decryption node calls the private key of the additive homomorphic encryption algorithm to perform a modular exponentiation inverse operation on the single aggregate ciphertext to decrypt it globally, recovering the comprehensive plaintext value. Based on the segmentation and separation rules consisting of the starting position and the length of the effective bit segment, a hierarchical mask matrix is ​​introduced to extract and separate each binary bit segment, and restore the global cumulative value corresponding to each level of management.

[0006] Preferably, step S101 includes the following sub-steps: Step S1011, obtain the hierarchical code and cell number corresponding to the lower-level node, and query the topology structure tree stored in the system to determine the topology depth and branch weight of the lower-level node in the aggregation network; Step S1012, calculate the starting position of the binary bit segment allocated to the lower-level node when uploading based on the topology depth and branch weight, and reserve a binary safety gap in combination with the historical maximum extreme value as an anti-overflow isolation bit to complete the mapping combination of the combined plaintext data packets.

[0007] Preferably, the additive homomorphic encryption algorithm used in step S102 is a public-key encryption algorithm based on the difficult problem of factoring large integer composite numbers. The lower-level node calls the public key to perform modular exponentiation and homomorphic mask product operation on the combined plaintext data packets to generate state ciphertext.

[0008] Preferably, after receiving the ciphertexts of the status sent by multiple lower-level nodes, the relay aggregation node performs a modular exponentiation operation on multiple ciphertexts belonging to the same reporting period in the ciphertext state without decryption by the private key, so as to aggregate and combine the algebraic summation values ​​of the plaintext data packets on the corresponding binary bit fields in the ciphertext state to generate a single aggregated ciphertext.

[0009] Preferably, in step S104, after the core decryption node recovers the comprehensive plaintext value by decrypting with the private key, it calls the locally stored hierarchical mask matrix, which contains binary extraction codes corresponding one-to-one with each level of management. The core decryption node performs a bitwise AND operation on the comprehensive plaintext value and the binary extraction codes in sequence, clears the bit segment data of non-target management levels, and reverses the right shift to align and restore the operation result, thus separating and extracting the independent accumulated plaintext values ​​of each level of management.

[0010] Preferably, after the core decryption node extracts the global cumulative value corresponding to each level of management, it continuously stores the global cumulative value into the historical database according to the time series, calculates the transient change rate and variance fluctuation of the global cumulative value between adjacent reporting periods; when the transient change rate continuously exceeds the set change threshold and the variance fluctuation shows a monotonically increasing trend, the core decryption node determines that there is an abnormal data flow injection disturbance in the current transmission channel, and sends a channel abnormality warning signal to the corresponding network node.

[0011] Preferably, the reporting period for the encrypted state of the lower-level node is 500ms to 2000ms, and the key length of the additive homomorphic encryption algorithm is greater than or equal to 1024 bits, and the time taken for the in-situ aggregation operation of the encrypted state of the relay aggregation node is less than or equal to 50ms.

[0012] Preferably, the relay aggregation node monitors the concurrent reporting traffic in the current aggregation channel. When the concurrent reporting traffic exceeds the traffic threshold, it broadcasts a bit width adjustment instruction to the lower-level node so that the lower-level node extends the binary bit width of the anti-overflow isolation bit by 2 bits to 4 bits to the higher bit side. The traffic threshold is dynamically calculated based on the sum of the average concurrent traffic during historical regular periods and three times the standard deviation.

[0013] Preferably, before constructing the combined plaintext data packet, the lower-level node performs normalization mapping on the collected sensitive state parameters, proportionally scaling the original values ​​of the sensitive state parameters to the range between 0 and 1, and then performs left shift mapping processing.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. In homomorphic encryption of sensitive data transmission, the data source node reconstructs multidimensional state parameters using bit segment isolation rules. A fixed length of isolation bits is configured between the effective word lengths of each state parameter, so that multiple independent indicators are converted and combined into a single comprehensive plaintext value. Then, comprehensive ciphertext is generated by encryption with a public key. This data form modulation mechanism is coupled with ciphertext state multiplication operation, so that the relay aggregation nodes on the transmission path can complete the in-situ accumulation of data of each node by algebraic multiplication of the ciphertext field without being configured with decryption permissions. From the mechanism level, the risk of leakage of plaintext decryption by relay nodes is eliminated, and the crosstalk of plaintext cross-section when multidimensional state parameters are superimposed in parallel in the ciphertext field is effectively blocked. This enables privacy security and relay statistical capabilities in the long-link multi-level transmission process to achieve an objective symbiotic state of non-interference.

[0015] 2. The core decryption node receives the aggregated ciphertext output by the relay aggregation node and decrypts it using the private key to recover the comprehensive plaintext value containing the global accumulation result. Then, according to the preset bit segment division rules, it uses reverse masking to extract and separate the precise global accumulation values ​​of the state parameters of each level of the node in different dimensions. This progressive interaction mode based on in-situ dynamic aggregation of ciphertext and global overall decryption transforms the decryption computation overhead of the central node from a linear growth state dependent on the number of concurrent terminal nodes to a constant low-load state only related to the level and regional distribution of the relay aggregation node. This objectively eliminates the core node computing power congestion problem caused by the concurrent accumulation of fragmented ciphertext during long-link, long-cycle data reporting. Attached Figure Description

[0016] Figure 1 This is a flowchart of the multi-level network sensitive data homomorphic encryption method of the present invention; Figure 2 This is a diagram illustrating the evolution of the encrypted state of sensitive data in this invention.

[0017] 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

[0018] 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.

[0019] A homomorphic encryption method for sensitive data transmission includes the following steps: Step S101: The lower-level node obtains the sensitive state parameters collected locally. Based on the starting position of the binary bit segment corresponding to the management level and the effective bit segment length determined by the historical maximum extreme value, it performs a left shift operation on the sensitive state parameters to map the sensitive state parameters into the corresponding binary bit segment. It also adds a preset bit width anti-overflow isolation bit to the high bit side of the binary bit segment and packages it to generate a combined plaintext data packet. In step S102, the lower-level node calls the public key of the additive homomorphic encryption algorithm to perform a high-dimensional hidden calculation on the combined plaintext data packet to generate the state ciphertext, and transmits the state ciphertext to the relay aggregation node of the receiving end through the configured multi-level network channel. In step S103, the relay aggregation node aggregates multiple state ciphertexts belonging to the same reporting cycle in the undecrypted ciphertext field using homomorphic modular multiplication, and generates a single aggregated ciphertext containing the global accumulation result of each binary bit segment. The single aggregated ciphertext is then transmitted upward to the core decryption node at the center. In step S104, the core decryption node calls the private key of the additive homomorphic encryption algorithm to perform a modular exponentiation inverse operation on the single aggregate ciphertext to decrypt it globally, recovering the comprehensive plaintext value. Based on the segmentation and separation rules consisting of the starting position and the length of the effective bit segment, a hierarchical mask matrix is ​​introduced to extract and separate each binary bit segment, and restore the global cumulative value corresponding to each level of management.

[0020] Preferably, step S1011 includes the following sub-steps: Step S1011, obtain the hierarchical code and cell number corresponding to the lower-level node, and query the topology structure tree stored in the system to determine the topology depth and branch weight of the lower-level node in the aggregation network; Step S1012, calculate the starting position of the binary bit segment allocated to the lower-level node when uploading based on the topology depth and branch weight, and reserve a binary safety gap in combination with the historical maximum extreme value as an anti-overflow isolation bit to complete the mapping combination of the combined plaintext data packets.

[0021] Preferably, the additive homomorphic encryption algorithm used in step S102 is a public-key encryption algorithm based on the difficult problem of factoring large integer composite numbers. The lower-level node calls the public key to perform modular exponentiation and homomorphic mask product operation on the combined plaintext data packets to generate state ciphertext.

[0022] Preferably, after receiving the ciphertexts of the status sent by multiple lower-level nodes, the relay aggregation node performs a modular exponentiation operation on multiple ciphertexts belonging to the same reporting period in the ciphertext state without decryption by the private key, so as to aggregate and combine the algebraic summation values ​​of the plaintext data packets on the corresponding binary bit fields in the ciphertext state to generate a single aggregated ciphertext.

[0023] Preferably, in step S104, after the core decryption node recovers the comprehensive plaintext value by decrypting with the private key, it calls the locally stored hierarchical mask matrix, which contains binary extraction codes corresponding one-to-one with each level of management. The core decryption node performs a bitwise AND operation on the comprehensive plaintext value and the binary extraction codes in sequence, clears the bit segment data of non-target management levels, and reverses the right shift to align and restore the operation result, thus separating and extracting the independent accumulated plaintext values ​​of each level of management.

[0024] Preferably, after the core decryption node extracts the global cumulative value corresponding to each level of management, it continuously stores the global cumulative value into the historical database according to the time series, calculates the transient change rate and variance fluctuation of the global cumulative value between adjacent reporting periods; when the transient change rate continuously exceeds the set change threshold and the variance fluctuation shows a monotonically increasing trend, the core decryption node determines that there is an abnormal data flow injection disturbance in the current transmission channel, and sends a channel abnormality warning signal to the corresponding network node.

[0025] Preferably, the reporting period for the encrypted state of the lower-level node is 500ms to 2000ms, and the key length of the additive homomorphic encryption algorithm is greater than or equal to 1024 bits, and the time taken for the in-situ aggregation operation of the encrypted state of the relay aggregation node is less than or equal to 50ms.

[0026] Preferably, the relay aggregation node monitors the concurrent reporting traffic in the current aggregation channel. When the concurrent reporting traffic exceeds the traffic threshold, it broadcasts a bit width adjustment instruction to the lower-level node so that the lower-level node extends the binary bit width of the anti-overflow isolation bit by 2 bits to 4 bits to the higher bit side. The traffic threshold is dynamically calculated based on the sum of the average concurrent traffic during historical regular periods and three times the standard deviation.

[0027] Preferably, before constructing the combined plaintext data packet, the lower-level node performs normalization mapping on the collected sensitive state parameters, proportionally scaling the original values ​​of the sensitive state parameters to the range between 0 and 1, and then performs left shift mapping processing.

[0028] Example 1: Deployed in a globally hierarchical interconnected network within the smart library cloud platform, its core operating logic is based on a distributed homomorphic computation architecture. During system operation, each school-level terminal application node acts as a data source collection unit, defining local sensitive reading status data as a numerical sequence. According to preset binary bit segment isolation rules, each status data is left-shifted and mapped to a specific binary space. Simultaneously, a preset length of anti-overflow isolation bits is superimposed on the high-order bits of each status bit segment, thereby constructing a combined plaintext data packet. This node calls the public key of the additive homomorphic encryption algorithm to perform modular exponentiation high-dimensional hiding calculation on the combined plaintext data packet, generating the state ciphertext, and transmitting the state ciphertext through multi-level network transmission links. The state ciphertext is sent to the local relay aggregation node. After receiving the state ciphertext from multiple lower-level nodes, the relay aggregation node performs modular multiplication in the ciphertext state, directly aggregating the multiple state ciphertexts in the same state domain to generate a single aggregated ciphertext. Finally, this single aggregated ciphertext is transmitted to the core decryption node of the provincial center platform via network layer routing. The core decryption node uses a pre-stored private key to perform global decryption, restoring the plaintext value containing the superposition result of multiple state parameters. It then extracts the global cumulative value corresponding to each binary bit segment through a mask matrix, thereby accurately achieving rapid aggregation and statistics of global state data while maintaining a high degree of privacy for the original data at all levels.

[0029] Example 2: This experiment aims to verify the performance of in-situ ciphertext aggregation in a multi-level interconnected network under complex transmission links. By setting datasets of different sizes, the computational stability and data reconstruction accuracy of the algorithm under high-concurrency data traffic are evaluated. The experimental platform consists of one provincial central control unit, four regional aggregation relay units, and 5000 terminal acquisition nodes. Each node synchronously transmits status data through a preset network interface. The experimental design covers multiple sets of concurrent tests from 500 to 5000 terminal nodes, with an incremental gradient of 500 nodes, for a total of 10 experimental sample groups. The range of sensitive state parameter values ​​collected by each node is set from 1 to 100. To ensure that numerical contamination does not occur in the ciphertext domain operation, the state bit segments are separated. A 6-bit fixed-length overflow prevention isolation bit is configured, and a 10-bit effective data bearer area is configured on the low-order side of each bit segment to construct a combined plaintext data packet. In the experiment, each regional aggregation relay unit performs multiplication and accumulation operations in ciphertext state. The numerical accuracy and aggregation response latency of the core decryption node after completing global decryption are statistically analyzed. The experimental data are as follows. According to the isomorphic algebra principle of additive homomorphic encryption, the modular multiplication operation of the ciphertext space corresponds to the algebraic addition operation of the plaintext space. The relay aggregation node completes the plaintext numerical summation while maintaining the data ciphertext form. It receives multiple state ciphertexts belonging to the same reporting period through the built-in large integer modular multiplication hardware acceleration unit, calculates the modular product of the state ciphertexts, and sets the set of ciphertexts received from lower-level nodes in the current reporting period as... Relay aggregation nodes compute single aggregated ciphertext The calculation formula is: ,in, The single aggregated ciphertext generated by the aggregation has a value less than positive integers, For the first The encrypted status uploaded by each lower-level node has a dimension of one and a value less than 1. positive integers, The modulus of the homomorphic encryption algorithm is a large integer composite number, taking the value of a positive integer with a word length of 1024 bits. This allows the binary segments corresponding to each lower-level node to be summed and superimposed within the ciphertext space, generating a single aggregated ciphertext containing the global sum of all binary segments. With 500 nodes, the aggregation response latency is 125.6ms, and the accumulation error is 0; with 1000 nodes, the aggregation response latency is 142.3ms, and the accumulation error is 0; with 1500 nodes, the aggregation response latency is 168.9ms, and the accumulation error is 0; with 2000 nodes, the aggregation response latency is 185.1ms. The cumulative error is 0; when the number of nodes is 2500, the aggregation response latency is 201.5ms, and the cumulative error is 0; when the number of nodes is 3000, the aggregation response latency is 228.7ms, and the cumulative error is 0; when the number of nodes is 3500, the aggregation response latency is 254.3ms, and the cumulative error is 0; when the number of nodes is 4000, the aggregation response latency is 279.8ms, and the cumulative error is 0; when the number of nodes is 4500, the aggregation response latency is 305.6ms, and the cumulative error is 0; when the number of nodes is 5000, the aggregation response latency is 331.2ms, and the cumulative error is 0.

[0030] Experimental results show that during the ciphertext aggregation process, as the number of terminal nodes increases, the computational load of the aggregated ciphertext remains basically constant. The original data of each node is accurately extracted after decryption. The above data verifies the effectiveness of the coupling between the data form modulation mechanism and the ciphertext state multiplication operation. Through the synergistic effect of bit segment isolation and ciphertext field accumulation, the relay aggregation node can complete global data statistics even without decryption permissions. Moreover, this processing mode controls the decryption computing power of the central node within a constant range, reducing the data transmission pressure in high-concurrency environments. This confirms the computational stability and data reconstruction accuracy of the encryption method in in-situ aggregation of multidimensional state parameters in complex network environments.

[0031] Example 3: This example combines Figures 1 to 2 The method of homomorphic encryption for sensitive data transmission is explained, such as... Figure 1As shown, the process includes step S101, whereby the lower-level node acquires the locally collected sensitive state parameters, performs a left shift operation on the sensitive state parameters according to the starting position of the binary bit segment corresponding to its management level and the effective bit segment length determined by the historical maximum extreme value, maps them into the corresponding binary bit segment, and adds a preset width of anti-overflow isolation bit to the high-order side of the binary bit segment to package and generate a combined plaintext data packet; step S102, whereby the lower-level node calls the public key of the additive homomorphic encryption algorithm to perform a modular exponentiation high-dimensional hiding calculation on the combined plaintext data packet to generate state ciphertext, and transmits the state ciphertext to the relay aggregation node through a multi-level network channel. In step S103, the relay aggregation node aggregates multiple state ciphertexts belonging to the same reporting period in the undecrypted ciphertext domain using homomorphic modular multiplication, generating a single aggregated ciphertext containing the global summation result of each binary bit segment, and transmits it upwards to the core decryption node; in step S104, the core decryption node calls the private key of the additive homomorphic encryption algorithm to perform a modular exponentiation operation on the single aggregated ciphertext to globally decrypt and recover the comprehensive plaintext value. Based on the segmentation separation rules consisting of the starting position and the effective bit segment length, a hierarchical mask matrix is ​​introduced to extract and separate each binary bit segment, restoring the global summation value corresponding to each management level, such as... Figure 2 As shown, the system acquires locally collected raw values ​​and performs normalization mapping and proportional scaling to generate sensitive state parameters. It determines the starting position of the binary bit segment, performs a left shift operation, and superimposes anti-overflow isolation bits on the high-order side to generate a combined plaintext data packet. It calls the public key to perform modular exponentiation high-dimensional hiding calculation on the combined plaintext data packet to output the state ciphertext. In the ciphertext domain, it aggregates the state ciphertexts belonging to the same reporting period in situ through homomorphic modular multiplication to generate a single aggregated ciphertext. It transmits the single aggregated ciphertext to the core decryption node to perform modular exponentiation inversion operation for global decryption to obtain the comprehensive plaintext value. It introduces a hierarchical mask matrix to extract and separate the comprehensive plaintext value to obtain the global cumulative value. When the transient change rate exceeds the change threshold and the variance fluctuation increases, it determines that there is an abnormal injection disturbance in the channel and outputs a channel abnormality warning signal.

[0032] Example 4: In a multi-level concurrent transmission scenario of a distributed smart library linkage service network, the provincial center platform performs real-time aggregation and statistics on sensitive reading status data uploaded by school-level terminal application nodes. When each level of sub-platform and lower-level node is in a high-frequency acquisition state, the sensitive status parameters of different levels of nodes fluctuate. During the process of the data integration module performing homomorphic modular multiplication in-situ aggregation of the state ciphertext using relay aggregation nodes, the accumulated value in the algebraic domain of the ciphertext exhibits non-linear growth, leading to data overflow in the lower-order segments and algebraic pollution in adjacent higher-order segments. This causes technical problems such as data bit misalignment and data distortion when the core decryption node reverses the decomposition. To clarify the processing path and eliminate... To mitigate the impact of bit interleaving, the method claimed in this invention employs a hierarchical mask matrix-based extraction procedure at the core decryption node. This core decryption node is deployed in a provincial data center, comprising a central processing unit (CPU) with floating-point capabilities and a dedicated storage array. Upon receiving a single aggregated ciphertext generated from the convergence of multiple network transmission links, the CPU invokes a pre-stored private key to perform a global decryption operation on the single aggregated ciphertext, generating a plaintext value containing the superposition of multi-dimensional state parameters. At this point, the hierarchical mask matrix acts as a control mechanism, intervening in the first preset time sequence. The storage array, based on historical maximum extreme values ​​and management hierarchy planning, pre-stores multiple independent and non-overlapping binary... The binary logic windows are formed by an array of AND gates. For the binary bit segments corresponding to the first management level, the central processing unit extracts a preset data segment from the 32nd to the 40th bit of the plaintext value. This preset data segment is then combined with a preset first-level mask operator to fill all corresponding bits with binary 1s and all other bits with binary 0s in a dedicated logic matrix. In-place AND gate operations are then performed. During this operation, to address the shift attenuation caused by modular multiplication and accumulation, the system automatically introduces carry-in parameters carried in the encrypted header by lower-level nodes as quantization compensation operators, which are superimposed on the output state of the AND gate operation to correct for the shift attenuation caused by data non-multiplication. The low-bit deviation caused by linear accumulation is addressed by the central processing unit switching to the preset data segment from bit 41 to bit 48 for the second management level. The second-level mask operator is then used to perform the same in-situ logic AND gate operation. In more detail, the carry record parameter carried by the lower-level node in the encrypted header is generated synchronously by the microprocessor of the lower-level node by monitoring the overflow register status of the fixed-point quantization arithmetic operation during the stage of generating combined plaintext data packets from local data. This carry record parameter uses a two-bit structured scalar, where the first bit represents the carry flag to the higher bit when the current bit segment undergoes algebraic accumulation, and the second bit stores the residual order when the carry occurs.This parameter is written into a customized reserved field in the multi-level network protocol header corresponding to the plaintext data packet. During long-link relay transmission, this header field flows along with the state ciphertext in plaintext form without private key encryption. This allows relay aggregation nodes to read and update the ciphertext algebraically based on the number of concurrent nodes when performing homomorphic modular multiplication. Finally, it is sent to the core decryption node along with the single aggregate ciphertext. After reading this header, the central processing unit directly converts its value into the corresponding binary quantization compensation matrix, accurately compensating for the algebraic penetration caused by low-bit interleaving and superposition on the high-bit window.

[0033] Through this hierarchical masking separation and extraction procedure, the multidimensional data stream, which was originally mixed and superimposed in a single comprehensive plaintext value, is accurately discretized and decomposed. Without key leakage, independent physical accumulation values ​​of each management level are generated. Even under the condition of 5000 concurrent nodes, the cooperation between the anti-overflow isolation bit and the hierarchical masking operator prevents data overflow from penetrating into adjacent bit segments, so that the aggregation accuracy of global reading state data reaches 100%. The neutral technical indicators output by the system at the end directly serve as the basis for the intelligent decision-making module. While ensuring the security of sensitive data in each sub-platform, it completes the value closed loop from in-situ dynamic aggregation of ciphertext to full-link data security governance.

[0034] Example 5: When the system faces the situation of the school-level terminal application node accessing the linkage service network for the first time, due to the difference between the node hardware clock and the channel noise, the hidden calculation of the state ciphertext is accompanied by the baseline delay deviation, which increases the verification load in the global data governance. When establishing the initial boundary, the central processing unit distributes the pre-calibration signal to the lower-level node, controls the lower-level node to send 100 empty data packets containing feature codes to the relay aggregation node. The relay aggregation node calculates the average transmission time based on the arrival time interval and writes the average transmission time as the reference offset into the storage array. Subsequently, in the stage of constructing the combined plaintext data packet, the central processing unit uses the reference offset as the hedging operator to apply the bit segment isolation rule. When left shifting and mapping, the shift depth corresponding to the reference offset is deducted in advance to correct the starting point of the hidden calculation, so that the single comprehensive plaintext value remains in the bit segment aligned state before the modular exponentiation calculation.

[0035] When the linkage service network is continuously operating and its transmission characteristics evolve due to device thermal drift, the system activates a dynamic timeliness correction unit. The relay aggregation node uses a fixed-length sliding time window to periodically acquire the response delay of 50 consecutive single aggregated ciphertexts and calculate the variance of the response delay. When the variance exceeds a first preset threshold, the central processing unit calls the fault-tolerant compensation operator. By adjusting the preset width of the anti-overflow isolation bit and synchronously changing the hierarchical mask matrix of the core decryption node, the ciphertext state multiplication and accumulation operation maintains boundary convergence in a fluctuating channel environment. The homomorphic domain aggregation numerical error of the multidimensional state parameters remains zero, avoiding algebraic interference caused by misalignment of sensitive reading state data. This keeps the concurrent transmission pressure at a low load. Based on the distributed communication state synchronization and protocol-driven register control principle, cross-node target parameter synchronous updates prevent data decomposition bit address misalignment. The relay aggregation node broadcasts a bit-width adjustment instruction to each lower-level node and synchronously sends a bit-width adjustment instruction, including an adjustment factor representing the binary bit-width increment, to the core decryption node deployed in the data center and including a vector processor. The core decryption node receives the bit width adjustment command and parses it to extract the adjustment factor. The binary extraction code in the hierarchical mask matrix is ​​recalculated, and the core decryption node extends the high-bit boundary of the binary extraction code corresponding to the original target management level to the higher bit side. 1 bit, of which the adjustment factor The dimension is one, and the value range is 2 to 4. The core decryption node stores the updated hierarchical mask matrix into the local storage array. It switches to take effect at the end of the current reporting cycle and is used to separate and extract the subsequently received single aggregate ciphertext through bitwise AND logical operations, maintaining the timing synchronization and data boundary convergence under the multi-node bit width adjustment state.

[0036] Example 6: When the system faces continuous high-load operation of the hierarchical interconnected network and a sudden surge in concurrent traffic to the rated boundary, bit signal offsets occur within the interconnected network due to transmission jitter. This leads to data misalignment during modular multiplication and aggregation of the state ciphertext. To determine the bit parameters and solidify the response boundary of the additive homomorphic encryption architecture, the central processing unit runs a gradient test program in a verification environment. The verification environment includes an arithmetic unit with 64-bit registers, and its simulation generates Gaussian white noise with a signal-to-noise ratio of 20dB to simulate wide area network channel interference. While keeping other network parameters unchanged, the verification... The module increases the physical width of the anti-overflow isolation bit from 2 bits to 12 bits in 1-bit increments, and simultaneously changes the length of the valid state bit segment from 4 bits to 16 bits. It records the error rate of decrypted plaintext and the computation time of the relay aggregation node. The trend shows that when the anti-overflow isolation bit is widened from 2 bits to 6 bits, the phenomenon of bit contamination caused by accumulated overflow bits gradually weakens, and the global accumulated value error rate of the decrypted output decreases to 0. However, when the anti-overflow isolation bit continues to expand to 12 bits, the total word length of the hardware registers limits the number of states that can be accommodated in a single combined plaintext data packet. The reduction of the number of state parameters from 4 to 2 caused a sudden decrease in the channel utilization efficiency of a single encryption. Based on the bidirectional constraint causal relationship, the system determined the fixed word length of the overflow prevention isolation bit to be 6 bits, and the fixed word length of the effective state bit segment to be 10 bits, thereby determining a stable working window between preventing overflow risk and maintaining channel capacity. When the parameter combination was subjected to a continuous 72-hour test in a full-load transmission simulation including 5000 nodes, the linkage network continuously processed 360 million combined plaintext data packets. The monitoring unit obtained the performance indicators during the transmission process, and the results showed that the relay aggregation node The homomorphic modular multiplication calculation latency is maintained within the range of 325.2ms to 331.8ms, and does not exhibit linear overflow with the increase in the number of concurrent nodes. The core decryption node uses a hierarchical mask matrix to extract the plaintext values ​​of the global decryption output in segments. The accumulated state results of each school-level terminal application node are the same as the original plaintext input matrix, and the accumulated value error of the entire link remains at 0. This monotonically convergent state of the value on the time axis and the constant distribution of the transmission load maintain the transmission stability and reproducibility of the additive homomorphic encryption architecture in the face of channel jitter and high-concurrency traffic peak shaping.

[0037] In situations where a multi-level interconnected network faces a sudden surge of large-scale status data from tens of thousands of campus-level terminal application nodes, uneven distribution of receiving bandwidth among the independent processing units within the relay aggregation node cluster leads to load imbalances in each dedicated buffer queue and induces local computing power congestion, disrupting the timing synchronization of homomorphic modular multiplication in-situ aggregation. To address this, the system deploys a multi-unit collaborative management procedure at the front end of the multi-level network channel. The provincial center platform collects the residual word length of the dedicated buffer queues from each relay aggregation node in real time and dynamically compares it with preset capacity thresholds. Specific relay aggregation nodes with residual word lengths exceeding a first preset threshold are selected as the current enabled targets. Simultaneously, the provincial center platform, based on each... The remaining word length of the current dedicated buffer queue of a unit is used to calculate the corresponding load saturation. Based on the inverse ratio of the load saturation, the generated state ciphertexts are distributed through route discretization and amortization, ensuring that the state ciphertexts maintain a reasonable traffic distribution state in the hardware channels within the cluster. When a specific relay aggregation node detects that the remaining word length of its own dedicated buffer queue is less than the second preset threshold, it automatically triggers the control switching logic and distributes a feedback suppression signal to the front-end gateway unit through the data bus. This causes the front-end gateway unit to adaptively switch the subsequent state ciphertexts to the idle alternative communication channel. Thus, at the method level, dynamic control of data flow and online offsetting of transmission dissipation are achieved, maintaining the continuity and stability of global reading state data aggregation and statistics in a changing environment.

[0038] 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.

[0039] 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 homomorphic encryption method for sensitive data transmission, characterized in that, Includes the following steps: Step S101: The lower-level node obtains the sensitive state parameters collected locally. Based on the starting position of the binary bit segment corresponding to the management level and the effective bit segment length determined by the historical maximum extreme value, it performs a left shift operation on the sensitive state parameters to map the sensitive state parameters into the corresponding binary bit segment. It also adds a preset bit width anti-overflow isolation bit to the high bit side of the binary bit segment and packages it to generate a combined plaintext data packet. In step S102, the lower-level node calls the public key of the additive homomorphic encryption algorithm to perform a high-dimensional hidden calculation on the combined plaintext data packet to generate the state ciphertext, and transmits the state ciphertext to the relay aggregation node of the receiving end through the configured multi-level network channel. In step S103, the relay aggregation node aggregates multiple state ciphertexts belonging to the same reporting cycle in the undecrypted ciphertext field using homomorphic modular multiplication, and generates a single aggregated ciphertext containing the global accumulation result of each binary bit segment. The single aggregated ciphertext is then transmitted upward to the core decryption node at the center. In step S104, the core decryption node calls the private key of the additive homomorphic encryption algorithm to perform a modular exponentiation inverse operation on the single aggregate ciphertext to decrypt it globally, recovering the comprehensive plaintext value. Based on the segmentation and separation rules consisting of the starting position and the length of the effective bit segment, a hierarchical mask matrix is ​​introduced to extract and separate each binary bit segment, and restore the global cumulative value corresponding to each level of management.

2. The homomorphic encryption method for sensitive data transmission according to claim 1, characterized in that, Step S101 includes the following sub-steps: Step S1011, obtain the hierarchical code and cell number corresponding to the lower-level node, and query the topology structure tree stored in the system to determine the topology depth and branch weight of the lower-level node in the aggregation network; Step S1012, calculate the starting position of the binary bit segment allocated to the lower-level node when uploading based on the topology depth and branch weight, and reserve a binary safety gap in combination with the historical maximum extreme value as an anti-overflow isolation bit to complete the mapping combination of the combined plaintext data packets.

3. The homomorphic encryption method for sensitive data transmission according to claim 1, characterized in that, In step S102, the additive homomorphic encryption algorithm used is a public-key encryption algorithm based on the difficult problem of factoring large integer composite numbers. The lower-level node calls the public key to perform modular exponentiation and homomorphic mask product operation on the combined plaintext data packets to generate state ciphertext.

4. The homomorphic encryption method for sensitive data transmission according to claim 1, characterized in that, After receiving the ciphertext of the status sent by multiple lower-level nodes, the relay aggregation node performs a modular exponentiation operation on multiple ciphertexts belonging to the same reporting period in the ciphertext state without decryption by the private key. This is done to aggregate and combine the algebraic sum of the plaintext data packets in the corresponding binary bit fields in the ciphertext state, generating a single aggregated ciphertext.

5. The homomorphic encryption method for sensitive data transmission according to claim 1, characterized in that, In step S104, after the core decryption node recovers the comprehensive plaintext value by decrypting with the private key, it calls the locally stored hierarchical mask matrix, which contains binary extraction codes that correspond one-to-one with each level of management. The core decryption node performs a bitwise AND operation on the comprehensive plaintext value and the binary extraction code in sequence, clears the bit field data of non-target management level, and reverses the right shift to align and restore the operation result, thus separating and extracting the independent accumulated plaintext values ​​of each management level.

6. The homomorphic encryption method for sensitive data transmission according to claim 1, characterized in that, After the core decryption node extracts the global accumulated value corresponding to each level of management, it continuously stores the global accumulated value into the historical database according to the time series. It calculates the transient change rate and variance fluctuation of the global accumulated value between adjacent reporting periods. When the transient change rate continuously exceeds the set change threshold and the variance fluctuation shows a monotonically increasing trend, the core decryption node determines that there is an abnormal data flow injection disturbance in the current transmission channel and sends a channel abnormality warning signal to the corresponding network node.

7. The homomorphic encryption method for sensitive data transmission according to claim 1, characterized in that, The reporting period for the encrypted state of the lower-level nodes is 500ms to 2000ms, and the key length of the additive homomorphic encryption algorithm is greater than or equal to 1024 bits. The in-situ aggregation operation of the encrypted state of the relay aggregation node takes less than or equal to 50ms.

8. A homomorphic encryption method for sensitive data transmission according to claim 1, characterized in that, The relay aggregation node monitors the concurrent reporting traffic in the current aggregation channel. When the concurrent reporting traffic exceeds the traffic threshold, it broadcasts a bit width adjustment instruction to the lower-level node so that the lower-level node extends the binary bit width of the anti-overflow isolation bit by 2 bits to 4 bits to the higher bit side. The traffic threshold is dynamically calculated based on the sum of the average concurrent traffic of historical regular periods and three times the standard deviation.

9. A homomorphic encryption method for sensitive data transmission according to claim 1, characterized in that, Before constructing the combined plaintext data packet, the lower-level node performs normalization mapping on the collected sensitive state parameters, proportionally scaling the original values ​​of the sensitive state parameters to the range between 0 and 1, and then performs left shift mapping processing.

Images