An improved Paillier dynamic operator method and system

Abstract

This invention discloses an improved Paillier dynamic operator method and system, which relates to the intersection of cryptography and machine learning. The method involves dynamically adapting encryption algorithms and parameters based on data type sensitivity to construct a dynamic hierarchical encryption protocol; selectively encrypting sparse gradients to encrypt the gradient coordinates with the largest absolute value of the gradient vector; and performing periodic and frequency-band encryption encoding on agricultural time series data, optimizing encryption efficiency. The method reduces invalid encryption operations and improves encryption efficiency through a reverse workflow of first sparse and then encrypt. It also customizes encoding for agricultural data, reduces the amount of encrypted data, and features a quantum-resistant design that can withstand quantum computer attacks.

Description

Technical Field

[0001] The present invention relates to the interdisciplinary field of cryptography and machine learning, and in particular to an improved Paillier dynamic operator method and system. Background Art

[0002] When federated learning is applied to agricultural scenarios (such as multi-region agricultural forecasting and joint pricing models), data privacy is crucial. To meet security requirements, the industry often introduces homomorphic encryption technologies (such as Paillier and BFV) to implement encrypted computations, or combines them with sparsification mechanisms to reduce communication costs. However, due to the complexity and diversity of agricultural data types (including multidimensional data such as time series, geographic data, and price data), existing general encryption mechanisms face multiple challenges in terms of performance, flexibility, and compatibility.

[0003] Traditional homomorphic encryption introduces excessively high computational overhead in federated learning and cannot meet the real-time requirements of edge nodes. Existing solutions use fixed encryption parameters and cannot dynamically adapt to data of different sensitivity levels. There are compatibility issues between the encryption gradient and the sparse transmission mechanism, resulting in an effective information loss rate greater than 35%. There is a lack of customized encoding solutions for agricultural data characteristics. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problems solved by the present invention are: the existing homomorphic encryption methods have too high computational overhead, cannot dynamically adapt to data of different sensitivity levels, lack customized encoding for agricultural data characteristics, and how to solve the compatibility problem between encrypted gradients and sparse transmission mechanisms.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: an improved Paillier dynamic operator method, comprising dynamically adapting encryption algorithms and parameters based on data type sensitivity to construct a dynamic hierarchical encryption protocol; encrypting the gradient coordinates with the largest absolute value of the gradient vector through sparse gradient selective encryption; performing periodic and frequency-band encryption encoding on agricultural time series data, and optimizing the encryption operation efficiency; data types include land GIS data, crop yield data, and income data based on agricultural scenarios; parameters include the polynomial degree and scaling coefficient in the CKKS encryption operator, the plaintext modulus in the BFV encryption operator, and the key length in the Paillier encryption operator.

[0007] As a preferred solution of the improved Paillier dynamic operator method described in the present invention, the dynamic adaptive encryption algorithm includes constructing a sensitivity adaptive encryption mechanism, grading different data types according to their privacy sensitivity, and selecting encryption operators in real time through scene demand perception.

[0008] As a preferred solution of the improved Paillier dynamic operator method described in the present invention, the sensitivity-adaptive encryption mechanism includes automatically selecting an encryption strategy based on the privacy sensitivity of the data, and dynamically adjusting the polynomial degree, scaling factor, plaintext modulus and key length based on the sensitivity.

[0009] As a preferred solution of the improved Paillier dynamic operator method described in the present invention, the sparse gradient selective encryption includes fusing gradient sparsification and selective encryption mechanisms, constructing a Top-K gradient selection mask, performing homomorphic encryption operations only on the Top-K most important gradient coordinates, screening the part with the largest absolute value of the gradient vector through a binary mask, i.e., a mask operation, outputting the local encryption function using the mask operation after element-by-element encryption, combining the output function with the gradient coordinates of the part with the largest absolute value of the gradient vector, and generating a partially encrypted sparse gradient vector.

[0010] As a preferred solution of the improved Paillier dynamic operator method described in the present invention, the construction of the Top-K gradient selection mask includes selecting the coordinate positions of the top K maximum values ​​based on the absolute value of the gradient vector, generating a binary mask matrix, marking the positions corresponding to important gradients as 1, and marking the positions corresponding to unimportant gradients as 0, identifying the high-information gradient areas to be encrypted, and generating a gradient selection mask.

[0011] As a preferred solution of the improved Paillier dynamic operator method described in the present invention, the frequency-band encryption coding includes seasonal decomposition of agricultural time series data, encryption of seasonal component components after seasonal decomposition of agricultural time series data based on Fourier transform, extraction of data with seasonal fluctuations, and direct transmission of trend component components after seasonal decomposition of agricultural time series data in plain text.

[0012] As a preferred solution of the improved Paillier dynamic operator method described in the present invention, the seasonal decomposition includes splitting the agricultural time series data according to the season, splitting it into trend components, seasonal components and residual components, and performing homomorphic encryption operations only on the short-term periodic repetitive fluctuation part of the seasonal component.

[0013] As a preferred solution of the improved Paillier dynamic operator method described in the present invention, the optimization of encryption operation efficiency includes introducing CUDA parallel acceleration for large-scale matrix multiplication operations under homomorphic encryption performed during the sparse gradient selective encryption process, and adopting the physical unclonable function PUF interface of the key distribution chip and the encrypted traffic compression encoder.

[0014] As a preferred solution of the improved Paillier dynamic operator method described in the present invention, the key distribution includes hierarchical distribution of keys, generating a master key through a central coordinator, generating hierarchical sub-keys in the key distribution stage, using bilinear pairing to verify the consistency of the ciphertext, and processing data interference through the monotonicity of the exponential function.

[0015] Another object of the present invention is to provide an improved Paillier dynamic operator system that can encrypt the gradient coordinates with the largest absolute value of the gradient vector through selective encryption of sparse gradients, thereby solving the compatibility problem between the encrypted gradients and the sparse transmission mechanism in the current homomorphic encryption technology.

[0016] As a preferred solution of the improved Paillier dynamic operator system described in the present invention, it includes: a sensitivity adaptation module, a sparse gradient selective encryption module, and a time series data encoding optimization module. The sensitivity adaptation module is used to dynamically select appropriate encryption algorithms and parameter configurations according to the privacy sensitivity of different data types, and bind the data encryption strategy to the usage scenario and sensitivity through the defined data sensitivity-encryption scheme mapping rule; the sparse gradient selective encryption module is used to perform Top-K importance screening on the gradient by integrating gradient sparsification and local homomorphic encryption technology, encrypt the coordinates of the part with the largest absolute value of the gradient vector, construct a mask matrix based on the absolute value of the gradient, mark the Top-K important coordinates, and use The public key performs homomorphic encryption on the selected coordinates to generate an encrypted sparse gradient; the time series data encoding optimization module includes an agricultural time series data encoding module and a hardware acceleration optimization module. The agricultural time series data encoding module performs hierarchical encryption on data with seasonal fluctuations in crop prices, decomposes the time series into trend components, seasonal components and residual components, performs Fourier transform on seasonal components with significant periodicity, extracts high-frequency components, and encrypts the high-frequency components using a homomorphic encryption method. The trend component is directly transmitted in plain text. The hardware acceleration optimization module targets high-load matrix operations under homomorphic encryption, and uses a CUDA parallel acceleration solution to distribute keys hierarchically, uses bilinear pairing to verify the consistency of the ciphertext, and processes interference data.

[0017] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of an improved Paillier dynamic operator method.

[0018] A computer-readable storage medium stores a computer program, which implements the steps of an improved Paillier dynamic operator method when executed by a processor.

[0019] Beneficial effects of the present invention: An improved Paillier dynamic operator method provided by the present invention dynamically adapts encryption algorithms and parameters based on data type sensitivity, constructs a dynamic hierarchical encryption protocol, achieves a balance between security and operational efficiency, integrates gradient sparsification and local homomorphic encryption technology, encrypts some gradient coordinates with high information volume, reduces communication and computing overhead, avoids resource waste, performs periodic and frequency-segmented encryption encoding on agricultural time series data, and provides customized encoding for agriculture, taking into account both security and transmission efficiency. The present invention achieves better results in dynamic hierarchical encryption, customized agricultural coding, and hardware acceleration optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 An overall flow chart of an improved Paillier dynamic operator method provided for the first embodiment of the present invention.

[0022] Figure 2 A sensitivity evaluation diagram of an improved Paillier dynamic operator method provided in the second embodiment of the present invention.

[0023] Figure 3 A dynamic parameter decision diagram of an improved Paillier dynamic operator method provided in the second embodiment of the present invention.

[0024] Figure 4 A gradient sparse mask matrix diagram of an improved Paillier dynamic operator method provided in the second embodiment of the present invention.

[0025] Figure 5 An encryption-federation collaborative architecture diagram of an improved Paillier dynamic operator method provided in the second embodiment of the present invention.

[0026] Figure 6 An overall schematic diagram of an improved Paillier dynamic operator system provided by the third embodiment of the present invention. DETAILED DESCRIPTION

[0027] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0028] Example 1, reference Figure 1 , as an embodiment of the present invention, provides an improved Paillier dynamic operator method, comprising:

[0029] S1: Dynamically adapt encryption algorithms and parameters based on the sensitivity of data types to build a dynamic hierarchical encryption protocol.

[0030] Furthermore, the dynamically adaptive encryption algorithm includes building a sensitivity-adaptive encryption mechanism, grading different data types according to their privacy sensitivity, and selecting encryption operators in real time through sensing the needs of the scenario.

[0031] It should be noted that the sensitivity-adaptive encryption mechanism includes automatically selecting encryption strategies based on the privacy sensitivity of the data, and dynamically adjusting the polynomial degree, scaling factor, plaintext modulus, and key length based on the sensitivity.

[0032] The sensitivity adaptive encryption mechanism is expressed as:

[0033] ;

[0034] in, It is a sensitivity adaptive encryption mechanism. For input data, There are three encryption operators: is the degree of the polynomial, is the scaling factor, is the plaintext modulus, is the key length, For land GIS data, For crop yield, For crop prices, the sensitivity-adaptive encryption mechanism is to strongly bind the data encryption strategy with the usage scenario and sensitivity, which can achieve the role of dynamic adjustment of security overhead.

[0035] It should also be noted that by constructing a dynamic hierarchical encryption protocol, based on a sensitivity-adaptive encryption mechanism, and dynamically selecting the optimal encryption operator and adjusting key parameters according to the privacy sensitivity of different data types, a strong binding between security policies and scenario requirements is achieved, solving the problem that traditional static encryption methods are difficult to balance security, efficiency and scenario adaptability. Without affecting system performance, security overhead is dynamically optimized to ensure strong encryption of highly sensitive data while avoiding excessive encryption of low-sensitivity data, achieving the best balance between precise protection and computing efficiency.

[0036] S2: Through sparse gradient selective encryption, the gradient coordinates with the largest absolute value of the gradient vector are encrypted.

[0037] Furthermore, sparse gradient selective encryption involves integrating gradient sparsification with a selective encryption mechanism. By constructing a Top-K gradient selection mask, homomorphic encryption is performed only on the Top-K most important gradient coordinates. The masking operation is performed by filtering the gradient vector with the largest absolute value through a binary mask. After element-by-element encryption, the masking operation is used to output the local encryption function. The output function is combined with the gradient coordinates with the largest absolute value of the gradient vector to generate a partially encrypted sparse gradient vector, which can be expressed as:

[0038] ;

[0039] in, is the generated encrypted sparse gradient, is a local encryption function, Is the sensitive data to be encrypted, is the public key, It is represented as the mask operation after element-by-element encryption. The matrix is ​​selected for the Top-K positions, and encryption operations are performed only on the gradients with high information content to avoid resource waste.

[0040] It should be noted that constructing the Top-K gradient selection mask includes selecting the coordinate positions of the top K maximum values ​​based on the absolute value of the gradient vector, generating a binary mask matrix, marking the positions corresponding to important gradients as 1, and marking the positions corresponding to unimportant gradients as 0, identifying the high-information gradient areas to be encrypted, and generating a gradient selection mask, which is expressed as:

[0041] ;

[0042] in, is the mask vector elements, After taking the absolute value of the vector, select the one with the largest absolute value. An indexed collection of elements, is the original vector.

[0043] It should also be noted that through the binary mask matrix and element-by-element encryption mask operation, a partially encrypted sparse gradient vector is generated, which significantly reduces the computational and communication overhead. By combining local encryption with sparsification, the resource waste problem existing in traditional full gradient encryption is solved. While ensuring the security of sensitive data, the encryption transmission efficiency is improved, achieving a balance between security and efficiency.

[0044] S3: Perform periodic and frequency-segmented encryption encoding on agricultural time series data, and optimize encryption operation efficiency.

[0045] Furthermore, the frequency-band encryption coding includes seasonal decomposition of agricultural time series data, encryption of the seasonal component components after seasonal decomposition of agricultural time series data based on Fourier transform, extraction of data with seasonal fluctuations, and direct transmission of the trend component components after seasonal decomposition of agricultural time series data in plain text.

[0046] Agricultural time series data is represented as:

[0047] ;

[0048] in, is the original time series data, at time point The actual data observed, It is the part of the trend component data that changes slowly over time. is the seasonal component, the part of the data that fluctuates periodically. It is the residual component, which is the random fluctuation, noise or anomaly in the data other than trend and seasonality.

[0049] It should be noted that seasonal decomposition includes splitting agricultural time series data according to seasonal cycles, splitting them into trend components, seasonal components and residual components, and only performing homomorphic encryption operations on the short-term periodic repeated fluctuations in the seasonal components.

[0050] It should also be noted that optimizing encryption operation efficiency includes introducing CUDA parallel acceleration for large-scale matrix multiplication operations under homomorphic encryption performed during the sparse gradient selective encryption process, and adopting the physical unclonable function PUF interface and encrypted traffic compression encoder.

[0051] It should also be noted that the key distribution includes hierarchical division of keys, generation of master keys by a central coordinator, generation of hierarchical sub-keys in the key distribution stage, verification of ciphertext consistency by bilinear pairing, and processing of data interference by exponential function monotonicity.

[0052] Hierarchical key distribution is expressed as:

[0053] ;

[0054] ;

[0055] in, For the generated key, is a pseudo-random function, is the master key, The user's regional node ID. is the timestamp, For the generated key, A function for generating a secondary key.

[0056] It should also be noted that through seasonal decomposition, a periodic and frequency-segmented encryption encoding method is proposed for agricultural time series data, and only the short-term periodic repeated fluctuation seasonal components are homomorphically encrypted, which solves the redundancy problem of full encryption calculation of traditional time series data. Combined with CUDA parallel acceleration to optimize encryption calculation, hierarchical key distribution is used to improve key management efficiency, optimize encryption efficiency and security, and significantly reduce encryption overhead while ensuring the privacy of agricultural data, realizing efficient, secure and coordinated time series data protection.

[0057] Example 2, reference Figure 2-Figure 5 , which is an embodiment of the present invention, provides an improved Paillier dynamic operator method. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0058] First, the experiment set up the same hardware environment and used the same set of standard machine learning tasks as the verification platform. Based on the MNIST handwriting dataset, the differences between the traditional Paillier and improved Paillier encryption schemes were verified. The experiment set up two comparison schemes, one for the control group and the other for the experimental group. The experimental group and the control group only differed in the encryption module, and the other parameters were completely the same. The experimental group evaluated the sensitivity of the data from multiple dimensions, such as Figure 2 Including privacy level, data value, legal requirements, attack cost, timeliness, the score of each dimension represents sensitivity, the higher the score, the stronger the sensitivity. Figure 3 The experimental group in this experiment dynamically adapts the encryption algorithm and parameters to select or compare different encryption schemes, mainly including CKKS, BFV and Paillier.

[0059] This experiment consists of four stages: the encryption stage, in which the training data gradient is encrypted using the traditional Paillier and improved Paillier respectively, the time required for encryption is recorded, and the average value is taken as the encryption time consumption indicator; the communication stage, in which the actual amount of communication data generated each time the model parameters are updated is recorded and normalized, and the relative bandwidth occupancy ratio of the improved scheme is calculated based on the standard of "the bandwidth occupied by the traditional scheme is 1x"; Figure 4 In the gradient information retention evaluation phase, after decrypting the encrypted gradient, the cosine similarity is compared with the unencrypted gradient. The similarity reflects the gradient information retention. The average similarity of all gradient vectors is selected as the information retention rate indicator. In the model accuracy evaluation phase, the complete model is trained under two sets of schemes respectively, and the final classification accuracy is evaluated on the test set. The encryption-federation collaborative architecture of the experimental group is as follows. Figure 5 As shown, a modular, layered layout is employed to form a dual-track operational system. Terminal data is uploaded to the central hub after triggering security reinforcement at regional nodes. Simultaneously, encryption policies are distributed and executed downward through hierarchical channels, combining the global nature of centralized control with distributed protection. Using the unencrypted training accuracy as a reference, the percentage of accuracy loss for each scheme was calculated. The experimental results are recorded and analyzed in Table 1.

[0060] Table 1 Experimental results record

[0061] index Paillier Improved Paillier Encryption takes time 48.7 12.2 Communication bandwidth usage 1x 0.38x Gradient information retention rate 61% 88% Model accuracy loss 5.2% 0.7%

[0062] As can be seen from the table, the improved Paillier encryption scheme of the present invention achieves a better balance than the traditional Paillier encryption scheme. First, the encryption efficiency is significantly improved, and the encryption time is reduced from 48.7 seconds to 12.2 seconds, an increase of about 4 times, which significantly shortens the training delay and reduces the computational delay of frequent gradient encryption in federated learning. Secondly, the communication bandwidth is greatly reduced, and the amount of communication data is reduced to 38% of the traditional scheme, which effectively reduces the network load in federated learning or distributed training and reduces the risk of network congestion. Thirdly, the gradient information is more complete, and the gradient information retention rate is increased from 61% to 88%, retaining more effective training information and improving model recoverability and training stability. Then, the model accuracy loss is smaller. Under the premise of ensuring data privacy, the model accuracy loss is reduced from 5.2% to 0.7%, which verifies that the encryption process has little impact on model performance.

[0063] In summary, the improved Paillier encryption method can ensure high privacy protection while taking into account high efficiency and high precision. It has comprehensive performance that is superior to existing technologies and can effectively support the security training needs in actual agricultural credit applications.

[0064] Example 3, reference Figure 6, which is an embodiment of the present invention, provides an overall process of an improved Paillier dynamic operator system, including a sensitivity adaptation module 100, a sparse gradient selective encryption module 200, and a time series data encoding optimization module 300.

[0065] Among them, S4: The sensitivity adaptation module 100 is used to dynamically select appropriate encryption algorithms and parameter configurations based on the privacy sensitivity of different data types, and bind the data encryption strategy with the usage scenario and sensitivity through the defined data sensitivity-encryption scheme mapping rules.

[0066] Among them, S5: Sparse gradient selective encryption module 200 is used to perform Top-K importance screening on gradients by integrating gradient sparsification and local homomorphic encryption technology, encrypt only coordinates with high information content, construct a mask matrix based on the absolute value of the gradient, mark the Top-K important coordinates, use the public key to perform homomorphic encryption on the selected coordinates, and generate encrypted sparse gradients.

[0067] Among them, S6: the time series data encoding optimization module 300 includes an agricultural time series data encoding module 301 and a hardware acceleration optimization module 302. The agricultural time series data encoding module 301 performs hierarchical encryption on data with seasonal fluctuations in crop prices, decomposes the time series into trend components, seasonal components and residual components, performs Fourier transform on seasonal components with significant periodicity, extracts high-frequency components, and encrypts the high-frequency components using homomorphic encryption methods. The trend components are directly transmitted in plain text. The hardware acceleration optimization module 302 targets high-load matrix operations under homomorphic encryption, based on the CUDA parallel acceleration solution, through hierarchical key distribution, and uses bilinear pairing to verify the consistency of the ciphertext and process interference data.

[0068] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the method of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0069] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0070] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting, or processing it in another suitable manner as necessary, and then storing it in a computer memory.

[0071] It should be understood that various aspects of the present invention may be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods may be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one or a combination of the following technologies known in the art may be used: a discrete logic circuit having logic gates for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gates, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc. It should be noted that the above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to be limiting. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art will understand that modifications or equivalent substitutions may 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, and such modifications are intended to be encompassed by the claims of the present invention.

Claims

1. An improved Paillier dynamic operator method, characterized in that: include: Dynamically adapt encryption algorithms and parameters based on the sensitivity of data types to build a dynamic hierarchical encryption protocol. Specifically, it includes: dynamically selecting encryption operators and adjusting the parameters involved in the selected operators according to the sensitivity of the data; Through sparse gradient selective encryption, the gradient coordinates with the largest absolute value of the gradient vector are encrypted. Specifically, this includes: integrating gradient sparsification and selective encryption mechanisms, constructing a Top-K gradient selection mask, performing homomorphic encryption operations only on the Top-K most important gradient coordinates, screening the part of the gradient vector with the largest absolute value through a binary mask, performing a mask operation, and using the mask operation after element-by-element encryption to output the local encryption function. The output function is combined with the gradient coordinates with the largest absolute value of the gradient vector to generate a partially encrypted sparse gradient vector; The agricultural time series data is encrypted and encoded periodically and in frequency segments, specifically including: seasonal decomposition of agricultural time series data, splitting agricultural time series data according to seasons into trend component, seasonal component and residual component, encrypting the seasonal component based on Fourier transform, and transmitting the trend component directly in plain text; the seasonal component is the part of the data that fluctuates periodically, the trend component is the part of the data that changes slowly over time, and the residual component is the random fluctuation, noise or anomaly in the data; Among them, the data types include land GIS data, crop yield data and income data based on agricultural scenarios; the parameters include the polynomial degree and scaling factor in the CKKS encryption operator, the plaintext modulus in the BFV encryption operator, and the key length in the Paillier encryption operator.

2. The improved Paillier dynamic operator method according to claim 1, wherein: The construction of the Top-K gradient selection mask includes: Based on the absolute value of the gradient vector, the coordinate positions of the top K maximum values ​​are selected to generate a binary mask matrix. The positions corresponding to important gradients are marked as 1, and the positions corresponding to unimportant gradients are marked as 0. This identifies the high-information gradient areas that need to be encrypted and generates a Top-K gradient selection mask.

3. An improved Paillier dynamic operator system, characterized by: It includes a sensitivity adaptation module (100), a sparse gradient selective encryption module (200), and a time series data encoding optimization module (300); The sensitivity adaptation module (100) is used to dynamically adapt encryption algorithms and parameters based on the sensitivity of data types to construct a dynamic hierarchical encryption protocol, specifically including: dynamically selecting encryption operators and adjusting parameters involved in the selected operators according to the sensitivity of the data; The sparse gradient selective encryption module (200) is used to encrypt the gradient coordinates with the largest absolute value of the gradient vector through sparse gradient selective encryption, specifically comprising: fusing gradient sparsification and selective encryption mechanisms, constructing a Top-K gradient selection mask, performing homomorphic encryption operations only on the Top-K most important gradient coordinates, screening the part with the largest absolute value of the gradient vector through a binary mask, performing a mask operation, outputting a local encryption function using the mask operation after element-by-element encryption, combining the output function with the gradient coordinates of the part with the largest absolute value of the gradient vector, and generating a partially encrypted sparse gradient vector; The time series data encoding optimization module (300) is used to perform periodic and frequency-segmented encryption encoding on agricultural time series data, specifically comprising: seasonal decomposition of agricultural time series data, splitting the agricultural time series data according to seasons into trend component components, seasonal component components, and residual component components, encrypting the seasonal component components based on Fourier transform, and directly transmitting the trend component components in plain text; wherein the seasonal component components are the parts of the data that fluctuate periodically, the trend component components are the parts of the data that change slowly over time, and the residual component components are random fluctuations, noise, or anomalies in the data; Among them, the data types include land GIS data, crop yield data and income data based on agricultural scenarios; the parameters include the polynomial degree and scaling factor in the CKKS encryption operator, the plaintext modulus in the BFV encryption operator, and the key length in the Paillier encryption operator.

4. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the improved Paillier dynamic operator method according to any one of claims 1 to 2 are implemented.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the improved Paillier dynamic operator method according to any one of claims 1 to 2 are implemented.

Images