A homomorphic encryption protection system for flower consumer private image features

Abstract

The application relates to the technical field of homomorphic encryption, and discloses a homomorphic encryption protection system for flower consumer privacy image features, which comprises the following steps: acquiring the historical actual quality of interactive flowers, the apoptosis metabolic rate constant and the absolute elapsed time, constructing a multi-dimensional initial continuous data tensor which is objectively attenuated with time, mapping the tensor into equivalent algebraic potential energy within the limit of the truncation order and performing dimension orthogonal elimination, combining Gaussian bottom noise to be packaged into polynomial ciphertext, using fitness group intelligence optimization to obtain optimal evolution damping, fusing the elapsed time to output stable state ciphertext under the cipher state, then performing orthogonal inner product with external seasonal market ciphertext, and calling a public key to perform algebraic dimension reduction convergence, and finally using an isolated private key to crack and reverse restore into a preference concentration on the terminal side. The application effectively suppresses the cipher state noise divergence caused by multi-dimensional asynchronous metabolism, and realizes the safe evolution and accurate evaluation of the privacy elements according to the natural law.

Description

Technical Field

[0001] This invention relates to the field of homomorphic encryption technology, and more specifically, to a homomorphic encryption protection system for the privacy profile characteristics of flower consumers. Background Technology

[0002] With the profound development of my country's digital economy, the flower consumption market has completed a comprehensive transformation from traditional offline stores to online digital business models. Emerging models such as flower e-commerce, community group buying, subscription services, and instant retail are experiencing explosive growth, driving the industry's market size to grow rapidly. According to publicly available industry data, the transaction volume of China's flower e-commerce market exceeded 120 billion yuan in 2025, with over 320 million online consumers. Fresh-cut flower consumption has also gradually shifted from traditional holiday and ritualistic consumption to everyday, therapeutic, and essential household consumption. Against this industry backdrop, consumer behavior data across the entire value chain—including browsing, saving, purchasing, and repurchase—of different flower categories such as roses, lisianthus, orchids, and succulents, along with corresponding consumption frequency, average order value, and scenario preferences, has become a core data asset for flower retail platforms to achieve precise marketing, personalized recommendations, supply chain optimization, and user lifecycle management.

[0003] It should be noted that the profile data of flower consumers is not ordinary behavioral data, but rather highly sensitive personal information. This type of data can not only directly reflect consumers' spending power, consumption habits, and aesthetic preferences, but also indirectly deduce consumers' lifestyle, emotional needs, holiday plans, family structure, and other private information. As sensitive personal information, its collection, storage, processing, and transfer are all subject to strict compliance supervision.

[0004] Traditional consumer data privacy protection methods are no longer adequate for the current business needs of the flower retail industry. Conventional solutions such as data anonymization and masking suffer from a trade-off between security and usability: overly coarse-grained anonymization can render profile data completely worthless, failing to support accurate personalized recommendations; overly fine-grained anonymization cannot withstand re-identification attacks from big data correlation analysis, making true privacy and security difficult to achieve. Furthermore, technologies such as multi-party secure computation and federated learning, in high-concurrency, low-latency online real-time recommendation scenarios, suffer from high multi-node communication overhead, low computational efficiency, and high deployment costs, making them unsuitable for the thousands of matching requests per second required by e-commerce platforms.

[0005] Against this backdrop, homomorphic encryption technology, with its core characteristic of directly performing algebraic operations in encrypted form and ensuring that the decrypted result is completely identical to the plaintext result, has become the optimal technical path for protecting the privacy profiles of flower consumers. Based on homomorphic encryption, platforms can complete the matching calculation between user profiles and seasonal market products without ever accessing users' plaintext privacy data, truly achieving data usability without visibility, fully preserving the business value of the data while meeting regulatory compliance requirements. However, in actual industry implementation, existing homomorphic encryption solutions face two major technical challenges in adapting to the flower consumption scenario, severely restricting the large-scale application of the technology.

[0006] First, existing flower consumer profiling systems generally employ static modeling logic, completely detaching flowers from their biological evolutionary attributes as living organisms, resulting in a severe disconnect between the profiling and users' actual needs. After cut flowers are separated from the parent plant, their ornamental value continuously declines due to physiological mechanisms such as water loss and ethylene autocatalytic reactions, following a first-order nonlinear reaction kinetic law. Furthermore, the apoptosis metabolic rate of different flower varieties exhibits extreme heterogeneity: the apoptosis metabolic rate constant for newly blooming roses can reach 0.25 / day, with a vase life of only 3-5 days; while the metabolic rate constant for Phalaenopsis orchids is only 0.03 / day, with a viewing period of several months—a difference of nearly an order of magnitude. Correspondingly, a user's consumption preference for a particular flower variety is not a fixed static constant, but a dynamic variable strongly correlated with the natural apoptosis pattern of that variety. That is, after purchasing cut roses, a user's immediate preference for roses rapidly declines as the flowers wither; while after purchasing Phalaenopsis orchids, the preference weight gradually decreases over time. However, existing profiling systems either use a fixed time decay window or manually set a uniform weight decay coefficient, which does not match the biological metabolic characteristics of different flower categories. This not only leads to distorted user profiles but also frequently results in ineffective marketing by repeatedly pushing the same type of products to users immediately after they have made a purchase, seriously damaging user experience and platform conversion efficiency.

[0007] Second, existing homomorphic encryption techniques cannot adapt to the dynamic nonlinear decay scenario of flower preferences, exhibiting fatal flaws such as geometric expansion of dense-state noise and imbalance in multi-dimensional noise accumulation. Homomorphic encryption's underlying operations are built upon polynomial algebraic rings, supporting only discrete addition and multiplication operations. To simulate the exponential decay of flower preferences over time, a Taylor series expansion must be used to convert the continuous exponential function into a discrete polynomial form to complete the computation in dense state. To ensure the accuracy of decay simulation, existing technologies often require increasing the truncation order of the Taylor expansion, which directly leads to an exponential increase in the number of polynomial multiplication operations. Each dense-state multiplication operation causes a geometric expansion of Gaussian noise in the ciphertext. More seriously, due to the heterogeneity of metabolic rates among different flower varieties, higher decay dimensions require higher expansion orders and more frequent dense-state multiplication operations, resulting in a much faster noise accumulation rate than lower decay dimensions, ultimately leading to a severe imbalance in the dense-state noise accumulation of multi-dimensional profiles. When the accumulated noise in a high-attenuation dimension exceeds the security threshold of the ciphertext modulus, noise overflow occurs, leading to completely incorrect decryption results and causing the collapse of the global cryptographic computation system. To avoid this problem, existing solutions can only choose one of two options: either significantly reduce the Taylor expansion order, sacrificing attenuation simulation accuracy and resulting in complete distortion of user profiles; or infinitely increase the ciphertext modulus, which directly causes an exponential decrease in computational efficiency and a surge in hardware computing power, making it completely unsuitable for the high-concurrency, low-latency real-time business scenarios of e-commerce platforms.

[0008] In addition, existing homomorphic encryption schemes have several derivative defects in this application scenario: most schemes do not have a dedicated dimensional normalization mechanism designed for the flower consumption scenario. The dimensional differences in flower quality, volume, and metabolic rate in different dimensions can lead to numerical overflow and precision loss in encrypted operations; after multi-dimensional ciphertext matching operations, the ciphertext order will continue to expand, requiring frequent relinearization operations, which further aggravates noise accumulation and computational overhead; most schemes adopt a centralized key management model, with user private keys kept by the platform, which poses a risk of key leakage and illegal data misuse, and fails to enable users to have independent control over their personal privacy data. Summary of the Invention

[0009] This invention provides a homomorphic encryption protection system for the privacy profile characteristics of flower consumers, which solves the technical problems mentioned in the background art.

[0010] This invention provides a homomorphic encryption protection system based on the privacy profile features of flower consumers, comprising:

[0011] We obtain the historical actual mass and apoptosis metabolic rate constant of flowers in various dimensions, and construct a multi-dimensional initial continuous data tensor by combining it with absolute elapsed time.

[0012] Based on a preset truncation order limit, the initial continuous data tensor is mapped to a plaintext mapping expression.

[0013] Obtain the dimension cancellation constant and floating-point precision multiplier, normalize the dimensions of the plaintext mapping expression, and encapsulate it into polynomial ciphertext by combining it with Gaussian noise.

[0014] Based on the cell apoptosis metabolic rate constants of each dimension, a fitness is constructed for swarm intelligence optimization to obtain the optimal evolutionary damping;

[0015] Obtain the current elapsed time, combine the optimal evolution damping with the current elapsed time to perform evaluation calculation on the polynomial ciphertext, and output the decay state ciphertext;

[0016] Receive the external seasonal market profile ciphertext, multiply it with the decay state ciphertext to generate a ciphertext tensor, and call the relinearization public key to perform ciphertext relinearization to obtain a two-dimensional matching ciphertext;

[0017] The private key polynomial is invoked to decrypt the two-dimensional matching ciphertext to obtain the decrypted plaintext. After eliminating the accumulated floating-point precision multipliers, the final preference intensity is output using the concentration reduction coefficient.

[0018] Furthermore, the process of obtaining the historical actual mass and apoptosis metabolic rate constant of flowers in various dimensions, combined with absolute elapsed time, to construct a multi-dimensional initial continuous data tensor includes:

[0019] Divide the historical actual mass by the standard human olfactory perception spatial volume reference constant to obtain the basic concentration value;

[0020] Multiply the negative one, the apoptosis metabolic rate constant, and the absolute elapsed time together, and use the product as the exponent of the natural constant to obtain the nonlinear decaying multiplier;

[0021] Multiplying the base concentration value by the nonlinear decay multiplier yields the initial continuous data tensor of the corresponding dimension.

[0022] Furthermore, the step of mapping the initial continuous data tensor to a plaintext mapping expression based on a preset truncation order limit includes:

[0023] Multiplying the apoptosis metabolic rate constant by the absolute elapsed time yields a dimensionless product term;

[0024] Introducing a summation index, the summation index raised to the power of negative one is divided by the factorial of the summation index, and then multiplied by the summation index raised to the power of the dimensionless product term to obtain the expanded algebraic term;

[0025] Let the summation index accumulate from zero to the truncation order limit, perform series summation on the expanded algebraic terms, and multiply the summation result by the basic concentration value to obtain the plaintext mapping expression.

[0026] Furthermore, the step of obtaining the dimensionless cancellation constant and floating-point precision multiplier, normalizing the dimensions of the plaintext mapping expression, and encapsulating it into polynomial ciphertext by combining Gaussian noise includes:

[0027] Divide the standard human olfactory perception spatial volume reference constant by the system reference mass constant to obtain the dimensionless cancellation constant;

[0028] The plaintext mapping formula, the dimensionless cancellation constant, and the floating-point precision multiplier at the initial moment are multiplied continuously, and the nearest integer is taken and then moduloed by the plaintext modulus range to obtain the dimensionless plaintext.

[0029] Multiply the uniform polynomial on the error-learning ring with the private key polynomial and take the opposite number, add the Gaussian noise, and add the product of the dimensionless plaintext and ciphertext modulus intervals divided by the nearest integer of the quotient of the plaintext modulus interval. Take the modulus of the whole product with respect to the ciphertext modulus interval to form the first component of the polynomial ciphertext.

[0030] The polynomial ciphertext is output by combining the first component of the polynomial ciphertext with the uniform polynomial as the second component.

[0031] Furthermore, the process of constructing fitness based on the apoptosis metabolic rate constants of each dimension to perform swarm intelligence optimization and obtain optimal evolutionary damping includes:

[0032] Initialize the candidate evolution damping and optimization step size for each dimension, and record the individual historical best position and the global historical best position;

[0033] The damped metabolic product is obtained by multiplying the candidate evolution damping of each dimension by the apoptosis metabolic rate constant, and the average value of the damped metabolic product of all dimensions is calculated.

[0034] The target fitness is obtained by subtracting the average value from the damping metabolic product of each dimension and then squaring the result. The sum of the squared terms of all dimensions is then taken as the reciprocal.

[0035] The endogenous inertia weight is obtained by dividing the current dimension’s apoptosis metabolic rate constant by the sum of the apoptosis metabolic rate constants of all dimensions, the individual’s historical best fitness is divided by the current target fitness, and the global historical best fitness is divided by the current target fitness.

[0036] The updated optimization evolution step size is obtained by multiplying the endogenous inertia weight by the optimization evolution step size, adding the individual cognitive weight by the difference between the individual's historical best position and the current candidate evolution damping, and adding the global social weight by the difference between the global historical best position and the current candidate evolution damping.

[0037] The updated optimization evolution step size is superimposed on the candidate evolution damper for optimization iteration until convergence and the optimal evolution damper is output.

[0038] Furthermore, the step of obtaining the current elapsed time, fusing the optimal evolutionary damping with the current elapsed time to perform evaluation calculations on the polynomial ciphertext, and outputting the decay state ciphertext includes:

[0039] The optimal evolutionary damping, the apoptosis metabolic rate constant, and the current elapsed time are multiplied to construct the evolutionary projection factor;

[0040] Based on the summation index, the summation index raised to the power of negative one is divided by the factorial of the summation index, and then multiplied by the summation index raised to the power of the evolutionary inference factor.

[0041] The summation index is accumulated from zero to the truncation order limit for series summation. The summation result is multiplied by the floating-point precision multiplier, and the nearest integer is taken. The modulus of the plaintext modulus interval is then taken to obtain the time-decayed plaintext polynomial.

[0042] The polynomial ciphertext is homomorphically multiplied with the time-decaying plaintext polynomial to obtain the decaying state ciphertext.

[0043] Furthermore, the step of receiving the external seasonal market profile ciphertext, multiplying it with the decay state ciphertext to generate a ciphertext tensor, and then calling the relinearization public key to perform ciphertext relinearization to obtain the two-dimensional matching ciphertext includes:

[0044] Perform homomorphic multiplication on the decay state ciphertext of all dimensions and the corresponding seasonal market profile ciphertext amplified by the floating-point precision multiplier, sum the results, and then take the modulus of the ciphertext modulus range to generate the ciphertext tensor containing zero-order ciphertext tensor terms, first-order ciphertext tensor terms, and second-order ciphertext tensor terms.

[0045] Invoke the relinearized public key that contains the zero-order public key component and the first-order public key component;

[0046] Multiply the second-order ciphertext tensor term by the zero-order public key component and add the zero-order ciphertext tensor term to form the first convergence component of the two-dimensional matching ciphertext.

[0047] Multiply the second-order ciphertext tensor term by the first-order public key component and add the first-order ciphertext tensor term to form the second convergence component of the two-dimensional matching ciphertext.

[0048] The first convergence component and the second convergence component are combined and then moduloed within the ciphertext modulus range to output the two-dimensional matched ciphertext.

[0049] Furthermore, the process of using the private key polynomial to decrypt the two-dimensional matching ciphertext to obtain the decrypted plaintext, eliminating the accumulated floating-point precision multipliers, and then outputting the final preference intensity using the concentration reduction coefficient includes:

[0050] Multiply the second convergence component of the two-dimensional matching ciphertext by the private key polynomial, add the first convergence component of the two-dimensional matching ciphertext, multiply the whole by the ratio of the plaintext modulus interval to the ciphertext modulus interval, take the nearest integer, and then take the modulus of the plaintext modulus interval to extract the decrypted plaintext.

[0051] The concentration reduction coefficient is constructed by dividing the system reference mass constant by the human standard olfactory perception spatial volume reference constant.

[0052] Divide the decrypted plaintext by the triple composite scaling factor generated by multiple homomorphic multiplications performed by the floating-point precision multiplier, and then multiply by the concentration restoration coefficient to output the final preference intensity.

[0053] The beneficial effects of this invention are as follows: by extracting the metabolic parameters of flowers in the environment and the absolute elapsed time to construct the underlying evolution tensor, and using the swarm intelligence algorithm to spontaneously obtain the optimal damping parameters, the problem of unbalanced expansion of noise in dense-state ring operation caused by multidimensional asynchronous decay is effectively alleviated. This realizes the adaptive steady-state evolution and direct matching of consumer flower preference information in spatiotemporal elapsed time, which not only ensures the concealment of preference data during the circulation period, but also greatly improves the robustness and compliance of the dense-state polynomial operation system throughout the entire life cycle. Attached Figure Description

[0054] Figure 1 This is a flowchart of the working process of a homomorphic encryption protection system for the privacy profile features of flower consumers according to the present invention. Detailed Implementation

[0055] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0056] This embodiment applies to a privacy computing device comprising a user interaction terminal, a cryptographic computing server, and a key management module. The privacy computing device includes, but is not limited to, enterprise-level servers, cloud computing clusters, and industrial control computers; the user interaction terminal includes, but is not limited to, mobile application interfaces, PC web pages, and smart retail terminal interactive screens, used to collect user flower interaction data and output the final preference strength; the cryptographic computing server is used to perform cryptographic polynomial operations and swarm intelligence optimization operations, supporting the BFV homomorphic encryption BEHZ version operation protocol based on ring error learning; the key management module is used to generate and distribute the public key, private key, and relinearization key required for homomorphic encryption, with key data transmitted to the corresponding computing module via a TLS encrypted channel.

[0057] like Figure 1 As shown, a homomorphic encryption protection system for flower consumer privacy profile features includes:

[0058] We obtain the historical actual mass and apoptosis metabolic rate constant of flowers in various dimensions, and construct a multi-dimensional initial continuous data tensor by combining it with absolute elapsed time.

[0059] Based on a preset truncation order limit, the initial continuous data tensor is mapped to a plaintext mapping expression.

[0060] Obtain the dimension cancellation constant and floating-point precision multiplier, normalize the dimensions of the plaintext mapping expression, and encapsulate it into polynomial ciphertext by combining it with Gaussian noise.

[0061] Based on the cell apoptosis metabolic rate constants of each dimension, a fitness is constructed for swarm intelligence optimization to obtain the optimal evolutionary damping;

[0062] Obtain the current elapsed time, combine the optimal evolution damping with the current elapsed time to perform evaluation calculation on the polynomial ciphertext, and output the decay state ciphertext;

[0063] Receive the external seasonal market profile ciphertext, multiply it with the decay state ciphertext to generate a ciphertext tensor, and call the relinearization public key to perform ciphertext relinearization to obtain a two-dimensional matching ciphertext;

[0064] The private key polynomial is invoked to decrypt the two-dimensional matching ciphertext to obtain the decrypted plaintext. After eliminating the accumulated floating-point precision multipliers, the final preference intensity is output using the concentration reduction coefficient.

[0065] The homomorphic encryption global parameters used in this system are pre-configured and verified, and all parameters meet the following mandatory constraints:

[0066] A circumradical ring is defined as a ring of polynomials generated by a circumradical polynomial of degree 2. The order of the cyclotomic polynomial It is a power of 2, and its value range is... Used to limit the operational dimension of a polynomial ring; plaintext modulus It is a prime number and satisfies The range of values ​​is The ciphertext modulus (q) is used to limit the numerical space of plaintext operations; the ciphertext modulus (q) is a positive integer and satisfies that (q / p) is an integer with a range of values ​​of 1 to 2. The bit length of the ciphertext modulus must meet the noise budget requirement corresponding to the homomorphic operation depth. The noise budget consumption of a single homomorphic multiplication operation shall not exceed 15% of the total noise budget, and the total noise budget threshold is [value missing]. ,in The standard deviation of Gaussian noise; floating-point precision multipliers. It is a positive integer, and its value range is 1. And satisfy This is used to scale floating-point plaintext values ​​to large integers while preserving computational precision.

[0067] The key management module completes the generation and distribution of all keys according to the following steps:

[0068] Private key polynomial In the circular ring Generated by uniform random sampling, with coefficients taking values ​​of 0 or 1, the order of the polynomial is the same as that of the cyclotomic polynomial. To maintain consistency, the key is securely stored offline by the key management module and is not disclosed to the public.

[0069] Public-key polynomials use a binary tuple format. ,in To divide the ring The coefficients of the uniform polynomial generated by uniform random sampling are in Evenly distributed within the range, ,in A small-coefficient noise polynomial is generated by sampling a discrete Gaussian distribution with a mean of 0 and a standard deviation of 3.2. The public key is made public and used for plaintext encryption operations.

[0070] The relinearized public key uses a binary format. Generated based on the squared terms of the private key polynomial, where To divide the ring The polynomial generated by uniform random sampling above, ,in The small-coefficient noise polynomial is generated by sampling a discrete Gaussian distribution with a mean of 0 and a standard deviation of 3.2. The relinearization public key is made public and used for order convergence operations after ciphertext multiplication.

[0071] Plaintext encoding uses the SIMD batch encoding rule of the BFV scheme, mapping single-dimensional plaintext values ​​to a single slot in a circular polynomial. Multi-dimensional data corresponds to multiple independent slots in the polynomial, with the number of slots corresponding to the order of the circular polynomial. The values ​​in each slot are kept consistent with 1 / 2, and the values ​​in each slot are calculated independently without interference. Negative numbers are filled using the two's complement rule during encoding to ensure that the plaintext values ​​remain closed in the polynomial ring operation.

[0072] The baseline concentration value is obtained by dividing the actual historical mass by the standard human olfactory perception spatial volume reference constant.

[0073] Historical actual quality refers to the fresh weight of a single type of flower corresponding to a single user interaction, denoted as The unit is kilograms, which can be synchronously obtained through flower transaction order data and category specification data associated with user interaction behavior. Each flower category corresponds to a unique dimensional index. , Let be a positive integer, and let be the total number of dimensions. , The maximum number of slots for SIMD encoding shall not exceed the limit; the standard human olfactory perception space volume reference constant refers to the standardized human olfactory perception space volume reference value, denoted as... The unit is cubic meters, with a fixed value of 0.005 cubic meters, corresponding to the standardized spatial volume of the human nasal cavity and olfactory perception area, used to convert absolute mass into spatial concentration value; the basic concentration value refers to the proportion of flower mass within a unit of perception space, with the unit being kilograms per cubic meter, representing the initial basic preference quantification value for the corresponding flower category, calculated using the following formula:

[0074]

[0075] in, For the first The basic concentration values ​​corresponding to each dimension.

[0076] Multiplying the negative one, the apoptosis metabolic rate constant, and the absolute elapsed time together, and using the product as the exponent of the natural constant, yields the nonlinear decaying multiplier.

[0077] The apoptosis metabolic rate constant refers to the first-order apoptosis response rate constant of a cut flower of a corresponding flower variety after in vitro transplantation, denoted as The unit is 1 / day, and the range of values ​​is... Metabolic rate constants can be obtained from publicly available databases of the biological characteristics of flower varieties. Standard values ​​for typical varieties include 0.25 for roses, 0.12 for lisianthus, and 0.03 for phalaenopsis orchids. Different varieties exhibit varying metabolic rate constants, representing the inherent property of flower quality decay over time. Absolute elapsed time refers to the continuous time length from the initial moment of interaction between the user and the corresponding flower to the current moment, denoted as _____. The unit is days, which can be obtained synchronously through the system clock. The time sampling accuracy can be adjusted to the hour or day level according to the calculation requirements. The sampling period is strictly matched with the time unit of the metabolic rate constant. If the sampling accuracy is adjusted to the hour level, the metabolic rate constant is synchronously converted to 1 / hour unit, and the conversion ratio is 1 / 24. The nonlinear decay multiplier is a dimensionless parameter with a value range of (0,1], which characterizes the degree of natural decay of flower quality over time. The calculation formula is:

[0078]

[0079] in, For the first Each dimension at time The corresponding nonlinear decaying multiplier has a dimensionless parameter as input to the exponent term.

[0080] Multiplying the base concentration value by the nonlinear decay multiplier yields the initial continuous data tensor for the corresponding dimension.

[0081] The initial continuous data tensor represents the corresponding flower category at time t. The preference values ​​are continuously quantified to provide continuous basic data for subsequent polynomial mapping. The unit is kilograms per cubic meter, and the calculation formula is as follows:

[0082]

[0083] in, For the first The initial continuous data tensor corresponding to each dimension is consistent with the total number of flower categories. The multi-dimensional tensor is mapped to the corresponding slots of the SIMD encoding according to the dimension index order to ensure the dimension alignment of subsequent dense state operations.

[0084] Multiplying the apoptosis metabolic rate constant by the absolute elapsed time yields a dimensionless product term.

[0085] The dimensionless product term is a dimensionless parameter. The time dimension is eliminated by multiplying the dimensional rate constant by time, providing a unitless input for subsequent series expansion and avoiding computational errors caused by dimensional differences. The absolute value of the dimensionless product term must be less than 1 to ensure that the truncation error of the Taylor series expansion converges to an acceptable range. The calculation formula is as follows:

[0086]

[0087] in, For the first Each dimension at time The corresponding dimensionless product term, when When the absolute value is greater than or equal to 1, the expansion accuracy is improved synchronously by adjusting the truncation order limit, and the minimum value of the truncation order limit is not less than 1. .

[0088] By introducing a summation index, we divide the summation index raised to the power of negative one by the factorial of the summation index, and then multiply it by the summation index raised to the power of the dimensionless product term to obtain the expanded algebraic term.

[0089] The summation index is denoted as , is a non-negative integer used for traversal counting in series expansion; the expanded algebraic terms are single terms in the Taylor series expansion of the exponential function, corresponding to the natural exponential function. The first Taylor expansion The term is used to convert continuous exponential functions into polynomial algebraic form, adapting to the algebraic ring operation rules of homomorphic encryption that only support addition and multiplication. The formula for calculating the expanded algebraic term is:

[0090]

[0091] in, For the first The first dimension, the first The expansion algebraic terms corresponding to the order are dimensionless parameters.

[0092] Let the summation index accumulate from zero to the limit of the truncation order, perform a series summation on the expanded algebraic terms, and multiply the summation result by the basic concentration value to obtain the plaintext mapping expression.

[0093] The limit of the truncation order is denoted as , is a positive integer, and its value range is . The order of the polynomial can be adjusted according to the upper limit of the polynomial order and the accuracy requirements of the homomorphic encryption scheme. This limits the highest order of the Taylor expansion and avoids the expansion of dense noise caused by excessively high polynomial order. The plaintext mapping is a polynomial approximation of the initial continuous data tensor, with units of kilograms per cubic meter, consistent with the dimensions of the initial continuous data tensor. The calculation formula is as follows:

[0094]

[0095] in, For the corresponding number The plaintext mapping of each dimension, the result of the series summation is a dimensionless parameter.

[0096] Dividing the standard human olfactory perception spatial volume reference constant by the system reference mass constant yields the dimensionless cancellation constant.

[0097] The system reference quality constant refers to the standardized quality reference value preset by the system, denoted as The unit is kilogram, with a fixed value of 1 kilogram. This is used to eliminate the dimension of mass, converting floating-point concentration values ​​into dimensionless integers to meet the integer arithmetic requirements of lattice cipher homomorphic encryption. The dimension cancellation constant is a dimensionless parameter used to cancel the volume and mass dimensions in the basic concentration value, achieving dimensionless processing of the plaintext. The calculation formula is:

[0098]

[0099] in, is a dimensionless cancellation constant.

[0100] The plaintext mapping formula at the initial moment, the dimensionless cancellation constant, and the floating-point precision multiplier are multiplied continuously. The nearest integer is then taken modulo the plaintext modulus range to obtain the dimensionless plaintext.

[0101] The initial moment refers to the moment when the user begins to interact with the flower, denoted as . Corresponding to absolute elapsed time ; Dimensionless plaintext is an integer, and its value range is . The plaintext space adapted for homomorphic encryption is calculated using the following formula:

[0102]

[0103] in, This indicates the operation of rounding to the nearest integer; For the first The dimensionless plaintext corresponding to each dimension, after scaling, must satisfy the following: If it exceeds this range, adjust the floating-point precision multiplier. The value of is selected to avoid the loss of plaintext information due to modulo operations.

[0104] Multiply the uniform polynomial on the error-learning ring with the private key polynomial and take the opposite, add Gaussian noise, and then multiply the product of the dimensionless plaintext and ciphertext moduli intervals by the nearest integer of the quotient of the plaintext moduli interval. Take the modulus of the whole product with respect to the ciphertext moduli interval to form the first component of the polynomial ciphertext.

[0105] Uniform polynomial exponential ring The polynomial generated by uniform random sampling is denoted as polynomial coefficients in Uniformly distributed within the range; Gaussian noise refers to a small-coefficient polynomial generated by sampling a discrete Gaussian distribution with a mean of 0 and a standard deviation of 3.2, denoted as This is used to ensure the semantic security of homomorphic encryption; the quotient of the ciphertext modulus interval divided by the plaintext modulus interval is an integer scaling factor, denoted as . This is used to map dimensionless plaintext to ciphertext space, preventing plaintext information from being obscured by noise during encrypted operations; the formula for calculating the first component of polynomial ciphertext is:

[0106]

[0107] in, The first component of the polynomial ciphertext is a circular ring. Polynomials over.

[0108] The first component of the polynomial ciphertext is combined with the uniform polynomial as the second component to output the polynomial ciphertext.

[0109] The polynomial ciphertext is in binary format, conforming to the standard ciphertext format of the BFV homomorphic encryption scheme, and can be directly used for subsequent encrypted homomorphic operations. The calculation formula is:

[0110]

[0111] in, For the first The initial polynomial ciphertext corresponding to each dimension is stored in the order of dimension index, and corresponds one-to-one with the SIMD encoding slot.

[0112] Initialize the candidate evolution damping and optimization step size for each dimension, and record the individual historical best position and the global historical best position.

[0113] Candidate evolution damping is denoted as , for the first In the nth iteration The parameters to be optimized in each dimension are dimensionless, with an initial value uniformly set to 1 and a range limited to [0.1, 10]. These parameters are used to adjust the decay rate of the corresponding dimension and balance the accumulation rate of dense-state noise in each dimension. The optimization evolution step size is denoted as . , for the first In the nth iteration The parameter update step size for each dimension is initialized to 0.1, and its range is limited to [-0.5, 0.5] by a speed clamping rule. This is used to control the parameter adjustment magnitude during optimization iterations; the iteration number index is denoted as... , a positive integer, represents the iteration round of the optimization search, with a maximum iteration count set to 200; the individual historical best position refers to the candidate evolutionary damping value corresponding to the highest fitness in the iteration history of a single dimension, with the initial value being the initial candidate evolutionary damping; the global historical best position refers to the set of candidate evolutionary damping values ​​corresponding to the highest fitness in the iteration history of all dimensions, with the initial value being the set of initial candidate evolutionary damping for all dimensions; the population size is set to the total number of dimensions. Each dimension corresponds to an independent optimization particle.

[0114] The damped metabolic product is obtained by multiplying the candidate evolution damping of each dimension by the apoptosis metabolic rate constant, and the average value of the damped metabolic product of all dimensions is calculated.

[0115] The damped metabolic product is a dimensionless parameter that characterizes the effective decay rate of the corresponding dimension after damping adjustment. It is used to measure the degree of balance in the decay rates of different dimensions. The smaller the variance of the damped metabolic product, the more balanced the decay rates of each dimension, and the more consistent the noise accumulation rate of the dense-state operation, which can effectively avoid premature overflow of dense-state noise in high-decay dimensions. The average value is the arithmetic mean of the damped metabolic products of all dimensions, used to calculate the deviation of the decay rate of each dimension from the overall average level. The calculation formula is:

[0116]

[0117] in, For the first In the nth iteration Damped metabolic product corresponding to each dimension; For the first The average value of all dimensionally damped metabolic products in each iteration; The index for the traversal summation of the variance mean is a positive integer.

[0118] The target fitness is obtained by subtracting the average value from the damped metabolic product of each dimension and then squaring the result. The sum of these squared terms for all dimensions is then taken as the reciprocal.

[0119] The target fitness is a dimensionless parameter with a value greater than 0. The value of the target fitness is inversely proportional to the variance of the damped metabolic product of each dimension. The smaller the variance, the higher the fitness, and the more balanced the decay rate of each dimension. This can effectively avoid premature overflow of dense-state noise in high-decay dimensions. The calculation formula is as follows:

[0120]

[0121] in, For the first The target fitness of the next iteration; For the first The state vector for each iteration consists of all candidate evolution dampers; It is a local constant, taking the value of This is used to avoid division by zero errors when the summation term is 0, and to ensure the numerical stability of fitness calculations.

[0122] The endogenous inertia weight is calculated by dividing the current dimension's apoptosis metabolic rate constant by the sum of all dimensions' apoptosis metabolic rate constants. The individual's historical best fitness is divided by the current target fitness as the individual's cognitive weight, and the global historical best fitness is divided by the current target fitness as the global social weight.

[0123] The intrinsic inertia weight is a dimensionless parameter with a value range of (0,1). It is used to adapt to dimensions with different metabolic rates. The higher the metabolic rate of a dimension, the larger the inertia weight, and the stronger the inheritance of parameter adjustments. The individual cognition weight is a dimensionless parameter used to control the proportion of the step size a particle takes to learn from its own historical best position. The closer the current fitness is to the individual's historical best fitness, the larger the weight. The global social weight is a dimensionless parameter used to control the proportion of the step size a particle takes to learn from the global best position of the group. The closer the current fitness is to the global historical best fitness, the larger the weight. The calculation formula is as follows:

[0124]

[0125] in, For the first The intrinsic inertia weights corresponding to each dimension; For the first In the nth iteration The individual cognitive weights corresponding to each dimension; For the first In the nth iteration The global social weights corresponding to each dimension; For the first Each dimension corresponds to the individual's best historical position; This is the best position in the overall historical context; The fitness corresponding to an individual's historical best position; The fitness is the fitness corresponding to the best position in the global history.

[0126] The updated optimization step size is obtained by multiplying the intrinsic inertia weight by the optimization step size, adding the individual cognitive weight multiplied by the difference between the individual's historical best position and the current candidate evolution damping, and adding the global social weight multiplied by the difference between the global historical best position and the current candidate evolution damping. The updated optimization step size needs to be limited to the range [-0.5, 0.5] by the velocity clamping rule to avoid excessively large step sizes that could cause particles to fly out of the effective boundary, ensuring the convergence of the optimization process. The calculation formula is as follows:

[0127]

[0128] in, For the updated optimization evolution step size; The best position in the global history The values ​​corresponding to each dimension.

[0129] The updated optimization step size is superimposed on the candidate evolution damper for optimization iteration until convergence and the optimal evolution damper is output.

[0130] The updated candidate evolution damping needs to be limited to the range [0.1, 10] by boundary constraint rules. If it exceeds the boundary, the value will be reset to the corresponding boundary value to ensure that the parameters iterate within the effective range. The convergence conditions of the optimization iteration include the number of iterations reaching the preset maximum number of iterations of 200, the global best fitness having no improvement for 30 consecutive iterations, and the absolute value of the optimization evolution step size in all dimensions being less than the convergence threshold. The iteration terminates when any condition is met; the final values ​​of each dimension output after convergence are the optimal evolution damping, denoted as . The calculation formula is:

[0131]

[0132] During the iteration process, after each iteration, the target fitness of all particles is calculated simultaneously. If the fitness of the current particle is higher than the individual's historical best fitness, the individual's historical best position and the individual's historical best fitness are updated. If the fitness of the current particle is higher than the global historical best fitness, the global historical best position and the global historical best fitness are updated. If the convergence condition is not met, the iteration count is incremented by 1, and the iteration steps are repeated. If the convergence condition is met, the iteration is terminated and the optimal evolution damping is output.

[0133] The evolutionary projection factor is constructed by multiplying the optimal evolutionary damping, the apoptosis metabolic rate constant, and the current elapsed time.

[0134] The current elapsed time is denoted as , refers to the time since the initial interaction The continuous time from the start of the calculation to the execution time of the dense-state operation, in days, is consistent with the time unit of the initial continuous data tensor construction stage; the evolutionary inference factor is a dimensionless parameter, which, after optimal evolutionary damping adjustment, can achieve equalization of the decay rate of each dimension, avoid extreme differences in the accumulation rate of dense-state operation noise in different dimensions, and ensure the stability of dense-state operation. The absolute value of the evolutionary inference factor must be less than 1 to ensure the convergence of the truncation error of the Taylor series expansion. The calculation formula is as follows:

[0135]

[0136] in, For the first The evolutionary inference factors corresponding to each dimension.

[0137] Based on the summation index, the summation index raised to the power of negative one is divided by the factorial of the summation index, and then multiplied by the summation index of the evolutionary derivation factor raised to the power of the summation index.

[0138] The summation index is consistent with the summation index in the plaintext mapping transformation process, and is a non-negative integer. This is used for traversal counting in series expansion, ensuring that the mathematical form of the expansion is consistent with the initial plaintext mapping, and avoiding operational errors. The formula for calculating a single term is:

[0139]

[0140] in, For the first The first dimension, the first The series expansion term corresponding to the order is a dimensionless parameter.

[0141] Let the summation index accumulate from zero to the limit of the truncation order and perform a series summation. Then multiply the summation result by the floating-point precision multiplier, take the nearest integer, and take the modulus of the plaintext modulus interval to obtain the time-decayed plaintext polynomial.

[0142] The time-decay plaintext polynomial is an integer polynomial over a circular ring, with coefficients ranging from 1 to 2. To ensure dimensionality matching in homomorphic multiplication operations and avoid precision loss during calculations, the scaling rules and modulus range are consistent with the initial dimensionless plaintext. The calculation formula is as follows:

[0143]

[0144] in, For the first Each dimension corresponds to a time-decaying plaintext polynomial. The multi-dimensional values ​​are mapped to the corresponding slots in the SIMD encoding according to the dimension index order, corresponding one-to-one with the dimensions of the initial polynomial ciphertext.

[0145] Perform a homomorphic multiplication operation between the polynomial ciphertext and the time-decaying plaintext polynomial to obtain the decaying state ciphertext.

[0146] The homomorphic multiplication operation follows the plaintext-ciphertext multiplication rules of the BFV homomorphic encryption scheme. The noise budget is updated synchronously during the operation, and the noise budget consumption of a single plaintext-ciphertext multiplication does not exceed 10% of the total noise budget. The output ciphertext remains in standard binary format and can be directly used for subsequent encrypted operations. The calculation formula is as follows:

[0147]

[0148] in, For the corresponding number The decaying state ciphertext in each dimension; This is the notation for the homomorphic multiplication operation between plaintext and ciphertext in the BFV homomorphic encryption scheme. The specific steps are as follows: multiply each of the two components of the ciphertext tuple by the plaintext polynomial, and then divide the ciphertext modulus. Take the modulo operation and output the new binary ciphertext. No relinearization operation is required during the operation.

[0149] Perform homomorphic multiplication of the decay state ciphertext of all dimensions with its corresponding seasonal market profile ciphertext amplified by floating-point precision multipliers and sum them. Then take the modulus of the ciphertext modulus range to generate a ciphertext tensor containing zero-order ciphertext tensor terms, first-order ciphertext tensor terms, and second-order ciphertext tensor terms.

[0150] The seasonal market profile ciphertext is BFV ciphertext in standard binary format, generated by the market server using encryption parameters, floating-point precision multipliers, and SIMD encoding rules completely consistent with the user side. The market profile data for the corresponding dimension has dimensionless weight parameters with values ​​ranging from [0,1], perfectly matching the dimensions and scaling rules of the user-side ciphertext, ensuring dimensional consistency of the homomorphic multiplication operation. After homomorphic multiplication of the two sets of binary format ciphertexts, a triplet ciphertext tensor containing zero-order, first-order, and second-order components is generated, where the zero-order ciphertext tensor term is denoted as... The first-order ciphertext tensor term is denoted as The second-order ciphertext tensor term is denoted as The components of the ciphertext tensor are all circular rings. The polynomial on the given surface has coefficients that take values ​​in the range of 1 / 2. The calculation formula is:

[0151]

[0152] in, For encrypted tensors; For the corresponding number A detailed profile of the seasonal market from multiple dimensions; This is the homomorphic multiplication operation symbol between two sets of ciphertexts in the BFV homomorphic encryption scheme, which generates a ciphertext tensor in triplet format after the operation.

[0153] Call the relinearized public key that contains the zero-order public key component and the first-order public key component.

[0154] The relinearized public key uses a binary format. ,in For zero-order public key components, It is a first-order public key component, pre-generated and publicly disclosed by the key management module. It is used to restore the high-order ciphertext tensor in triplet format to the standard ciphertext in binary format, reduce the ciphertext order, and control the noise accumulation of encrypted operations.

[0155] Multiplying the second-order ciphertext tensor term by the zero-order public key component and adding the zero-order ciphertext tensor term constitutes the first convergence component of the two-dimensional matching ciphertext.

[0156] The first convergence component is a circular ring. The polynomial on the given surface has coefficients that take values ​​in the range of 1 / 2. This is used to incorporate information from second-order ciphertext tensors into lower-order components, achieving convergence of ciphertext orders. The calculation formula is:

[0157]

[0158] in, This is the first convergence component of the two-dimensional matching ciphertext.

[0159] Multiplying the second-order ciphertext tensor term by the first-order public key component and adding the first-order ciphertext tensor term constitutes the second convergence component of the two-dimensional matching ciphertext.

[0160] The second convergence component is a circular ring. The polynomial on the given surface has coefficients that take values ​​in the range of 1 / 2. Together with the first convergent component, it forms the relinearized binary ciphertext component, which maintains the same format as the initial ciphertext. The calculation formula is:

[0161]

[0162] in, This is the second convergence component of the two-dimensional matching ciphertext.

[0163] The first and second convergence components are combined and then moduloed within the ciphertext modulus range to output a two-dimensional matching ciphertext.

[0164] The two-dimensional matching ciphertext is a BFV ciphertext in standard binary format, which can be directly decrypted through the standard decryption process. It represents the ciphertext matching result between user preferences and seasonal market profiles. After relinearization, the noise budget is updated synchronously to ensure that the remaining noise budget meets the decryption requirements. The calculation formula is as follows:

[0165]

[0166] in, This is a two-dimensional matching ciphertext.

[0167] Multiply the second convergence component of the two-dimensional matching ciphertext by the private key polynomial, add the first convergence component of the two-dimensional matching ciphertext, multiply the whole by the ratio of the plaintext modulus interval to the ciphertext modulus interval, take the nearest integer, and then take the modulus of the plaintext modulus interval to extract the decrypted plaintext.

[0168] The decryption process follows the standard decryption rules of the BFV homomorphic encryption scheme. Random encrypted components of the ciphertext are eliminated using a private key polynomial. After scaling and modulo operations, the plaintext result of the matching operation is restored. After decryption, the plaintext needs to be validated. If the plaintext exceeds the range [0, p-1], it is considered noise overflow and the result is invalid. The calculation formula is:

[0169]

[0170] in, To decrypt the plaintext, is an integer, and its value range is 1. .

[0171] The concentration reduction coefficient is constructed by dividing the system reference mass constant by the standard human olfactory perception spatial volume reference constant.

[0172] The concentration reduction coefficient is a dimensionless parameter used in the reverse restoration of the initial plaintext processing stage to perform dimensional cancellation. It restores the dimensionless decrypted plaintext to a physically meaningful preferred concentration value, maintaining the same dimension as the initial base concentration value. The calculation formula is as follows:

[0173]

[0174] in, The concentration reduction coefficient is denoted as .

[0175] Divide the decrypted plaintext by the triple composite scaling factor generated by multiple homomorphic multiplications performed by floating-point precision multipliers, and then multiply by the concentration reduction coefficient to output the final preference intensity.

[0176] The triple composite scaling factor is generated by the cumulative multiplication of three floating-point precision multipliers: initial plaintext scaling, time-decrease plaintext polynomial scaling, and market profile ciphertext scaling. It is denoted as... This is used to eliminate integer scaling during the calculation process and restore the true floating-point value; the final preference intensity unit is kilograms per cubic meter, which represents the user's comprehensive preference quantification value for the current seasonal flower category, and the calculation formula is:

[0177]

[0178] in, To determine the final preference strength, the output needs to be validated. If the value is less than 0, it should be reset to 0.

[0179] Throughout the entire computation process, the remaining noise budget after each homomorphic operation is monitored in real time. When the remaining noise budget is less than 20% of the total noise budget, a ciphertext refresh operation is triggered, and the ciphertext noise is refreshed using a homomorphic encryption bootstrapping process. For all division operations, a minimum value protection constant is set to avoid division by zero errors. For all boundary scenarios, corresponding numerical constraint rules are set to ensure the numerical stability and validity of the computation process.

[0180] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A homomorphic encryption protection system for the privacy profile characteristics of flower consumers, characterized in that, Configured for execution: We obtain the historical actual mass and apoptosis metabolic rate constant of flowers in various dimensions, and construct a multi-dimensional initial continuous data tensor by combining it with absolute elapsed time. Based on a preset truncation order limit, the initial continuous data tensor is mapped to a plaintext mapping expression. Obtain the dimension cancellation constant and floating-point precision multiplier, normalize the dimensions of the plaintext mapping expression, and encapsulate it into polynomial ciphertext by combining it with Gaussian noise. Based on the cell apoptosis metabolic rate constants of each dimension, a fitness is constructed for swarm intelligence optimization to obtain the optimal evolutionary damping; Obtain the current elapsed time, combine the optimal evolution damping with the current elapsed time to perform evaluation calculation on the polynomial ciphertext, and output the decay state ciphertext; Receive the external seasonal market profile ciphertext, multiply it with the decay state ciphertext to generate a ciphertext tensor, and call the relinearization public key to perform ciphertext relinearization to obtain a two-dimensional matching ciphertext; The private key polynomial is invoked to decrypt the two-dimensional matching ciphertext to obtain the decrypted plaintext. After eliminating the accumulated floating-point precision multipliers, the final preference intensity is output using the concentration reduction coefficient.

2. The homomorphic encryption protection system for flower consumer privacy profile features according to claim 1, characterized in that, The process involves obtaining the historical actual mass and apoptosis metabolic rate constant of flowers in various dimensions, and combining this with absolute elapsed time to construct a multi-dimensional initial continuous data tensor, including: Divide the historical actual mass by the standard human olfactory perception spatial volume reference constant to obtain the basic concentration value; Multiply the negative one, the apoptosis metabolic rate constant, and the absolute elapsed time together, and use the product as the exponent of the natural constant to obtain the nonlinear decaying multiplier; Multiplying the base concentration value by the nonlinear decay multiplier yields the initial continuous data tensor of the corresponding dimension.

3. A homomorphic encryption protection system for flower consumer privacy profile features according to claim 2, characterized in that, The process of mapping the initial continuous data tensor to a plaintext mapping expression based on a preset truncation order limit includes: Multiplying the apoptosis metabolic rate constant by the absolute elapsed time yields a dimensionless product term; Introducing a summation index, the summation index raised to the power of negative one is divided by the factorial of the summation index, and then multiplied by the summation index raised to the power of the dimensionless product term to obtain the expanded algebraic term; Let the summation index accumulate from zero to the truncation order limit, perform series summation on the expanded algebraic terms, and multiply the summation result by the basic concentration value to obtain the plaintext mapping expression.

4. A homomorphic encryption protection system for flower consumer privacy profile features according to claim 3, characterized in that, The process of obtaining the dimensionless cancellation constant and floating-point precision multiplier, normalizing the dimensions of the plaintext mapping expression, and encapsulating it into polynomial ciphertext using Gaussian noise includes: Divide the standard human olfactory perception spatial volume reference constant by the system reference mass constant to obtain the dimensionless cancellation constant; The plaintext mapping formula, the dimensionless cancellation constant, and the floating-point precision multiplier at the initial moment are multiplied continuously, and the nearest integer is taken and then moduloed by the plaintext modulus range to obtain the dimensionless plaintext. Multiply the uniform polynomial on the error-learning ring with the private key polynomial and take the opposite number, add the Gaussian noise, and add the product of the dimensionless plaintext and ciphertext modulus intervals divided by the nearest integer of the quotient of the plaintext modulus interval. Take the modulus of the whole product with respect to the ciphertext modulus interval to form the first component of the polynomial ciphertext. The polynomial ciphertext is output by combining the first component of the polynomial ciphertext with the uniform polynomial as the second component.

5. A homomorphic encryption protection system for flower consumer privacy profile features according to claim 4, characterized in that, The fitness constructed based on the apoptosis metabolic rate constants of each dimension for swarm intelligence optimization to obtain optimal evolutionary damping includes: Initialize the candidate evolution damping and optimization step size for each dimension, and record the individual historical best position and the global historical best position; The damped metabolic product is obtained by multiplying the candidate evolution damping of each dimension by the apoptosis metabolic rate constant, and the average value of the damped metabolic product of all dimensions is calculated. The target fitness is obtained by subtracting the average value from the damping metabolic product of each dimension and then squaring the result. The sum of the squared terms of all dimensions is then taken as the reciprocal. The endogenous inertia weight is obtained by dividing the current dimension’s apoptosis metabolic rate constant by the sum of the apoptosis metabolic rate constants of all dimensions, the individual’s historical best fitness is divided by the current target fitness, and the global historical best fitness is divided by the current target fitness. The updated optimization evolution step size is obtained by multiplying the endogenous inertia weight by the optimization evolution step size, adding the individual cognitive weight by the difference between the individual's historical best position and the current candidate evolution damping, and adding the global social weight by the difference between the global historical best position and the current candidate evolution damping. The updated optimization evolution step size is superimposed on the candidate evolution damper for optimization iteration until convergence and the optimal evolution damper is output.

6. A homomorphic encryption protection system for flower consumer privacy profile features according to claim 5, characterized in that, The process of obtaining the current elapsed time, fusing the optimal evolutionary damping with the current elapsed time to perform evaluation calculations on the polynomial ciphertext, and outputting the decay state ciphertext includes: The optimal evolutionary damping, the apoptosis metabolic rate constant, and the current elapsed time are multiplied to construct the evolutionary projection factor; Based on the summation index, the summation index raised to the power of negative one is divided by the factorial of the summation index, and then multiplied by the summation index raised to the power of the evolutionary inference factor. The summation index is accumulated from zero to the truncation order limit for series summation. The summation result is multiplied by the floating-point precision multiplier, and the nearest integer is taken. The modulus of the plaintext modulus interval is then taken to obtain the time-decayed plaintext polynomial. The polynomial ciphertext is homomorphically multiplied with the time-decaying plaintext polynomial to obtain the decaying state ciphertext.

7. A homomorphic encryption protection system for flower consumer privacy profile features according to claim 6, characterized in that, The process of receiving the external seasonal market profile ciphertext, multiplying it with the decay state ciphertext to generate a ciphertext tensor, and then using the relinearization public key to perform ciphertext relinearization to obtain the two-dimensional matching ciphertext includes: Perform homomorphic multiplication on the decay state ciphertext of all dimensions and the corresponding seasonal market profile ciphertext amplified by the floating-point precision multiplier, sum the results, and then take the modulus of the ciphertext modulus range to generate the ciphertext tensor containing zero-order ciphertext tensor terms, first-order ciphertext tensor terms, and second-order ciphertext tensor terms. Invoke the relinearized public key that contains the zero-order public key component and the first-order public key component; Multiply the second-order ciphertext tensor term by the zero-order public key component and add the zero-order ciphertext tensor term to form the first convergence component of the two-dimensional matching ciphertext. Multiply the second-order ciphertext tensor term by the first-order public key component and add the first-order ciphertext tensor term to form the second convergence component of the two-dimensional matching ciphertext. The first convergence component and the second convergence component are combined and then moduloed within the ciphertext modulus range to output the two-dimensional matching ciphertext.

8. A homomorphic encryption protection system for flower consumer privacy profile features according to claim 7, characterized in that, The process of using the private key polynomial to decrypt the two-dimensional matching ciphertext to obtain the decrypted plaintext, eliminating the accumulated floating-point precision multipliers, and then using the concentration reduction coefficient to output the final preference intensity includes: Multiply the second convergence component of the two-dimensional matching ciphertext by the private key polynomial, add the first convergence component of the two-dimensional matching ciphertext, multiply the whole by the ratio of the plaintext modulus interval to the ciphertext modulus interval, take the nearest integer, and then take the modulus of the plaintext modulus interval to extract the decrypted plaintext. The concentration reduction coefficient is constructed by dividing the system reference mass constant by the human standard olfactory perception spatial volume reference constant. Divide the decrypted plaintext by the triple composite scaling factor generated by multiple homomorphic multiplications performed by the floating-point precision multiplier, and then multiply by the concentration restoration coefficient to output the final preference intensity.

Images