A multi-port coupon pushing method and device

Abstract

This invention discloses a method and apparatus for multi-port coupon delivery, relating to the field of intelligent marketing technology. The method includes: inputting a multimodal feature tensor into a SNN decision engine, generating SNN synaptic weight parameters based on the spatiotemporal features of user historical behavior, and dynamically adjusting port priorities through synaptic plasticity rules to generate a dynamic port strategy matrix; based on the dynamic port strategy matrix, preloading coupon strategies through a memristor array, and predicting user trajectories using a quantum city spatiotemporal model to generate real-time push instructions; executing the real-time push instructions, and embedding a topological photonic crystal verification code in the coupon propagation chain, verifying the authenticity of user redemption through zero-knowledge proof, and generating a redemption voucher chain. This invention solves the feature distortion problem caused by data heterogeneity in edge computing by utilizing the spatiotemporal coding characteristics of spiking neural networks to eliminate environmental noise and fuse multi-source features.

Description

Technical Field

[0001] This invention relates to the field of intelligent marketing technology, and in particular to a method and apparatus for pushing coupons across multiple platforms. Background Technology

[0002] In recent years, with the rapid development of mobile internet and IoT technologies, the field of intelligent marketing is undergoing a transformation from traditional one-way push notifications to multimodal, intelligent decision-making. Regarding coupon push technology, existing technologies mainly revolve around user profiling and static rule engines, such as recommendation devices based on collaborative filtering algorithms and distributed marketing platforms based on federated learning. The introduction of edge computing has made real-time data processing possible, and the low latency of 5G networks further improves the timeliness of marketing decisions.

[0003] However, existing technologies still have significant shortcomings: First, regarding adaptability to dynamic environments, push notification devices based on static rules or shallow machine learning struggle to effectively capture the spatiotemporal correlation characteristics of user behavior, resulting in an inability to achieve precise bandwidth allocation and priority adjustment in multi-port collaborative scenarios. Second, at the security verification level, existing federated learning schemes mostly employ classical encryption algorithms, which are vulnerable to quantum computing attacks, and lack a dynamic binding mechanism between verification behavior and identity parameters, leading to a persistently high rate of coupon misuse. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a multi-port coupon push method to solve the problems of poor adaptability to dynamic environments and weak security verification mechanisms in existing technologies.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for pushing coupons via multiple ports, comprising: collecting multi-source heterogeneous data; preprocessing the multi-source heterogeneous data using a neuromorphic processing architecture to generate a multimodal feature tensor; inputting the multimodal feature tensor into an SNN decision engine; generating SNN synaptic weight parameters based on the spatiotemporal features of user historical behavior; dynamically adjusting port priorities through synaptic plasticity rules to generate a dynamic port policy matrix; preloading coupon policies through a memristor array based on the dynamic port policy matrix; predicting user trajectories using a quantum city spatiotemporal model to generate real-time push instructions; executing the real-time push instructions; embedding a topological photonic crystal verification code in the coupon propagation link; verifying the authenticity of user redemption through zero-knowledge proof to generate a redemption voucher chain; updating the SNN synaptic weight parameters through quantum-safe federated learning based on the redemption voucher chain; synchronizing the updated parameters to a 5G base station edge server and a supermarket local server; and completing cross-port push of coupons through a multi-port collaborative delivery engine.

[0008] As a preferred embodiment of the multi-port coupon push method of the present invention, the multi-source heterogeneous data includes spatiotemporal positioning signals, radio frequency signals, platform transaction logs, and port real-time load data.

[0009] As a preferred embodiment of the multi-port coupon push method of the present invention, the specific steps for generating the multimodal feature tensor are as follows:

[0010] The system eliminates environmental noise in radio frequency signals by using a neuromorphic processing architecture, performs drift correction on spatiotemporal positioning signals, filters outliers in platform transaction logs, and generates purified multi-source heterogeneous data.

[0011] The purified multi-source heterogeneous data is input into the feature extraction layer of the neuromorphic processing architecture to generate spatiotemporal feature vectors of user behavior, product interaction feature vectors, and consumer preference feature vectors.

[0012] By using the pulse fusion layer of the neuromorphic processing architecture, the spatiotemporal feature vectors of user behavior, the feature vectors of product interaction, and the feature vectors of consumer preferences are concatenated into tensors to generate a multimodal feature tensor.

[0013] As a preferred embodiment of the multi-port coupon push method described in this invention, the steps of inputting multimodal feature tensors into an SNN decision engine, generating SNN synaptic weight parameters based on the spatiotemporal features of user historical behavior, and dynamically adjusting port priorities through synaptic plasticity rules to generate a dynamic port strategy matrix are as follows.

[0014] The multimodal feature tensor is input into the spatiotemporal feature encoding layer of the SNN decision engine. The spatiotemporal features of the user's historical behavior are extracted by the pulse time window segmentation algorithm to generate the encoded spatiotemporal feature vector. Based on the correlation between the frequency of user behavior and the timing of port interaction, the SNN synaptic weight parameters are generated.

[0015] The SNN synaptic weight parameters are input into the synaptic plasticity control layer, and the port priority is dynamically adjusted by the Gaussian pulse time-dependent plasticity rule to output the optimized SNN synaptic weight distribution.

[0016] Based on the optimized SNN synaptic weight distribution, tensor outer product operation is performed in conjunction with real-time port load data to generate a dynamic port policy four-dimensional tensor, and a dynamic port policy matrix is ​​generated through multimodal fusion dimensionality reduction.

[0017] As a preferred embodiment of the multi-port coupon push method of the present invention, the specific steps for generating the real-time push instruction are as follows:

[0018] Based on the row vector physical port mapping relationship of the dynamic port policy matrix, the priority parameters of the coupon policy are linearly mapped to the conduction values ​​of the memristor array, and the policy tensor is generated by the resistance state modulation of the dual memristor cross array.

[0019] Input the strategy tensor into the quantum city spatiotemporal model, perform grid-based encoding and spatiotemporal entanglement calculation on the user's historical trajectory, and generate a predicted coordinate sequence tensor.

[0020] The predicted coordinate sequence tensor and the policy tensor are subjected to four-dimensional spatiotemporal convolution operation. The comprehensive score tensor of the candidate nodes is generated by the attention weighted kernel function, and the set of effective candidate nodes is selected according to the dynamic fusion threshold.

[0021] When the quantum clock deviation in the set of valid candidate nodes meets the synchronization tolerance threshold, the dynamic port strategy matrix calculates the bandwidth allocation coefficient of the target port and generates a real-time push command.

[0022] As a preferred embodiment of the multi-platform coupon push method of the present invention, the specific steps for generating the verification voucher chain are as follows:

[0023] A topological photonic crystal verification code is embedded in the coupon propagation link corresponding to the real-time push command, and a decoding verification interface is pre-configured.

[0024] When a user terminal accesses the decoding verification interface, it decodes and verifies the optical feature information using a zero-knowledge proof protocol and generates an interactive proof file.

[0025] The interactive proof documents are matched spatiotemporally with the coupon redemption request, and a redemption voucher chain with a multi-level verification data structure is generated.

[0026] As a preferred embodiment of the multi-port coupon push method described in this invention, the method involves updating the SNN synaptic weight parameters based on the verification voucher chain using quantum-safe federated learning and synchronizing them to the 5G base station edge server and the supermarket's local server. The cross-port push of coupons is then completed through a multi-port collaborative delivery engine. The specific steps are as follows:

[0027] Based on the verification certificate chain, spatiotemporal feature verification parameters are generated by quantum-safe hashing algorithm and spatiotemporal convolution operation;

[0028] Based on spatiotemporal feature verification parameters, the SNN synaptic weight parameters are updated through a quantum-safe federated learning protocol.

[0029] The updated SNN synaptic weight parameters are synchronized to the 5G base station edge server and the supermarket local server through the quantum key distribution channel to form a dual-end dynamic parameter pool.

[0030] The multi-port collaborative delivery engine parses the dynamic parameter pool of both ends, generates cross-end push instructions that match time and space, and executes coupon push.

[0031] Secondly, this invention provides a multi-port coupon push device, comprising a data fusion module, a decision generation module, a strategy preloading module, a verification module, and a federated optimization module. The data fusion module collects multi-source heterogeneous data, preprocesses the data using a neuromorphic edge computing architecture, and generates a multimodal feature tensor. The decision generation module inputs the multimodal feature tensor into an SNN decision engine, generates SNN synaptic weight parameters based on the spatiotemporal features of user historical behavior, and dynamically adjusts port priorities using synaptic plasticity rules to generate a dynamic port strategy matrix. The strategy preloading module... The system consists of three modules: a module for generating real-time push instructions; a module for executing real-time push instructions; a module for verifying coupons; a module for executing coupons; a module for optimizing coupons; and a module for optimizing coupons. The module is designed to preload coupon policies using a memristor array based on a dynamic port policy matrix, and predict user trajectories using a quantum city spatiotemporal model. The module executes the real-time push instructions and embeds a topological photonic crystal verification code into the coupon propagation chain. It verifies the authenticity of user redemption using zero-knowledge proofs and generates a redemption voucher chain. The module is designed to update SNN synaptic weight parameters based on the redemption voucher chain using quantum-safe federated learning, and synchronize this update to the 5G base station edge server and the supermarket's local server. This multi-port collaborative delivery engine enables cross-platform coupon push.

[0032] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the multi-port coupon push method as described in the first aspect of the present invention.

[0033] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the multi-port coupon push method as described in the first aspect of the present invention.

[0034] The beneficial effects of this invention are as follows: by utilizing the spatiotemporal coding characteristics of spiking neural networks to eliminate environmental noise and fuse multi-source features, the feature distortion problem caused by data heterogeneity in edge computing is solved; furthermore, by dynamically adjusting port strategies through the synaptic plasticity rules of the SNN decision engine, the problem that static rule engines cannot respond to spatiotemporal behavior changes in real time is solved, and millisecond-level precise scheduling of multi-port resources is achieved. Attached Figure Description

[0035] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 A flowchart for a method of pushing coupons across multiple platforms.

[0037] Figure 2 A flowchart illustrating the generation of multimodal feature tensors for a coupon push method that utilizes multiple delivery platforms.

[0038] Figure 3 A flowchart illustrating the generation of a dynamic port strategy matrix for a multi-port coupon push method.

[0039] Figure 4 A flowchart for generating real-time push instructions for a coupon push method that is deployed across multiple platforms. Detailed Implementation

[0040] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0041] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0042] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0043] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a method for pushing coupons across multiple platforms, including the following steps:

[0044] S1: Collect multi-source heterogeneous data, preprocess the multi-source heterogeneous data through a neuromorphic edge computing architecture, and generate multimodal feature tensors.

[0045] S1.1: Multi-source heterogeneous data includes spatiotemporal positioning signals, radio frequency signals, platform transaction logs, and real-time port load data.

[0046] It should be noted that the spatiotemporal positioning signal refers to the spatiotemporal positioning signal of the user's mobile terminal, which obtains the user's real-time latitude and longitude coordinates and timestamp through GPS or Beidou navigation, and supplements the meter-level accuracy location information in indoor scenarios with indoor Bluetooth beacon or UWB positioning technology; the radio frequency signal refers to the radio frequency signal of the supermarket IoT device, which achieves sub-meter-level spatial positioning through signal strength fingerprint database matching; the platform transaction log refers to the transaction log of the third-party payment platform, which is stored in a time-series database and establishes an association index with the spatiotemporal positioning signal; the port real-time load data refers to the real-time monitoring index set of bandwidth utilization, TCP connection number and data packet queue delay of the 5G base station edge server communication port, which is collected through the SNMP protocol and normalized to a load coefficient in the range of 0 to 1.

[0047] S1.2: Eliminate environmental noise in radio frequency signals through a neuromorphic processing architecture, perform drift correction on spatiotemporal positioning signals, filter outliers in platform transaction logs, and generate purified multi-source heterogeneous data.

[0048] The specific process includes: the neuromorphic processing architecture processes radio frequency (RF) signals using the spatiotemporal coding characteristics of spiking neural networks; it identifies pulse timing patterns in RF signals through an adaptive pulse firing rate adjustment mechanism; and it separates effective signals from environmental electromagnetic noise using the membrane potential integral characteristics of LIF neurons. The adaptive pulse firing rate adjustment mechanism identifies and eliminates environmental electromagnetic interference; it corrects positioning deviations through synaptic weight updates; and it dynamically adjusts synaptic connection strength based on the deviation between the timestamp of the positioning signal reported by the user's mobile terminal and the base station reference clock, achieved through pulse timing-dependent plasticity rules. In platform transaction log processing, abnormal transaction records are detected and filtered using pulse firing pattern recognition. The processed RF signal, corrected spatiotemporal positioning signal, and purified platform transaction logs are aligned in the pulse coding space of the neuromorphic processing architecture to generate purified multi-source heterogeneous data that retains spatiotemporal correlation features and semantic consistency.

[0049] S1.3: Input the purified multi-source heterogeneous data into the feature extraction layer of the neuromorphic processing architecture to generate spatiotemporal feature vectors of user behavior, product interaction feature vectors, and consumer preference feature vectors.

[0050] The specific process includes: when the neuromorphic processing architecture extracts features from the purified multi-source heterogeneous data, it analyzes the movement trajectory patterns in radio frequency signals and spatiotemporal positioning signals, and uses a pulse temporal window segmentation algorithm to generate spatiotemporal feature vectors of user behavior; at the same time, it processes the product identification sequence and purchase records in the platform's transaction logs, and constructs product interaction feature vectors based on the statistical characteristics of pulse issuance frequency; and combines the distribution characteristics of transaction amount and time interval patterns, it uses pulse phase coding technology to form consumer preference feature vectors; the entire architecture achieves synchronous alignment of the time dimensions of each feature vector through a hierarchical pulse issuance mechanism, and finally outputs three types of structured data in parallel: user behavior spatiotemporal feature vectors, product interaction feature vectors, and consumer preference feature vectors.

[0051] Furthermore, the neuromorphic processing architecture utilizes the spatiotemporal coding characteristics of spiking neural networks to process multi-source heterogeneous data. It employs an adaptive pulse firing rate adjustment mechanism to eliminate environmental electromagnetic interference in radio frequency signals, while simultaneously establishing a pulse timing-based coordinate compensation model to correct drift errors in spatiotemporal positioning signals. Pulse firing pattern recognition is used to filter abnormal records in the platform's transaction logs. The purified data is then input into the feature extraction layer. The spatiotemporal feature encoding unit parses movement trajectory patterns and generates user behavior spatiotemporal feature vectors through pulse timing window segmentation. The product interaction analysis unit processes purchase records and constructs product interaction feature vectors based on pulse firing frequency. The consumer preference extraction unit integrates transaction features and uses pulse phase encoding to form consumer preference feature vectors. In the pulse fusion layer, a phase synchronization mechanism aligns the time dimensions of each feature vector. A multidimensional tensor splicing method integrates movement trajectory, product preference, and consumption habit features. The synaptic connection matrix eliminates redundancy in the spliced ​​features. The final trained architecture comprises a three-layer spiking neural network, with synaptic weights optimized through a backpropagation temporal dependency algorithm. Deployed on a neuromorphic chip, it achieves real-time processing capabilities for synaptic operations, with feature extraction errors controlled within 3%.

[0052] S1.4: Through the pulse fusion layer of the neuromorphic processing architecture, the spatiotemporal feature vectors of user behavior, product interaction feature vectors, and consumption preference feature vectors are concatenated into tensors to generate a multimodal feature tensor.

[0053] The specific process includes: the pulse fusion layer of the neuromorphic processing architecture receives spatiotemporal feature vectors of user behavior, product interaction, and consumption preference; aligns the time dimensions of the three feature vectors through a pulse phase synchronization mechanism; uses a multidimensional tensor concatenation method in pulse coding space to fuse the spatiotemporal feature vectors of user behavior, product interaction, and consumption preference into tensors; the synaptic connection matrix in the pulse fusion layer adjusts the weights of the concatenated features to eliminate redundant information between features; and finally outputs a multimodal feature tensor containing spatiotemporal behavior features, product preference features, and consumption pattern features, with each dimension of features maintaining semantic consistency and spatiotemporal correlation.

[0054] S2: Input the multimodal feature tensor into the SNN decision engine, generate SNN synaptic weight parameters based on the spatiotemporal features of user historical behavior, and dynamically adjust port priorities through synaptic plasticity rules to generate a dynamic port strategy matrix.

[0055] S2.1: Input the multimodal feature tensor into the spatiotemporal feature encoding layer of the SNN decision engine, extract the spatiotemporal features of the user's historical behavior through the pulse time window segmentation algorithm, generate the encoded spatiotemporal feature vector, and generate the SNN synaptic weight parameters based on the correlation between the frequency of user behavior and the timing of port interaction.

[0056] The specific process includes: multimodal feature tensors are fed into the spatiotemporal feature encoding layer of the SNN decision engine; a pulse time window segmentation algorithm decomposes user historical behavior data into discrete pulse event sequences; and LIF neurons are used to perform phase encoding on the spatiotemporal coordinates. The encoded spatiotemporal feature vectors are then processed by a synaptic weight generation unit, which uses Gaussian pulse temporal dependence plasticity rules to analyze the temporal correlation between user behavior pulse firing rate and port activation events, thereby deriving SNN synaptic weight parameters that reflect the strength of the behavior-port association. During the weight parameter generation process, 5G base station positioning data and supermarket indoor Bluetooth beacon information are simultaneously integrated to ensure accurate mapping between spatiotemporal features and physical port locations. The final output SNN synaptic weight parameters include multi-dimensional spatiotemporal association features such as user movement trajectory patterns, dwell time distribution, and port interaction frequency.

[0057] User behavior frequency data is extracted from the spatiotemporal feature vector of user behavior in the multimodal feature tensor, specifically from spatiotemporal positioning signals and statistics of user historical access frequency and interaction time interval recorded in platform transaction logs.

[0058] Port interaction time-series data is extracted from real-time port load data in multi-source heterogeneous data, specifically from TCP connection timestamps and interaction event interval sequences recorded by the communication ports of 5G base station edge servers.

[0059] The SNN decision engine consists of a spatiotemporal feature encoding layer, a synaptic plasticity regulation layer, and a multimodal fusion dimensionality reduction layer. It is trained using a pulse time window segmentation algorithm and Gaussian pulse temporal dependence plasticity rules and needs to be deployed in conjunction with the neuromorphic computing chip hardware of a 5G base station edge server.

[0060] Spatiotemporal coordinates refer to the precise positioning data of a user's mobile terminal in three-dimensional physical space and time dimension, which is obtained through the fusion positioning technology of 5G base station positioning system and indoor Bluetooth beacon in shopping malls.

[0061] S2.2: Input the SNN synaptic weight parameters into the synaptic plasticity control layer, dynamically adjust the port priority through Gaussian pulse time-dependent plasticity rules, and output the optimized SNN synaptic weight distribution.

[0062] The specific process includes: after the SNN synaptic weight parameters are input into the synaptic plasticity control layer, the Gaussian pulse time-series dependent plasticity rule analyzes the time interval distribution of the input pulse sequence to obtain the time difference between adjacent pulse firing; the time difference is weighted according to the Gaussian distribution function, with recent interaction events receiving higher weights and historical events having their weights decaying over time; the synaptic plasticity control layer maps the weight allocation results to port priority scores, strengthening the synaptic connections of neural pathways corresponding to high-frequency interaction ports and weakening the pathway connections of low-frequency ports; during the dynamic adjustment process, the pulse firing rate is synchronized with the port load status in real time to ensure that the weight distribution matches the current network conditions; finally, the optimized SNN synaptic weight distribution is output, which simultaneously reflects user behavior habits and port resource status, forming adaptive decision parameters.

[0063] Port priority refers to the service priority level of each communication port of the 5G base station edge server in the coupon push task. It is dynamically derived by analyzing the correlation between user behavior frequency and port interaction timing through the SNN decision engine and combining the Gaussian impulse timing dependency plasticity rule.

[0064] The synaptic plasticity regulation layer is an extension of the spatiotemporal feature encoding layer built into the SNN decision engine, serving as a functional component in the spiking neural network specifically responsible for dynamically adjusting the strength of synaptic connections.

[0065] The time interval distribution of a pulse sequence refers to the statistical characteristics of the time difference between user interaction events with different ports. Specifically, it is represented by the probability density function in a spiking neural network that records the time difference between adjacent interaction events, and follows a Gaussian distribution.

[0066] S2.3: Based on the optimized SNN synaptic weight distribution, tensor outer product operation is performed using real-time port load data to generate a dynamic port policy four-dimensional tensor. Then, a dynamic port policy matrix is ​​generated through multimodal fusion dimensionality reduction. The expression is:

[0067] ;

[0068] in, express The dynamic port policy four-dimensional tensor generated in real time. Represents a time variable. express The 3D tensor of synaptic weights learned by the time-lapse spiking neural network. This represents the tensor outer product operation. express A two-dimensional matrix of port load data collected in real time. It represents the Hadamah accumulation. This represents the natural exponential function. This represents the time interval between the current moment and the time when the policy was generated. The time constant representing the memory decay rate of the control strategy. Noise injection intensity coefficient (0 < 0) represents the noise injection intensity coefficient to prevent strategy rigidity. <1), This represents a random noise tensor that follows a standard normal distribution.

[0069] The specific process includes: after aligning the optimized SNN synaptic weight distribution with the real-time port load data in the spatiotemporal dimensions, performing a tensor outer product operation; combining the three-dimensional tensor of synaptic weights with the two-dimensional matrix of port loads through a fully connected operation to form a dynamic port policy four-dimensional tensor containing four-dimensional features of time, space, weight, and load; the multimodal fusion dimensionality reduction process uses the Hadamard product to perform feature cross-calculation on the four-dimensional tensor, and the natural exponential function performs a nonlinear transformation on the time decay factor to control the memory retention strength of historical policies; the noise injection intensity coefficient adjusts the superposition ratio of the standard normal distribution random noise tensor to prevent policy optimization from getting trapped in local optima; the finally generated dynamic port policy matrix fully retains the key decision features while meeting the requirements of real-time computing efficiency.

[0070] The synaptic weight 3D tensor is output by the synaptic plasticity control layer of the SNN decision engine and is generated by analyzing the correlation between user behavior frequency and port interaction timing through Gaussian impulse temporal dependence plasticity rules.

[0071] The port load 2D matrix is ​​obtained by normalizing port bandwidth utilization and TCP connection count data collected in real time from the 5G base station edge server.

[0072] S3: Based on the dynamic port strategy matrix, coupon strategies are preloaded through the memristor array and combined with the quantum city spatiotemporal model to predict user trajectories and generate real-time push instructions.

[0073] S3.1: Based on the row vector physical port mapping relationship of the dynamic port policy matrix, the priority parameters of the coupon policy are linearly mapped to the conduction values ​​of the memristor array, and the policy tensor is generated by the resistance state modulation of the dual memristor cross array.

[0074] The specific process includes: after establishing a one-to-one mapping relationship between the row vectors of the dynamic port strategy matrix and the physical ports, the priority parameters of the coupon strategy are converted into normalized values ​​in the 0-1 range through a linear transformation function; the memristor array receives the normalized priority parameters and converts each normalized priority parameter value into the conductivity value of the corresponding memristor unit, with the conductivity value being directly proportional to the priority; the dual memristor cross array adopts resistive cross modulation technology, adjusting the memristor resistance value through a voltage pulse sequence, so that adjacent units form a complementary conductivity mode; the spatial distribution pattern of the conductivity state characterizes the spatiotemporal preference characteristics of the coupon distribution strategy, and finally generates a strategy tensor containing port priority, distribution timing, and resource allocation strategy.

[0075] The row vector physical port mapping relationship refers to the one-to-one correspondence between each row vector of the dynamic port policy matrix and the physical communication port of the 5G base station edge server. It is derived by analyzing the spatiotemporal correlation between user behavior characteristics and port load status through the SNN decision engine.

[0076] The priority parameter refers to the execution priority level of the coupon push task on different communication ports, which is derived by converting the row vectors of the dynamic port strategy matrix to the physical port mapping relationship.

[0077] S3.2: Input the policy tensor into the quantum city spatiotemporal model, perform gridded encoding and spatiotemporal entanglement calculation on the user's historical trajectory, and generate a predicted coordinate sequence tensor, the expression of which is:

[0078] ;

[0079] in, Represents the predicted coordinate sequence tensor. This represents the spatiotemporal convolutional neural network operator. This represents the octree gridding coding function. Represents a user's unique identifier. This represents the user's historical trajectory data. Represents the spatiotemporal convolution kernel tensor. This represents a four-dimensional spatiotemporal convolution operation. This represents the policy tensor generated by the memristor array. Indicates the dynamic attenuation coefficient (0.1≤ ≤10), Indicates the index of the currently processed trajectory segment. This represents the total number of trajectory segments. Indicates the first The degree of quantum entanglement of a trajectory segment Indicates the first Spatial coordinate sequence matrix of trajectory segments Represents the urban spatiotemporal tensor. Represents the spacetime entanglement operator, The topological feature extraction function representing the spatiotemporal trajectory of user behavior. This represents the relative entropy measure between the candidate node characteristics and the query conditions.

[0080] The specific process includes: after the policy tensor is input into the quantum city spatiotemporal model, the octree gridding encoding function divides the user's historical trajectory data matrix into spatiotemporal cube units of equal volume, each unit containing location coordinates and timestamp information; the spatiotemporal convolutional neural network operator performs four-dimensional spatiotemporal convolution operations on the gridded trajectory, and the spatiotemporal convolution kernel tensor slides synchronously in the three-dimensional spatial and temporal dimensions to extract movement pattern features; the quantum entanglement degree calculation unit evaluates the spatial distribution correlation of each trajectory segment, and establishes quantum correlations for cross-regional movement through the spatiotemporal entanglement operator; the dynamic decay coefficient adjusts the influence weight of historical trajectory data on the current prediction, and the policy tensor generated by the memristor array provides port resource constraints; the final output prediction coordinate sequence tensor integrates spatiotemporal convolution features, quantum entanglement correlations, and resource optimization strategies.

[0081] Furthermore, the training process of the quantum city spatiotemporal model divides massive user historical trajectory data into standardized spatiotemporal cube units using an octree gridding encoding function. Each unit accurately records movement characteristics within a 10m×10m×1m spatial range and a 5-minute time window. Then, the spatiotemporal convolutional neural network operator uses a four-dimensional convolutional kernel that slides synchronously in the three spatial and temporal dimensions. The kernel weights are optimized using a gradient descent algorithm to extract the spatiotemporal correlation patterns of the movement trajectories. Simultaneously, the quantum entanglement calculation unit analyzes the spatial distribution density and temporal overlap of trajectory segments, establishing a quantitative model reflecting user movement patterns. The sub-state correlation matrix is ​​used, where the cross-region correlation strength is calculated by weighting with an exponential decay function. During training, the dynamic decay coefficient automatically adjusts the influence weight of historical trajectories according to the real-time network load, and the policy tensor generated by the memristor array is incorporated into the model as a resource constraint. The final output predicted coordinate sequence tensor integrates spatiotemporal convolution features, quantum entanglement correlation, and resource optimization strategies, and maintains the model's timeliness through a 24-hour incremental update mechanism. The entire training process ensures a 93% sparsity of the quantum correlation matrix while reducing trajectory prediction error by 42% and keeping resource allocation matching error within 8%.

[0082] Historical trajectory data refers to the continuous sequence of location coordinates and their corresponding timestamps recorded by the user's mobile terminal in the spatiotemporal dimension, including GPS / BeiDou positioning data, indoor Bluetooth beacon positioning records, and spatiotemporal correlation data in the platform transaction logs.

[0083] S3.3: Perform a four-dimensional spatiotemporal convolution operation on the predicted coordinate sequence tensor and the policy tensor. Generate a comprehensive score tensor for candidate nodes using an attention-weighted kernel function, and filter the set of valid candidate nodes based on a dynamic fusion threshold. The expression is as follows:

[0084] ;

[0085] in, This represents the comprehensive score tensor of the candidate nodes. Indicates causal constraint. This represents the tensor shrinking operator under causal constraints. This indicates the total number of heads of attention. Indicates the index of the current attention head. Represents the spacetime curvature gradient operator. This represents a feature extraction function based on the spatiotemporal curvature gradient. This represents the operation of the hyperbolic space tensor product. Representing three-dimensional hyperbolic space, This represents the policy tensor projection function based on three-dimensional hyperbolic space. Indicates the first Feature dimension scaling factor for each attention head Represents the manifold compression function. This indicates dynamic deformability. Represents the policy tensor. This indicates that a spatiotemporal convolution operation is performed on the policy tensor using a dynamically deformable convolution kernel. This represents a spatiotemporal feature fusion operation based on depthwise separable convolution.

[0086] The specific process includes: aligning the predicted coordinate sequence tensor and the policy tensor in four-dimensional spatiotemporal dimensions; performing spatiotemporal convolution on the policy tensor using a dynamically deformable convolution kernel to generate a feature representation adapted to local spatiotemporal characteristics; integrating the feature extraction results of multiple attention heads using an attention weighting kernel function, with each attention head using a different feature dimension scaling factor to transform the input features; analyzing the geometric distribution characteristics of the predicted coordinate sequence tensor using a spatiotemporal curvature gradient operator; calculating the nonlinear correlation between nodes based on the hyperbolic space tensor product operation in three-dimensional hyperbolic space; ensuring that temporal dependencies are not destroyed using a tensor shrinking operator under causal constraints; and reducing the dimensionality of high-dimensional features using a manifold compression function; finally, the comprehensive score tensor of the generated candidate nodes is filtered by a dynamic fusion threshold, retaining nodes with score tensors higher than the dynamic fusion threshold to form a set of valid candidate nodes.

[0087] The dynamic fusion threshold is dynamically set based on the statistical distribution characteristics of the comprehensive score tensor of the candidate nodes, and is determined by analyzing the sum of the mean and standard deviation of the score tensor.

[0088] S3.4: When the quantum-clock offset in the set of valid candidate nodes meets the synchronization tolerance threshold, the dynamic port strategy matrix calculates the bandwidth allocation coefficient of the target port and generates a real-time push command, expressed as:

[0089] ;

[0090] in, Represents the broadband allocation coefficient (0≤ ≤1), Represents the nonlinear attenuation coefficient (0 < <1), Indicates quantum clock bias, This indicates the synchronization tolerance threshold. Indicates the curvature index. Indicates the port's base bandwidth. Represents the gradient field of the policy matrix. Indicates the number of packets at odd positions. This represents the load balancing sensitivity index. This represents the total number of data packets. Represents the hyperbolic tangent function. Represents the spatial dimension feature tensor. Represents the time-dimensional feature tensor.

[0091] The specific process includes: after the quantum-clock deviation in the set of effective candidate nodes reaches the synchronization tolerance threshold, the dynamic port strategy matrix converts the ratio of quantum-clock deviation to synchronization tolerance threshold into a nonlinear decay coefficient using a hyperbolic tangent function; the gradient field of the strategy matrix extracts the port priority distribution characteristics, and the curvature index adjusts the sensitivity of bandwidth allocation to spatiotemporal geometric features; the ratio of odd-position packet count to total packet count reflects the instantaneous load status of the port, and the load balancing sensitivity index controls the bandwidth adjustment magnitude; the port baseline bandwidth, combined with the nonlinear decay coefficient, curvature adjustment factor, and load balancing parameters, is normalized to generate the bandwidth allocation coefficient for the target port; finally, real-time push instructions are generated based on the bandwidth allocation coefficient to ensure accurate matching between the coupon distribution task and the network resource status.

[0092] Quantum-clock skew refers to the time synchronization difference between the clock signal of the 5G base station edge server and the quantum state evolution process of the user's mobile terminal. It is naturally generated and recorded when the quantum city spatiotemporal model processes user trajectory data.

[0093] The synchronization tolerance threshold is dynamically derived based on the time-frequency fault tolerance specifications of the 5G network synchronization signal matrix and the entangled state stability requirements of the quantum city spatiotemporal model.

[0094] S4: Execute real-time push instructions and embed topological photonic crystal verification codes into the coupon propagation chain. Verify the authenticity of user redemption through zero-knowledge proof and generate a redemption voucher chain.

[0095] S4.1: Embed a topological photonic crystal verification code in the coupon propagation link corresponding to the real-time push command, and pre-configure a decoding verification interface.

[0096] The specific process includes: in the coupon propagation link corresponding to the real-time push command, the topological photonic crystal verification code is implanted with carrier phase through photonic bandgap modulation technology to form an optical signature matching the link topology; the decoding verification interface is pre-installed in the optical signal processing unit of the 5G base station edge server, and the photonic crystal features are identified using a Mach-Zehnder interferometer structure; the binding relationship between the lattice constant of the verification code and the coupon identifier is established through nonlinear optical transformation to ensure that the propagation process is verifiable; the photonic crystal defect mode corresponds to a specific decoding key, which triggers the photoelectric conversion detection of the verification interface when the coupon arrives at the user terminal; the complete transmission link maintains the phase consistency of the topological photonic crystal verification code, while maintaining the real-time response capability of the decoding verification interface.

[0097] S4.2: When the user terminal accesses the decoding verification interface, it decodes and verifies the optical feature information through the zero-knowledge proof protocol and generates an interactive proof document.

[0098] The specific process includes: when the user terminal accesses the decoding verification interface, the zero-knowledge proof protocol extracts the optical feature information of the topological photonic crystal verification code and verifies the authenticity of the photonic bandgap mode through a non-interactive proof process; the verification process uses elliptic curve cryptography to construct a commitment scheme, converting the photonic crystal parameters into verifiable mathematical propositions; the decoding verification interface generates challenge parameters through a quantum random number generator, and the user terminal analyzes the response proof based on the challenge parameters; the interactive proof file records the complete Sigma protocol interaction process, including the three elements of initial commitment, challenge parameters, and response proof; the successfully verified interactive proof file is automatically appended with a digital signature, forming an auditable verification credential, ensuring that the integrity and authenticity of the coupon propagation chain are cryptographically proven.

[0099] Optical feature information refers to the unique photonic bandgap mode and phase distribution characteristics formed by the topological photonic crystal verification code in the propagation link, which are collected and extracted by the optical signal processing unit of the 5G base station edge server.

[0100] S4.3: Perform spatiotemporal matching between the interactive proof documents and the coupon redemption request, and generate a redemption voucher chain with a multi-level verification data structure.

[0101] The specific process includes: aligning the timestamps and geographic coordinates of the interactive proof document and the coupon redemption request using a spatiotemporal matching engine to establish a spatiotemporal correlation mapping; constructing a multi-level verification data structure using a Merkle tree, with the digital signature of the interactive proof document, the coupon identifier, and the redemption request parameters as leaf nodes; generating the redemption certificate chain using blockchain technology, with each block containing the hash value of the previous block, the timestamp of the current verification data, and the hierarchical verification result; triggering the execution of a smart contract based on the spatiotemporal matching result, writing a new redemption certificate under the conditions of photonic crystal verification and zero-knowledge proof; and finally generating a redemption certificate chain that completely records the trusted data of the entire process from optical feature verification to coupon redemption, forming an immutable redemption audit trail.

[0102] A coupon redemption request refers to a verification instruction submitted by a user terminal in a consumption scenario, which includes a coupon identifier, redemption timestamp, and geolocation information. It is generated either by the user actively triggering the consumption behavior or by the POS terminal automatically detecting and generating the request.

[0103] S5: Based on the verification voucher chain, it updates the SNN synaptic weight parameters through quantum-safe federated learning and synchronizes them to the 5G base station edge server and the supermarket local server. It then completes the cross-terminal push of coupons through a multi-port collaborative delivery engine.

[0104] S5.1: Based on the verification certificate chain, spatiotemporal feature verification parameters are generated through quantum-secure hashing algorithm and spatiotemporal convolution operation. The expression is:

[0105] ;

[0106] in, Represents the spatiotemporal feature verification parameters. Represents a standard hash function. Indicates the chain identifier for reconciliation vouchers. Represents a timestamp. Represents geographic coordinates, Indicates the size of the sliding time window. This indicates the sequence number of the current write-off event within the time window. Differential units representing the characteristics of write-off events The differential unit representing the time variable. Differential units representing spatial dimensions Indicates the first The dynamic weighting coefficient of each write-off event (0.24≤ ≤1).

[0107] The specific process includes: after the verification voucher chain is input into the quantum-safe hash algorithm, the standard hash function performs one-way encryption on the verification voucher chain identifier; spatiotemporal convolution operation processes the verification events by sequence number within a sliding time window, and the differential unit extracts the change features of the timestamp sequence and geographic coordinates respectively; the dynamic weight coefficient is automatically adjusted according to the distribution density of the verification events in the time window, with recent events receiving higher weights; the quantum-safe hash algorithm performs tensor product operation on the hash result and the spatiotemporal differential features to generate spatiotemporal feature verification parameters that integrate time dimension, spatial dimension, and voucher integrity; the expression fully describes the calculation process of hash encryption, spatiotemporal differentiation, and dynamic weighting, ensuring that the output spatiotemporal feature verification parameters have both quantum-resistant and spatiotemporal correlation characteristics.

[0108] S5.2: Verify parameters based on spatiotemporal features and update SNN synaptic weight parameters through a quantum-safe federated learning protocol.

[0109] The specific process includes: after the spatiotemporal feature verification parameters are input into the quantum-safe federated learning protocol, the privacy of the parameters is protected by homomorphic encryption technology, and distributed analysis is performed in the encrypted state; a quantum key distribution network establishes a secure communication channel to ensure that the parameter exchange between the SNN decision engine and the edge nodes cannot be eavesdropped; the federated averaging algorithm aggregates the spatiotemporal feature gradients from multiple edge nodes, and a quantum random number generator is used to add differential privacy noise; the update process retains the pulse timing dependence of the SNN synaptic weight parameters, and the federated learning results are converted into neuromorphically tractable weight adjustment quantities through pulse firing rate encoding; the finally generated updated SNN synaptic weight parameters fuse global spatiotemporal feature patterns, while simultaneously meeting the requirements of quantum security and the plasticity of spiking neural networks.

[0110] S5.3: The updated SNN synaptic weight parameters are synchronized to the 5G base station edge server and the supermarket local server through the quantum key distribution channel to form a dual-end dynamic parameter pool.

[0111] The specific process includes: when the updated SNN synaptic weight parameters are transmitted through the quantum key distribution channel, a quantum random key is generated using the BB84 protocol to encrypt the parameters once; the 5G base station edge server and the supermarket local server establish a secure channel through quantum entangled photon pairs, and the key distribution process has information theory security; the dual-end dynamic parameter pool is maintained using a blockchain structure, and each synchronization operation generates a new block to record the parameter version and timing mark; after receiving the encrypted parameters, the edge server restores the SNN synaptic weight parameters through a quantum decryption module and verifies the quantum signature to ensure integrity; when the supermarket local server updates the parameter pool synchronously, it performs the same quantum decryption and verification process, and finally the dual-end dynamic parameter pool maintains strict parameter consistency and timing synchronization.

[0112] S5.4: The multi-port collaborative delivery engine parses the dynamic parameter pool of both ends, generates cross-end push instructions that match time and space, and executes coupon push.

[0113] The specific process includes: a multi-port collaborative delivery engine extracts the latest SNN synaptic weight parameters and spatiotemporal feature verification parameters from the dual-end dynamic parameter pool, and reconstructs user behavior patterns through a spiking neural network decoder; a spatiotemporal matching algorithm aligns the port status data of the 5G base station edge server with the coupon inventory data of the supermarket's local server using tensors, and selects the delivery path combination with the highest spatiotemporal feature matching degree and balanced resource load through tensor outer product operation combined with attention mechanism weighted scoring, thus obtaining the optimal delivery path; the cross-end push instruction generation process integrates quantum clock synchronization signals and port priority parameters to construct a composite instruction that includes time windows, geofencing, and resource allocation strategies; the coupon push execution unit dynamically adjusts the transmission power and modulation mode based on the instruction parameters to ensure the completion of millimeter wave and Sub-6GHz dual-band collaborative transmission within the specified spatiotemporal range; and finally, the delivery results are fed back to the dual-end dynamic parameter pool in real time, forming a closed-loop optimization mechanism.

[0114] This embodiment also provides a multi-port coupon push device, including: a data fusion module, a decision generation module, a strategy preloading module, a redemption verification module, and a federated optimization module. The data fusion module collects multi-source heterogeneous data, preprocesses the multi-source heterogeneous data using a neuromorphic edge computing architecture, and generates a multimodal feature tensor. The decision generation module inputs the multimodal feature tensor into an SNN decision engine, generates SNN synaptic weight parameters based on the spatiotemporal features of user historical behavior, and dynamically adjusts port priorities through synaptic plasticity rules to generate a dynamic port strategy matrix. The strategy preloading module... The system is designed to generate real-time push instructions based on a dynamic port policy matrix, preload coupon policies using a memristor array, and predict user trajectories using a quantum city spatiotemporal model. A verification module executes these instructions and embeds a topological photonic crystal verification code into the coupon propagation chain. Zero-knowledge proofs are used to verify the authenticity of user verification, generating a verification voucher chain. A federated optimization module updates SNN synaptic weight parameters based on the verification voucher chain using quantum-safe federated learning and synchronizes the updates to 5G base station edge servers and supermarket local servers. This multi-port collaborative delivery engine enables cross-platform coupon pushes.

[0115] This embodiment also provides a computer device applicable to the multi-port coupon push method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the multi-port coupon push method proposed in the above embodiment.

[0116] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0117] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the coupon push method for multi-port delivery as proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0118] In summary, this invention solves the feature distortion problem caused by data heterogeneity in edge computing by: eliminating environmental noise and fusing multi-source features using the spatiotemporal coding characteristics of spiking neural networks; furthermore, by dynamically adjusting port strategies using the synaptic plasticity rules of the SNN decision engine, it solves the problem that static rule engines cannot respond to spatiotemporal behavior changes in real time, and achieves millisecond-level precise scheduling of multi-port resources.

[0119] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for pushing coupons across multiple platforms, characterized in that: include, Collect multi-source heterogeneous data, preprocess the multi-source heterogeneous data through a neuromorphic processing architecture, and generate multimodal feature tensors; The multimodal feature tensor is input into the SNN decision engine. Based on the spatiotemporal features of the user's historical behavior, SNN synaptic weight parameters are generated. Then, the port priorities are dynamically adjusted through synaptic plasticity rules to generate a dynamic port policy matrix. The specific steps are as follows. The multimodal feature tensor is input into the spatiotemporal feature encoding layer of the SNN decision engine. The spatiotemporal features of the user's historical behavior are extracted by the pulse time window segmentation algorithm to generate the encoded spatiotemporal feature vector. Based on the correlation between the frequency of user behavior and the timing of port interaction, the SNN synaptic weight parameters are generated. The SNN synaptic weight parameters are input into the synaptic plasticity control layer, and the port priority is dynamically adjusted by the Gaussian pulse time-dependent plasticity rule to output the optimized SNN synaptic weight distribution. Based on the optimized SNN synaptic weight distribution, tensor outer product operation is performed in combination with real-time port load data to generate a dynamic port policy four-dimensional tensor, and a dynamic port policy matrix is ​​generated by multimodal fusion dimensionality reduction. Based on a dynamic port strategy matrix, coupon strategies are preloaded using a memristor array, and user trajectories are predicted using a quantum city spatiotemporal model to generate real-time push notifications. The specific steps are as follows. Based on the row vector physical port mapping relationship of the dynamic port policy matrix, the priority parameters of the coupon policy are linearly mapped to the conduction values ​​of the memristor array, and the policy tensor is generated by the resistance state modulation of the dual memristor cross array. Input the strategy tensor into the quantum city spatiotemporal model, perform grid-based encoding and spatiotemporal entanglement calculation on the user's historical trajectory, and generate a predicted coordinate sequence tensor. The predicted coordinate sequence tensor and the policy tensor are subjected to four-dimensional spatiotemporal convolution operation. The comprehensive score tensor of the candidate nodes is generated by the attention weighted kernel function, and the set of effective candidate nodes is selected according to the dynamic fusion threshold. When the quantum-clock deviation in the set of valid candidate nodes meets the synchronization tolerance threshold, the dynamic port strategy matrix calculates the bandwidth allocation coefficient of the target port and generates a real-time push command. The system executes real-time push notifications and embeds a topological photonic crystal verification code into the coupon propagation chain. Zero-knowledge proofs are used to verify the authenticity of user redemption, generating a redemption voucher chain. The specific steps are as follows. A topological photonic crystal verification code is embedded in the coupon propagation link corresponding to the real-time push command, and a decoding verification interface is pre-configured. When a user terminal accesses the decoding verification interface, it decodes and verifies the optical feature information using a zero-knowledge proof protocol and generates an interactive proof file. The interaction verification documents are matched spatiotemporally with the coupon redemption request, and a redemption voucher chain with a multi-level verification data structure is generated. Based on the verification voucher chain, the SNN synaptic weight parameters are updated through quantum-safe federated learning and synchronized to the 5G base station edge server and the supermarket local server. The coupons are then pushed across terminals through a multi-port collaborative delivery engine.

2. The coupon push method with multi-port delivery as described in claim 1, characterized in that: The multi-source heterogeneous data includes spatiotemporal positioning signals, radio frequency signals, platform transaction logs, and real-time port load data.

3. The coupon push method with multi-port delivery as described in claim 1, characterized in that: The specific steps for generating the multimodal feature tensor are as follows: The system eliminates environmental noise in radio frequency signals by using a neuromorphic processing architecture, performs drift correction on spatiotemporal positioning signals, filters outliers in platform transaction logs, and generates purified multi-source heterogeneous data. The purified multi-source heterogeneous data is input into the feature extraction layer of the neuromorphic processing architecture to generate spatiotemporal feature vectors of user behavior, product interaction feature vectors, and consumer preference feature vectors. By using the pulse fusion layer of the neuromorphic processing architecture, the spatiotemporal feature vectors of user behavior, the feature vectors of product interaction, and the feature vectors of consumer preferences are concatenated into tensors to generate a multimodal feature tensor.

4. The coupon push method with multi-port delivery as described in claim 3, characterized in that: The aforementioned system, based on the verification voucher chain, updates the SNN synaptic weight parameters through quantum-safe federated learning and synchronizes them to the 5G base station edge server and the supermarket's local server. A multi-port collaborative delivery engine then pushes coupons across multiple platforms. The specific steps are as follows: Based on the verification certificate chain, spatiotemporal feature verification parameters are generated by quantum-safe hashing algorithm and spatiotemporal convolution operation; Based on spatiotemporal feature verification parameters, the SNN synaptic weight parameters are updated through a quantum-safe federated learning protocol. The updated SNN synaptic weight parameters are synchronized to the 5G base station edge server and the supermarket local server through the quantum key distribution channel to form a dual-end dynamic parameter pool. The multi-port collaborative delivery engine parses the dynamic parameter pool of both ends, generates cross-end push instructions that match time and space, and executes coupon push.

5. A coupon push device for multi-port distribution, based on the coupon push method for multi-port distribution according to any one of claims 1 to 4, characterized in that: It includes a data fusion module, a decision generation module, a policy preloading module, a verification module, and a federated optimization module. The data fusion module is used to collect multi-source heterogeneous data, preprocess the multi-source heterogeneous data through a neuromorphic edge computing architecture, and generate multimodal feature tensors. The decision generation module is used to input multimodal feature tensors into the SNN decision engine, generate SNN synaptic weight parameters based on the spatiotemporal features of user historical behavior, and dynamically adjust port priorities through synaptic plasticity rules to generate a dynamic port strategy matrix. The strategy preloading module is used to preload coupon strategies based on a dynamic port strategy matrix through a memristor array, and combine it with a quantum city spatiotemporal model to predict user trajectories and generate real-time push instructions. The verification module is used to execute real-time push instructions and embed topological photonic crystal verification codes in the coupon propagation chain. It verifies the authenticity of user verification through zero-knowledge proof and generates a verification voucher chain. The federated optimization module is used to update the SNN synaptic weight parameters based on the verification voucher chain through quantum-safe federated learning, and synchronize them to the 5G base station edge server and the supermarket local server. It then completes the cross-terminal push of coupons through a multi-port collaborative delivery engine.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the multi-port coupon push method according to any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the multi-port coupon delivery method according to any one of claims 1 to 4.

Images