Intelligent marketing terminal software and hardware collaborative adaptation method and system

Abstract

The application discloses a smart marketing terminal software and hardware collaborative adaptation method and system, and belongs to the technical field of intelligent terminals, comprising the following steps: obtaining a hierarchical function module graph, a directional function linkage link and a function authority control domain by dynamically analyzing a data stream root node through a data interaction dynamic analysis mechanism; constructing an intelligent neural decision topology network by combining an enhanced deep reinforcement learning path planning algorithm, a disk adaptive enhanced learning network and an asynchronous scheduling dynamic evolution criterion, and simultaneously converting the function authority control domain into a function coupling strength embedded network to output an optimized intelligent neural decision topology network; then, based on the optimized network, an adaptation architecture is intelligently generated through a hierarchical interaction mapping and an asynchronous disk strategy intelligent generation mechanism; finally, the adaptation architecture is applied and verified, and if the adaptation requirement is not met, the analysis is re-performed, and if the adaptation requirement is met, the adaptation is completed; the adaptation precision of the subsystem and the data disk stability are improved, and the application is suitable for a smart marketing terminal data processing scene.

Description

Technical Field

[0001] This invention belongs to the field of smart terminal technology, specifically a method and system for the collaborative adaptation of software and hardware in smart marketing terminals. Background Technology

[0002] With the rapid development of smart marketing terminals, their data processing capabilities and functional complexity are constantly increasing. The asynchronous data storage system, as a core component of smart marketing terminals, is responsible for the asynchronous storage, processing, and interaction of terminal data. Due to differences in hardware configurations and software environments among different smart marketing terminals, the asynchronous data storage system needs to be effectively adapted to the terminal's hardware and software environment to ensure the stability and efficiency of data processing. In existing technologies, the adaptation of asynchronous data storage systems for smart marketing terminals mostly adopts a static configuration approach, lacking dynamic adjustment capabilities and struggling to cope with dynamic changes in the terminal's hardware and software environment. Furthermore, the adaptation process does not fully consider the interrelationships and access control between functional modules, resulting in low adaptation accuracy and poor coordination, affecting the overall performance of the smart marketing terminal. Therefore, a method is needed to achieve dynamic and accurate adaptation of the asynchronous data storage system for smart marketing terminals to the terminal's hardware and software environment to solve the problems existing in current technologies. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes a method and system for collaborative adaptation of software and hardware in intelligent marketing terminals. It employs a dynamic data interaction parsing mechanism to deconstruct the root node of the data flow hierarchically, obtaining a hierarchical functional module graph, directional functional linkage links, and functional permission control domains. Combining an enhanced deep reinforcement learning path planning algorithm, an adaptive reinforcement learning network for data storage, and asynchronous scheduling dynamic evolution criteria, it constructs an intelligent neural decision-making topology network. Simultaneously, it transforms the functional permission control domain into a functional coupling strength embedding network, outputting an optimized intelligent neural decision-making topology network. Based on this optimized network, an adaptation architecture is output through a hierarchical interactive mapping and asynchronous data storage strategy intelligent generation mechanism. Finally, the adaptation architecture is applied and verified; if the adaptation requirements are not met, re-parsing is performed; otherwise, adaptation is complete. This improves the subsystem's adaptation accuracy and data storage stability, making it suitable for intelligent marketing terminal data processing scenarios.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] Methods for software and hardware co-adaptation of intelligent marketing terminals include:

[0006] S1: In response to the adaptation requirements of the intelligent marketing terminal data asynchronous data drop-to-the-pan system, the system performs targeted hierarchical deconstruction from the root node of the data flow of the subsystem through a data interaction dynamic parsing mechanism to obtain the hierarchical functional module map, targeted functional linkage links and functional permission control domains.

[0007] S2: Based on the hierarchical functional module map and directional functional linkage links, an intelligent neural decision-making topology network is constructed through an enhanced deep reinforcement learning path planning algorithm, a disk-based adaptive reinforcement learning network, and an asynchronous scheduling dynamic evolution criterion. At the same time, the functional permission control domain is transformed into functional coupling strength and synchronously embedded into the interaction associations between the upper and lower level modules and the corresponding functions at the same level in the intelligent neural decision-making topology network. The tightness of the interaction association is adjusted by the functional coupling strength value, and the optimized intelligent neural decision-making topology network is output.

[0008] S3: Based on the optimized intelligent neural decision topology network, it outputs an adaptive architecture for the asynchronous disk drop system by combining hierarchical interactive mapping with an asynchronous disk drop strategy intelligent generation mechanism.

[0009] S4: Apply the generated adaptation architecture to the intelligent marketing terminal data asynchronous disk system to verify whether it can meet the adaptation requirements in S1. If it does, the adaptation is completed; if it does not, the data interaction and dynamic parsing are performed again based on the verification results.

[0010] Specifically, step S1 includes:

[0011] In response to the adaptation requirements of the intelligent marketing terminal's asynchronous data transfer system, a dynamic data interaction and parsing mechanism has been initiated.

[0012] A data flow root node identification model is constructed. By monitoring the data flow input points of the intelligent marketing terminal in real time, the data flow root node is identified using the identification model, and the data features of the data flow root node are extracted. The data features include the data type, transmission frequency, and data volume characteristics of the data flow root node.

[0013] Based on the data characteristics, the data stream is deconstructed layer by layer by a hierarchical perceptron, and the functional module attributes contained in each layer are extracted to generate a hierarchical functional module map. The hierarchical functional module map is used to characterize the hierarchical structure and module attributes of each functional module in the intelligent marketing terminal data asynchronous delivery system.

[0014] Based on the generated hierarchical functional module graph, the calling relationships between functional modules at different levels and within the same level are extracted using a graph traversal algorithm to form a directional functional linkage link; the directional functional linkage link is used to characterize the interaction relationship and data flow between each functional module.

[0015] Based on the preset module access control strategy in the asynchronous data drop-off system of intelligent marketing terminals, and combined with the module attributes of each functional module in the generated hierarchical functional module map, a functional permission control domain is generated. The functional permission control domain is represented by a permission mapping table, which is used to characterize the operation permission range and control rules of each functional module.

[0016] Specifically, the functional permission control domain includes a data encryption permission subdomain, a disk placement priority permission subdomain, and a data access permission subdomain; the data encryption permission subdomain is used to define the encryption level of different types of data; the disk placement priority permission subdomain is used to define the order rules for data placement on disk; and the data access permission subdomain is used to define the operation permissions of different user roles on the data.

[0017] Specifically, the construction of the intelligent neural decision-making topology network includes:

[0018] An enhanced deep reinforcement learning path planning algorithm is used to train a pre-built path decision model with a hierarchical functional module graph and directional functional linkage links as input, and outputs a preliminary network topology. The preliminary network topology is based on the hierarchical relationship and linkage links of functional modules to build an initial connection framework between functional modules.

[0019] The initial network topology is optimized using a disk-based adaptive reinforcement learning network. The connection weights between different functional modules are adjusted through a preset reward function. At the same time, combined with the asynchronous scheduling dynamic evolution criterion, the network structure is dynamically adjusted during network training based on the module interaction efficiency and data disk-based requirements, ultimately resulting in an intelligent neural decision-making topology network.

[0020] Specifically, the intelligent neural decision-making topology network includes an input layer, a hidden layer, and an output layer; the input layer receives feature data of hierarchical functional module maps and directional functional linkage links, the hidden layer realizes functional association calculation through the connection weights between neuron nodes, and the output layer outputs the disk placement decision result.

[0021] Specifically, the enhanced deep reinforcement learning path planning algorithm is as follows:

[0022] Using a hierarchical functional module graph and directional functional linkage links as input, the construction process of the intelligent neural decision-making topology network is modeled as a Markov decision process. The state space is defined as the state information of each functional module in the hierarchical functional module graph, including the load status and data transmission volume of the functional module. The action space is defined as adding or adjusting the connection relationships between functional modules in the intelligent neural decision-making topology network. The reward function is designed based on the performance indicators of the constructed intelligent neural decision-making topology network, including the network's transmission efficiency and stability. Through iterative optimization, the optimal path planning scheme is found to construct the basic framework of the intelligent neural decision-making topology network.

[0023] Specifically, the asynchronous scheduling dynamic evolution criteria include:

[0024] The first volume of non-real-time data is written to disk in batches using an asynchronous disk writing criterion.

[0025] The second volume real-time data adopts an incremental asynchronous disk write-to-disk criterion.

[0026] Emergency data is prioritized for disk storage; the emergency data is determined based on business continuity, data security, or decision-making timeliness.

[0027] Specifically, the process of transforming the functional permission control domain into functional coupling strength is as follows:

[0028] Based on the different user roles defined in the functional permission control domain, the permission dependency relationship between functional modules is analyzed; the permission dependency relationship means that the effectiveness or normal execution of the operation permission of one functional module depends on the operation result of another functional module.

[0029] Based on the permission dependency relationship obtained from the analysis, if the operation permission of one functional module depends on the operation result of another functional module, it is determined that there is permission coupling between the two corresponding functional modules; the permission coupling is used to characterize the interaction relationship formed between functional modules due to permission association.

[0030] For a given permission coupling relationship, each pair of functional modules with permission coupling is assigned a corresponding value, namely the functional coupling strength, based on the tightness and complexity of the permission dependency; the value range of the functional coupling strength is [0,1].

[0031] Specifically, step S3 includes:

[0032] Based on the optimized intelligent neural decision-making topology network, an asynchronous disk persistence strategy intelligent generation mechanism is used to generate a data disk persistence task scheduling sequence according to the interaction and correlation between functional modules in the optimized intelligent neural decision-making topology network and the functional coupling strength transformed from the functional permission control domain, and to assign disk persistence priorities to different tasks; the allocation of disk persistence priorities matches the degree of module interaction represented by the functional coupling strength; the asynchronous disk persistence strategy intelligent generation mechanism includes strategy generation and strategy optimization; the strategy generation generates a preliminary disk persistence strategy based on the output results of the intelligent neural decision-making topology network; the strategy optimization combines the terminal hardware storage capacity and network bandwidth to optimize the preliminary disk persistence strategy;

[0033] Based on the generated data write-to-disk task scheduling sequence, a hierarchical interactive mapping mechanism is used to map the hierarchical functional module map and directional functional linkage links in the optimized intelligent neural decision-making topology network to the hardware resource layer. The hardware resource layer then searches for the optimal computing unit that matches each functional module. This hierarchical interactive mapping mechanism includes a mapping from the data acquisition layer to the cache scheduling layer and a mapping from the cache scheduling layer to the disk write-to-disk layer. The mapping from the data acquisition layer to the cache scheduling layer is used to determine the data caching strategy; the mapping from the cache scheduling layer to the disk write-to-disk layer is used to determine the data writing method.

[0034] A permission configuration table is generated by combining the functional permission control domain, and access control policies are set for the matched computing units through the permission configuration table.

[0035] The data write-to-disk task scheduling sequence, write-to-disk priority, hardware computing unit matching results, and access control policies are integrated to form an adaptive architecture for the asynchronous data write-to-disk system. The adaptive architecture includes the correspondence between software functional modules and hardware resources, the data write-to-disk scheduling mechanism, and the permission control rules.

[0036] The intelligent marketing terminal hardware and software collaborative adaptation system includes: a data parsing module, a decision network construction module, an adaptation architecture generation module, and a verification feedback module;

[0037] The data parsing module responds to the adaptation requirements of the asynchronous data drop-off system of the intelligent marketing terminal and completes the directional hierarchical deconstruction of the root node of the data flow through the dynamic parsing mechanism of data interaction.

[0038] The decision network construction module is used to construct and optimize the intelligent neural decision topology network;

[0039] The adaptation architecture generation module is used to generate an adaptation architecture based on the optimized intelligent neural decision topology network.

[0040] The verification feedback module is used to verify whether the generated adaptation architecture meets the adaptation requirements.

[0041] Compared with the prior art, the beneficial effects of the present invention are:

[0042] This invention proposes a smart marketing terminal software and hardware collaborative adaptation system, and optimizes and improves its architecture, operation steps and processes. The system has the advantages of simple process, low investment and operation costs and low production and operation costs.

[0043] This invention proposes a method for the collaborative adaptation of intelligent marketing terminal software and hardware. This method accurately deconstructs the root node of the data flow through a dynamic parsing mechanism of data interaction, and the obtained hierarchical functional module map and directional functional linkage links provide a precise foundation for subsequent network construction. Furthermore, it combines algorithms such as enhanced deep reinforcement learning to construct an intelligent neural decision-making topology network and embeds the interaction relationship of functional coupling strength adjustment modules, which effectively improves the rationality of network construction and the accuracy of functional matching, and solves the problems of fuzzy functional association and low matching efficiency in traditional adaptation.

[0044] This invention proposes a method for the collaborative adaptation of software and hardware in intelligent marketing terminals. The adaptation architecture generated based on an optimized intelligent neural decision-making topology network can accurately meet the requirements of the asynchronous data storage system of intelligent marketing terminals. Furthermore, through a closed-loop verification mechanism, if the adaptation architecture does not meet the requirements, it can be re-analyzed and adjusted to ensure the applicability and stability of the final architecture. This process not only ensures the efficiency of data storage but also flexibly responds to possible changes in system requirements, improving the overall quality of subsystem adaptation and the reliability of subsequent operation. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the intelligent marketing terminal software and hardware co-adaptation method of the present invention;

[0046] Figure 2 This is a flowchart illustrating the principle of the intelligent marketing terminal software and hardware co-adaptation method of the present invention.

[0047] Figure 3 This is a diagram of the hardware and software collaborative adaptation system architecture for the intelligent marketing terminal of this invention. Detailed Implementation

[0048] Example 1:

[0049] Please see Figure 1 and Figure 2 The present invention provides an embodiment of a method for software and hardware co-adaptation of intelligent marketing terminals, comprising the following steps:

[0050] S1: In response to the adaptation requirements of the intelligent marketing terminal data asynchronous data drop-to-the-pan system, the system performs targeted hierarchical deconstruction from the root node of the data flow of the subsystem through a data interaction dynamic parsing mechanism to obtain the hierarchical functional module map, targeted functional linkage links and functional permission control domains.

[0051] The functional permission control domain includes a data encryption permission subdomain, a disk placement priority permission subdomain, and a data access permission subdomain. The data encryption permission subdomain is used to define the encryption level of different types of data. The disk placement priority permission subdomain is used to define the order rules for data placement on disk. The data access permission subdomain is used to define the operation permissions of different user roles on the data.

[0052] Furthermore, the data interaction dynamic parsing mechanism includes a data protocol parsing unit, a data flow feature extraction unit, and a node association identification unit; the data protocol parsing unit is used to parse the transmission protocol format of the data flow root node; the data flow feature extraction unit is used to extract the data type, transmission frequency, and data volume features in the data flow root node; the node association identification unit is used to identify the interaction relationship between the data flow root node and downstream nodes, and the three work together to complete the targeted hierarchical deconstruction.

[0053] Furthermore, the data flow root node includes the sales data collection source, the member data entry source, and the marketing activity data generation source.

[0054] Furthermore, the directional hierarchical deconstruction refers to starting from the aforementioned source node and analyzing the relay nodes, processing nodes, and storage nodes in the data transmission path layer by layer to form a multi-level deconstruction result.

[0055] Furthermore, the hierarchical functional module diagram includes a data acquisition layer module, a data caching layer module, a data processing layer module, and a data storage layer module. The modules are arranged sequentially according to the data transmission order, and each module contains at least one specific functional unit.

[0056] Furthermore, the directional functional linkage link refers to the directed association link formed by the modules in the hierarchical functional module map according to the data processing flow, specifically including a complete link of data acquisition, data caching, data preprocessing, asynchronous scheduling, disk writing, and verification feedback.

[0057] S2: Based on the hierarchical functional module map and directional functional linkage links, an intelligent neural decision-making topology network is constructed through an enhanced deep reinforcement learning path planning algorithm, a disk-based adaptive reinforcement learning network, and an asynchronous scheduling dynamic evolution criterion. At the same time, the functional permission control domain is transformed into functional coupling strength and synchronously embedded into the interaction associations between the upper and lower level modules and the corresponding functions at the same level in the intelligent neural decision-making topology network. The tightness of the interaction association is adjusted by the functional coupling strength value, and the optimized intelligent neural decision-making topology network is output.

[0058] The intelligent neural decision-making topology network includes an input layer, a hidden layer, and an output layer. The input layer receives feature data from the hierarchical functional module map and the directional functional linkage links. The hidden layer realizes functional association calculation through the connection weights between neuron nodes. The output layer outputs the disk placement decision result.

[0059] The asynchronous scheduling dynamic evolution criteria include: the first volume of non-real-time data adopts the batch asynchronous disk writing criterion; the second volume of real-time data adopts the incremental asynchronous disk writing criterion; and the urgent data adopts the priority scheduling disk writing criterion.

[0060] Among them, the first volume of non-real-time data is large volume non-real-time data, which refers to a single data volume of at least A, such as a single data volume of several GB or more, and does not need to be immediately written to disk after being generated. It is a type of data that can be processed centrally during idle periods of system resources or according to a preset cycle. It is commonly found in scenarios such as historical transaction statistics data of smart marketing terminals and monthly marketing activity review data.

[0061] Among them, the second volume real-time data is small volume real-time data, which refers to a single data volume of B, such as a single data volume of only a few KB to a few MB, and needs to be written to disk within milliseconds after the data is generated, so as to ensure that subsequent business links can be processed in a timely manner based on the data. It is commonly found in scenarios such as real-time transaction order snapshot data and user instant interaction behavior data in smart marketing terminals.

[0062] Urgent data refers to data that has extremely high requirements for business continuity, data security, or decision-making timeliness. It needs to break through the conventional disk storage scheduling order and prioritize the use of system resources to complete disk storage in order to avoid business losses or data risks due to delays. It is commonly found in scenarios such as high-value customer order data, payment transaction confirmation data, and system fault warning log data in intelligent marketing terminals.

[0063] Furthermore, the enhanced deep reinforcement learning path planning algorithm is used to optimize the interaction paths of functional modules in the intelligent neural decision-making topology network; the disk-based adaptive reinforcement learning network is used to dynamically adjust network parameters according to the data disk-based requirements; and the asynchronous scheduling dynamic evolution criterion is used to guide the dynamic evolution of the network structure.

[0064] Furthermore, the intelligent neural decision-making topology network is used to simulate the collaborative decision-making relationship between various functional modules; the functional coupling strength is a value obtained based on the functional permission control domain quantization, used to characterize the tightness of the interaction and association between functional modules.

[0065] Furthermore, the disk-based adaptive reinforcement learning network takes data size, data real-time requirements, and data type as input features, and dynamically adjusts the disk-based strategy through continuous learning. When the input features change, the disk-based parameters output by the network are updated adaptively.

[0066] Furthermore, the adjustment of the tightness of the interaction association by the functional coupling strength value specifically means that the higher the functional coupling strength value, the greater the connection weight of the corresponding interaction association in the topology network, the higher the data transmission priority on the path, and the smaller the transmission delay.

[0067] S3: Based on the optimized intelligent neural decision topology network, it outputs an adaptive architecture for the asynchronous disk drop system by combining hierarchical interactive mapping with an asynchronous disk drop strategy intelligent generation mechanism.

[0068] Furthermore, hierarchical interaction mapping refers to mapping the optimized intelligent neural decision-making topology network to the hardware and software environment of the intelligent marketing terminal in a hierarchical manner; the asynchronous disk placement strategy intelligent generation mechanism is used to generate an adapted asynchronous data placement strategy based on the interaction associations in the network.

[0069] Furthermore, the adaptation architecture includes a hardware adaptation layer, a software adaptation layer, and an interface adaptation layer. The hardware adaptation layer is used to adapt to the terminal's storage device and processor, the software adaptation layer is used to adapt to the operating system and applications, and the interface adaptation layer is used to adapt to the data transmission interface.

[0070] S4: Apply the generated adaptation architecture to the intelligent marketing terminal data asynchronous disk system to verify whether it can meet the adaptation requirements in S1. If it does, the adaptation is completed; if it does not, the data interaction and dynamic parsing are performed again based on the verification results.

[0071] Furthermore, the verification includes functional verification and performance verification. Functional verification is used to check whether the adapted architecture has implemented all functions for asynchronous data persistence to disk, and performance verification is used to test whether the speed, success rate and resource utilization of data persistence to disk meet the requirements.

[0072] Furthermore, based on the verification results, the data interaction dynamic parsing is re-performed, including: when the verification fails, extracting the problem features found during the verification process, using them as new parsing parameters to input into the data interaction dynamic parsing mechanism, and re-executing steps S1 to S3 until the adaptation architecture meets the adaptation requirements in S1.

[0073] The steps in S1 include:

[0074] S1.1: In response to the adaptation requirements of the intelligent marketing terminal's asynchronous data transfer system, a dynamic data interaction and parsing mechanism is initiated;

[0075] S1.2: Construct an identification model for the root node of the data flow. By monitoring the data flow input points of the intelligent marketing terminal in real time, the identification model is used to identify the root node of the data flow and extract the data features of the root node. The data features include the data type, transmission frequency and data volume characteristics of the root node of the data flow.

[0076] Furthermore, the specific steps in S1.2 include:

[0077] (1) Clarify the type definition of data flow root nodes. Combined with the business scenarios of intelligent marketing terminals, determine the business attributes and data output characteristics of the three types of data flow root nodes: sales data collection source, member data entry source, and marketing activity data generation source. For example, sales data collection source is mostly associated with transaction behavior, while member data entry source is related to user information registration.

[0078] (2) Collect sample data of data flow root nodes. For the three types of data flow root nodes, select data flow input instances that are clearly marked as root nodes as sample data from the historical operation records of the intelligent marketing terminal data asynchronous drop-down system. At the same time, collect a quantitative number of data flow input instances that are not root nodes as control samples.

[0079] (3) Clean the collected sample data to remove outliers, such as abnormal fluctuations in transmission frequency or data volume that significantly exceeds the reasonable range; then standardize the data to adjust the data types, transmission frequencies and data volume characteristics of different magnitudes to a unified analysis dimension to avoid the model judgment being affected by differences in data magnitude.

[0080] (4) Extract the feature vector of the sample data. For each sample data, distinguish whether it is transaction data, user data or activity data from the data type dimension; count the number of data inputs per unit time from the transmission frequency dimension; calculate the amount of data input each time from the data volume dimension, and combine the information of these three dimensions into the feature vector of the sample data as the input parameters for model training.

[0081] (5) Load the recurrent neural network architecture, set the number of nodes and connection method of the input layer, hidden layer and output layer, wherein the input layer corresponds to the dimension of the feature vector, and the output layer corresponds to the category judgment result of the root node. The recurrent neural network is the prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0082] (6) The recurrent neural network is trained using the training set after the input parameters are divided to obtain a preliminary recognition model;

[0083] (7) Analyze the incorrect identification cases in the validation set after the input parameters are divided in the preliminary identification model, and determine whether the data type features are not distinguished enough, the transmission frequency feature weight is not set reasonably, or the data volume features are not effectively captured. Adjust the feature extraction layer or hidden layer structure of the preliminary identification model in a targeted manner, retrain the preliminary identification model and verify it until the model's recognition accuracy meets the actual application requirements and a well-trained identification model is obtained.

[0084] (8) Integrate the recognition model into the data flow monitoring module of the intelligent marketing terminal so that it can receive dynamic data from the data flow input point in real time;

[0085] (9) The recognition model extracts features from the input real-time data stream, compares the learned data stream root node feature patterns, determines whether the input point is a data stream root node, and marks its type.

[0086] (10) For the identified root node of the data stream, continuously monitor and record the specific category of its data type, count the transmission frequency per unit time, calculate the volume of each data transmission, and form a complete set of data features of the root node to provide basic data for subsequent data stream hierarchical deconstruction.

[0087] Furthermore, when calculating the transmission frequency, it is determined by counting the number of data inputs within a fixed time window; in calculating the loss function for training the recognition model, the model parameters are adjusted based on the difference between the predicted results and the actual labels.

[0088] S1.3: Based on the data characteristics, the data stream is deconstructed layer by layer through a hierarchical perceptron, and the functional module attributes contained in each layer are extracted to generate a hierarchical functional module map; the hierarchical functional module map is used to characterize the hierarchical structure and module attributes of each functional module in the intelligent marketing terminal data asynchronous delivery system;

[0089] Furthermore, the specific steps in S1.3 include:

[0090] (1) Supplement the data features of the extracted data stream root node with business dimensions and format adaptation, including: starting from the business logic of the intelligent marketing terminal data asynchronous data drop system, matching the corresponding business scenario labels for different data types, such as labeling transaction data with real-time settlement association, labeling membership data with user profile construction association, and labeling marketing activity data with activity effect statistics association, to ensure that data features are deeply bound to the system business requirements; at the same time, convert the data features into an input format that can be recognized by the hierarchical perceptron, such as dividing them into three levels of labels according to the transmission frequency: high-frequency real-time transmission, medium-frequency timed transmission, and low-frequency batch transmission, and dividing them into three types of processing requirement labels according to the data volume: small volume instant processing, medium volume cache processing, and large volume fragment processing, so that the data features retain the original numerical information and have clear business semantics;

[0091] (2) Combining the hierarchical transmission characteristics of the data flow with the business hierarchy of the system functional modules, design the network architecture of the hierarchical perceptron; the network architecture of the hierarchical perceptron is divided into a first input layer, a first hidden layer and a first output layer. The first hidden layer sets up a multi-level structure according to the transmission path of the data flow in the system. The first hidden layer corresponds to the deconstruction requirements of the access data, the second hidden layer corresponds to the data preprocessing, the third hidden layer corresponds to the data caching layer, the fourth hidden layer corresponds to the data processing layer, and the fifth hidden layer corresponds to the data storage layer, so as to ensure that the hierarchy of the hierarchical perceptron corresponds one-to-one with the hierarchy of the actual functional modules of the system; in the parameter initialization stage, the initial weights are set for the business objectives of different hidden layers, and the activation function of each hidden layer is set at the same time. The function type suitable for hierarchical classification is selected so that the hierarchical perceptron can accurately capture the feature mapping relationship of different levels;

[0092] (3) Collect the data flow transmission logs and corresponding function module call records of the asynchronous data drop system of the intelligent marketing terminal in the historical operation, and construct a training dataset. Each sample in the dataset contains a combination of data features and corresponding hierarchical function module labels. For example, high-frequency real-time transmission + small volume instant processing + transaction data corresponds to the first hidden layer - real-time transaction access label. At the same time, the training dataset is divided into training set and validation set according to the proportion. The training set is input into the hierarchical perceptron, and the weight parameters of each hidden layer are iteratively adjusted through the backpropagation algorithm. After each iteration, the difference between the hierarchical module prediction results output by the hierarchical perceptron and the actual labels is compared, the error value is calculated, and the weights of each layer are corrected according to the error value. The hierarchical recognition accuracy of the hierarchical perceptron is monitored in real time on the validation set. When the accuracy is stable at the preset threshold for several consecutive rounds, such as above 95%, and the recognition deviation of different hierarchical modules is controlled within the preset allowable range, the training is stopped, the calibration of the hierarchical perceptron is completed, and the trained hierarchical perceptron is obtained. The backpropagation algorithm is the existing technology in this field and is not an inventive solution of this application. It will not be described in detail here.

[0093] (4) Input the preprocessed real-time data stream features into the trained hierarchical perceptron and start the layer-by-layer deconstruction process, including: starting from the first hidden layer, the perceptron determines the specific hierarchical module type that the data stream should access based on the transmission frequency and data type characteristics of the data stream; then the data stream features enter the second hidden layer, and the perceptron determines the module to be called based on the data volume characteristics and preprocessing requirements, while analyzing the relationship between the module and the first hidden layer; then, the deconstruction of the third, fourth, and fifth hidden layers is completed in sequence according to the same logic. Each hidden layer outputs the corresponding functional module type, module calling conditions, and interaction path with the previous layer module by the perceptron, forming a complete data stream hierarchical deconstruction record to ensure that each data transmission link can accurately match the corresponding functional module;

[0094] (5) For each functional module identified after layer-by-layer deconstruction, detailed attribute information is extracted from three dimensions: business attributes, technical attributes, and interaction attributes. In terms of business attributes, the business scenario, core business objectives, and business priorities corresponding to the module are recorded. In terms of technical attributes, the processing capabilities, dependent hardware resources, and technical implementation methods of the module are extracted. In terms of interaction attributes, the input data format, output data format, interaction protocol with upstream and downstream modules, and data transmission latency requirements of the module are clarified. All extracted attribute information is standardized. The input data format includes JSON or CSV format, and the interaction protocol of upstream and downstream modules includes HTTP protocol and MQ message queue protocol.

[0095] (6) Based on the module hierarchy determined by the layer-by-layer deconstruction and the extracted module attributes, construct a hierarchical functional module map, which specifically includes: in the vertical dimension of the map, arrange the modules of each level in the order of the hierarchical structure of the first hidden layer, use vertical lines to represent the upper and lower level associations between modules, and mark the associated transmission paths and protocols; in the horizontal dimension, arrange the modules in the same level from high to low according to business priority, and use horizontal dashed lines to represent the cooperation relationship between modules in the same level; then, mark the standardized business attributes, technical attributes and interaction attributes on each module node to ensure the integrity of the map information; finally, compare and verify the constructed map with the actual functional module configuration table of the system to check whether there are any missing modules, hierarchical errors or attribute labeling errors; for the problems found, trace back the deconstruction process and attribute extraction link of the hierarchical perceptron, correct the errors and improve the map again until the map can accurately and comprehensively represent the hierarchical structure and module attributes of each functional module in the system, and obtain the hierarchical functional module map.

[0096] S1.4: Based on the generated hierarchical functional module graph, the calling relationships between functional modules at different levels and within the same level are extracted using a graph traversal algorithm to form a directional functional linkage link; the directional functional linkage link is used to characterize the interaction relationship and data flow between each functional module;

[0097] Furthermore, the specific steps in S1.4 include:

[0098] (1) Perform structured preprocessing on the generated hierarchical functional module graph to clarify the definition and association rules of each element in the graph. Each functional module is defined as an independent node. The node attributes include the module name, the level to which it belongs, the business priority, and the interaction attributes. The interaction attributes include the input and output data format and the interaction protocol. The labeled hierarchical associations and same-level collaboration relationships between modules are defined as initial edges. The edge attributes record the association type, transmission path, and protocol. At the same time, delete invalid and redundant edges in the graph, such as duplicate associations caused by labeling errors.

[0099] (2) Based on the structural characteristics and calling relationships of the hierarchical functional module graph, extract the requirements, select a hybrid traversal algorithm that combines depth-first search and breadth-first search, and configure the core parameters of the hybrid traversal algorithm, including: 1. Set the traversal starting point, taking all modules of the data access layer as the initial starting point to ensure that the traversal covers all possible calling links; 2. Define the traversal termination condition, when traversing all modules of the data storage layer, or traversing a node with no subsequent related modules, stop the traversal of that branch; 3. Set the association weight threshold, assign association weight to each edge according to the frequency of interaction between modules and the importance of business.

[0100] (3) Start the depth-first search algorithm, starting from each module node of the first hidden layer, and traverse the cross-level call relationship layer by layer downwards. During the traversal, if any node has multiple lower-level associations, the depth-first search algorithm will delve into each branch one by one, extract all possible cross-level call relationships, and mark the main link or backup link for each call chain to ensure that no key cross-level calls are missed. Specifically, the depth-first search algorithm is the existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0101] (4) After completing the cross-level traversal, the breadth-first search algorithm is started to extract the collaborative call relationship between modules within each level. During the traversal, the breadth-first search algorithm records all related nodes of each module at the same level to ensure that all horizontal collaborative calls within the level are covered, avoiding omissions due to differences in module position or collaboration frequency, and finally forming a complete set of collaborative call relationships within each level. Specifically, the breadth-first search algorithm is the existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0102] (5) Integrate the extraction results of cross-level call relationships and collaborative call relationships, and sort them into targeted functional linkage links according to the actual data flow processing flow. This includes: First, taking the cross-level call chain as the main framework, each cross-level call chain is taken as a main link. Then, the same-level collaborative relationship is taken as a branch link and attached to the corresponding node of the main link. Then, the directional information is marked for each link to clarify the call direction, data flow direction and call triggering conditions between modules. At the same time, the links are classified according to business scenarios to obtain the initially constructed targeted functional linkage links.

[0103] (6) Perform multi-dimensional verification on the initially constructed directional functional linkage links to ensure the accuracy and integrity of the links. The multi-dimensional verification includes business logic verification, module coverage verification, and performance rationality verification.

[0104] Furthermore, business logic verification compares each link with the actual business process of the asynchronous data drop-off system of the intelligent marketing terminal to check whether it conforms to the data processing logic; module coverage verification checks whether all functional modules at all levels are included in at least one link. If there are uncovered modules, the process needs to be backtracked to supplement their call relationships with other modules and improve the link; performance rationality verification analyzes whether the call frequency and processing capacity of the modules in the link match.

[0105] (7) For problems found in multi-dimensional verification, such as missing links, incorrect directions, and performance mismatch, adjust the parameters of the hybrid traversal algorithm or supplement the module association information, re-optimize the link structure, and finally form an accurate, complete, and directional functional linkage link that meets business and performance requirements.

[0106] S1.5: Based on the preset module access control strategy in the intelligent marketing terminal data asynchronous data drop system, and combined with the module attributes of each functional module in the generated hierarchical functional module map, a functional permission control domain is generated; the functional permission control domain is represented by a permission mapping table, which is used to characterize the operation permission range and control rules of each functional module.

[0107] Furthermore, the preset module access control policies include: clearly defining the responsibilities of all preset roles in the system, such as system administrators, operations and maintenance personnel, marketing operators, and ordinary users. For example, system administrators are responsible for configuring all modules, while marketing operators can only operate the data collection and activity data modules; defining the types of operations that each role can perform, such as query, add, modify, delete, execute, and configure. For example, ordinary users only have query permissions, while operations and maintenance personnel have execute and configuration permissions; and determining the scope of the functional modules explicitly associated in the policy to avoid omissions.

[0108] Furthermore, the specific steps in S1.5 include:

[0109] (1) The preset module access control strategy in the asynchronous data drop-off system of intelligent marketing terminals is fully decomposed, and the abstract strategy terms are transformed into specific rules that can be implemented.

[0110] (2) Standardize the decomposed strategies, unify the description of role names, standardize the definition of operation types, and establish a preliminary correspondence table of roles-operations-modules;

[0111] (3) From the generated hierarchical functional module map, extract the core attributes related to permission control for each functional module and classify them according to the degree of permission correlation, including: first, extract basic attributes, such as the module's level, core module functions, and data types processed by the module. Among them, the core module functions include data encryption, disk scheduling, and data query. The data types processed by the module include transaction sensitive data, basic member data, and marketing statistics. Then, extract interaction attributes, such as the module's input and output data formats and the linkage configuration of permissions affected by interaction attributes. Finally, extract technical attributes, including the module's deployment location and security level. Among them, the deployment location includes local servers and cloud nodes, and the security level includes high security level, medium security level, and low security level. Technical attributes determine the access method restrictions of permissions. Classify and organize the extracted attributes according to basic attributes, interaction attributes, and technical attributes, and establish an attribute file for each module.

[0112] (4) Combining the decomposed module access control strategy and the extracted module attributes, formulate role-module-permission mapping rules, including: matching basic permissions based on role responsibilities and module functions; adjusting the permission strictness based on the module's data processing data type and security level; supplementing the linkage permission rules based on the module's interaction attributes; and finally, verifying the formulated mapping rules.

[0113] (5) Based on different dimensions of access control, the functional access control domain is divided into a data encryption access subdomain, a disk storage priority access subdomain, and a data access permission subdomain. The content of each subdomain is filled in with mapping rules and module attributes, including: Constructing the data encryption access subdomain: For each module that needs to perform data encryption, the encryption level is defined based on the sensitivity of the data type processed by the module. For example, transaction data is set to Level 1 encryption and member sensitive information is set to Level 2 encryption. At the same time, the encryption operation permissions of different roles are configured. For example, the system administrator has the key generation and management permissions for Level 1 and Level 2 encryption, while the operation and maintenance personnel only have the encryption execution permissions for Level 1 and Level 2 encryption and no key management permissions. The encryption algorithm corresponding to the encryption level is specified. For example, Level 1 encryption uses the RSA algorithm and Level 2 encryption uses the AES algorithm. Ensure that encryption permissions match data security requirements; construct a disk placement priority permission subdomain: based on the business priority attributes of modules, define disk placement priority rules, configure priority adjustment permissions for different roles, such as system administrators can adjust the disk placement priority of all modules, marketing operators can only adjust the priority of marketing-related modules, and can only adjust between medium and low priorities, clarify the disk placement resource allocation corresponding to the priority, and ensure that disk placement scheduling meets the business urgency; construct a data access permission subdomain: combined with the security level and deployment location attributes of modules, define the access permission scope of different roles, clarify access log recording requirements, and ensure that data access is traceable and controllable. Among them, RSA and AES algorithms are existing technologies in this field and are not inventive solutions of this application, and will not be described in detail here.

[0114] Furthermore, the access log recording requirements include that all roles accessing high-security modules must record the access time, access IP address, and operation content.

[0115] Furthermore, the access permissions for different roles include: for the high-security local transaction data module, only system administrators and operations personnel are allowed to access it via the intranet; for the medium-security cloud marketing data module, marketing operators are allowed to access it via VPN, and access operation permissions are configured, such as ordinary users can only access query operations for all modules and can only view non-sensitive fields.

[0116] (6) Integrate the contents of the three subdomains to form a complete functional permission control domain, and present it in the form of a permission mapping table.

[0117] The construction of the intelligent neural decision-making topology network includes:

[0118] A1: An enhanced deep reinforcement learning path planning algorithm is used to train a pre-built path decision model with a hierarchical functional module graph and directional functional linkage links as input, and outputs a preliminary network topology structure; the preliminary network topology structure is based on the hierarchical relationship and linkage links of functional modules to build an initial connection framework between functional modules.

[0119] Furthermore, the specific steps of A1 include:

[0120] (1) Transform the hierarchical functional module map and the directional functional linkage link into structured input data that can be recognized by the enhanced deep reinforcement learning algorithm. Specifically, for the hierarchical functional module map, extract the core features of each functional module and quantify them. For the directional functional linkage link, extract the link information and encode it.

[0121] Furthermore, the link information includes: the start and end modules of the link, and the link transmission success rate; encoding is to convert each link into a link feature group containing the start index, end index, transmission frequency value, and success rate value.

[0122] (2) Based on the path planning requirements of the asynchronous data drop-off system of intelligent marketing terminals, a network architecture for the path decision model is constructed; the path decision model adopts a deep neural network structure, wherein the deep neural network is the existing technology in this field and is not an inventive solution of this application, and will not be described in detail here;

[0123] (3) For the constructed path decision model, configure the core parameters of the enhanced deep reinforcement learning path planning algorithm, and clarify the learning rules and decision logic of the algorithm, including: defining the agent and the environment, wherein the constructed path decision model is used as the agent, responsible for outputting the path selection in each training iteration; the actual operation scenario of the intelligent marketing terminal data asynchronous drop system is used as the environment, which contains all functional modules and linkage links, and can provide feedback rewards or penalties based on the path selection of the agent; secondly, define the state, action and reward, wherein the state is the feature set of the environment at any time, including the feature vector of the currently available modules, the feature group of the currently active links, and the system The current load conditions, such as module occupancy and link transmission pressure, are considered. The action is the operation that the agent can perform, i.e., selecting a path from all possible paths in the current state. The reward is the feedback from the environment to the agent's action. If the selected path meets the system constraints, such as transmission latency less than the threshold, success rate higher than the threshold, and can improve data persistence efficiency, such as shortening persistence time and reducing data loss rate, a positive reward is given; otherwise, a negative reward is given. The reward value is positively correlated with the path performance. Finally, the learning parameters of the algorithm are set. In this invention, the learning rate is set to 0.001, the discount factor is set to 0.9, and the exploration rate is set to 0.9, which are gradually reduced with training iterations.

[0124] (4) Input the preprocessed hierarchical functional module map and the directional functional linkage link data into the constructed path decision model, start the training iteration process of the enhanced deep reinforcement learning path planning algorithm, and optimize the model parameters in batches. Each iteration is executed according to the following process: environment initialization, generating the initial state at the current moment, such as the module and link characteristics when the system is idle, and the initial load situation, and input the initial state into the agent; secondly, the agent selects an action according to the current state and the exploration rate. If the exploration rate is high, it randomly selects an untried path to explore new possibilities. If the exploration rate is low, it selects the path with the highest historical reward. Utilizing known experience, the environment then simulates the system operation process based on the path chosen by the agent, calculating the actual performance indicators of the path, such as transmission delay, success rate, and disk write efficiency, and assigning corresponding reward values ​​to the agent according to the reward rules. Next, the agent updates the model parameters through temporal difference learning based on the reward values ​​fed back by the environment and the state at the next moment. Finally, it determines whether the iteration should terminate. If the number of iterations reaches a preset maximum value, or the average reward value of the model in multiple consecutive iterations tends to stabilize (i.e., fluctuation is less than 5%), training stops; otherwise, it proceeds to the next iteration until the termination condition is met, resulting in an optimized path decision model.

[0125] Furthermore, updating the model parameters through temporal difference learning specifically involves: calculating the difference between the predicted reward and the actual reward using the loss function, and adjusting the weights and biases of the hidden and output layers through backpropagation of the optimizer, so that the model can choose the path with higher reward in the next similar state. Here, temporal difference is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0126] (5) Input the complete hierarchical functional module map and the directional functional linkage link data into the optimized path decision model, and output the selection probability of all possible paths from the model. Generate a preliminary network topology based on the probability distribution.

[0127] Further, a preliminary network topology is generated based on the probability distribution, including: selecting core paths: choosing paths with a selection probability higher than 80% as the core links of the topology; determining module node positions: arranging all functional module nodes vertically according to the hierarchical structure of the first hidden layer, and horizontally arranging module nodes at the same level according to business priority; then, drawing link connections: connecting module nodes corresponding to the core paths with directed line segments, marking the transmission frequency and success rate of the links on the line segments, and marking backup paths with lower selection probabilities, such as 50%-80%, with dashed lines to form a complete network topology diagram; finally, validating the preliminary network topology by simulating the actual data transmission scenario of the asynchronous data drop-to-disk system for intelligent marketing terminals, inputting different types of data into the preliminary network topology, and testing the data transmission latency, success rate, and drop-to-disk efficiency. If the test results meet the system's preset indicators, such as latency less than 100 milliseconds, success rate higher than 99%, and drop-to-disk efficiency improved by at least 50%, the preliminary network topology is valid; if not, returning to the model training stage, adjusting algorithm parameters or optimizing the model structure, retraining, and regenerating and validating the topology until the preset indicator requirements are met.

[0128] A2: The initial network topology is optimized using a disk-based adaptive reinforcement learning network. The connection weights between different functional modules are adjusted through a preset reward function. At the same time, combined with the asynchronous scheduling dynamic evolution criterion, the network structure is dynamically adjusted during network training based on the module interaction efficiency and data disk-based requirements, ultimately resulting in an intelligent neural decision-making topology network.

[0129] Furthermore, the specific steps of A2 include:

[0130] (1) Construct the basic architecture of the disk-based adaptive reinforcement learning network and import the preliminary network topology generated earlier into the disk-based adaptive reinforcement learning network;

[0131] Furthermore, the disk-based adaptive reinforcement learning network adopts an architecture of input layer - adaptive feature extraction layer - weight adjustment layer - structure evolution layer - output layer. The input layer is responsible for receiving the core information of the initial network topology, including functional module node data and link data; the adaptive feature extraction layer automatically captures key features related to module interaction and data disk insertion through convolutional neural network units, without the need for manual preset feature dimensions; the weight adjustment layer introduces a gradient descent optimization unit to adjust the connection weights between modules based on the reward function feedback; the structure evolution layer is equipped with a dynamic decision unit to adjust the network structure according to the asynchronous scheduling dynamic evolution criterion; the output layer outputs the optimized module connection relationships and weight configurations.

[0132] Furthermore, when importing the initial network topology, the module nodes need to be mapped to the node library of the disk-based adaptive reinforcement learning network in the original hierarchical order. The core links and backup links in the initial network topology are marked as the initial active links and the initial links to be activated, respectively. At the same time, the initial connection weight of each link is recorded to ensure that the information of the initial topology is completely transmitted to the optimized network.

[0133] (2) To address the core requirements of the asynchronous data drop-off system for intelligent marketing terminals, a drop-off-oriented reward function is designed to quantify and evaluate the effectiveness of the connection between modules;

[0134] Furthermore, the disk-oriented reward function is constructed from three dimensions: First, the disk-writing efficiency dimension, which calculates the ratio of the actual amount of data successfully written to disk per unit time to the theoretical maximum amount of data to be written to disk. The higher the ratio, the higher the reward for this dimension. Second, the module interaction stability dimension, which statistically analyzes the transmission success rate and latency fluctuation range of the inter-module links within a preset time window. If the success rate is higher than a preset threshold, such as 99.5%, and the latency fluctuation is less than the threshold, such as ±5 milliseconds, a positive reward is given for this dimension; otherwise, a negative reward is given. Third, the resource utilization dimension, which calculates the CPU utilization, memory utilization, and disk I / O utilization of the functional modules. If each indicator is within a preset reasonable range, such as a CPU utilization of 30%-60%, a reward is given for this dimension; if it exceeds the range, the reward is deducted. The rewards from the three dimensions are added together according to their weights to obtain the final reward value. A positive reward value indicates that the current connection configuration is effective, while a negative value indicates that optimization is needed.

[0135] (3) Initiate the training process of the disk-based adaptive reinforcement learning network, and adjust the connection weights between functional modules through the reward function feedback, specifically including:

[0136] In each iteration of network training, the actual data transmission scenario of the asynchronous data drop-off system for intelligent marketing terminals is simulated: different types of data streams are input into the imported initial network topology, and the actual operating indicators of each link are recorded, such as drop-off efficiency, transmission success rate, latency, and module resource usage; the actual operating indicators are input into the calibrated reward function to calculate the reward value corresponding to each link; secondly, the link connection weights are adjusted according to the reward value: for links with positive reward values, their connection weights are increased according to the reward value, with the increase being positively correlated with the reward value, thus enhancing the priority of the link in subsequent data transmission; for links with negative reward values, their connection weights are decreased according to the absolute value of the reward value, with the decrease being positively correlated with the absolute value. If the weight drops to a preset minimum value, such as below 0.1, it is marked as a link to be eliminated; after each round of weight adjustment, the data transmission scenario is re-simulated to calculate a new reward value, and it is determined whether the weights converge, i.e., in multiple consecutive iterations, the weight change is less than a preset threshold, such as 0.5%. If convergence is not achieved, iterative adjustment continues until convergence is obtained, resulting in the optimized connection weight configuration.

[0137] (4) While adjusting the connection weights, the network structure is dynamically adjusted based on the asynchronous scheduling dynamic evolution principle, according to the module interaction efficiency and data disk persistence requirements. Specifically, this includes:

[0138] The core judgment indicators for defining the dynamic evolution criteria of asynchronous scheduling are defined, specifically including: module interaction efficiency, fluctuation in data persistence demand, and link redundancy. Secondly, structural adjustments are performed according to the criteria and different scenarios: Scenario 1: Low module interaction efficiency and high data persistence demand indicate that the current links cannot meet the demand, requiring the addition of new links; Scenario 2: High module interaction efficiency but high link redundancy indicates that redundant links are consuming resources, requiring link deletion: the lowest-weighted link to be phased out is removed from the network structure, and the input / output link configuration of related modules is updated to avoid data transmission interruptions; Scenario 3: Large fluctuations in data persistence demand require dynamic adjustment of module node configuration: at module levels with sudden increases in data volume, temporary backup module nodes are added to share the data processing pressure. After the demand recovers, the backup module nodes are marked as dormant to avoid resource waste. After each structural adjustment, the overall network performance indicators are recalculated. If the indicators improve, the adjustment results are retained; if the indicators decline, the adjustment is rolled back to ensure that structural adjustments always optimize network performance.

[0139] Furthermore, module interaction efficiency is equal to the ratio of the actual transmission rate of the link to the theoretical maximum transmission rate; data disk write demand fluctuation is equal to the percentage difference between the current disk write volume and the historical average disk write volume for the same period; link redundancy is equal to the ratio of the number of parallel links between the same module to the number of necessary links.

[0140] (5) After completing the connection weight adjustment and network structure adjustment, an intelligent neural decision-making topology network is generated, and its performance is verified through multi-dimensional testing.

[0141] The enhanced deep reinforcement learning path planning algorithm is as follows:

[0142] Using a hierarchical functional module graph and directional functional linkage links as input, the construction process of the intelligent neural decision-making topology network is modeled as a Markov decision process. The state space is defined as the state information of each functional module in the hierarchical functional module graph, including the load status and data transmission volume of the functional module. The action space is defined as adding or adjusting the connection relationship between functional modules in the intelligent neural decision-making topology network. The reward function is designed based on the performance indicators of the constructed intelligent neural decision-making topology network, including the network's transmission efficiency and stability. Through iterative optimization, the optimal path planning scheme is found to construct the basic framework of the intelligent neural decision-making topology network. Markov decision-making is prior art in this field and is not an inventive solution of this application, so it will not be elaborated here.

[0143] The process of transforming the functional permission control domain into functional coupling strength is as follows:

[0144] B1: Based on the different user roles defined in the functional permission control domain, analyze the permission dependency relationship between functional modules; the permission dependency relationship means that the effectiveness or normal execution of the operation permission of one functional module depends on the operation result of another functional module.

[0145] B2: Based on the permission dependency relationship obtained from the analysis, if the operation permission of one functional module depends on the operation result of another functional module, it is determined that there is permission coupling between the two corresponding functional modules; the permission coupling is used to characterize the interaction relationship formed between functional modules due to permission association;

[0146] B3: For a given permission coupling relationship, assign a corresponding value to each pair of functional modules with permission coupling based on the tightness and complexity of permission dependencies, i.e., functional coupling strength; the value range of the functional coupling strength is [0, 1].

[0147] The steps in S3 include:

[0148] S3.1: Based on the optimized intelligent neural decision-making topology network, through the asynchronous disk persistence strategy intelligent generation mechanism, according to the interaction and relationship between each functional module in the optimized intelligent neural decision-making topology network and the functional coupling strength transformed from the functional permission control domain, a data disk persistence task scheduling sequence is generated, and disk persistence priority is assigned to different tasks.

[0149] The allocation of disk placement priority needs to match the tightness of module interaction as represented by the functional coupling strength;

[0150] The asynchronous disk persistence strategy intelligent generation mechanism includes strategy generation and strategy optimization; the strategy generation generates a preliminary disk persistence strategy based on the output of the intelligent neural decision topology network; the strategy optimization combines the terminal hardware storage capacity and network bandwidth to optimize the preliminary disk persistence strategy, resulting in an optimized disk persistence strategy.

[0151] Furthermore, the specific steps in S3.1 include:

[0152] (1) Load the asynchronous disk-based intelligent generation mechanism architecture and import the data, including the optimized intelligent neural decision topology network data and the functional coupling strength data of the functional permission control domain transformation;

[0153] (2) Based on the intelligent generation mechanism of asynchronous disk writing strategy, obtain the optimized disk writing strategy;

[0154] (3) Based on the optimized disk write strategy, generate the final data write task scheduling sequence and assign a clear write priority to each task, specifically including:

[0155] Based on the execution time window, relationships, and module hierarchy of tasks in the optimized disk write strategy, all tasks to be written to disk are sorted by time sequence and execution dependency to form a linear scheduling sequence. The linear scheduling sequence includes information such as the unique identifier of each task, corresponding module, data type and data volume, execution time window, associated tasks, allocated storage device, and estimated bandwidth usage.

[0156] A three-dimensional scoring method is used to determine the priority level, including: the first dimension is the weight of the module interaction link, for example, core path tasks get 3 points and non-core path tasks get 1 point; the second dimension is the functional coupling strength, for example, highly coupled module group tasks get 3 points, medium coupled module group tasks get 2 points, and low coupled module group tasks get 1 point; the third dimension is the data sensitivity, for example, high sensitivity gets 3 points, medium sensitivity gets 2 points, and low sensitivity gets 1 point. The scores of the three dimensions are added together to obtain the total score, where 8-9 points is high priority, 5-7 points is medium priority, and 3-4 points is low priority.

[0157] Each task in the scheduling sequence is assigned a priority based on its total score, and the final output is a complete data disk dumping task scheduling sequence and disk dumping priority configuration.

[0158] S3.2: Based on the generated data disk task scheduling sequence, the hierarchical functional module map and directional functional linkage links in the optimized intelligent neural decision topology network are mapped to the hardware resource layer through a hierarchical interactive mapping mechanism. The computing unit that best matches each functional module is found in the hardware resource layer.

[0159] The layered interactive mapping mechanism includes a mapping from the data acquisition layer to the cache scheduling layer and a mapping from the cache scheduling layer to the disk write layer; the mapping from the data acquisition layer to the cache scheduling layer is used to determine the data caching strategy; the mapping from the cache scheduling layer to the disk write layer is used to determine the data writing method.

[0160] S3.3: Combine the functional permission control domain to generate a permission configuration table, and set access control policies for the matched computing units through the permission configuration table;

[0161] Furthermore, permission configuration refers to mapping the permission rules in the functional permission control domain to the permission control modules of software and hardware, including mapping the encryption permission level to the key management policy of the encryption module, and mapping the disk write priority permission to the priority queue of the scheduling module.

[0162] S3.4: Integrate the data write-to-disk task scheduling sequence, write-to-disk priority, hardware computing unit matching results, and access control policies to form an adaptive architecture for the asynchronous data write-to-disk system; the adaptive architecture includes the correspondence between software functional modules and hardware resources, the data write-to-disk scheduling mechanism, and the permission control rules.

[0163] Example 2:

[0164] Please see Figure 3 Another embodiment of the present invention provides: a smart marketing terminal software and hardware co-adaptation system, comprising:

[0165] Data parsing module, decision network construction module, adaptation architecture generation module, verification and feedback module;

[0166] The data parsing module responds to the adaptation requirements of the asynchronous data drop-off system of the intelligent marketing terminal, and completes the targeted hierarchical deconstruction of the root node of the data flow through the dynamic parsing mechanism of data interaction; the data parsing module includes a root node identification unit, a hierarchical deconstruction unit, a linkage link extraction unit, and an access control domain generation unit;

[0167] The root node identification unit is used to build an identification model, monitor the data flow input point in real time to identify the root node and extract data features; the hierarchical deconstruction unit is used to deconstruct the data flow layer by layer through a hierarchical perceptron based on the data features of the data flow root node, extract functional module attributes and generate a hierarchical functional module graph; the linkage link extraction unit is used to extract the call relationship between modules based on the hierarchical functional module graph through a graph traversal algorithm to form a directional functional linkage link; the permission control domain generation unit combines the module attributes in the hierarchical functional module graph with the preset access control policy to generate a functional permission control domain that represents the scope of module operation permissions and control rules.

[0168] The decision network construction module is used to build and optimize intelligent neural decision topology networks; the decision network construction module includes an initial network generation unit, a network optimization unit, and a coupling strength embedding unit.

[0169] The initial network generation unit utilizes an enhanced deep reinforcement learning path planning algorithm, taking a hierarchical functional module graph and directional functional linkage links as inputs, to train a path decision model and output a preliminary network topology. The network optimization unit adjusts the module connection weights through a disk-based adaptive reinforcement learning network, dynamically adjusts the network structure by combining asynchronous scheduling dynamic evolution criteria, and optimizes the preliminary network topology. The coupling strength embedding unit transforms the functional permission control domain into functional coupling strength, embeds the module interactions in the network, and adjusts the interaction tightness through numerical adjustment, ultimately outputting an optimized intelligent neural decision topology network.

[0170] The adaptation architecture generation module is used to generate an adaptation architecture based on the optimized intelligent neural decision topology network; the adaptation architecture generation module includes a disk placement strategy generation unit, a mapping configuration unit, and an architecture integration unit;

[0171] The disk persistence strategy generation unit is used to generate a data persistence task scheduling sequence and assign priorities based on the module interaction and functional coupling strength through an asynchronous disk persistence strategy intelligent generation mechanism. The mapping configuration unit is used to map the module graph and linkage links in the network to the hardware resource layer to match the computing units through layered interactive mapping, and generate a permission configuration table and set access control policies in combination with the functional permission control domain. The architecture integration unit integrates the task scheduling sequence, priority, hardware matching results and access control policies to form an adaptive architecture for the asynchronous data persistence system.

[0172] The verification feedback module is used to verify whether the generated adaptation architecture meets the adaptation requirements, forming a closed-loop logic; the verification feedback module includes an architecture application unit and a verification feedback unit.

[0173] The architecture application unit applies the adapted architecture to the intelligent marketing terminal's asynchronous data drop-off system; the verification feedback unit verifies whether the adapted architecture meets the adaptation requirements. If it does, the adaptation is completed; if not, the data parsing module is triggered to re-interact and dynamically parse the data based on the verification results until the adapted architecture meets the requirements.

[0174] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the present invention. All of these variations are within the protection scope of the present invention.

Claims

1. A method for software and hardware collaborative adaptation of intelligent marketing terminals, characterized in that, The application comprises the following steps: S1: In response to the adaptation requirement of the intelligent marketing terminal data asynchronous landing subsystem, the data stream root node of the subsystem is directionally deconstructed through a data interaction dynamic analysis mechanism to obtain a hierarchical function module graph, a directional function linkage link, and a function permission control domain; S2: Based on the hierarchical function module graph and the directional function linkage link, an intelligent neural decision topology network is constructed through an enhanced deep reinforcement learning path planning algorithm, a landing adaptive enhanced learning network, and an asynchronous scheduling dynamic evolution criterion. Meanwhile, the function permission control domain is converted into a function coupling strength and embedded in the corresponding interaction relationship between the upper and lower modules and the same level function in the intelligent neural decision topology network. The function coupling strength value is used to adjust the closeness of the interaction relationship, and the optimized intelligent neural decision topology network is outputted; S3: Based on the optimized intelligent neural decision topology network, an adaptive architecture of the data asynchronous landing subsystem is outputted through a hierarchical interaction mapping combined with an asynchronous landing strategy intelligent generation mechanism; S4: The generated adaptive architecture is applied to the intelligent marketing terminal data asynchronous landing subsystem to verify whether it can meet the adaptation requirement in S1. If it meets the requirement, the adaptation is completed. If it does not meet the requirement, the data interaction dynamic analysis is re-performed based on the verification result; The construction of the intelligent neural decision topology network comprises the following steps: An enhanced deep reinforcement learning path planning algorithm is used to train a pre-constructed path decision model with the hierarchical function module graph and the directional function linkage link as inputs, and an initial network topology structure is outputted. The initial network topology structure is based on the hierarchical relationship and linkage link of the function modules to construct an initial connection framework between the function modules. The hierarchical function module graph is used to represent the hierarchical structure and module attributes of each function module in the intelligent marketing terminal data asynchronous landing subsystem. The directional function linkage link is a directed association link formed by each module in the hierarchical function module graph according to the data processing flow, and is used to represent the interaction relationship and data flow direction between the function modules. A landing adaptive enhanced learning network is used to optimize the initial network topology structure. The connection weights between different function modules are adjusted through a preset reward function. Meanwhile, an asynchronous scheduling dynamic evolution criterion is combined to dynamically adjust the network structure according to the module interaction efficiency and data landing demand during the network training process, and finally an intelligent neural decision topology network is obtained. The landing adaptive enhanced learning network is used to dynamically adjust the network parameters according to the data landing demand. The asynchronous scheduling dynamic evolution criterion is used to guide the dynamic evolution of the network structure. The enhanced deep reinforcement learning path planning algorithm comprises the following steps: The construction process of the intelligent neural decision topology network is modeled as a Markov decision process with the hierarchical function module graph and the directional function linkage link as inputs; the state space is defined as the state information of each function module in the hierarchical function module graph, which includes the load condition and data transmission volume of the function module; the action space is defined as adding or adjusting the connection relationship between the function modules in the intelligent neural decision topology network; the reward function is designed according to the performance indicators of the constructed intelligent neural decision topology network, including the transmission efficiency and stability of the network, and through iterative optimization, the optimal path planning scheme is found to construct the basic framework of the intelligent neural decision topology network; The asynchronous scheduling dynamic evolution criterion comprises: The first volume of non-real-time data adopts a batch asynchronous disk writing criterion; The second volume of real-time data adopts an incremental asynchronous disk writing criterion; The emergency data adopts a priority scheduling disk writing criterion; the emergency data is determined according to business continuity, data security or decision timeliness; The process of converting the function permission control domain into the function coupling strength comprises: According to the operation permissions of different user roles on each function module defined in the function permission control domain, the permission dependency relationship between the function modules is analyzed; the permission dependency relationship means that the operation permission of one function module is valid or normally executed on the premise of the operation result of another function module; the function permission control domain is represented by a permission mapping table, which is used to represent the operation permission range and control rules of each function module; According to the analyzed permission dependency relationship, if the operation permission of one function module depends on the operation result of another function module, it is determined that there is a permission coupling between the corresponding two function modules; the permission coupling is used to represent the interaction relationship between the function modules due to the permission association; According to the closeness and complexity of the permission dependency, a corresponding numerical value, i.e., the function coupling strength, is given to each pair of function modules with permission coupling; the value range of the function coupling strength is [0, 1].

2. The intelligent marketing terminal software and hardware cooperative adaptation method of claim 1, wherein, The steps of the S1 comprise: In response to the adaptive needs of the intelligent marketing terminal data asynchronous disk writing subsystem, a data interaction dynamic analysis mechanism is started; An identification model of the data stream root node is constructed, the data stream root node is identified by using the identification model through real-time monitoring of the data stream input point of the intelligent marketing terminal, and the data characteristics of the data stream root node are extracted; the data characteristics include the data type, transmission frequency and data volume characteristics of the data stream root node; Based on the data characteristics, the data stream is deconstructed layer by layer by a hierarchical perception unit, the function module attributes contained in each layer are extracted, and a hierarchical function module graph is generated; the hierarchical function module graph is used to represent the hierarchical structure and module attributes of each function module in the intelligent marketing terminal data asynchronous disk writing subsystem; According to the generated hierarchical function module graph, the calling relationship between the function modules in different levels and the same level is extracted by a graph traversal algorithm to form a directional function linkage link; the directional function linkage link is used to represent the interaction relationship and data flow direction between the function modules. The function permission control domain is generated based on a preset module access control strategy in the intelligent marketing terminal data asynchronous landing subsystem and in combination with module attributes of each function module in the generated hierarchical function module graph; the function permission control domain is represented by a permission mapping table and is used to represent operation permission ranges and control rules of each function module. 3.The intelligent marketing terminal software and hardware cooperative adaptation method of claim 2, characterized in that, The function permission control domain includes a data encryption permission subdomain, a landing priority permission subdomain, and a data access permission subdomain; the data encryption permission subdomain is used to define encryption levels of different types of data; the landing priority permission subdomain is used to define rules for the order of data landing; and the data access permission subdomain is used to define operation permissions of different user roles on data.

4. The intelligent marketing terminal software and hardware cooperative adaptation method of claim 3, wherein, The intelligent neural decision topology network includes an input layer, a hidden layer, and an output layer; the input layer receives feature data of the hierarchical function module graph and the directional function linkage, the hidden layer performs function association calculation through connection weights between neuron nodes, and the output layer outputs a landing decision result.

5. The intelligent marketing terminal software and hardware collaborative adaptation method of claim 4, wherein, The step S3 includes: Based on the optimized intelligent neural decision topology network, a data landing task scheduling sequence is generated through an asynchronous landing strategy intelligent generation mechanism according to interaction associations between function modules in the optimized intelligent neural decision topology network and function coupling strengths converted from the function permission control domain, and landing priorities are allocated to different tasks; the allocation of the landing priorities matches the close degree of module interaction represented by the function coupling strengths; the asynchronous landing strategy intelligent generation mechanism includes strategy generation and strategy optimization; the strategy generation generates a preliminary landing strategy based on an output result of the intelligent neural decision topology network; and the strategy optimization optimizes the preliminary landing strategy in combination with terminal hardware storage capacity and network bandwidth; Based on the generated data landing task scheduling sequence, the hierarchical function module graph and the directional function linkage in the optimized intelligent neural decision topology network are mapped to a hardware resource layer through a hierarchical interaction mapping mechanism to find optimal matching calculation units for the function modules in the hardware resource layer; the hierarchical interaction mapping mechanism includes mapping of a data collection layer to a cache scheduling layer and mapping of the cache scheduling layer to a disk landing layer; the mapping of the data collection layer to the cache scheduling layer is used to determine a data caching strategy; and the mapping of the cache scheduling layer to the disk landing layer is used to determine a data writing mode; A permission configuration table is generated in combination with the function permission control domain, and an access control strategy is set for the matched calculation units through the permission configuration table; The data asynchronous landing subsystem adaptation architecture is formed by integrating the data landing task scheduling sequence, the landing priority, the hardware calculation unit matching result, and the access control strategy; the adaptation architecture includes a corresponding relationship between software function modules and hardware resources, a data landing scheduling mechanism, and a permission control rule.

6. The system for the software and hardware collaborative adaptation of the intelligent marketing terminal, which is used for realizing the method for the software and hardware collaborative adaptation of the intelligent marketing terminal according to any one of claims 1-5, characterized in that, It includes: a data analysis module, a decision network construction module, an adaptation architecture generation module, and a verification feedback module; The data analysis module dynamically analyzes data flow root nodes through a data interaction dynamic analysis mechanism in response to adaptation requirements of the intelligent marketing terminal data asynchronous landing subsystem. The decision network construction module is configured to construct and optimize an intelligent neural decision topology network. The adaptive architecture generation module is configured to generate an adaptive architecture according to the optimized intelligent neural decision topology network. The verification feedback module is configured to verify whether the generated adaptive architecture meets the adaptive requirement.

Images