Data service system inspired by brain and method thereof

Abstract

The invention relates to a brain inspired data service system and a brain inspired data service method, which are characterized in that a layered architecture is constructed, the layered architecture comprises a data node cluster deployed in a mixed manner, the data node cluster specifically comprises a core node, an acquisition node and a gateway node, and the core node, the acquisition node and the gateway node are respectively used for processing core service data, real-time edge data and cross-domain connection; the small-world connection layer realizes same-domain efficient communication through high-clustering local connection, and breaks a data island through dynamic long-range cross-domain connection; the intelligent control layer integrates a DMN control center, a collaborative state monitoring engine, a prediction analysis module and a reinforcement learning driven resource scheduler. The system initializes a topological structure through a small-world network manager, and when the communication frequency of a cross-domain node exceeds a threshold value, long-range connection is dynamically inserted to ensure that the hop count of a cross-domain path is stably not higher than a preset value. The DMN control center realizes dual-mode switching based on a system load threshold; in an idle period, data pre-cleaning, knowledge graph construction, predictive pre-fetching and resource pre-allocation are executed; and optimizing routing in combination with a graph traversal algorithm in a task period.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of distributed computing and artificial intelligence, and relates to a brain-inspired data service system and method thereof. BACKGROUND

[0002] With the rapid development of big data, cloud computing, Internet of Things and artificial intelligence technology, modern data service systems are facing the dual challenges of architectural complexity and service intelligence. In large enterprises or organizational environments, data is usually stored in multiple independent department systems, forming a serious data island phenomenon. Although the traditional centralized architecture is simple to manage, as the data scale expands, the center node is prone to become a performance bottleneck and a single point of failure, and it is difficult to meet the demand for cross-department data collaborative analysis. Although the existing distributed system relieves the storage pressure through node expansion, there are still obvious defects in data discovery efficiency, service path optimization, and resource dynamic scheduling.

[0003] The current mainstream data service system generally adopts a passive response mode, only performs corresponding operations when receiving user requests, and lacks the ability to actively integrate and preprocess data resources. This mechanism leads to three outstanding problems: first, cross-domain data access needs to go through multiple layers of transfer, and the communication delay is significantly increased; second, the allocation of system resources lacks foresight and cannot be dynamically adjusted according to business fluctuations; third, a large number of computing resources are in an idle state during off-peak hours, causing serious resource waste. More importantly, the existing system lacks the self-learning and adaptive ability of biological neural networks, making it difficult to establish an intelligent data correlation network and a predictive service mechanism. SUMMARY

[0004] The purpose of the present application is to provide a brain-inspired data service system and method thereof, which has the advantages of reducing cross-domain data access delay, implementing resource dynamic scheduling, actively integrating and preprocessing data resources, and improving system response efficiency and resource utilization.

[0005] In order to solve the above technical problems, the present application solves the problems by the following technical scheme: a brain-inspired data service system, comprising: a data node cluster composed of a plurality of distributed data nodes for storing, processing, analyzing data or providing data services, the data nodes being at least one of servers, containers, microservices, logical units or AIAgent, including core nodes deployed in a cloud server cluster, edge nodes deployed in edge computing devices and gateway nodes deployed in a containerized platform, corresponding to business domain core data processing, edge data collection and cross-domain data connection requirements respectively; Small-world connection layer: a small-world network model is used to construct the connection topology between data nodes, including a small-world network manager, local high clustering connection and long-range cross-domain connection, wherein the local high clustering connection realizes the dense interconnection of data nodes in the same service domain, and the long-range cross-domain connection dynamically maintains the cross-domain shortcut through the small-world network manager; Intelligent control layer: containing DMN control hub and subordinate state monitoring engine, prediction analysis module and resource scheduler, as the system control hub simulating the default mode network function of the brain; Infrastructure layer: deployed in a hybrid architecture of cloud server cluster, containerized platform, edge computing device and physical server; The intelligent control layer collects the communication hop count and delay index of the small-world connection layer in real time through the state monitoring engine, and drives the prediction analysis module to generate topology optimization instructions;The DMN control hub automatically triggers mode switching based on the real-time data of the state monitoring engine through a threshold comparison algorithm, and pushes the hot area weight to the small-world network manager.

[0006] By adopting the above technical scheme, by constructing a data node cluster containing a hybrid deployment: cloud core, edge node, containerized gateway, and introducing a small-world connection layer and DMN intelligent control layer inspired by the brain, the resource island and efficiency bottleneck problem of the traditional system is solved, and the technical effect is to realize the precise adaptation of resources according to the scene: core processing, edge collection, cross-domain connection, utilize the small-world topology characteristics: local efficient aggregation and global short path accessibility, significantly improve the data transmission efficiency, and through the global perception and intelligent regulation of the DMN control hub such as hot area weight pushing, enhance the adaptability and cross-domain service capability of the system, lay a foundation for building an efficient, intelligent and robust data service system.

[0007] The application further provides that the small-world network manager of the small-world connection layer realizes: Initialize the network topology based on the Watts Strogatz model variant algorithm, and introduce long-range cross-domain connection through random reconnection edge; According to the preset period, analyze the node communication log, when the communication frequency between cross-domain nodes exceeds the threshold, dynamically insert long-range connection through service grid, for reducing the communication path hop count.

[0008] By adopting the technical scheme, the construction and dynamic adjustment mechanism of the small-world network manager are specifically defined, long-range connections are dynamically inserted according to a communication frequency threshold based on the Watts-Strogatz variant initialization, and the technical effects are to ensure that the network has high clustering at the beginning, and can intelligently optimize the topology structure according to the actual cross-domain communication demand, dynamically establish a direct'shortcut' through the service grid technology, effectively reduce the originally long communication path hop count, significantly reduce the delay of cross-business domain data access, improve the response capability of the system to the dynamic data access mode, and solve the problem of low cross-domain efficiency caused by data islands.

[0009] The application further provides that the DMN control hub of the intelligent control layer cooperates through the following modules: A state monitoring engine: real-time collection of CPU load, network throughput, container resource utilization indicators of the data node, and generation of a system heat map; A prediction analysis module: configured with a prediction engine, using an LSTM model to predict real-time data collection peaks of the edge node, and using a Transformer model to analyze cross-domain query patterns; A resource scheduler: based on a reinforcement learning algorithm, dynamically adjusting the resource quota of the core node and the edge node, automatically migrating the CPU core number / memory quota based on the containerization platform when the data collection amount of the edge node suddenly increases, and dynamically expanding the computing resources of the edge node by 120% of the current load of the edge node.

[0010] By adopting the technical scheme, the internal modules of the intelligent control layer, the cooperative working mode and specific functions of the state monitoring engine, the prediction analysis module and the resource scheduler are described in detail, including the collection of multi-dimensional indicators, the use of LSTM / Transformer prediction, and the dynamic scheduling of resources based on reinforcement learning, and the technical effects are to realize fine, real-time monitoring and visual heat map of the system state including CPU, network and container resources, to capture short-term edge peaks using advanced prediction models such as LSTM, to analyze long-term cross-domain patterns using Transformer, to anticipate demand changes, and to realize intelligent, elastic migration and on-demand expansion of resources between core and edge nodes through the resource scheduler driven by reinforcement learning, to effectively cope with sudden loads, and to significantly improve resource utilization and overall system stability.

[0011] The application further provides that the DMN control module supports dynamic switching between idle mode and task mode: When entering idle mode, when the system load, CPU utilization and network I / O are continuously below the threshold for a preset time, the following operations are triggered: The prediction engine is called to analyze historical data access logs and node communication frequency to generate a future data demand prediction result; Perform data cleaning, deduplication, index optimization background tasks; Push the predicted hot area information to the small world connection layer, for pre-adjusting the high clustering connection weight of the local router; When entering the task mode, the routing optimization module is enabled, the real-time load state of the cross-domain scheduler in the small world connection layer is combined, and the optimal task execution path between data service nodes is calculated through a graph traversal algorithm.

[0012] By adopting the above technical solutions, the switching trigger conditions and respective core operations of the DMN control hub in the idle mode and the task mode are limited when the system load, CPU utilization and network I / O continuous preset time are lower than the threshold value, the prediction engine is called in the idle mode to analyze historical logs to generate future demand prediction, perform data cleaning, index optimization and other background tasks, and push hot area information to the connection layer for pre-adjusting the local router weight; in the task mode, the routing optimization module is enabled, the real-time load is combined, and the optimal path is calculated through a graph traversal algorithm, and the technical effects are that the background thinking function of the brain default mode network is simulated, the idle resources are actively utilized for prediction analysis and optimization reserve value in the idle time, and the system is quickly switched to the efficient execution state when the task arrives, and the intelligence, resource utilization and task response speed of the system are greatly improved.

[0013] The application is further provided that: the DMN control hub triggers the following background tasks in the idle mode: Perform data cleaning, deduplication and normalization processing; Perform index and metadata optimization; Integrate cross-domain data association relationships based on a knowledge graph to form a high-level knowledge representation to support complex queries; Perform predictive data prefetching to cache high-frequency access data to hot area nodes; The DMN control hub pre-fetches data that may be accessed in the future to adjacent data nodes and pre-allocates computing resources according to the output result of the prediction analysis module in the idle mode.

[0014] By adopting the technical scheme, the specific background tasks executed by the DMN in the idle mode are further refined, including executing data cleaning, deduplication and normalization processing, performing indexing and metadata optimization, integrating cross-domain data association relationships based on a knowledge graph to form high-level knowledge representation to support complex queries, executing predictive data prefetching to cache high-frequency access data to hot zone nodes, and according to the output of the prediction analysis module, pre-fetching data that may be accessed in the future to adjacent nodes and pre-allocating computing resources, and the technical effect is that the key maintenance and optimization work is performed during the system low load period, especially by constructing and updating the knowledge graph to integrate cross-domain associations and predictive prefetching and pre-allocation, the subsequent complex query response capability, data access speed and resource preparation degree are significantly improved, the user request delay is effectively reduced, and the system throughput is improved.

[0015] The application further provides that the prediction engine adopts a double model fusion algorithm: For short-term prediction, less than a preset time, an LSTM model is used to capture burst access patterns based on time series data. For long-term prediction, greater than the preset time, a Transformer model is used to mine cross-cluster data association requirements in combination with the topological relationship of the small world connection layer.

[0016] By adopting the above technical scheme, it is clear that the prediction engine adopts a double model fusion algorithm, i.e., for short-term prediction, less than a preset time, an LSTM model is used to capture burst access patterns based on time series data, and for long-term prediction, greater than the preset time, a Transformer model is used to mine cross-cluster data association requirements in combination with the topological relationship of the small world connection layer, and the technical effect is that the most suitable model is selected according to the different prediction time scales, LSTM is good at capturing short-term burst traffic such as edge device data flood, and Transformer can effectively analyze long-term dependencies and mine cross-business domain data association, and this targeted strategy combines the advantages of the two models to significantly improve the prediction accuracy, providing more reliable support for resource pre-allocation, data prefetching and network optimization.

[0017] The application further provides that the small world connection layer includes a local router and a cross-domain scheduler, which respectively process intra-cluster and cross-domain requests. The local router processes local data requests of data service nodes within the intra-cluster based on high-cluster connection relationships. The cross-domain scheduler forwards data requests across clusters through long-range connections, and when a topology update is triggered, the long-range connection topology of the cross-domain scheduler is adjusted first to ensure that the number of hops of the cross-domain request path is stable and does not exceed a preset number of hops. The topology update is triggered by the following events: Data access hot zone changes output by the DMN control module in the prediction mode. The number of consecutive failures of the long-range connection of the cross-domain dispatcher is greater than or greater than a preset number of times. The high-cluster connection load rate of the local router is continuously greater than a preset proportion within a preset time.

[0018] By adopting the technical solution, the internal components of the small-world connection layer are defined, the local router based on the high-cluster connection processes the intra-cluster request, the cross-domain dispatcher processes the cross-cluster request through the long-range connection, and the optimization mechanism of the cross-domain dispatcher preferentially adjusts the topology of the cross-domain dispatcher to ensure that the hop number of the cross-domain request path is stable and does not exceed a preset hop number, and specific events triggering the topology update are specified, for example, the hot area change predicted by the DMN, the number of consecutive failures of the long-range connection of the cross-domain dispatcher is greater than a preset number of times, and the high-cluster connection load rate of the local router continuously exceeds the limit, and the technical effect is to realize clear division of routing responsibilities, preferentially ensure efficient and stable cross-domain paths through dynamic topology update driven by events, ensure that the network structure can be quickly and accurately optimized in the case of hot spot migration, connection failure or local congestion, maintain low-latency and high-reliability cross-domain communication, and improve the service quality and robustness in a dynamic environment.

[0019] A brain-inspired data service system method, comprising the following steps: A hierarchical initialization step: deploying containerized data nodes in the infrastructure layer, generating an initial topology through the small-world network manager of the small-world connection layer, and starting full-link monitoring by the state monitoring engine of the intelligent control layer; A cross-layer cooperative processing step: when the external access layer receives a cross-department query request, the DMN control hub of the intelligent control layer calls the pre-computed results of the prediction analysis module, directly reaches the sales and financial gateway node through the long-range cross-domain connection of the small-world connection layer, and the edge node synchronously provides real-time device data; A hierarchical adaptive optimization step: the intelligent control layer triggers the small-world connection layer to adjust the long-range connection distribution according to the state monitoring data, and simultaneously guides the infrastructure layer to perform container resource reallocation.

[0020] By adopting the technical solution, the claim describes a hierarchical method applying the above system, which includes three core steps: a hierarchical initialization step of deploying containerized nodes and generating an initial topology to start monitoring, a cross-layer cooperative processing step of calling pre-computed results through long-range connection to directly reach target nodes such as sales and financial gateway nodes and integrate edge real-time data when the access layer receives a cross-department request, and a hierarchical adaptive optimization step of triggering the connection layer to adjust the topology such as long-range connection distribution according to monitoring data and guiding the infrastructure layer to perform container resource reallocation, and the technical effect is to realize fast deployment and start of the system through hierarchical division and cooperative processes, efficiently process complex cross-domain requests, and dynamically trigger linkage optimization of each layer to form end-to-end closed-loop self-optimization capability, so as to ensure that the system continuously adapts to changes in business requirements and data load fluctuations, and overall improve the agility, efficiency and intelligent level.

[0021] The application is further provided that: the layered adaptive optimization includes: a small-world connection layer: when the communication between the production data node and the R&D data node lasts for more than a preset number of days exceeding the system average, a long-range connection is added between the two through a small-world network manager, and the number of hops of the communication path is shortened; an intelligent control layer: a prediction analysis module pre-allocates a preset proportion of database connection resources for the monthly financial settlement task in advance based on historical data; an infrastructure layer: when the resource scheduler detects that the CPU utilization rate of the edge node exceeds a preset percentage, a number of new instances are automatically started in the container platform for load balancing; By adopting the above technical solution, the operation examples in the layered adaptive optimization steps are specified, for example, the small-world connection layer adds a direct long-range connection through the manager when the communication between the production and R&D nodes lasts for more than a preset number of days exceeding the system average, greatly shortening the number of hops, such as from 5 hops to 1 hop, the intelligent control layer prediction analysis module pre-allocates a preset proportion, such as 50%, of database connection resources for the monthly financial settlement task in advance based on historical data, and the infrastructure layer resource scheduler automatically starts a new container instance for load balancing when it detects that the CPU utilization rate of the edge node exceeds a preset percentage, which has the technical effect of providing a specific implementation scheme for layered optimization, greatly shortening the path by automatically establishing key connections, reserving resources in advance based on prediction to avoid contention, and flexibly expanding to cope with bottlenecks, which together significantly reduces the delay of the key path, prevents performance degradation, and ensures the stability and scalability of the service under load pressure.

[0022] The application is further provided that: the intelligent control layer further includes a learning and adaptation engine for: analyzing historical access patterns and updating an access frequency model; evaluating task execution efficiency and optimizing resource scheduling strategies; adjusting LSTM / Transformer model parameters according to real-time feedback; synchronously updating a knowledge base to reflect the latest data association relationships.

[0023] The learning and adaptation engine updates the access pattern model according to the task execution log, and optimizes the long-range connection distribution of the small-world connection layer; When the knowledge association relationship in the knowledge base changes, the index reconstruction and metadata update of the data node are triggered, and the pre-computed query strategy is adjusted synchronously.

[0024] By adopting the technical scheme, the role of the learning and adaptation engine in the intelligent control layer is emphasized for analyzing historical access patterns and updating the frequency model, evaluating task execution efficiency to optimize resource scheduling strategies, adjusting LSTM or Transformer model parameters according to real-time feedback, synchronously updating the knowledge base to reflect the latest data association relationship, and correlating the influence on the connection layer and the data node, that is, the learning engine updates the access pattern model to optimize the small-world long-range connection distribution according to the task log, triggers the index reconstruction and metadata update of the data node when the knowledge base association relationship changes, and synchronously adjusts the pre-computed query strategy, the technical effect of which is to give the system the ability to continuously evolve, automatically optimize core components such as updating model parameters, improving scheduling strategies, and adjusting network connections by continuously learning historical patterns and feedback, and intelligently trigger downstream synchronous updates when the knowledge or association changes, so that the system can continuously adapt to new data patterns, business requirements, and performance targets, and maintain long-term running efficiency and intelligent service level.

[0025] The present application has the following technical effects due to the adoption of the above technical scheme. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is a system overall architecture diagram; Figure 2 is a small-world network connection topology diagram; Figure 3 is a DMN internal workflow diagram. DETAILED DESCRIPTION

[0027] The present application will be further described in detail below in combination with the drawings and examples.

[0028] Example 1 In the prior art, with the wide application of big data and cloud computing technology, enterprise data service systems are facing problems such as cross-department data integration difficulties, low resource utilization, and response delay, and traditional architectures rely on centralized nodes to process requests, resulting in high load on core nodes, and edge data cannot be processed in time, and existing systems cannot effectively utilize resources for preprocessing during idle periods, and cross-domain queries require multiple jumps, increasing communication delay, for example, the production data and financial data of a manufacturing enterprise belong to different server clusters, and when generating a comprehensive report, manual coordination of data interfaces is required, resulting in a response time exceeding the business tolerance threshold.

[0029] The inventor observes that the biological nervous system has the characteristics of adaptive connection and idle period self-optimization, proposes a working mechanism simulating the default mode network of the brain, first analyzes the spatiotemporal distribution characteristics of cross-domain data access, finds that local high-frequency communication coexists with cross-domain low-frequency access, secondly, studies the high clustering and short path characteristics of the small world network, tries to apply it to the data node connection topology, further designs a hierarchical control architecture, performs predictive resource pre-allocation during the idle period, reduces the response delay when the task is triggered, and realizes the direct routing of cross-domain requests by forming a dynamic connection network between edge nodes and core nodes.

[0030] The application proposes a brain-inspired data service system, including a data node cluster, a small world connection layer, an intelligent control layer, an external access layer and an infrastructure layer. The data node cluster is composed of distributed servers, containers, microservices, etc., covering core nodes, edge nodes and gateway nodes. The small world connection layer adopts a topology structure combining high clustering and long-range connection. The intelligent control layer includes state monitoring, predictive analysis and resource scheduling modules. The external access layer provides standardized interfaces. The infrastructure layer adopts a hybrid architecture.

[0031] The data node cluster refers to a distributed computing unit divided according to business needs, which can be quickly deployed using containerization technology, and is used to handle data service needs in different business domains. The small world connection layer refers to a connection topology construction module based on complex network theory, which maintains communication efficiency within the business domain through local dense connection and realizes direct routing across domains through random long-range connection. The intelligent control layer refers to a central control system simulating the default mode of the brain, which drives system optimization through real-time monitoring and predictive analysis. The infrastructure layer refers to a combination of physical and virtualized resources that support system operation, such as deploying core nodes on cloud servers and edge nodes on industrial gateway devices.

[0032] During system operation, the hybrid architecture of the infrastructure layer provides elastic resource support. When the external access layer receives user requests, the small world connection layer selects local routing or cross-domain scheduling according to the request type. The intelligent control layer continuously monitors node status and performs data preprocessing and resource pre-allocation during the idle period. For example, when frequent interactions between production nodes and quality nodes are detected, a long-range connection channel is automatically established. When the monthly financial settlement task is triggered, the pre-allocated database connection resources can be immediately put into use, avoiding resource contention.

[0033] This scheme realizes the decoupling of control and execution through a hierarchical architecture, avoiding the bottleneck of a single control node. The small world connection topology maintains local efficiency while reducing cross-domain hop count, reducing redundant links compared to the traditional full connection mode. The predictive resource scheduling mechanism of the intelligent control layer changes the traditional post-response to pre-preparation, effectively utilizing idle system resources.

[0034] The application realizes path optimization of cross-domain data service, shortens the average response time of cross-department query to a business acceptable range, improves the efficiency of sudden task processing through a dynamic resource pre-allocation mechanism, improves the resource utilization rate of edge nodes, automatically executes data optimization tasks during idle periods to avoid waste of computing resources, and forms a continuously self-optimizing service capability.

[0035] The small-world network manager of the small-world connection layer initializes the network topology based on a Watts Strogatz model variant algorithm, introduces long-range cross-domain connections through random edge reconnection, analyzes node communication logs at a preset period, and dynamically inserts long-range connections through a service mesh when the communication frequency between cross-domain nodes exceeds a threshold, to reduce the communication path hop count.

[0036] The Watts Strogatz model variant algorithm is a topology generation algorithm improved based on the small-world network theory, which can specifically introduce long-range connections in an initial high-clustering network through a random edge reconnection mechanism, balance local connection density and global connectivity efficiency by adjusting a reconnection probability parameter, and replace part of the connections in the initial ring topology with a probability, for example, replace local connections with cross-domain connections with a preset probability, to form a network structure with high clustering and low path length. The service mesh dynamically inserts long-range connections refers to real-time adjustment of node-to-node communication links through service mesh technology, for example, when it is detected that the cross-domain communication frequency exceeds a set threshold, a direct connection channel is automatically established between the corresponding nodes.

[0037] In the network initialization phase, an improved Watts Strogatz algorithm is used to construct a basic topology, cross-domain connections are introduced through a random reconnection mechanism while preserving the local high-clustering characteristics, communication logs of cross-domain nodes are periodically analyzed, for example, node-to-node interaction frequency is counted in units of hours or days, when the communication frequency of a specific node pair exceeds a preset threshold, the dynamic link configuration function of the service mesh is triggered, at this time, the system automatically establishes a long-range connection between the corresponding nodes, bypasses the original intermediate nodes to form a direct path, and reduces the cross-domain communication hop count from multiple hops in the original topology to a single hop.

[0038] Traditional systems usually use fixed topologies or simple distributed architectures, and cross-domain communication paths rely on statically configured intermediate nodes, which cannot be dynamically optimized according to actual communication needs. However, the present scheme can automatically identify high-frequency cross-domain interaction scenarios and optimize connection topologies through periodic communication analysis and service mesh dynamic adjustment mechanisms, effectively solving the problems of high cross-domain communication delay and path redundancy in traditional architectures.

[0039] The application can automatically optimize the cross-domain communication path according to the actual business demand, significantly reduce the data transmission delay in the high-frequency cross-domain interaction scene, reduce the intermediate forwarding link by dynamically inserting the long-range connection, avoid the waste of bandwidth resources caused by path redundancy, and improve the response efficiency of the cross-domain data service.

[0040] The DMN control hub of the intelligent control layer cooperates through the following modules: the state monitoring engine collects the CPU load, network throughput, and container resource utilization rate indicators of the data nodes in real time and generates a system heat map; the prediction analysis module is configured with a prediction engine, which uses an LSTM model to predict the real-time data collection peak of the edge node and analyzes the cross-domain query mode through a Transformer model; and the resource scheduler dynamically adjusts the resource quota of the core node and the edge node based on a reinforcement learning algorithm, automatically migrates the CPU core number or memory quota based on the containerized platform when the data collection amount of the edge node suddenly increases, and dynamically expands the computing resources of the edge node by 120% of the current load of the edge node.

[0041] The state monitoring engine refers to a module for continuously collecting system running indicators, which can be implemented by using a Prometheus monitoring framework combined with a custom indicator collector, and periodically acquires resource consumption data of each node through exposed API interfaces. The LSTM model in the prediction analysis module refers to a long short-term memory neural network, which can be implemented by using a three-layer network structure built with a TensorFlow framework to process periodic fluctuation characteristics in time series data. The Transformer model refers to a deep learning architecture based on a self-attention mechanism, which can be implemented by fine-tuning a pre-trained model in a HuggingFace library to capture associated patterns in cross-domain queries. The reinforcement learning algorithm in the resource scheduler refers to a dynamic decision-making method based on environmental feedback, which can be implemented by using a Q-learning algorithm combined with API interfaces of a container orchestration platform to optimize resource allocation strategies through a reward function.

[0042] The state monitoring engine periodically collects running indicators of each node and generates a visual heat map, in which the color depth reflects the resource consumption intensity of different regions. The prediction analysis module predicts the data collection peak in the next five minutes based on the historical load curve of the edge node in the heat map, and analyzes the semantic association in cross-department query requests through a Transformer model. When a sudden traffic occurs in the edge node, the resource scheduler calculates the optimal expansion scheme based on the reinforcement learning strategy, such as increasing the CPU core number of a certain edge node from 4 cores to 5 cores and the memory quota from 8 GB to 9.6 GB in the containerized platform, with an expansion range accurately matching the 120% growth demand of the current load.

[0043] The traditional system usually adopts a static threshold alarm mechanism in resource scheduling, which cannot cope with burst traffic scenarios. However, the present scheme can predict the trend of resource demand change in advance by combining the dual model architecture of time series prediction and semantic analysis, and realize precise elastic expansion and contraction operation combined with reinforcement learning algorithm, avoiding the response delay problem caused by manual intervention. The present application can automatically identify the mutation trend of data collection quantity in the edge computing scene, timely adjust the calculation resource configuration to avoid node overload, and when the edge node of the logistics warehouse system has a surge of order data during the promotion period, the system can automatically increase the processing capacity to above the preset safety threshold, ensuring the integrity and timeliness of real-time collection of equipment data, and avoiding the overuse of resources of the core node affecting other business operations.

[0044] The DMN control module supports dynamic switching between idle mode and task mode: when entering idle mode, when the system load, CPU utilization, and network I / O are continuously below the threshold for a preset time, the following operations are triggered: calling the prediction engine to analyze historical data access logs and node communication frequency to generate future data demand prediction results; performing data cleaning, deduplication, and index optimization background tasks; pushing the predicted hot area information to the small world connection layer to pre-adjust the high clustering connection weight of the local router; when entering task mode, the routing optimization module is enabled, and the real-time load state of the cross-domain scheduler in the small world connection layer is combined to calculate the optimal task execution path between data service nodes through graph traversal algorithm.

[0045] Idle mode refers to a low-power running state automatically triggered when the system resource utilization is below the preset threshold, which can be implemented based on a sliding window-based load monitoring algorithm, and is used to execute background optimization tasks during off-peak hours. Task mode refers to a high-performance running state activated in response to real-time requests, which is implemented through a dynamic path planning algorithm, such as an improved version of Dijkstra's algorithm to calculate the optimal task path. The prediction engine refers to a data access prediction module based on machine learning models, which can be implemented using a fusion architecture of LSTM neural networks and Transformer models to identify periodic access patterns and cross-domain associated demands. Data prefetching refers to a cache mechanism that loads high-frequency access data to target nodes in advance based on prediction results, which is implemented through a distributed cache management protocol, such as using LRU cache replacement strategy and prefetch queue to work together.

[0046] When the system detects that the CPU utilization rate and network traffic indicators are continuously below the set threshold for a preset length of time, it automatically switches to idle mode. At this time, the prediction engine performs time series analysis on historical access logs to identify a set of data objects that may be frequently accessed in the future. The data cleaning task marks and merges redundant data in the storage node. The index optimization module reconstructs the B+ tree index structure to improve query efficiency. The predicted hot zone information is pushed to the local router through the message queue, allowing it to adjust the connection weight distribution strategy in advance. When external requests arrive, the system immediately switches to task mode. The routing optimization module uses the breadth-first search algorithm to traverse the topology graph based on the load state of the current cross-domain scheduler, and selects the task execution path with the least number of hops and balanced load.

[0047] The traditional system only maintains the basic running state during the idle period, and does not effectively utilize idle resources to perform data optimization tasks, resulting in resource waste and response delay. The mode switching mechanism in the prior art relies on manual configuration strategy and cannot dynamically adjust the running state according to real-time load. The present scheme actively completes data preprocessing and resource pre-allocation during the system idle period through automatic threshold monitoring and prediction-driven background task execution, significantly reducing the service response time in task mode.

[0048] The present application realizes the dual improvement of system resource utilization and response efficiency. The index optimization performed in idle mode improves cross-domain query speed, the data prefetching mechanism reduces the access delay of high-frequency access data, and the dynamic path planning in task mode effectively avoids network congestion, ensuring stable and controllable end-to-end processing time of cross-department data requests. The system can automatically switch between running modes according to real-time load status, ensuring service quality while reducing energy consumption and idle computing resources.

[0049] In idle mode, the following background tasks are triggered: data cleaning, deduplication, and normalization processing are performed; index and metadata optimization is performed; cross-domain data association relationships are integrated based on knowledge graph to form high-level knowledge representation to support complex queries; predictive data prefetching is performed to cache high-frequency access data to hot zone nodes; based on the output results of the prediction analysis module, data that may be accessed in the future is prefetched to adjacent data nodes, and computing resources are pre-allocated.

[0050] Data cleaning refers to automatically identifying and correcting errors or redundant information in the data through preset rules, which can be implemented by regular expression matching combined with machine learning models to improve data quality and consistency. Index and metadata optimization refers to restructuring the database index structure, which can be implemented through B+ tree balancing algorithm or columnar storage optimization technology to speed up query response speed. Knowledge graph integration cross-domain data association refers to building an entity relationship network through a graph database, which can be implemented using Neo4j or JanusGraph to discover potential connections across business domains. Predictive data prefetching refers to loading data in advance based on access pattern prediction results, which can be implemented using LRU cache replacement algorithm combined with time series prediction model to reduce subsequent query delay.

[0051] When the system enters idle mode, the background task execution process is automatically started. The data cleaning module traverses the storage nodes, identifies abnormal data formats through the rule engine, and performs standardized processing. The index optimization engine analyzes query logs, reconstructs the index structure of high-frequency access fields, and synchronously updates cross-domain entity relationships. The knowledge graph engine establishes new semantic association paths, and the predictive analysis module generates prefetch instructions based on historical access patterns, distributes predicted data copies to the cache area of the target node, and pre-allocates computing resources between adjacent nodes through the container orchestration platform to dynamically adjust resource quotas, ensuring that the prefetched data has instant processing capabilities.

[0052] The traditional system only maintains the basic running state during the idle period, and does not effectively utilize computing resources for data optimization. The existing data preprocessing technology requires manual triggering and lacks cross-domain association analysis, and cannot form a global optimization strategy. The present scheme realizes the collaborative optimization of system resources through automated background tasks, completes data quality improvement and access path pre-optimization during off-peak periods, and overcomes the lagging defects of the traditional responsive processing mode.

[0053] The present application realizes the full utilization of computing resources during the idle period of the system, effectively improves the data query efficiency and accuracy, and continuously maintains the cross-domain data association relationship to enhance the response capability of complex queries. The predictive data prefetching mechanism significantly reduces the service delay in high-concurrency scenarios, and the pre-allocated computing resource strategy guarantees the availability of data prefetching, forming a complete offline optimization and online service collaborative mechanism.

[0054] The prediction engine adopts a dual-model fusion algorithm. A short-term prediction adopts an LSTM model when the prediction time is less than a preset time, to capture sudden access patterns based on time series data. A long-term prediction adopts a Transformer model when the prediction time is greater than the preset time, to mine cross-cluster data correlation requirements in combination with the topology relationship of the small-world connection layer. The LSTM model refers to a long short-term memory neural network model, which can be specifically implemented by a recurrent neural network with a forgetting gate, an input gate and an output gate, to capture short-term dependencies and sudden access patterns in time series data. The Transformer model refers to a deep learning model based on a self-attention mechanism, which can be specifically implemented by an encoder structure composed of a multi-head attention layer and a feedforward neural network layer, to process long-distance dependencies and cross-cluster data correlation requirements. The dual-model fusion algorithm refers to a technical solution of assigning prediction tasks of different time scales to different models for processing. Specifically, the prediction task types can be divided by a preset time threshold, and a model switching mechanism can be established to implement the dual-model fusion algorithm, to balance short-term fluctuation characteristics and long-term correlation patterns.

[0055] When the prediction time range is below the preset time threshold, the system automatically calls the LSTM model to process time series data. The model processes time series features layer by layer through a recurrent neural network structure, and captures the short-term fluctuation patterns of device sensor data acquisition, user query frequency and other indicators using memory cells. When the prediction time range exceeds the preset time threshold, the system switches to the Transformer model, which analyzes the topological connection relationship of cross-cluster nodes through a self-attention mechanism, identifies potential data correlation requirements between different business domains, and generates a final output through a weighted fusion module, to guide resource pre-allocation and data prefetching strategies.

[0056] Traditional prediction methods usually use a single model to process prediction tasks of different time scales, resulting in short-term prediction ignoring cross-domain correlation features and long-term prediction being difficult to capture sudden fluctuations. The present solution uses a dual-model division mechanism to maintain the sensitivity of short-term prediction while effectively using topological relationships to enhance the correlation analysis capability of long-term prediction, solving the problem of insufficient adaptability of a single model.

[0057] The present application can automatically select the optimal model according to the time characteristics of the prediction task, improving the resource utilization rate of edge nodes and the efficiency of cross-domain data scheduling. The short-term prediction module accurately captures real-time data acquisition peaks, avoiding resource shortages caused by sudden loads. The long-term prediction module optimizes cross-cluster resource pre-allocation through topological relationship analysis, reducing cross-domain query delays.

[0058] The small-world connection layer includes a local router and a cross-domain scheduler, which respectively process intra-cluster and cross-cluster requests; the local router processes local data requests of data service nodes within the intra-cluster based on high-cluster connection relationships; the cross-domain scheduler forwards cross-cluster data requests through long-range connections, and when a topology update is triggered, the long-range connection topology of the cross-domain scheduler is adjusted in priority to ensure that the number of hops of the cross-domain request path is stable and does not exceed a preset number of hops; wherein the topology update is triggered by the following events: data access hot area changes output by the DMN control module in the prediction mode; the number of consecutive failures of the long-range connection of the cross-domain scheduler is greater than or equal to a preset number; and the high-cluster connection load rate of the local router is continuously greater than a preset proportion within a preset time.

[0059] The local router is a communication control unit deployed within a business domain, which can be implemented by a routing device based on the OSPF protocol, and realizes low-latency communication of nodes within the domain by maintaining high-cluster connection relationships, for reducing the transmission delay of data interaction within the same business domain. The cross-domain scheduler is a relay device for cross-business domain communication, which can be implemented by a service mesh component extended from the BGP protocol, and shortens the cross-domain communication path by dynamically maintaining long-range connections, for solving the problem of low cross-domain access efficiency caused by data islands. The topology update trigger event is a monitoring mechanism for network state changes, which can use a sliding window algorithm to count the number of connection failures, and detect load rate anomalies through time series analysis, for realizing adaptive adjustment of network structure to maintain service quality.

[0060] The local router forms an intra-domain communication backbone network by maintaining high-cluster connection relationships, and when receiving an intra-cluster request, directly transmits it through a pre-established physical link. The cross-domain scheduler continuously monitors the state of the long-range connection, and when detecting that the number of hops of the cross-domain communication path exceeds a threshold, it triggers a topology adjustment operation of the cross-domain scheduler in priority, for example, when the DMN control module predicts that the sales data hot area migrates, the cross-domain scheduler automatically establishes a new long-range connection between the sales domain and the finance domain, while removing inefficient redundant connections. The local router and the cross-domain scheduler form a hierarchical control architecture, which maintains the communication efficiency within the domain while optimizing the cross-domain path selection.

[0061] The traditional data system uses a static routing strategy, which leads to a decrease in cross-domain communication efficiency with business fluctuations. The present scheme realizes continuous optimization of the cross-domain path through a dynamic topology update mechanism. In the prior art, cross-domain connection adjustment usually requires manual intervention. The present scheme realizes automatic network reconstruction based on preset trigger conditions, effectively coping with sudden cross-domain access demands.

[0062] The application can automatically identify high-load communication paths and implement targeted optimization, significantly reduce communication delay in cross-domain data access scenarios, ensure communication efficiency through hierarchical processing mechanism, dynamically adjust cross-domain connection topology to effectively avoid network congestion, improve the response stability of cross-department data services, and eliminate the influence of single-point failure on the overall performance of the system in time based on the automatic update mechanism triggered by the preset conditions, and ensure the continuous and efficient operation of the data service system in complex business scenarios.

[0063] The brain-inspired data service system method includes a hierarchical initialization step, a cross-layer collaborative processing step, and a hierarchical adaptive optimization step: the hierarchical initialization step deploys containerized data nodes in the infrastructure layer, generates an initial topology through the small-world network manager of the small-world connection layer, and starts full-link monitoring by the state monitoring engine of the intelligent control layer; the cross-layer collaborative processing step calls the pre-computed results of the prediction analysis module through the DMN control hub of the intelligent control layer when the cross-department query request is received in the external access layer, directly reaches the sales and financial gateway node through the long-range cross-domain connection of the small-world connection layer, and the edge node synchronously provides real-time device data; the hierarchical adaptive optimization step triggers the small-world connection layer to adjust the long-range connection distribution according to the state monitoring data of the intelligent control layer, and simultaneously guides the infrastructure layer to perform container resource reallocation.

[0064] The hierarchical initialization step refers to the process of establishing the system infrastructure, which can specifically deploy data nodes by using a container orchestration tool, for example, manage node instances in different locations through a Kubernetes cluster. This step lays the foundation for dynamic resource scheduling. The cross-layer collaborative processing step refers to a mechanism in which multiple hierarchical components cooperate to respond to complex requests. This mechanism can specifically implement cross-domain connection routing through service mesh technology, effectively solving the problem of cross-department data access delay in traditional systems. The hierarchical adaptive optimization step refers to the process of continuous self-regulation of the system. This process can specifically use a reinforcement learning algorithm to dynamically adjust resource quotas, achieving dual improvement of resource utilization and service quality.

[0065] In the system startup phase, the containerized platform can be configured to deploy core nodes in the cloud and lightweight container instances in edge devices. When receiving a cross-domain query, the prediction analysis module can predict the target node based on historical access patterns, for example, directly routing financial data requests to a preset gateway container instance. During the running process, the state monitoring engine can periodically collect network delay data, and when it detects that the communication hop count between specific nodes exceeds the standard, it triggers the horizontal expansion operation of the container instance. This hierarchical collaboration mechanism ensures that the data processing path always remains in the optimal state, while avoiding system shocks caused by single-level adjustments.

[0066] The traditional centralized system needs to forward the request layer by layer when querying across departments, while the method realizes direct routing across domains through pre-computed paths, reduces intermediate links, the static resource allocation mode in the prior art cannot cope with sudden load, while the method realizes elastic scaling of computing resources through hierarchical adaptive optimization, and each level of the conventional distributed system operates independently, while the method realizes overall collaborative optimization through cross-level transmission of state monitoring data.

[0067] The application realizes end-to-end direct processing of cross-domain data requests, effectively reduces query response delay, and dynamic resource reallocation mechanism ensures service stability of edge nodes under sudden load, avoids data processing interruption caused by insufficient resources, and multi-level collaborative optimization enables the system to automatically adapt to changes in business requirements, improving the processing efficiency of complex query tasks.

[0068] The specific implementation of the hierarchical adaptive optimization step includes the collaborative optimization mechanism of the small-world connection layer, the intelligent control layer and the infrastructure layer. When the communication between the production data node and the research and development data node lasts for more than a preset number of days and exceeds the system average, the small-world network manager adds a long-range connection between them and shortens the communication path hop count; the prediction analysis module pre-allocates a preset proportion of database connection resources for the monthly financial settlement task based on historical data; when the resource scheduler detects that the CPU utilization rate of the edge node exceeds a preset percentage, it automatically starts several new instances in the containerization platform for load balancing.

[0069] Hierarchical adaptive optimization refers to the system solving cross-domain communication and resource allocation problems through collaborative adjustment of different levels. Specifically, it can be implemented by dynamically adjusting the connection topology and resource allocation algorithm. The small-world connection layer optimizes the communication path by adding cross-domain long-range connections, such as triggering topology update when the communication frequency between nodes exceeds a threshold. The intelligent control layer responds to periodic task requirements through predictive resource pre-allocation, such as pre-allocating database connection resources based on historical access patterns. The infrastructure layer realizes load balancing through dynamic scaling of containerized instances, such as automatically starting new instances when the resource utilization rate of the edge node reaches a preset threshold.

[0070] When the communication frequency between the production and research and development data nodes is continuously higher than the system average, the small-world network manager automatically establishes a cross-domain long-range connection, shortening the communication path that originally needs to be forwarded through multiple intermediate nodes to a direct link. The intelligent control layer pre-allocates database connection pool resources before the monthly financial settlement task starts by analyzing historical task execution logs, avoiding resource contention during the task peak period. The resource scheduler of the infrastructure layer monitors the load of the edge node in real time, and when it detects that the CPU utilization rate exceeds a preset threshold, it automatically starts a new instance in the containerization platform, migrating part of the computing tasks to the new instance for execution.

[0071] In some embodiments, the preset number of days can be, for example, 30 days, the system average can be dynamically calculated by a sliding window algorithm; the preset proportion of database connection resources can be, for example, 40% of the total connection pool; the preset percentage can be, for example, 80% of the CPU utilization threshold; the number of new instances can be dynamically calculated according to the current load and resource margin.

[0072] The traditional system usually adopts a fixed routing strategy when optimizing cross-domain communication, and cannot dynamically adjust the topology structure according to the actual communication mode. The present scheme realizes adaptive optimization of cross-domain connection through a hierarchical collaborative mechanism. Existing resource allocation methods mostly use static quotas or passive response expansion. The present scheme realizes active resource supply through predictive pre-allocation and dynamic instance creation. Existing load balancing techniques rely on fixed thresholds to trigger expansion. The present scheme combines real-time monitoring and containerization technology to realize elastic resource scheduling.

[0073] The present application effectively reduces the path hop count and transmission delay of cross-domain communication, improves the resource supply efficiency of periodic tasks, enhances the dynamic load processing capability in the edge computing scenario, and optimizes the communication path to reduce the cross-department data interaction delay. Predictive resource pre-allocation avoids resource bottlenecks during task execution, and dynamic scaling of containerized instances ensures the stability of the system in high-load scenarios.

[0074] The intelligent control layer also includes a learning and adaptation engine for analyzing historical access patterns and updating access frequency models, evaluating task execution efficiency and optimizing resource scheduling strategies, adjusting LSTM or Transformer model parameters based on real-time feedback, and synchronously updating the knowledge base to reflect the latest data association relationships. The learning and adaptation engine updates the access pattern model based on task execution logs and optimizes the long-range connection distribution of the small-world connection layer. When the knowledge association relationships in the knowledge base change, trigger index reconstruction and metadata update of the data nodes, and synchronously adjust the pre-computed query strategy.

[0075] The learning and adaptation engine refers to a computing module with autonomous learning and dynamic adjustment capabilities, which can be implemented using a machine learning framework combined with a rule engine. It optimizes strategies by continuously analyzing system operation data. The access frequency model refers to a probability distribution model that describes data access patterns, which can be constructed using a hidden Markov model or time series analysis algorithm to predict future data access hotspots. The resource scheduling strategy refers to a decision rule for dynamically allocating computing resources, which can be implemented using a reinforcement learning algorithm combined with a queuing theory model to adjust resource quotas based on real-time load status. The knowledge base refers to a structured storage system that stores data association relationships, which can be implemented using a graph database or triple storage architecture to maintain cross-domain data association relationships through knowledge graph technology.

[0076] The learning and adaptation engine continuously collects task execution logs and system running indicators, updates the access mode model through a combination of offline training and online inference, automatically generates long-range connection optimization suggestions and pushes them to the small-world connection layer when a significant change in the access frequency of a specific data node is detected, automatically starts the index reconstruction process when a knowledge base update event is triggered, recalculates the correlation weights between data nodes, and generates an optimized query execution plan based on the new correlation relationships, and the model parameter adjustment process uses an online learning mechanism to combine real-time feedback data with historical training data to iteratively update the weight parameters of the prediction model through a gradient descent algorithm.

[0077] The traditional system lacks a continuous learning mechanism and cannot dynamically optimize resource allocation strategies based on running data, while the present scheme realizes automatic tuning of system parameters and real-time synchronization of the knowledge base by introducing a learning and adaptation engine. The resource scheduling strategies in the prior art are mostly static configurations and are difficult to adapt to dynamic load changes, while the present scheme realizes dynamic optimization of resource allocation through reinforcement learning algorithms.

[0078] The present application can effectively solve the problems of low resource utilization and poor cross-domain query efficiency of traditional data service systems, continuously optimize system configuration through a self-learning mechanism, reduce manual maintenance costs, eliminate data island phenomena through a real-time synchronization mechanism of the knowledge base, improve the response speed of cross-domain data correlation queries, and enhance the adaptability of the system to sudden access patterns through a dynamic model parameter adjustment function, ensuring that the prediction accuracy continues to improve over time.

Claims

1. A brain-inspired data service system, characterized in that include: Data node cluster: composed of several distributed data nodes, used to store, process, analyze data or provide data services. The data node is at least one of a server, container, microservice, logic unit or AIAgent, including core nodes deployed in cloud server clusters, edge nodes deployed in edge computing devices and gateway nodes deployed in containerized platforms, corresponding to the core data processing, edge data collection and cross-domain data connection requirements of the business domain respectively; Small-world connection layer: This layer uses a small-world network model to build the connection topology between data nodes, including a small-world network manager, local high-cluster connections, and long-range cross-domain connections. Local high-cluster connections enable dense interconnection of data nodes in the same business domain, while long-range cross-domain connections dynamically maintain cross-domain shortcuts through the small-world network manager. Intelligent control layer: includes the DMN control center and its subordinate status monitoring engine, prediction analysis module and resource scheduler. It serves as the system control center to simulate the brain's default mode network function. External access layer: provides user application access interface and protocol adapter. Infrastructure layer: Deployed on a hybrid architecture of cloud server clusters, containerized platforms, edge computing devices, and physical servers; The intelligent control layer collects the communication hop count and delay indicators of the small-world connection layer in real time through the status monitoring engine, driving the prediction analysis module to generate topology optimization instructions; the DMN control center automatically triggers mode switching through the threshold comparison algorithm based on the real-time data of the status monitoring engine, and pushes the hot zone weights to the small-world network manager.

2. A brain-inspired data service system according to claim 1, characterized in that: The small-world network manager of the small-world connection layer implements: Initialize the network topology based on a variant of the Watts-Strogatz model and introduce long-range cross-domain connections by randomly rewiring edges. Node communication logs are analyzed according to a preset period. When the communication frequency between cross-domain nodes exceeds the threshold, long-distance connections are dynamically inserted through the service grid to reduce the number of communication path hops.

3. The brain-inspired data service system according to claim 2, characterized in that: The DMN control center of the intelligent control layer works together through the following modules: Status monitoring engine: collects data node CPU load, network throughput, and container resource utilization indicators in real time to generate a system heat map; Prediction and Analysis Module: This module is equipped with a prediction engine that uses the LSTM model to predict the real-time data collection peak of edge nodes and the Transformer model to analyze cross-domain query patterns. Resource Scheduler: Dynamically adjusts the resource quotas of core nodes and edge nodes based on reinforcement learning algorithms. When the amount of data collected by edge nodes suddenly increases, it automatically migrates computing resources based on the CPU core number / memory quota of the containerized platform and dynamically expands the computing resources by 120% of the current load of the edge nodes to the edge nodes.

4. The brain-inspired data service system according to claim 3, characterized in that: The DMN control module supports dynamic switching between idle mode and task mode: When entering idle mode, if the system load, CPU utilization, or network I / O stays below the preset threshold for a continuous period, the following actions are triggered: Call the prediction engine to analyze historical data access logs and node communication frequency to generate future data demand prediction results; Perform background tasks of data cleaning, deduplication, and index optimization; Push the predicted hotspot information to the small-world connection layer to pre-adjust the high clustering connection weights of the local router; When entering task mode, the routing optimization module is enabled. Combined with the real-time load status of the cross-domain scheduler in the small-world connection layer, the optimal task execution path between data service nodes is calculated through a graph traversal algorithm.

5. The brain-inspired data service system according to claim 4, characterized in that: The DMN control hub triggers the following background tasks in idle mode: Perform data cleaning, deduplication, and normalization; Perform index and metadata optimization; Integrate cross-domain data associations based on knowledge graphs to form high-level knowledge representations to support complex queries; Perform predictive data prefetching to cache frequently accessed data to hot zone nodes; In idle mode, the DMN control center pre-fetches data that may be accessed in the future to adjacent data nodes based on the output results of the prediction analysis module and pre-allocates computing resources.

6. The brain-inspired data service system according to claim 5, characterized in that: The prediction engine uses a dual-model fusion algorithm: For short-term prediction, when the time is less than the preset time, the LSTM model is used to capture the burst access pattern based on time series data; For long-term predictions, when the time is greater than the preset time, the Transformer model is used to combine the topological relationship of the small-world connection layer to mine cross-cluster data association requirements.

7. The brain-inspired data service system according to claim 3, characterized in that: The small-world connection layer includes a local router and a cross-domain scheduler, which handle same-cluster and cross-domain requests respectively; The local router processes local data requests from data service nodes in the same cluster based on high cluster connection relationships; The cross-domain scheduler forwards cross-cluster data requests through long-distance connections. When a topology update is triggered, the long-distance connection topology of the cross-domain scheduler is adjusted first to ensure that the number of hops in the cross-domain request path is stable and does not exceed the preset number of hops. Topology updates are triggered by the following events: The DMN control module outputs data access hot zone changes in prediction mode; The number of consecutive failures of long-distance connections of the cross-domain scheduler is greater than or equal to the preset number; The high cluster connection load rate of the local router continues to be greater than the preset ratio within the preset time.

8. A brain-inspired data service system method, applied to a brain-inspired data service system according to any one of claims 1 to 9, characterized in that: The following steps are involved: Hierarchical initialization steps: Deploy containerized data nodes at the infrastructure layer, generate the initial topology through the small-world network manager at the small-world connection layer, and initiate full-link monitoring with the status monitoring engine at the intelligent control layer. Cross-layer collaborative processing steps: When the external access layer receives a cross-department query request, the DMN control center of the intelligent control layer calls the pre-calculated results of the prediction analysis module, directly connects to the sales and financial gateway node through the long-distance cross-domain connection of the small-world connection layer, and the edge node synchronously provides real-time device data; Hierarchical adaptive optimization steps: The intelligent control layer triggers the small-world connection layer to adjust the long-range connection distribution based on status monitoring data, and at the same time guides the infrastructure layer to reallocate container resources.

9. The method according to claim 8, characterized in that The hierarchical adaptive optimization includes: Small-world connection layer: When the communication between production data nodes and R&D data nodes exceeds the system average for a preset number of days, the small-world network manager creates a new long-range connection between the two and shortens the number of hops in the communication path. Intelligent control layer: The predictive analysis module pre-allocates a preset proportion of database connection resources for monthly financial settlement tasks based on historical data; Infrastructure layer: When the resource scheduler detects that the CPU utilization of the edge node exceeds the preset percentage, it automatically starts several new instances in the containerized platform for load balancing.

10. The method according to claim 9, characterized in that The intelligent control layer also includes a learning and adaptation engine that analyzes historical access patterns and updates access frequency models; evaluates task execution efficiency and optimizes resource scheduling strategies; adjusts LSTM / Transformer model parameters based on real-time feedback; and synchronously updates the knowledge base to reflect the latest data associations. The learning and adaptation engine updates the access pattern model based on task execution logs and optimizes the long-range connection distribution of the small-world connection layer; When the knowledge association relationship in the knowledge base changes, the index reconstruction and metadata update of the data node are triggered, and the pre-calculation query strategy is adjusted synchronously.

Images