6G communication data processing method and system based on LSTM neural network model

Abstract

The invention relates to a 6G communication data processing method and system based on an LSTM neural network model. The method comprises the following steps: splitting a data packet under the limitation of a granularity range based on the popularity and the popularity stability of the data packet, forming the data packet from a part with stable popularity, putting the data packet into a stable data queue, and putting the other part of data packet into an unstable data queue; determining the popularity of each data packet in the stable data queue and the unstable data queue based on an associated popularity determination model; and if the popularity of the data packet belongs to the popularity range of the intelligent access node, locally storing the data packet. According to the method, the data storage position is associated with the service heat range determined based on the big data instead of cold and hot association of the data, so that the data is comprehensively distributed in the whole system in combination with the big data use information, and the comprehensive service efficiency and capability of 6G communication are improved.

Description

[Technical Field] This invention belongs to the field of intelligent communication technology, and in particular relates to a method and system for 6G communication data processing based on an LSTM neural network model. [Background Technology] Currently, 5G mobile communication technology has entered the stage of large-scale commercial use, achieving significant breakthroughs in peak data rates, connection density, and air interface latency. To address the increasing complexity of networks, existing technologies have begun to introduce software-defined networking and network function virtualization as underlying support, and to achieve a certain degree of resource coordination through centralized network management controllers. In terms of intelligence, academia and industry are actively exploring the application of machine learning, especially deep learning models, to network traffic prediction, fault detection, and some self-optimization scenarios. Specifically, existing communication architectures typically employ relatively rigid designs, with their computing, storage, and communication resources tightly coupled physically and logically, resulting in insufficient flexibility in resource allocation. At the resource management level, most systems still rely on reactive strategies based on preset thresholds or static optimization algorithms based on historical averages, lacking the ability to proactively perceive and adapt to dynamic changes in the network environment. Furthermore, although the concept of edge computing has been proposed and implemented, collaboration between the cloud, edge, and terminal is mostly limited to simple task offloading and data transmission, failing to achieve deep integration and global optimization of computing power, data, intelligence, and management. The application of artificial intelligence (AI) technology in networks often exists as isolated "plug-in" components, operating by collecting data, training models centrally, and redeploying them. This results in long model update cycles and a lack of rapid self-adjustment based on local environments, failing to develop endogenous, sustainably evolving intelligence inherent in the network architecture. Existing AI-based network optimization solutions mostly employ an "offline training, online inference" model, failing to achieve differentiated model customization and adaptive adjustments. Their intelligent decision-making relies on centralized, large-scale data training, making it unable to quickly respond to sudden changes in the local network environment, leading to rigid system intelligence and an inability to co-evolve with the environment. When facing complex tasks across domains and layers, it struggles to achieve globally optimal intelligent resource coordination and allocation, resulting in low overall resource utilization efficiency. Systems often lack a unified intelligent kernel capable of sensing the environment, predicting trends, and flexibly configuring wireless interface parameters and distributed computing resources accordingly, limiting the network's adaptability to unknown business scenarios and extreme performance requirements.

[0001] Driven by the dual waves of artificial intelligence and computing power, the communications field is on the eve of a paradigm shift. Current cutting-edge research is no longer satisfied with simple rule engines, but is widely exploring the introduction of advanced AI technologies such as large-scale deep learning models, reinforcement learning, and federated learning into network management and optimization. Cloud computing centers provide near-unlimited centralized computing power for training massive global network models; the widespread adoption of edge computing nodes provides massive terminal computing power for distributed intelligence. At the hardware level, GPUs, NPUs, and dedicated AI accelerator cards are integrated from the cloud to base stations and even terminal devices, providing unprecedented hardware support for complex AI inference tasks. From a technological perspective, building a fully perceptive, real-time decision-making, and self-evolving "intrinsic intelligence" network has become an industry consensus and is considered a core characteristic of 6G. This invention, based on LSTM prediction results and considering the local service characteristics of local nodes, dynamically unloads and distributes data content across the cloud, edge, and endpoint based on user usage information. It migrates some real-time tasks to adjacent lightly loaded edge nodes, ultimately ensuring that each physical node stores the most relevant and frequently accessed data within its service scope. In other words, within the same layer, data is stored locally in the most appropriate location. Across different layers, data can be stored locally with a wider range of accessibility, resulting in the largest range of accessibility stored at the central node, a moderate range at the edge nodes, and the smallest range at the intelligent access nodes. This approach differs from existing technologies that uniformly differentiate between hot and cold data, caching hot data at the endpoint, storing warm data at the edge nodes, and archiving cold data in the cloud center. Instead, this method links the data storage location to its actual service accessibility, rather than its inherent hot / cold correlation. Combined with the granularity of each node's hardware and software resources, this allows data to be comprehensively distributed across the entire system based on its big data usage information, improving the overall service efficiency and capabilities of 6G communication. [Summary of the Invention] To address the aforementioned problems in the prior art, this invention proposes a 6G communication data processing method and system based on an LSTM neural network model, the system comprising: Intelligent access nodes, edge nodes, and cloud center nodes; Each intelligent access node and edge node is associated with a granularity range and a popularity range. The intelligent access node is used to collect data to obtain data packets, which are then buffered, organized, and sent to the edge node. Based on the popularity and stability of the data packets, the data packets are split under the granularity range constraint. The data packets with stable popularity are placed into a stable data queue, and the other data packets are placed into an unstable data queue. The popularity of each data packet in the stable and unstable data queues is determined based on the associated popularity determination model. If the popularity of a data packet falls within the popularity range of the intelligent access node, it is stored locally and used as part of the air interface dataset. The edge node receives data packets from the intelligent access node, buffers and stores the data packets, and then sends them to the cloud center node; it determines the popularity of the data packets based on the association popularity determination model, and stores the data packets locally based on the popularity of the data packets to form edge data; The edge node is used to determine representative data blocks from the data packets before pushing the data packets, and push the representative data blocks to the smart access node to obtain the popularity of the representative data blocks to determine whether to push the data packets to the smart access node. The heat determination model, granularity range, and heat range are different for each smart access node or edge node.

[0002] Furthermore, there are one or more intelligent access nodes and one or more edge nodes; the intelligent access nodes and edge nodes send, receive, and push data based on their relationship with each other.

[0003] Furthermore, the air interface dataset refers to the collection and processing of raw data obtained from the interface between the terminal and the base station or access point via radio wave transmission in 6G wireless communication.

[0004] Furthermore, the heat determination module uses an LSTM inference engine.

[0005] Furthermore, the heat determination model for the association between intelligent access nodes and edge nodes is deployed in their local storage space.

[0006] Furthermore, the process involves: constructing a general popularity determination model based on an LSTM inference engine; training and validating the general popularity determination model using global historical usage data; and sending the trained and validated general popularity determination model to each edge node and smart access node. Based on the general popularity determination model, the edge nodes and smart access nodes fine-tune the popularity determination model using local historical usage data, thereby obtaining a popularity determination model associated with the edge node or smart access node itself.

[0007] Furthermore, the packet splitting based on packet popularity and its stability specifically includes the following steps: Step S11: Use the heat determination model associated with the current node to determine the heat of data in the data packet; the current node is the local edge node or smart access node; Step S12: Obtain the granularity range corresponding to the current node; the granularity range is the size range of the splittable data packets of the current node, involving the data packet size between the lower threshold and the upper threshold; Step S13: Place the data packet into the data queue to be split; initialize the stable data queue and the unstable data queue to be empty; Step S14: Obtain the data packet to be split from the head of the data packet to be split queue as the current data packet; if the current data packet is splittable, pre-split the data packet to obtain the first sub-data packet and the second sub-data packet; otherwise, put the current data packet into the unstable data queue. Step S15: Determine the heat of the first sub-data packet and the second sub-data packet respectively; Step S16: Determine whether the current data is stable in terms of popularity based on the current data packet's popularity, the first sub-data packet's popularity, and the second sub-data packet's popularity; if so, push the current data packet to the stable data queue; otherwise, place the first sub-data packet and the second sub-data packet into the tail of the data queue to be split, one after the other. The determination of whether the data is stable based on the popularity of the data packet, the popularity of the first sub-data packet and the second sub-data packet is specifically as follows: determine whether the popularity of the three is consistent. If they are, the data is determined to be stable; otherwise, it is determined to be unstable. Step S17: Determine if the queue to be split is empty. If it is, use the stable data queue and the unstable data queue as the splitting result; otherwise, return to step S14.

[0008] A server for 6G communication data processing based on an LSTM neural network model includes a processor coupled to a memory. The memory stores program instructions, and the 6G communication data processing system based on the LSTM neural network model is implemented when the program instructions stored in the memory are executed by the processor.

[0009] A 6G communication data processing platform based on an LSTM neural network model, wherein the system is used to implement the 6G communication data processing system based on the LSTM neural network model.

[0010] A cloud service system for 6G communication data processing based on an LSTM neural network model includes a processor coupled to a memory. The memory stores program instructions, and when the program instructions stored in the memory are executed by the processor, the 6G communication data processing system based on the LSTM neural network model is implemented.

[0011] The beneficial effects of this invention include: (1) Dynamically unload and distribute data content between the cloud, edge, and terminal. This ensures that each physical node stores the most relevant and popular data within its service scope. In other words, within the same level, data is stored locally in the most appropriate location; while between different levels, data can be stored locally with a wider range of popularity. Combined with the granularity of each node's own hardware and software resources, this allows data to be integrated into the overall distribution of big data usage information in the system, improving the overall service efficiency and capability of 6G communication.

[0012] (2) Based on the service characteristics of the node itself, LSTM neural network model training is carried out to guide the storage of data packets locally and the trial reception of pushed data. Through the consistency analysis of heat and heat stability, the data packets are iteratively divided and the division is weighed to obtain the final division balance. Thus, without needing to delve into the data content, the data is distributed in a comprehensive and balanced manner within the system. [Attached Image Description] The accompanying drawings, which are provided to further illustrate the invention and form part of this application, are not intended to unduly limit the invention. In the drawings: Figure 1 This is a schematic diagram of the 6G communication data processing method based on the LSTM neural network model provided by the present invention.

Detailed Implementation Methods

[0013] As attached Figure 1 As shown, data is collected to obtain data packets, which are then buffered and organized before being sent to edge nodes. Based on the popularity and stability of the data packets, the packets are split within a granular range. The packets with stable popularity are placed into a stable data queue, while the other packets are placed into an unstable data queue. The popularity of each data packet in the stable and unstable data queues is determined based on an association popularity determination model. If the popularity of a data packet falls within the popularity range of the intelligent access node, it is stored locally and used as part of the air interface dataset. The system includes: intelligent access nodes, edge nodes, and cloud center nodes; wherein: there are one or more intelligent access nodes and one or more edge nodes; each intelligent access node and edge node is associated with a granularity range and a heat range; wherein: the granularity range is related to the access capability and storage resource size of the intelligent access node; the heat range is used to limit the heat fluctuation range of the air interface dataset carried in the intelligent access node; when the heat range is large, the heat fluctuation range it carries is large, the amount of interaction with the edge node is large, and the required local carrying capacity is large; therefore, the heat range is an important balancing means for edge nodes and intelligent access nodes; for example, when an edge node is about to be overloaded, some real-time data can be offloaded to the cloud or migrated to adjacent lightly loaded edge nodes in advance, realizing the redistribution of air interface datasets in the system, so as to achieve efficient utilization of all data and resource consumption; of course, each node prioritizes storing the data with the highest heat value for it; however, it is obvious that for the same data packet, its heat value is different for different nodes, because the user groups and service areas it serves are different and dynamically changing; The intelligent access node is used to collect data to obtain data packets, buffer and organize the real-time data packets and send them to the edge nodes; it splits the data packets based on the popularity and stability of the real-time data packets; and it stores the splitting results locally to form an air interface dataset. Preferably, the data packet is a real-time data packet; The air interface dataset refers to the collection and processing of raw data obtained from the interface between terminals such as mobile phones and sensors and base stations or access points via radio wave transmission in 6G wireless communication. It contains a multi-dimensional and massive amount of data. Therefore, how to efficiently store and utilize it is crucial to the capabilities of hardware and software devices in the 6G system. The real-time data acquisition process specifically involves: continuously collecting wireless channel information, reference signal received power, service data stream, user identifier, real-time task data, etc., through dedicated data acquisition probes deployed on intelligent access nodes, forming real-time data packets; the real-time data packets correspond to the data acquisition method; due to differences in data sources, organization logic, time, format, and acquisition methods, the size of real-time data packets varies considerably; Preferred method: Deploy dedicated data acquisition probes at intelligent access nodes to continuously and massively collect wireless channel information, reference signal received power, service data streams, user identifiers, etc., thereby supporting the formation of a multi-dimensional, massive air interface dataset; Preferably, the intelligent access node is a user terminal or a base station; The process of splitting data packets based on their real-time popularity and stability includes the following steps: Step S11: Use the heat determination model associated with the current node to determine the heat of real-time data in the data packet; For example: when the current node is smart access node A: the heat determination module associated with the smart access node is used to determine the heat of the real-time data; when the current node is smart access node B: the heat determination module associated with the smart access node is used to determine the heat of the real-time data; when the current node is edge node C: the heat determination module associated with the smart access node is used to determine the heat of the real-time data. Preferably: The current node's associated heat determination model is deployed in the current node's local storage space; input features are extracted from data packets to obtain predicted data packet heat; historical usage data from the local air interface dataset is used to train the associated heat determination model to obtain the real-time heat of data packets relative to the local node; that is, different data packets may yield the same or different heat determination results after being predicted by different heat determination models; this is related to the service status and big data trends of the current node; the historical usage data includes data content, access frequency, source, organization logic, time, existence form and collection method, resource utilization, data access pattern, dependency relationship, communication overhead, distribution of users or requests, temporal locality, historical migration effect and priority, etc. Preferred approach: Data packets are composed of different data blocks due to differences in source, organization logic, collection events, existence format, collection method, etc.; when extracting input features from data packets, if the data packet size is large, representative data blocks in the data packet can be selected for input feature extraction; the representative data block can be the largest data block in the data packet; of course, other methods of determining representative data blocks can also be used; of course, the popularity of each data block can be determined sequentially, and then the weighted sum of the popularity of each data block can be used as the popularity value of the data packet; When the target data packet is program data, computational load migration can be performed. The historical usage data at this time includes: usage frequency, execution time, resource utilization, data access patterns, dependencies, communication overhead, distribution of users or requests, temporal locality, historical migration effects and priorities, etc. The extracted feature data includes source, execution time, resource utilization, data access patterns, dependencies, communication overhead, distribution of users or requests, temporal locality, historical migration effects and priorities, existence form and collection method, etc. Preferably, the heat determination module uses an LSTM inference engine; the loss function of the LSTM inference engine is the mean squared error commonly used in regression, and cross-entropy is used for classification, employing the Adam optimizer; further, the input layer of the LSTM inference engine contains two inputs, the first input being temporal features and the second input being static features; for the first input, an LSTM layer is set to extract temporal features, and for the second input, features are extracted through several fully connected layers; then, an attention layer or multiple LSTM layers are stacked to capture information from key time steps; then, global average pooling is used to vectorize the temporal features; finally, the outputs of the temporal branch and the static branch are concatenated together, and then fused through a fully connected layer, outputting the heat value through the output layer; Step S12: Obtain the granularity range corresponding to the current node; the granularity range is the size range of the splittable data packets of the current node, involving the data packet size between the lower threshold and the upper threshold; when the data packet size is less than the lower threshold, splitting will no longer be considered; after all, the smaller the data packet is split, the worse its logic is, and it may also lead to more access loss due to the worse logic. Step S13: Place the real-time data packet into the data queue to be split; that is, in the initial state, the data queue to be split contains the unsplit real-time data packet; initialize the stable data queue and the unstable data queue to be empty; Preferred approach: Before performing this step, first determine whether the real-time data packet is less than or equal to the upper limit threshold of the granularity range. If it is, use the real-time data packet as the splitting result and end the process; in this case, do not perform splitting; otherwise, proceed to the next step; Further approach: Push the real-time data packet to the stable data queue and use it as the splitting result. Step S14: Obtain the data packet to be split from the head of the data packet to be split as the current data packet; if the current data packet is splittable, pre-split the real-time data packet to obtain the first sub-data packet and the second sub-data packet; otherwise, put the current data packet into the unstable data queue; the unstable data queue is actually a data packet queue with uncertain heat stability. Preferred: The principle of splitting is that the current data packet size is greater than the upper limit threshold of the granularity range, and the sizes of the first sub-data packet and the second sub-data packet obtained after splitting are both greater than the lower limit threshold of the first granularity range; Preferred method: Segmentation based on the source, organization logic, time, format, and collection method of the current data packet; Step S15: Determine the heat of the first sub-data packet and the second sub-data packet respectively; Step S16: Determine whether the current data is stable in terms of popularity based on the current data packet's popularity, the first sub-data packet's popularity, and the second sub-data packet's popularity; if so, push the current data packet to the stable data queue; otherwise, place the first sub-data packet and the second sub-data packet into the tail of the data queue to be split, one after the other. The determination of whether the data is stable based on the popularity of the data packet, the popularity of the first sub-data packet and the second sub-data packet is specifically as follows: determine whether the popularity of the three is consistent. If they are, the data is determined to be stable; otherwise, it is determined to be unstable. Further: The determination of whether the data is hot based on the heat of the data packet, the heat of the first sub-data packet and the heat of the second sub-data packet is specifically as follows: the heat variance is calculated using the following formula (1). If the heat variance is less than the heat variance threshold, the data is determined to be hot and stable; otherwise, it is determined to be unstable. Wherein: r1, r2, and r3 are the heat of the data packet, the heat of the first sub-data packet and the heat of the second sub-data packet, respectively. (1); (2); Preferably, the heat variance threshold is a preset value, for example, 0.05~0.1; Preferred: The identifiers of real-time data packets or sub-data packets are placed in the queue to be split and the stable data queue; Step S17: Determine if the queue to be split is empty. If it is, use the stable data queue and the unstable data queue as the splitting result; otherwise, return to step S14. The process of storing local data based on the splitting results to form an air interface dataset specifically involves: obtaining the popularity of each data packet in the stable data queue and the unstable data queue; if the popularity of a data packet falls within the popularity range of the intelligent access node, then the data packet is stored locally and used as part of the air interface dataset; at this time, regardless of whether the data packet is in the stable data queue or the unstable data queue, as long as its popularity is clearly excluded from belonging to the current intelligent access node, it is stored locally. Alternatively: The method of storing local data based on the splitting results to form an air interface dataset specifically involves: determining the proportion of the data packet size in the unstable data queue to the real-time data packet size; when the proportion is less than the proportion threshold, storing the real-time data packet locally; when the real-time data packet can basically achieve heat stability within the granularity range after several splits, then completing the storage of the real-time data packet, thereby preserving the most complete logic of the real-time data packet. Alternative: The method of storing local data based on the splitting results to form an air interface dataset specifically involves: obtaining the heat value of each data packet in the stable data queue; if the heat value of a data packet falls within the heat value range of the intelligent access node, then the data packet is stored locally and used as part of the air interface dataset; this alternative method can be used when the storage space of the intelligent access node is limited; of course, other alternative methods are also possible, for example, for cases where the ratio is greater than or equal to the ratio threshold, all data packets in the unstable queue are saved, while each data packet in the stable queue is distinguished by heat value, and only data packets whose heat value range is within the heat value range associated with the current node are saved; The edge nodes are used for regional collaboration and optimization, aggregating data from multiple intelligent access nodes. Each edge node deploys an associated heat determination model to predict the heat of data packets sent by intelligent access nodes within its aggregated region, thereby redistributing data between edge nodes and cloud center nodes. Specifically, the edge nodes receive data packets from intelligent access nodes, buffer and store the data packets, and then send them to the cloud center nodes. They determine the heat of the data packets based on the heat determination model and store them locally based on the heat of the data packets to form edge data. Through the edge-to-edge collaboration interface, they communicate directly with other edge nodes to achieve wider-area load balancing and resource sharing.

[0014] When a smart access node buffers and organizes real-time data packets before sending them to an edge node, it does so by storing the packets in a local buffer pool. If the buffering time reaches its limit or the buffer pool is full, the packet is then sent to an edge node that is associated with the smart access node. Each smart access node can be associated with one or more edge nodes. Sending data packets to the cloud center node after buffering is similar; if the buffering time reaches its limit or the buffer pool is full, the packet is sent to the cloud center node. Therefore, the size of the data packets received by the edge node and the cloud center node is relatively fixed. However, this size differs and is related to the size of the buffer pool set locally by the node. Due to the limited processing capacity of the smart access node, its local buffer pool size is usually smaller than that of the edge node. Preferred method: The data packets sent by the intelligent access node to the edge node are raw data packets without heat judgment; the intelligent access node directly buffers, organizes and sends the data packets without judgment, and at the same time distinguishes the heat of data to determine whether to store them locally. Alternatively: The edge node sets up a receive buffer pool to buffer data packets from the intelligent access node. When the buffering conditions are met, it determines the heat of all data packets in the buffer pool and stores them locally based on the heat of all data packets to form edge data. Further: When the heat of all data packets does not fall within the heat range of the edge node, it determines the heat of each data packet in turn and stores it locally based on the heat of each data packet to form edge data. It can be seen that the edge node's heat filtering of data packets is a coarse-grained filtering. The edge nodes are used to determine representative data blocks from data packets before pushing them, and push these representative data blocks (or representative data block features related to the input vector of the popularity determination model) to the intelligent access nodes to obtain the popularity of the representative data blocks. Based on the popularity of the representative data blocks, they determine whether to push the data packet to the intelligent access nodes. After the intelligent access nodes determine the popularity of the representative data block relative to itself, they send it to the edge nodes. In this way, the edge nodes can obtain the relative hotspot of the representative data and make subsequent decisions on whether to push it. Obviously, when the relative popularity is within the popularity range of the edge access nodes to which the data packet is to be pushed, the data packet is pushed; otherwise, the data packet is not pushed. Alternative: Edge access nodes can use the same processing method as the real-time data packets obtained from local probing to determine whether to save the data packets pushed by the edge access node locally; that is, the data packets to be pushed by the edge access node are regarded as real-time data packets, and the data packets are split based on the popularity and stability of the real-time data packets, so as to selectively store them locally based on the splitting results. The cloud center node receives data packets from edge nodes and stores them in its local cloud storage space; that is, the cloud center does not determine the popularity of the packets. The cloud center pushes data packets to edge nodes to complete data exchange between them. Generally, the cloud center pushes data based on the attributes of the edge nodes, but it can also use a similar approach, attempting to push key blocks before pushing data packets to edge nodes. The cloud center node can also push data packets to edge nodes in a similar way. Edge nodes select data packets that match their popularity range for reception, or partially receive them using a first-level splitting method. Preferably, the edge node is also used to periodically use the associated local heat determination module to determine the heat of locally stored data packets, and when the edge node is in an overloaded or critically overloaded state, based on the relationship between the heat and the heat range of the edge node, select data packets that are far from the heat range and migrate them to adjacent edge nodes or cloud nodes. Similarly, the smart access node is also used to periodically determine the popularity of locally stored data packets using the associated local popularity determination module, and when the smart access node is overloaded or critically overloaded, it selects data packets that are far from the popularity range based on the relationship between the popularity and the popularity range of the smart access node, and migrates them to adjacent smart access nodes or edge nodes. The system dynamically offloads and distributes data content across the cloud, edge, and endpoint based on LSTM predictions. For example, if LSTM predicts that an edge node is about to become overloaded, it will offload some non-real-time computing tasks to the cloud or migrate some real-time tasks to adjacent lightly loaded edge nodes. Ultimately, this ensures that each physical node stores the most relevant and popular data within its service range. In other words, within the same layer, data is stored locally in the most appropriate location. Between different layers, data can be stored locally with a wider range of popularity, resulting in the largest range of popularity for central nodes, a moderate range for edge nodes, and the smallest range for intelligent access nodes. This approach differs from existing technologies that cache hot data on endpoints, store warm data on edge nodes, and archive cold data in the cloud center. It links the location of data storage to its actual service popularity range, rather than its inherent hot / cold correlation. Preferably, the system further includes an intelligent management module, which is used to realize endogenous intelligence; it is responsible for creating and generating a general popularity determination model, which uses global historical usage data to train the general popularity determination model, thereby using global data, using accumulated knowledge for reasoning and decision-making, and distributing the trained general popularity determination model to each edge node and intelligent access node; the edge nodes and intelligent access nodes fine-tune the popularity determination model based on the general popularity determination model using local historical usage data; Based on the same inventive concept, the present invention also provides a 6G communication data processing system based on an LSTM neural network model, the system being used to implement the above-mentioned 6G communication data processing method based on an LSTM neural network model; Based on the same inventive concept, the present invention also provides a 6G communication data processing server based on an LSTM neural network model, wherein the server is equipped with the aforementioned monitoring server for implementing the aforementioned 6G communication data processing method based on an LSTM neural network model; the monitoring server is a cloud server. Based on the same inventive concept, the present invention also provides a 6G communication data processing device based on an LSTM neural network model, the device being used to implement the above-mentioned 6G communication data processing method based on an LSTM neural network model. Based on the same inventive concept, the present invention also provides a 6G communication data processing platform based on an LSTM neural network model, the platform being used to implement the above-mentioned 6G communication data processing method based on an LSTM neural network model; A computer program (also referred to as a program, software, software application, script, or code) can be written in any form of programming language, including assembly or interpreted languages, declarative or procedural languages, and can be deployed in any form, including as a standalone program or as a module, component, subroutine, object, or other unit suitable for use in a computing environment. A computer program may, but does not necessarily, correspond to a file in a file system. A program can be stored as part of a file that holds other programs or data (e.g., one or more scripts stored in a markup language document), in a single file dedicated to said program, or in multiple co-located files (e.g., a file storing one or more modules, subroutines, or code portions). A computer program can be deployed to execute on a single computer or on multiple computers located at a single site or distributed across multiple sites and interconnected by a communications network.

[0015] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0016] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0017] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The function specified in one or more boxes.

[0018] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0019] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A 6G communication data processing system based on an LSTM neural network model, characterized in that, The system includes: intelligent access nodes, edge nodes, and cloud center nodes; Each intelligent access node and edge node is associated with a granularity range and a popularity range. The intelligent access node is used to collect data to obtain data packets, which are then buffered, organized, and sent to the edge node. Based on the popularity and stability of the data packets, the data packets are split under the granularity range constraint. The data packets with stable popularity are placed into a stable data queue, and the other data packets are placed into an unstable data queue. The popularity of each data packet in the stable and unstable data queues is determined based on the associated popularity determination model. If the popularity of a data packet falls within the popularity range of the intelligent access node, it is stored locally and used as part of the air interface dataset. The edge node receives data packets from the intelligent access node, buffers and stores the data packets, and then sends them to the cloud center node. The popularity of data packets is determined based on the association popularity determination model, and the popularity of data packets is used for local storage to form edge data; The edge node is used to determine representative data blocks from the data packets before pushing the data packets, and push the representative data blocks to the smart access node to obtain the popularity of the representative data blocks to determine whether to push the data packets to the smart access node. The heat determination model, granularity range, and heat range are different for each smart access node or edge node.

2. The 6G communication data processing method based on the LSTM neural network model according to claim 1, characterized in that, The intelligent access node can be one or more, and the edge node can be one or more; the intelligent access node and the edge node send, receive and push data based on their relationship with each other.

3. The 6G communication data processing method based on the LSTM neural network model according to claim 2, characterized in that, The air interface dataset refers to the collection and processing of raw data obtained from the interface between the terminal and the base station or access point via radio wave transmission in 6G wireless communication.

4. The 6G communication data processing method based on the LSTM neural network model according to claim 2, characterized in that, The heat determination module uses an LSTM inference engine.

5. The 6G communication data processing method based on the LSTM neural network model according to claim 4, characterized in that, The heat determination model for the association between intelligent access nodes and edge nodes is deployed in their local storage space.

6. The 6G communication data processing method based on the LSTM neural network model according to claim 5, characterized in that, in: A general popularity determination model is built based on the LSTM inference engine. The model is trained and validated using global historical usage data. The trained and validated model is then sent to each edge node and smart access node. The edge nodes and smart access nodes fine-tune the popularity determination model based on the model using local historical usage data, thereby obtaining a popularity determination model associated with the edge node or smart access node itself.

7. The 6G communication data processing method based on the LSTM neural network model according to claim 6, characterized in that, The packet splitting based on packet popularity and its stability specifically includes the following steps: Step S11: Use the heat determination model associated with the current node to determine the heat of data in the data packet; the current node is the local edge node or smart access node; Step S12: Obtain the granularity range corresponding to the current node; the granularity range is the size range of the splittable data packets of the current node, involving the data packet size between the lower threshold and the upper threshold; Step S13: Place the data packet into the data queue to be split; initialize the stable data queue and the unstable data queue to be empty; Step S14: Obtain the data packet to be split from the head of the data packet to be split queue as the current data packet; if the current data packet is splittable, pre-split the data packet to obtain the first sub-data packet and the second sub-data packet; otherwise, put the current data packet into the unstable data queue. Step S15: Determine the heat of the first sub-data packet and the second sub-data packet respectively; Step S16: Determine whether the current data is stable in terms of popularity based on the current data packet's popularity, the first sub-data packet's popularity, and the second sub-data packet's popularity; if so, push the current data packet to the stable data queue; otherwise, place the first sub-data packet and the second sub-data packet into the tail of the data queue to be split, one after the other. The determination of whether the data is stable based on the popularity of the data packet, the popularity of the first sub-data packet and the second sub-data packet is specifically as follows: determine whether the popularity of the three is consistent. If they are, the data is determined to be stable; otherwise, it is determined to be unstable. Step S17: Determine if the queue to be split is empty. If it is, use the stable data queue and the unstable data queue as the splitting result; otherwise, return to step S14.

8. A server for 6G communication data processing based on an LSTM neural network model, characterized in that, The system includes a processor coupled to a memory, the memory storing program instructions, which, when executed by the processor, implement the 6G communication data processing system based on the LSTM neural network model as described in any one of claims 1-7.

9. A 6G communication data processing platform based on an LSTM neural network model, characterized in that, The system is used to implement the 6G communication data processing system based on the LSTM neural network model as described in any one of claims 1-7.

10. A cloud service system for 6G communication data processing based on an LSTM neural network model, characterized in that, The system includes a processor coupled to a memory, the memory storing program instructions, which, when executed by the processor, implement the 6G communication data processing system based on the LSTM neural network model as described in any one of claims 1-7.

Images