Network data transmission management method and system based on artificial intelligence

Abstract

The invention provides a network data transmission management method and system based on artificial intelligence, and the method comprises the steps: collecting network node response data through transmitting a stress probe packet carrying a dyeing mark, so as to train an artificial intelligence model which can map a data flow feature into a node transmission stress memory state; before the actual service is transmitted, the model is utilized to predict a path stress map to be triggered by the service data flow on a predetermined path. And then, the system generates a neutralization sequence packet for counteracting the stress state according to the atlas, sends the neutralization sequence packet in advance according to a multi-hop delivery strategy, actively adjusts internal processing states of all nodes on a path, and pre-optimizes a network environment for an upcoming service data flow. According to the invention, passive response is changed into active intervention, so that prospective optimization management of network transmission performance is realized.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and specifically relates to a network data transmission management method and system based on artificial intelligence. Background Art

[0002] With the rapid development of applications such as cloud computing, big data, high-definition video streaming, and the Internet of Things, modern networks are carrying demanding traffic with high burst rates, large bandwidth, and low latency. To ensure the quality of service (QoE) for massive user base, network management technologies are constantly evolving, resulting in the emergence of a range of management mechanisms, including Quality of Service (QoS), traffic engineering, and congestion control. These technologies significantly alleviate network congestion and improve resource utilization by prioritizing packets, planning routes, and dynamically adjusting transmission rates based on network load.

[0003] However, these mainstream network management approaches inherently follow a reactive logic. Whether it's QoS queue management or TCP congestion control window adjustment, their triggering conditions are already signals of performance degradation within the network, such as buffer overflows, increased latency, or packet loss. This means that management actions always lag behind the actual occurrence of problems, acting as a post-empty remediation measure for network deterioration. When a high-burst data flow enters the network, it first impacts the internal processing resources (such as caches and CPUs) of network nodes along the way. This cumulative impact leads to performance degradation, which in turn triggers the adjustment mechanism. This delayed adjustment inevitably introduces initial performance jitter and latency. For emerging applications that are extremely sensitive to latency, even millisecond-level jitter can significantly degrade the user experience. Summary of the Invention

[0004] The present invention provides an artificial intelligence-based network data transmission management method and system to solve the above technical problems.

[0005] In a first aspect, the present invention provides a method for managing network data transmission based on artificial intelligence, the method comprising the following steps:

[0006] By sending stress probe packets carrying dye markers in the target network, the transmission performance response data of network nodes is actively induced and collected. Based on the response data, an artificial intelligence model is trained to map data flow characteristics into the transmission stress memory state of network nodes.

[0007] Before transmitting the service data flow on the target network, based on the data flow characteristics of the service data flow and the predetermined transmission path, an artificial intelligence model is used to predict the transmission stress memory state that will be generated by each network node on the predetermined transmission path, forming a path stress map;

[0008] Generate a neutralization sequence packet for offsetting the transmission stress memory state according to the path stress map, and determine a multi-hop delivery strategy for the neutralization sequence packet to synchronously offset the transmission stress memory state of multiple network nodes on a predetermined transmission path;

[0009] According to the multi-hop delivery strategy and along the predetermined transmission path, the neutralization sequence packet is sent in the target network to actively adjust the internal processing state of all network nodes on the predetermined transmission path.

[0010] Optionally, the method of actively inducing and collecting transmission performance response data of network nodes by sending a stress probe packet carrying a dye mark in the target network, and training an artificial intelligence model based on the response data to map data flow characteristics to the transmission stress memory state of the network node includes the following steps:

[0011] generating a stress probe package, the stress probe package including a stimulation probe for actively inducing a transferred stress memory state and a measurement probe for measuring the impact of the transferred stress memory state, wherein the stimulation probe carries a dye marker;

[0012] Send stress probe packets to the target network and synchronously collect node internal data generated by network nodes on the path when processing the stimulus probes through the out-of-band management channel;

[0013] Recording transmission performance response data of network nodes measured by measurement probes;

[0014] The dye markers carried by the stimulation probes are used to temporally correlate and causally bind the node internal data with the transmission performance response data to form training data pairs.

[0015] The training data pairs are input into the preset artificial intelligence model, and the training of the artificial intelligence model is completed through supervised learning. The artificial intelligence model is a hybrid architecture of graph neural network and long short-term memory network.

[0016] Optionally, the coloring mark carried by the stimulation probe is a unique identification code written in an extended header of a data packet header of the stimulation probe. Using the coloring mark carried by the stimulation probe to temporally correlate and causally bind node internal data with transmission performance response data to form a training data pair includes the following steps:

[0017] Configure deep packet inspection capabilities for all network nodes in the target network to identify unique identification codes;

[0018] When a network node detects a data packet containing a unique identification code, it triggers the recording of timestamps and the collection of node internal data;

[0019] Use the recorded target timestamp as the time base for the node's internal data;

[0020] At the same time, the transmission performance response data of the network nodes measured by the measurement probes are aligned with the time reference;

[0021] Based on the alignment results, the node internal data corresponding to the target timestamp is matched with the transmission performance response data to establish a causal binding between the data and form a training data pair.

[0022] Optionally, generating a neutralization sequence package for offsetting the transmission stress memory state according to the path stress map includes the following steps:

[0023] Analyze the path stress map to obtain the target transmission stress memory state corresponding to each target network node on the predetermined transmission path;

[0024] For each target transmission stress memory state, one or more reverse excitation patterns required to offset the stress state are calculated using a reverse resonance generation algorithm;

[0025] Mapping the reverse stimulus pattern to the data packet characteristics required for the neutralization sequence packet, the data packet characteristics including the data packet size, the service type field value, and the inter-packet sending interval;

[0026] If the transmission stress memory state is a composite state, a sequence consisting of multiple data sub-packets with different data packet characteristics is generated;

[0027] The data packet or data sub-packet containing the data packet characteristics is encoded and encapsulated to generate a neutralization sequence packet.

[0028] Optionally, mapping the reverse excitation pattern to data packet features required for the neutralization sequence packet includes the following steps:

[0029] Separate stress response characteristics with specific behavioral patterns from the transferred stress memory state;

[0030] Calculate theoretical waveforms that are opposite in phase to the stress response characteristics, or sequences of events that are antagonistic in behavioral patterns;

[0031] Discretize the theoretical waveform or event sequence into data packet sending events;

[0032] Determine the size of the corresponding data packet and the value of the Differentiated Services Code Point field in the IP header for each send event;

[0033] The sending time interval is determined between adjacent sending events, and all sending events are combined into a data packet feature for implementing the reverse excitation mode according to the time interval.

[0034] Optionally, the multi-hop delivery strategy for determining the neutralization sequence packet to synchronously offset the transmission stress memory states of multiple network nodes on a predetermined transmission path comprises the following steps:

[0035] For any target network node on the predetermined transmission path, the node that receives the data packet before the target network node and is immediately adjacent to the target network node is marked as the upstream node of the target network node, and the node that receives the data packet after the target network node and is immediately adjacent to the target network node is marked as the downstream node of the target network node;

[0036] Calculate the propagation delay of the neutralization sequence packet sent to offset the stress state of the upstream node when it reaches the target network node;

[0037] Based on propagation delay and using artificial intelligence model to predict the secondary transmission stress memory state generated by the target network node when the neutralization sequence packet is used as a new stimulus source;

[0038] Traversing all target network nodes with upstream nodes along a predetermined transmission path, and sequentially adjusting data packet features of corresponding neutralization sequence packets according to a secondary transmission stress memory state;

[0039] The sending timing and structure of all adjusted neutralization sequence packets are integrated to form a multi-hop delivery strategy for the neutralization sequence packets.

[0040] Optionally, traversing all target network nodes having upstream nodes along a predetermined transmission path and sequentially adjusting data packet features of corresponding neutralization sequence packets according to the secondary transmission stress memory state comprises the following steps:

[0041] Number all target network nodes with upstream nodes on the predetermined transmission path in ascending order, and select the node with the smallest node number as the current target node;

[0042] Obtain target data packet characteristics of a target neutralization sequence packet sent to an upstream node of a current target node;

[0043] Input the target data packet characteristics into the artificial intelligence model, and output the secondary transmission stress memory state generated by the target neutralization sequence packet on the current target node through the artificial intelligence model;

[0044] Retrieve the original transmission stress memory state of the current target node from the path stress map;

[0045] Performing vector superposition of the original transmission stress memory state's feature vector and the secondary transmission stress memory state's feature vector to generate a composite transmission stress memory state vector;

[0046] Regenerate the data packet characteristics of the target neutralization sequence packet based on the state vector of the composite transmission stress memory state and using the reverse resonance generation algorithm;

[0047] The next node is selected as the current target node according to the node number, and the above data packet feature adjustment steps are repeated until all nodes with node numbers have been selected as the current target nodes.

[0048] Optionally, the step of performing vector superposition of the feature vector of the original transmission stress memory state and the feature vector of the secondary transmission stress memory state to generate a composite transmission stress memory state vector comprises the following steps:

[0049] Determining a first confidence weight for a characteristic vector of an original transmission stress memory state according to characteristic stability of the service data stream;

[0050] determining a second confidence weight for a feature vector of the secondary transmission stress memory state based on the structural complexity of the target neutralization sequence package;

[0051] A weighted sum operation is performed on the feature vector of the original transmission stress memory state and the feature vector of the secondary transmission stress memory state in combination with the first confidence weight and the second confidence weight, and a result of the weighted sum operation is used as a composite transmission stress memory state vector.

[0052] In a second aspect, the present invention also provides an artificial intelligence-based network data transmission management system, comprising a memory, a processor, and a computer program stored on the memory and runnable on the processor. When the processor executes the computer program, it implements the artificial intelligence-based network data transmission management method as described in the first aspect.

[0053] In a third aspect, the present invention further provides a computer-readable storage medium having instructions stored thereon, which, when executed by a processor, configures the processor to execute the artificial intelligence-based network data transmission management method described in the first aspect.

[0054] The beneficial effects of the present invention are:

[0055] The present invention quantifies the internal processing state of network nodes into a predictable transmission stress memory state for the first time by actively sending stress probe packets and building an artificial intelligence model, thereby being able to foresee the potential impact of data flows on network paths. The core advantage of the present invention lies in its forward-looking prediction and intervention capabilities. It no longer waits for problems such as network congestion or performance degradation to occur before remediating them. Instead, before the business data stream is transmitted, it accurately understands the pressure distribution that the entire path will be under by predicting the path stress map formed. Based on this, neutralization sequence packets that can offset the negative effects are generated and accurately delivered, which is equivalent to clearing silt and removing obstacles in advance for the upcoming data peak, smoothing the transmission process of the data stream from the source. Compared with traditional traffic shaping or congestion control algorithms, the present invention puts management opportunities in front, changing passive adaptation to active shaping, which can significantly reduce latency, jitter and packet loss rate. In particular, for latency-sensitive businesses such as cloud gaming, telemedicine, and industrial control, it provides unprecedented service quality assurance and realizes refined, intelligent and forward-looking management of network transmission performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a flowchart of an artificial intelligence-based network data transmission management method in one embodiment of the present application.

[0057] Figure 2 This is a schematic diagram of the model structure of the artificial intelligence model in one of the implementation methods of this application.

[0058] Figure 3 This is a schematic diagram of the artificial intelligence model training process in one of the implementation methods of this application.

[0059] Figure 4 A schematic diagram of the process of generating a multi-hop delivery strategy in one embodiment of the present application. DETAILED DESCRIPTION

[0060] The following will be combined with the accompanying drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the embodiments described are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of this application.

[0061] The terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of this application can be implemented in an order other than that illustrated or described herein, and that the objects distinguished by "first," "second," and the like are generally of the same type, and do not limit the number of objects; for example, the first object can be one or more. In addition, the term "and / or" in the specification and claims refers to at least one of the connected objects, and the character " / " generally indicates that the objects connected are in an "or" relationship.

[0062] Figure 1 FIG. 1 is a flow chart of a method for managing network data transmission based on artificial intelligence in one embodiment. Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 1 At least part of the steps in the above process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but may be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps. Figure 1 As shown, the network data transmission management method based on artificial intelligence disclosed in the present invention specifically includes the following steps:

[0063] S101. Actively induce and collect the transmission performance response data of network nodes by sending stress probe packets carrying dye markers in the target network, and train an artificial intelligence model based on the response data to map data stream features into the transmission stress memory state of network nodes.

[0064] To build an AI model capable of accurately predicting network behavior, the first step is proactive data collection and model training. The principle is that rather than passively waiting for network problems to occur, it's better to proactively create a controlled stress testing environment to reveal the underlying operating principles of network devices. In practice, a specially constructed stress probe packet is injected into the target network. This probe packet consists of two parts: a stimulus probe, which stresses the device processing, and a unique identifier (a dye mark) embedded in its packet header; and a measurement probe that follows to assess performance changes. When the stimulus probe carrying the dye mark passes through a network node along its path, the node's built-in deep packet inspection recognizes the mark and immediately triggers a snapshot of the node's current internal state, such as CPU utilization and memory queue length. Simultaneously, the measurement probe records end-to-end performance metrics such as latency and jitter. Through dye markup, microscopic node internal states are precisely aligned in time with macroscopic transmission performance data, establishing a direct causal relationship and forming high-quality training data pairs. By inputting a large amount of such data pairs into a preset hybrid artificial intelligence model, the model can learn the complex mapping relationship from data flow features F and network topology T to network node transmission stress memory state S through supervised learning. Its learning goal can be expressed as mastering function The result of this step is a highly sensitive predictive model that, like an experienced network doctor, can diagnose the internal stress a data flow will cause on devices along the network path based solely on its characteristics.

[0065] S102. Before the target network transmits the service data stream, based on the data flow characteristics of the service data stream and the predetermined transmission path, an artificial intelligence model is used to predict the transmission stress memory state that will be generated by each network node on the predetermined transmission path, thereby forming a path stress map.

[0066] Among them, after obtaining an accurate prediction model, the next step is to use the model to generate a path stress map depicting the future network state before the actual business data transmission begins. The principle of this process is to expand the prediction capability from a single node to the entire transmission path to form a global, forward-looking risk view. The implementation method is to first analyze the inherent characteristics of the business data flow to be transmitted, such as video conferencing traffic presenting a stable small packet flow, while large file downloads appear as a continuous large packet flow, and these characteristics are quantified as the input vector of the model. At the same time, determine the network path that the business flow is scheduled to pass through, for example, from server A to user C, it needs to pass through routers R1 and R2, and this path information is encoded into a graph structure. Subsequently, for each network node i on the path, its topology information in the path P is input into the trained artificial intelligence model g together with the business data flow feature F. The model will output a corresponding transmission stress memory state prediction vector for each node , the prediction process can be expressed as . The predicted state vectors of all nodes on the path (R1, R2, etc.) Taken together, they form a complete path stress map. This step results in a detailed early warning map that not only indicates which node will experience a problem but also details the specific nature of the problem—for example, R1 might experience a cache overflow, while R2 might face scheduler congestion. This provides irreplaceable decision-making basis for subsequent refined proactive intervention.

[0067] S103. Generate a neutralization sequence packet for offsetting the transmission stress memory state based on the path stress map, and determine a multi-hop delivery strategy for the neutralization sequence packet to synchronously offset the transmission stress memory states of multiple network nodes on a predetermined transmission path.

[0068] Among them, according to the potential problems revealed by the path stress map, the next key step is to generate neutralization sequence packages to offset these negative states. Its core principle is reverse engineering and resonance offset, that is, if a specific data flow pattern causes adverse stress, then there must be another one or more data flow patterns that can produce the opposite effect, thereby playing a neutralizing or soothing role. The specific implementation method is to enable a reverse resonance generation algorithm for each stress state to be offset at each node in the path stress map. The goal of this algorithm is to solve an optimization problem and find a set of neutralization sequence package features that can produce a reverse effect. , so that when the neutralization sequence packet flows through node i, the stress state it generates This search process can be formalized as solving . The calculated neutralization sequence packet characteristics will be concretized into a series of physical properties of data packets, including the size of each data packet, the service level field value of the IP header, and the inter-packet sending interval accurate to microseconds. After these properties are encoded and encapsulated, they form a neutralization sequence packet that can accurately target the specific stress state of a specific network node. The ultimate effect of this step is to create a series of digital antidotes. They are not general traffic control tools, but active intervention tools with specific therapeutic effects tailored for each node on the path, preparing for the active and fine-tuning of the network status.

[0069] S104. Send the neutralization sequence packet in the target network according to the multi-hop delivery strategy and along the predetermined transmission path, so as to actively adjust the internal processing states of all network nodes on the predetermined transmission path.

[0070] Among them, in order to ensure that these neutralization sequence packets can achieve the expected synergistic effect in a complex multi-hop network environment, a sophisticated multi-hop delivery strategy must be designed and implemented for them. The principle of this step is that the secondary effects generated by the neutralization sequence packet itself when it propagates in the network must be considered, that is, the neutralization packet used to regulate the downstream node will also affect its status when it flows through the upstream node. This chain reaction must be included in the calculation to prevent the intervention measures themselves from causing new problems. In specific implementation, the strategy will be iteratively adjusted starting from the first node along the direction of the predetermined transmission path. First, the neutralization sequence packet designed for the first node is calculated. The AI ​​model was then used to predict the Secondary stress state generated when flowing through the second node Then, compare this secondary stress with the original stress state of the second node. Perform weighted superposition to form a compound stress state , and then recalculate and optimize the neutralization sequence package for the second node based on this more accurate composite state This iterative adjustment process can be expressed as follows: for any node k on the path, its final composite stress state is Its original stress With all upstream neutralization packages (in The sum of the secondary stresses produced: ,in The weights are adjustable. This cycle repeats until the end of the path, ultimately forming a multi-hop delivery strategy that includes the precise transmission timing and final structure of all neutralization sequence packets. This step ultimately achieves a surgical network intervention, ensuring global optimization and harmony of the entire regulation behavior. This allows the network to enter an ideal, proactively optimized state before the service data stream arrives, thereby ensuring efficient and stable transmission.

[0071] In one embodiment, a stress probe packet carrying a dye marker is sent in a target network to actively induce and collect transmission performance response data of network nodes, and an artificial intelligence model is trained based on the response data to map data stream features to the transmission stress memory state of the network nodes, including the following steps:

[0072] generating a stress probe package, the stress probe package including a stimulation probe for actively inducing a transferred stress memory state and a measurement probe for measuring the impact of the transferred stress memory state, wherein the stimulation probe carries a dye marker;

[0073] Send stress probe packets to the target network and synchronously collect node internal data generated by network nodes on the path when processing the stimulus probes through the out-of-band management channel;

[0074] Recording transmission performance response data of network nodes measured by measurement probes;

[0075] The dye markers carried by the stimulation probes are used to temporally correlate and causally bind the node internal data with the transmission performance response data to form training data pairs.

[0076] The training data pairs are input into the preset artificial intelligence model, and the training of the artificial intelligence model is completed through supervised learning. The artificial intelligence model is a hybrid architecture of graph neural network and long short-term memory network.

[0077] In this embodiment, referring to Figure 2The artificial intelligence model in this embodiment is a complex architecture designed specifically for predicting the transmission stress memory state of network nodes. Its full name is the Spatiotemporal Graph Attention and Long Short-Term Memory Network. The core idea is to decouple and deeply integrate the spatial information of the network topology with the temporal information of the data flow. At the model input layer, three types of data are required: the first is the "static characteristics of the network nodes" that define the inherent properties of the network nodes (such as hardware configuration and cache size); the second is the "transmission path topology structure" (usually an adjacency matrix) that describes the connection relationship between each node on the predetermined transmission path; and the third is the "service data flow timing characteristics" that include information such as packet size, service type field, and inter-packet sending interval. After the model is started, the data flow is divided into two paths and enters parallel processing modules: one path, the static features of the node and the path topology structure are sent to the graph attention network (GAT) module. As a spatial feature extractor, this module can dynamically assign different attention weights to each node on the path based on the importance of its neighboring nodes, thereby learning complex spatial dependencies and generating a unique "spatially aware node embedding" vector for each node on the path; the other path, the temporal features of the business data flow are sent to the long short-term memory network (LSTM) module. As a temporal feature extractor, this module effectively captures the long-term dependencies in the data flow sequence through its gating mechanism, compresses and refines the dynamic stimulation pattern of the entire data flow into a "temporal stimulus embedding" vector representing the overall characteristics.

[0078] Next, the model enters the critical feature fusion stage. For each target node on the predetermined path, the model concatenates its corresponding "spatial perception node embedding" with the "temporal stimulus embedding" representing the entire data stream, forming a comprehensive spatiotemporal feature vector that contains both node location information and traffic impact information. Finally, this high-dimensional fusion vector is fed into a multi-layer perceptron (MLP) as a decoder. Through a series of nonlinear transformations, it is mapped into the final prediction result: a multidimensional "transmission stress memory state vector." Each dimension of this vector corresponds to a specific, quantifiable performance metric, such as predicted CPU load increase, memory utilization, queuing delay, or packet loss rate. This allows for accurate prediction of the future state of network nodes under specific data flow impacts.

[0079] Reference Figure 3Generating a stress probe packet is the first step in data collection. The principle behind this is to construct a composite tool that can actively apply controllable stress and instantly quantify its impact. This probe packet is not a single data packet, but rather consists of two tightly coupled components: the front-end is the stimulus probe, acting like a pebble dropped into calm waters. It aims to actively induce fluctuations in the processing resources within network nodes through specific traffic patterns, such as short bursts of high concurrency or large packets, thereby revealing their behavioral characteristics under stress. The back-end is the measurement probe, acting like a sensitive stopwatch, following the stimulus probe to precisely measure changes in network performance caused by the stimulus probe, such as increased latency or jitter. To ensure causal relationships can be traced, a unique identifier is embedded in the stimulus probe's packet header extension field, serving as a coloring marker. This marker serves as a key link to all subsequent data. The result of this step is a standardized set of experimental tools for active probing, laying the foundation for subsequent controlled and reproducible performance testing in real-world network environments.

[0080] The next step is to inject the generated stress probe packets into the target network and simultaneously collect the device's internal response data. The core principle is to separate the measurement behavior from the measured object to ensure data purity. In practice, the stress probe packets are sent along the normal service data path to simulate the trajectory of real traffic. Simultaneously, an independent out-of-band management channel is established for direct communication with each network node along the path. Network nodes are pre-configured with a special monitoring agent that continuously monitors passing data packets using deep packet inspection. Upon detecting a stimulus probe carrying a specific dye marker, the monitoring agent is immediately triggered to capture a snapshot of the node's current internal operating state, capturing key metrics such as CPU load, memory usage, and queue depth. This collected internal data is assigned a unique identifier and transmitted back to the data collection center via the out-of-band management channel. This step provides direct, raw evidence of the network node's internal state in response to a specific stimulus, without disrupting normal network traffic.

[0081] While collecting the internal data of the node, it is also necessary to record the changes in the macro transmission performance reflected by the measurement probe. The principle of this step is to quantify the observable impact of the stress state induced by the stimulation probe on the external transmission quality. In specific implementation, when the measurement probe following the stimulation probe arrives at the end of the path or the designated monitoring point, the information it carries is used to calculate key performance indicators. For example, by recording the sending time of the measurement probe and receiving time , the end-to-end transmission delay L it experiences can be accurately calculated, and the calculation method is: . Similarly, by analyzing the changes in the intervals between the arrival times of multiple consecutive measurement probes, transmission jitter data can be obtained. If a measurement probe that is expected to arrive fails to arrive, it is recorded as a packet loss event. All of these calculated performance response data, such as latency, jitter, and packet loss rate, are strictly associated with the unique identification code of the stimulus probe that triggered them. The result of this process is a series of accurately quantified network performance response data, which provides a direct measurement standard for evaluating the correlation between internal state changes of nodes and external performance.

[0082] After data collection is complete, temporal correlation and causal binding must be performed to form training data pairs for model learning. This step utilizes coloring tags as universal indices to integrate raw data scattered across different dimensions and collection points into structured information with clear causal logic. During implementation, the data processing program iterates through all collected unique identifiers. For each unique identifier, the program retrieves two types of associated data from the database: one is the internal state dataset collected by all relevant nodes along the path triggered by that tag, including CPU load, memory usage, and other information; the other is macroscopic transmission performance response data, such as latency, measured by measurement probes. Using the identifiers as anchors, the program matches these two previously independent data sets, establishing a direct causal relationship between node internal state changes and the resulting network performance. This binding process ensures that each generated data pair truly reflects the complete physical process from stimulus to response. The ultimate result of this step is the generation of a large number of high-quality training samples, each of which clearly links the internal state of a node to the output label transmission performance, providing learning examples for subsequent supervised learning.

[0083] The final step is to input the constructed training data pairs into a pre-set AI model for training. This approach utilizes a supervised learning paradigm, allowing the model to automatically learn and generalize the complex nonlinear mapping patterns between data flow features and the stress memory states of network nodes by repeatedly observing a large number of causal examples. In practice, a hybrid architecture combining a graph neural network and a long-short-term memory network is employed. During training, the path topology information and node internal state data from the data pairs are fed into the graph neural network component to capture the spatial dependencies between the interactions between nodes. The time-varying feature sequences of the stimulus probes themselves are processed by the long-short-term memory network component to understand the temporal dynamics of traffic patterns. Based on the input, the model generates a predicted stress state and compares it with the true stress state labels from the data pairs, calculating the prediction error between the two. This error is then propagated back into the model layer by layer through a backpropagation algorithm, fine-tuning tens of thousands of connection weights within the network to bring the next prediction closer to the true value. This iterative optimization process continues until the model's overall prediction accuracy meets the preset convergence criteria. The ultimate result of this step is a trained, generalizable intelligent model that captures the underlying laws of network transmission and can accurately predict stress states for unprecedented traffic patterns.

[0084] In one embodiment, the coloring mark carried by the stimulation probe is a unique identification code written in an extension header of a data packet header of the stimulation probe. Using the coloring mark carried by the stimulation probe to temporally correlate and causally bind node internal data with transmission performance response data to form a training data pair includes the following steps:

[0085] Configure deep packet inspection capabilities for all network nodes in the target network to identify unique identification codes;

[0086] When a network node detects a data packet containing a unique identification code, it triggers the recording of timestamps and the collection of node internal data;

[0087] Use the recorded target timestamp as the time base for the node's internal data;

[0088] At the same time, the transmission performance response data of the network nodes measured by the measurement probes are aligned with the time reference;

[0089] Based on the alignment results, the node internal data corresponding to the target timestamp is matched with the transmission performance response data to establish a causal binding between the data and form a training data pair.

[0090] In this implementation, to achieve accurate data collection, the first step is to configure and enable deep packet inspection (DPI) on all key nodes in the target network. Its core principle is to empower network devices with a perception capability that goes beyond traditional data forwarding, enabling them to see through the contents of passing data packets, similar to security scanners, rather than simply examining their recipients. Specifically, this involves deploying an efficient pattern matching engine within each network node's operating system or dedicated hardware. This engine is configured with a specific recognition rule: it searches for a unique identifier, which serves as a color marker, within a reserved extension field in the packet header. Once the value of this field in the packet exactly matches the preset identifier, the engine immediately triggers an interrupt signal. This step effectively transforms passive network devices, once solely responsible for sending and receiving data, into intelligent, distributed sensor outposts. This lays a solid foundation for subsequent event-triggered, high-precision data collection, enabling the capture of transient network behavior.

[0091] When a network node identifies a probe packet carrying a unique identifier through deep packet inspection (DPI), it immediately triggers timestamp recording and synchronizes data collection within the node. This step aims to capture the precise moment of the stimulus and firmly bind the device's internal state to the time coordinate, ensuring immediacy and accuracy of data collection. In practice, when the DPI engine issues a trigger signal, the node's internal monitoring agent immediately performs two concurrent operations: first, it invokes the system kernel's high-precision clock to record the current timestamp with nanosecond accuracy. Second, almost simultaneously, the monitoring agent issues queries to various submodules of the system, rapidly capturing a series of key internal operating metrics, such as current CPU load, memory buffer occupancy, packet processing queue length, and hardware interrupt frequency. This series of operations is designed to be atomic to minimize latency. The result is a snapshot of the node's internal state, with a precise time fingerprint, that truly reflects the device's original operating conditions at the moment of response to the specific stimulus.

[0092] Using the recorded target timestamp as a unified time reference for all subsequent internal data is crucial for ensuring data consistency and comparability. This approach provides a common anchor for scattered data points from different modules and dimensions, preventing temporal distortion during data fusion. In specific implementation, the snapshot timestamp captured in the previous step serves as a primary key or core index. All internal node data collected at that point in time, whether CPU utilization, memory usage, or queue information, is uniformly tagged with this timestamp. This means that these various metrics are collectively defined as a collection of events occurring at the timestamp. When stored, they are organized into a structured record identified by a timestamp, ensuring the integrity and indivisibility of this internal state snapshot. This step effectively consolidates raw monitoring data, which may have been collected asynchronously and in various formats, into structured data units with a unified time coordinate, greatly simplifying the complexity of subsequent multi-source data correlation and analysis.

[0093] At the same time, the macro network performance data measured by the measurement probes must also be aligned with this time base. The principle of this step is to link externally observable performance with internally invisible device states in the time dimension, thereby establishing a potential logical relationship between the two. During implementation, the data processing center will collect the performance data recorded by the measurement probes at the network endpoints, for example, by recording their sending time and receiving time to calculate the end-to-end delay. Since the measurement probes are sent in pair with the stimulation probes with unique identification codes, the calculated delay is considered to be the final result caused by a series of node internal state changes caused by the stimulation probes on the path. Therefore, this performance data will be traced back and associated with the timestamps recorded by the stimulation probes at each node.

[0094] Finally, based on the precise alignment results, the node internal data corresponding to the target timestamp is matched with the transmission performance response data, establishing a direct causal binding, thus forming the final training data pair. The principle of this step is to solidify the results of all the previous preparation work into a standard input format that can be directly used by the machine learning model. Specifically, for each unique stimulus event, the system will pair all the node internal state data sets associated with the timestamp of its triggering with the macro network performance response data aligned with it, forming a structured training sample, which can be represented as a data tuple. This process is repeated for thousands of active detection events, thereby constructing a large training dataset that covers rich scenarios.

[0095] In one embodiment, generating a neutralization sequence package for counteracting a transmission stress memory state according to a pathway stress map comprises the following steps:

[0096] Analyze the path stress map to obtain the target transmission stress memory state corresponding to each target network node on the predetermined transmission path;

[0097] For each target transmission stress memory state, one or more reverse excitation patterns required to offset the stress state are calculated using a reverse resonance generation algorithm;

[0098] Mapping the reverse stimulus pattern to the data packet characteristics required for the neutralization sequence packet, the data packet characteristics including the data packet size, the service type field value, and the inter-packet sending interval;

[0099] If the transmission stress memory state is a composite state, a sequence consisting of multiple data sub-packets with different data packet characteristics is generated;

[0100] The data packet or data sub-packet containing the data packet characteristics is encoded and encapsulated to generate a neutralization sequence packet.

[0101] In this implementation, in order to generate an accurate neutralization sequence package, the path stress map must be deeply analyzed to obtain the exact target transmission stress memory state of each network node. The principle of this process is to convert the macro, full-path risk warning into a series of actionable intervention targets for specific nodes. During implementation, the path stress map data structure is traversed, and for each target network node on the predetermined transmission path, the path stress map is used to generate a target transmission stress memory state. , extract the corresponding multidimensional stress state vector This vector is the product of the AI ​​model's prediction, and each dimension quantifies a specific internal pressure, such as [CPU load increment, cache queue growth rate, scheduler latency]. The parsing process involves more than simply reading the data; it also involves interpreting the state vector to determine its primary conflict, such as a computing resource shortage or a memory resource bottleneck. The ultimate effect of this step is to break down a complex, global prediction into a series of clear, independent task lists. This clearly defines the specific stress state that needs to be offset or alleviated for each device on the path, allowing subsequent intervention measures to be targeted and highly targeted.

[0102] Next, for each target transmission stress memory state parsed from the spectrum, an inverse resonance generation algorithm is used to calculate the reverse excitation pattern necessary to offset the state. The core principle is based on the symmetry of the system, that is, if a specific flow pattern can induce a certain resonant stress response in the device, then there must be another pattern that can produce an intervention effect with the opposite phase, thereby achieving negative offset. The specific implementation method is to use the known target stress state vector as input to solve an optimization problem. The goal of this problem is to find an abstract excitation pattern feature vector When used as input to an AI model, the resulting predicted stress The vector sum of the target stress can approach zero. This solution process can be formalized as finding the optimal To minimize the objective function: Through iterative calculation, the final result is It is the most effective reverse incentive model in theory.

[0103] After obtaining the abstract reverse excitation pattern, it needs to be mapped to the specific data packet characteristics required for the neutralization sequence packet. The principle of this step is to convert the theoretical mathematical description into physical world parameters that can be recognized and processed by network equipment. The specific implementation method is to establish a mapping rule base or a lightweight conversion model that can translate each dimension in the reverse excitation pattern vector into a set of executable data packet parameters. For example, the component representing high-frequency disturbances in the vector may be mapped to an extremely short inter-packet sending interval; the component representing the impact on the cache may be mapped to a series of data packets of a specific size; and the component representing the change in processing priority may be mapped to a specific service type field (ToS / DSCP) value in the IP packet header. Therefore, an abstract It will be decoded into a specific parameter set, including the data packet size, service type field value, and the interval between packets.

[0104] In some cases, the transmission stress memory state faced by a network node may be a complex state resulting from the superposition of multiple stress factors. In this case, it is necessary to generate a sequence consisting of multiple data sub-packets with different packet characteristics. The principle is that addressing complex problems requires a combination of tools; single intervention measures cannot simultaneously address two different types of stress, such as CPU overload and queue congestion. Specifically, the complex stress state vector is first decomposed into multiple basic, orthogonal stress components, such as pure computational stress and pure cache stress. Then, for each basic stress component, the aforementioned reverse resonance generation and feature mapping process is independently performed to obtain the corresponding sub-packet feature set used to offset that component. Finally, these sub-packets corresponding to different basic stresses are combined according to a specific timing and logic to form a complete neutralization sequence containing multiple different types of data sub-packets.

[0105] Finally, the data packets or sub-packets containing all packet characteristics are encoded and encapsulated to generate a neutralized sequence packet that can be transmitted across the network. This step transforms the design drawings into a practical industrial product, constructing the parameterized packet description into a binary data stream that complies with network protocol standards. Specifically, this involves calling the underlying network programming interface to create packets one by one based on the set of packet characteristics determined in the previous step. For each sub-packet, the program precisely sets the Type of Service field in the IP header to the specified value and generates a payload of a specified size (the content is generally unimportant; the size is critical). These generated packets are then passed to a high-precision packet scheduler, which strictly adheres to the calculated inter-packet interval to control the timing of each packet's transmission. This step ultimately completes the final transformation from concept to reality, producing one or more binary bit streams carrying specific regulatory tasks, ready for injection into the target network. This provides the ultimate physical tool for proactive and precise network state management.

[0106] In one embodiment, mapping the reverse excitation pattern to the data packet characteristics required for the neutralization sequence packet includes the following steps:

[0107] Separate stress response characteristics with specific behavioral patterns from the transferred stress memory state;

[0108] Calculate theoretical waveforms that are opposite in phase to the stress response characteristics, or sequences of events that are antagonistic in behavioral patterns;

[0109] Discretize the theoretical waveform or event sequence into data packet sending events;

[0110] Determine the size of the corresponding data packet and the value of the Differentiated Services Code Point field in the IP header for each send event;

[0111] The sending time interval is determined between adjacent sending events, and all sending events are combined into a data packet feature for implementing the reverse excitation mode according to the time interval.

[0112] In this implementation, in order to achieve accurate offset of network stress, it is necessary to separate and identify the basic behavioral patterns that constitute the state from the complex and general transmission stress memory state. The principle is that any complex system response can be regarded as the superposition of several simpler and more typical basic response patterns. By deconstructing the multidimensional data vector representing the stress state, its internal dominant components can be understood. An effective implementation method is to use principal component analysis to decompose the original stress state vector into a series of orthogonal eigenvectors. The weighted sum of . Each eigenvector It represents a specific behavior pattern, such as high-frequency resource competition jitter and slow cache accumulation effect, and its corresponding weight After identifying the basic stress response characteristics, the next step is to calculate the theoretically effective countermeasures for each behavioral pattern.

[0113] This solution can be a theoretical waveform with completely opposite phases, or a sequence of events with antagonistic effects in behavioral patterns. Its core principle is to actively introduce an intervention force that is equal in magnitude and opposite in direction to the negative effect to achieve neutralization. In specific implementation, if the isolated stress response characteristics It is a waveform that changes with time, such as the periodic fluctuation of CPU load, so its ideal offset waveform is In theory, it is a simple inversion of the waveform, that is, If the behavior pattern manifests as discrete events, such as bursts of packet loss, then the antagonistic sequence of events might be to pre-trigger a slight traffic policing to smooth out subsequent traffic spikes.

[0114] After obtaining the theoretical cancellation waveform or event sequence, it must be discretized into a series of data packet sending events that can be executed by network devices. The principle is that the nature of digital network communication is discrete and cannot directly transmit continuous analog waveforms. Therefore, the continuous theoretical model must be converted into a series of independent actions that occur at specific time points. In specific implementation, the continuous theoretical cancellation waveform will be sampled at equal intervals. By selecting a very small time step , we can get a series of time points , and the amplitude of the waveform at these time points Each sampling point are defined as an independent data packet sending event, which occurs at , and its intervention intensity is determined by the amplitude Decision. This step transforms a continuous, unimplementable theoretical plan into a discrete, actionable sequence of instructions that details the precise timing at which intervention actions need to be initiated.

[0115] Next, for each discretized sending event, it is necessary to determine the specific physical properties of the corresponding data packet, mainly the size of the data packet and the value of the Differentiated Services Code Point field in the IP header. The principle of this step is to give actual physical meaning to the previously defined intervention intensity and convert it into a parameter that can have a substantial impact on the internal processing logic of the network node. A feasible implementation method is to convert the sampling amplitude into a specific data packet size through a preset mapping function. and Differentiated Services Code Point field values , where positive amplitudes may correspond to larger packets to increase processing load, and negative amplitudes may correspond to small packets with high-priority code points to preempt scheduling resources. In this way, different parts of the theoretical waveform are cleverly encoded into the packet size and priority.

[0116] Finally, the precise sending time interval is determined between all adjacent sending events, and the attributes of all events are integrated to form the final data packet feature set used to implement the reverse excitation mode. The principle of this step is to assemble all design elements into a complete, executable engineering blueprint to ensure that the sending rhythm of the neutralization sequence packet is completely consistent with the theoretical model. In practice, the sending time interval between two adjacent sending events j and j+1 is directly determined by the sampling time step in the discretization step. This interval is the key to maintaining the cancellation waveform shape. The final data packet feature set is a complete list of all data packet specifications and timings, in the form of , along with a global transmission interval parameter. This signature set is the final product, a comprehensive description of how to generate and send a series of data packets to reproduce the theoretically calculated reverse incentive pattern on the target network node. This step results in a precise set of instructions that can be directly read and executed by the packet-sending program, completing the entire process from abstract theory to concrete practice.

[0117] In one embodiment, determining a multi-hop delivery strategy for neutralizing sequence packets to synchronously offset transmission stress memory states of multiple network nodes on a predetermined transmission path includes the following steps:

[0118] For any target network node on the predetermined transmission path, the node that receives the data packet before the target network node and is immediately adjacent to the target network node is marked as the upstream node of the target network node, and the node that receives the data packet after the target network node and is immediately adjacent to the target network node is marked as the downstream node of the target network node;

[0119] Calculate the propagation delay of the neutralization sequence packet sent to offset the stress state of the upstream node when it reaches the target network node;

[0120] Based on propagation delay and using artificial intelligence model to predict the secondary transmission stress memory state generated by the target network node when the neutralization sequence packet is used as a new stimulus source;

[0121] Traversing all target network nodes with upstream nodes along a predetermined transmission path, and sequentially adjusting data packet features of corresponding neutralization sequence packets according to a secondary transmission stress memory state;

[0122] The sending timing and structure of all adjusted neutralization sequence packets are integrated to form a multi-hop delivery strategy for the neutralization sequence packets.

[0123] In this embodiment, referring to Figure 4 First, in order to establish a clear causal analysis chain, it is necessary to label the roles of all network nodes on the predetermined transmission path. The principle is to clearly define the order and adjacency between nodes, which is the basis for subsequent hop-by-hop impact analysis. The specific implementation method is to label any target network node on the path that is selected as the current analysis focus. , check its position in the transmission path sequence. The node that is immediately before in the sequence and the data packet will reach it first is marked as the upstream node. Conversely, the node that is immediately before in the sequence Afterwards, the node that the data packet reaches later is marked as a downstream node. This marking process is performed along the entire path, and all intermediate nodes except the first and last nodes are determined to have their unique upstream and downstream neighbors.

[0124] Before making multi-hop strategy adjustments, it is necessary to accurately calculate the propagation delay that the neutralization sequence packet sent to offset the stress state of the upstream node will experience when it reaches the current target network node. The principle is that the propagation of any intervention measure takes time. This time difference determines the synchronization of the state evolution of the upstream and downstream nodes and is a key parameter for predicting chain reactions. During implementation, the single-hop network delay between the upstream node and the current target network node will be obtained through active detection or the use of existing network monitoring data. This delay includes the propagation time of the data packet on the physical link, the queuing and processing time at the upstream node. Therefore, the neutralization sequence packet generated for the upstream node will not arrive instantly after leaving the upstream node. , but the time that will be delayed.

[0125] Next, based on the calculated propagation delay, the trained artificial intelligence model is used to predict the secondary transmission stress memory state generated by the neutralization sequence packet designed for the upstream node on the current target network node. The principle is that any data stream injected into the network, regardless of its original intention is good or bad, is a new source of excitation for the equipment it passes through, and will inevitably trigger a new stress response. In specific implementation, the data packet characteristics of the neutralization sequence packet originally designed to offset the stress of the upstream node will be input into the artificial intelligence model as the new excitation source characteristics. At the same time, the topology information of the current target network node is also input, and the model will predict the neutralization sequence packet flowing through the network. This prediction takes into account the propagation delay, ensuring that the analysis is based on the impact of the arrival of the neutralization packet.

[0126] The system then traverses the predetermined transmission path hop by hop, adjusting the data packet characteristics of each node's corresponding neutralization sequence packet based on the predicted secondary transmission stress memory state. This principle operates as a feedforward compensation approach, preemptively modifying the solution before a problem occurs to offset known negative chain reactions. This process begins at the second node in the path and continues to the last node. For the current target node, its original stress state to be offset is vector-superimposed with the secondary stresses generated by all upstream neutralization sequence packets, forming a composite target stress state. Based on this more comprehensive and accurate composite stress state, the inverse resonance generation algorithm is re-applied to calculate new, adjusted neutralization sequence packet characteristics. This iterative adjustment process ensures that interventions at each node take into account all upstream influences.

[0127] Finally, the transmission timing of all hop-by-hop adjusted neutralization sequence packets is integrated with the final structure to form a complete, globally coordinated multi-hop delivery strategy. The principle is to assemble all local, optimized solutions into a globally optimal, executable overall plan. Specifically, the characteristics of the neutralization sequence packets determined by each node along the path and their scheduled delivery times (taking into account propagation delay and processing timing) are aggregated and sorted to form a detailed execution plan. This plan not only specifies the specific content of each neutralization sequence packet (packet size, priority, and packet interval), but also precisely defines the time window and relative order of their transmission from the source. The ultimate result of this step is a final set of action instructions that jointly implements the proactive adjustment of the internal processing state of all nodes along the entire transmission path.

[0128] In one embodiment, traversing all target network nodes having upstream nodes along a predetermined transmission path and sequentially adjusting data packet features of corresponding neutralization sequence packets according to the secondary transmission stress memory state comprises the following steps:

[0129] Number all target network nodes with upstream nodes on the predetermined transmission path in ascending order, and select the node with the smallest node number as the current target node;

[0130] Obtain target data packet characteristics of a target neutralization sequence packet sent to an upstream node of a current target node;

[0131] Input the target data packet characteristics into the artificial intelligence model, and output the secondary transmission stress memory state generated by the target neutralization sequence packet on the current target node through the artificial intelligence model;

[0132] Retrieve the original transmission stress memory state of the current target node from the path stress map;

[0133] Performing vector superposition of the original transmission stress memory state's feature vector and the secondary transmission stress memory state's feature vector to generate a composite transmission stress memory state vector;

[0134] Regenerate the data packet characteristics of the target neutralization sequence packet based on the state vector of the composite transmission stress memory state and using the reverse resonance generation algorithm;

[0135] The next node is selected as the current target node according to the node number, and the above data packet feature adjustment steps are repeated until all nodes with node numbers have been selected as the current target nodes.

[0136] In this implementation, to ensure the rigor and orderliness of the multi-hop delivery strategy, the nodes on the transmission path must first be arranged in an orderly manner. The principle is to transform a complex mesh or chain structure into a linear, step-by-step process sequence, just like processing products step by step on an assembly line. The specific implementation method is to assign a unique number that increases from one to all network nodes with upstream nodes on the predetermined transmission path (i.e., starting from the second node to the last node) along the direction of the data flow. Subsequently, the node with the smallest number is selected as the starting point of the calculation process, i.e., the current target node. The effect of this step is to establish a clear and unchangeable iterative framework, which stipulates the order that all subsequent adjustments and calculations must follow, ensuring that the adjustment results of the upstream nodes can be used as known inputs for the adjustment of downstream nodes, thereby avoiding circular dependencies or calculation confusion.

[0137] After determining the current target node, it is necessary to obtain the source data that will affect it—that is, the target data packet characteristics of the neutralization sequence packet designed for its upstream node. The principle is that any prediction must be based on precise input. To calculate the secondary impact of upstream intervention measures on the current node, the specific form of the intervention must be determined. During implementation, the program retrieves the complete set of neutralization sequence packet characteristics calculated for the upstream node of the current target node from memory or a database. In the initial iteration phase, these are the characteristics calculated based on the original stress; in subsequent iterations, these will be optimized characteristics adjusted for the upstream chain reaction. This feature set describes every detail of the upstream neutralization sequence packet, including packet size, priority, and delivery cadence.

[0138] After obtaining the packet characteristics of the upstream neutralization sequence packet, a pre-trained AI model is used to output the secondary transmission stress memory state it will generate for the current target node. This approach leverages the model's powerful nonlinear mapping capabilities to simulate and predict the impact a specific data flow pattern will have on a network node's internal resources (such as CPU and memory). Specifically, the target packet characteristics acquired in the previous step are fed into the AI ​​model along with the current target node's topological information. After complex internal calculations, the model outputs a multidimensional vector that accurately describes the secondary stress state in numerical terms. To fully understand the challenges facing the current target node, the original transmission stress memory state must be retrieved from a pre-generated path stress map. The principle is that the stress a node ultimately experiences is the sum of its inherent problems and external secondary influences; neither is negligible.

[0139] During implementation, the program uses the current target node's ID as an index to query the path stress map data structure and extract a feature vector representing the node's original predicted stress. This vector, predicted at the beginning of the process based on analysis of future business data flows, represents the expected problems at the node in the absence of any intervention. The feature vectors of the original and secondary transmission stress memory states are then superimposed to generate a composite transmission stress memory state vector. This principle is similar to the synthesis of forces in physics: when an object is acted upon by multiple forces simultaneously, its final state of motion is determined by the vector sum of these forces. The same is true for the state of a network node. Specifically, a simple vector addition operation is performed, adding the vector representing the original stress to the vector representing the secondary stress to generate a new composite stress vector. This newly generated vector incorporates the effects of both stresses in every dimension, comprehensively describing the true and complete challenge facing the current target node.

[0140] After obtaining a precise composite stress state, the inverse resonance generation algorithm is used to regenerate an optimized set of packet signatures for the neutralization sequence packets for the current target node. The principle is that since the intervention target has been updated, the solution must also be adjusted to ensure precise effectiveness. During implementation, the newly generated composite stress state vector is fed into the inverse resonance generation algorithm. The algorithm optimizes and calculates a set of updated packet signatures that best offsets this composite state. Compared to the original solution, the newly generated signature set may include subtle but crucial changes in packet size, priority, or transmission interval. These changes are intended to additionally offset secondary effects from upstream.

[0141] Finally, the process selects the next node as the new current target node, following the node number sequence, and repeats the entire packet feature adjustment process. The principle is to ensure that this chain reaction of progressive compensation is propagated along the entire path, forming a globally coordinated optimization chain. The specific implementation is a loop structure. After recalculating the neutralization packet features for the current node, the program increments the node number by one and selects the next node on the path as the new target node. The process then returns to the second step, obtaining the optimized features just generated for the previous node as the new upstream stimulus, calculating the composite stress for the new target node, and optimizing its neutralization solution. This loop continues until all numbered nodes have been processed as current target nodes. The ultimate effect of this step is to ensure that all neutralization strategies along the entire path are iteratively optimized, forming a coherent, multi-hop delivery strategy that minimizes internal conflicts.

[0142] In one embodiment, superimposing the feature vector of the original transmission stress memory state with the feature vector of the secondary transmission stress memory state to generate a composite transmission stress memory state vector comprises the following steps:

[0143] Determining a first confidence weight for a characteristic vector of an original transmission stress memory state according to characteristic stability of the service data stream;

[0144] determining a second confidence weight for a feature vector of the secondary transmission stress memory state based on the structural complexity of the target neutralization sequence package;

[0145] A weighted sum operation is performed on the feature vector of the original transmission stress memory state and the feature vector of the secondary transmission stress memory state in combination with the first confidence weight and the second confidence weight, and a result of the weighted sum operation is used as a composite transmission stress memory state vector.

[0146] In this embodiment, before performing vector superposition, it is first necessary to determine a confidence weight for the feature vector representing the original transmission stress memory state. The core principle is that not all predictions are equally reliable, and the confidence of the prediction depends on the stability of its source data. If the characteristics of the upcoming business data stream are very stable and predictable, such as a constant bit rate video stream, then the prediction of the original stress state caused by it is more credible and should be given a higher weight. Conversely, if the data stream is extremely unstable and full of bursts, such as typical web browsing traffic, then the uncertainty of the prediction is higher and its weight should be lowered accordingly. A specific implementation method is to first calculate the coefficient of variation of the key characteristics of the business data stream (such as the packet arrival rate) as a quantitative instability indicator. Subsequently, the instability indicator can be mapped to a first confidence weight through a decreasing function. , for example using an exponential decay function ,in It is an indicator of instability.

[0147] Similarly, it is also necessary to determine the exclusive confidence weight for the characteristic vector representing the secondary transmission stress memory state. The principle of this step is that the accuracy of the prediction of the side effects caused by the neutralization sequence packet itself is not 100%, and it depends on the structural complexity of the neutralization sequence packet itself. A neutralization sequence packet with a simple structure and regular pattern (such as sending data packets of the same size at a constant rate) is easier to be accurately predicted by the model in the network, so the confidence of the prediction of the secondary stress state caused by it is high and it should be given a higher weight. On the contrary, a neutralization sequence packet designed to offset the complex state and with an extremely complex structure itself has increased nonlinearity and uncertainty in its behavior, and the model's prediction of its side effects is relatively small, and the weight should also be reduced accordingly. In specific implementation, the structural complexity of the neutralization sequence packet can be quantified by calculating the information entropy of the characteristic sequence of the data packet. Then, a decreasing function is used to convert the complexity index into a second confidence weight. ,For example ,in is a complexity indicator.

[0148] Finally, combining the first confidence weight and the second confidence weight determined in the first two steps, a weighted summation operation is performed on the feature vector of the original transmission stress memory state and the feature vector of the secondary transmission stress memory state, and the result of the operation is used as the final composite transmission stress memory state vector. The principle is that by introducing weights, the simple vector addition is upgraded to a confidence-based intelligent fusion, so that more reliable information occupies a more dominant position in the final decision. The specific implementation method is to perform a weighted vector summation. If the original stress vector is , the secondary stress vector is , then the composite stress vector obtained after weighted summation is The formula The result is no longer a simple superposition of the two stress states, but rather a fusion dynamically adjusted based on the confidence of their respective predictions. This step produces a composite stress state description that is more robust and closer to reality than simple addition, providing a more accurate and reliable optimization target for the subsequent regeneration of neutralization sequence packages based on this state.

[0149] The present invention also discloses an artificial intelligence-based network data transmission management system, comprising a memory, a processor, and a computer program stored on the memory and runnable on the processor. When the processor executes the computer program, it implements the artificial intelligence-based network data transmission management method as described in any of the above embodiments.

[0150] Among them, the processor can adopt a central processing unit (CPU). Of course, according to actual usage, other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be adopted. The general-purpose processor can adopt a microprocessor or any conventional processor, etc., and this application does not impose any restrictions on this.

[0151] Among them, the memory can be an internal storage unit of a computer device, such as a hard disk or memory of a computer device, or an external storage device of a computer device, such as a plug-in hard disk, smart memory card (SMC), secure digital card (SD) or flash memory card (FC) equipped on the computer device. In addition, the memory can also be a combination of an internal storage unit and an external storage device of a computer device. The memory is used to store computer programs and other programs and data required by the computer device. The memory can also be used to temporarily store data that has been output or is to be output. This application does not impose any restrictions on this.

[0152] The present invention also discloses a computer-readable storage medium having instructions stored thereon. When the instructions are executed by a processor, the processor is configured to execute the artificial intelligence-based network data transmission management method described in any one of the above embodiments.

[0153] Among them, the computer program can be stored in a machine-readable medium, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or certain middleware, etc. The machine-readable medium includes any entity or device that can carry computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the machine-readable medium includes but is not limited to the above-mentioned components.

[0154] Among them, through this computer-readable storage medium, the artificial intelligence-based network data transmission management method in the above embodiment is stored in the computer-readable storage medium, and is loaded and executed on the processor to facilitate the storage and application of the above method.

[0155] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of protection of the present application is limited to these examples. In line with the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0156] The one or more embodiments of this application are intended to encompass all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of this application should be included in the scope of protection of this application.

Claims

1. A network data transmission management method based on artificial intelligence, characterized in that: The steps include: By sending stress probe packets carrying dye markers in the target network, the transmission performance response data of network nodes is actively induced and collected. Based on the response data, an artificial intelligence model is trained to map data flow characteristics into the transmission stress memory state of network nodes. Before transmitting the service data flow on the target network, based on the data flow characteristics of the service data flow and the predetermined transmission path, an artificial intelligence model is used to predict the transmission stress memory state that will be generated by each network node on the predetermined transmission path, forming a path stress map; Generate a neutralization sequence packet for offsetting the transmission stress memory state according to the path stress map, and determine a multi-hop delivery strategy for the neutralization sequence packet to synchronously offset the transmission stress memory state of multiple network nodes on a predetermined transmission path; According to the multi-hop delivery strategy and along the predetermined transmission path, the neutralization sequence packet is sent in the target network to actively adjust the internal processing state of all network nodes on the predetermined transmission path.

2. The network data transmission management method based on artificial intelligence according to claim 1 is characterized in that: The method includes the following steps: actively inducing and collecting transmission performance response data of network nodes by sending stress probe packets carrying dye markers in the target network, and training an artificial intelligence model based on the response data to map data flow features to the transmission stress memory state of the network nodes. generating a stress probe package, the stress probe package including a stimulation probe for actively inducing a transferred stress memory state and a measurement probe for measuring the impact of the transferred stress memory state, wherein the stimulation probe carries a dye marker; Send stress probe packets to the target network and synchronously collect node internal data generated by network nodes on the path when processing the stimulus probes through the out-of-band management channel; Recording transmission performance response data of network nodes measured by measurement probes; The dye markers carried by the stimulation probes are used to temporally correlate and causally bind the node internal data with the transmission performance response data to form training data pairs. The training data pairs are input into the preset artificial intelligence model, and the training of the artificial intelligence model is completed through supervised learning. The artificial intelligence model is a hybrid architecture of graph neural network and long short-term memory network.

3. The network data transmission management method based on artificial intelligence according to claim 2 is characterized in that: The coloring mark carried by the stimulation probe is a unique identification code written in the extended header of the data packet header of the stimulation probe. The method of using the coloring mark carried by the stimulation probe to temporally associate and causally bind the node internal data with the transmission performance response data to form a training data pair includes the following steps: Configure deep packet inspection capabilities for all network nodes in the target network to identify unique identification codes; When a network node detects a data packet containing a unique identification code, it triggers the recording of timestamps and the collection of node internal data; Use the recorded target timestamp as the time base for the node's internal data; At the same time, the transmission performance response data of the network nodes measured by the measurement probes are aligned with the time reference; Based on the alignment results, the node internal data corresponding to the target timestamp is matched with the transmission performance response data to establish a causal binding between the data and form a training data pair.

4. The network data transmission management method based on artificial intelligence according to claim 1 is characterized in that: The step of generating a neutralization sequence package for offsetting the transmission stress memory state according to the path stress map comprises the following steps: Analyze the path stress map to obtain the target transmission stress memory state corresponding to each target network node on the predetermined transmission path; For each target transmission stress memory state, one or more reverse excitation patterns required to offset the stress state are calculated using a reverse resonance generation algorithm; Mapping the reverse stimulus pattern to the data packet characteristics required for the neutralization sequence packet, the data packet characteristics including the data packet size, the service type field value, and the inter-packet sending interval; If the transmission stress memory state is a composite state, a sequence consisting of multiple data sub-packets with different data packet characteristics is generated; The data packet or data sub-packet containing the data packet characteristics is encoded and encapsulated to generate a neutralization sequence packet.

5. The network data transmission management method based on artificial intelligence according to claim 4 is characterized in that: Mapping the reverse excitation pattern to the data packet characteristics required for the neutralization sequence packet includes the following steps: Separate stress response characteristics with specific behavioral patterns from the transferred stress memory state; Calculate theoretical waveforms that are opposite in phase to the stress response characteristics, or sequences of events that are antagonistic in behavioral patterns; Discretize the theoretical waveform or event sequence into data packet sending events; Determine the size of the corresponding data packet and the value of the Differentiated Services Code Point field in the IP header for each send event; The sending time interval is determined between adjacent sending events, and all sending events are combined into a data packet feature for implementing the reverse excitation mode according to the time interval.

6. The method for managing network data transmission based on artificial intelligence according to claim 4, characterized in that: The multi-hop delivery strategy for determining a neutralization sequence packet to synchronously offset the transmission stress memory states of multiple network nodes on a predetermined transmission path comprises the following steps: For any target network node on the predetermined transmission path, the node that receives the data packet before the target network node and is immediately adjacent to the target network node is marked as the upstream node of the target network node, and the node that receives the data packet after the target network node and is immediately adjacent to the target network node is marked as the downstream node of the target network node; Calculate the propagation delay of the neutralization sequence packet sent to offset the stress state of the upstream node when it reaches the target network node; Based on propagation delay and using artificial intelligence model to predict the secondary transmission stress memory state generated by the target network node when the neutralization sequence packet is used as a new stimulus source; Traversing all target network nodes with upstream nodes along a predetermined transmission path, and sequentially adjusting data packet features of corresponding neutralization sequence packets according to a secondary transmission stress memory state; The sending timing and structure of all adjusted neutralization sequence packets are integrated to form a multi-hop delivery strategy for the neutralization sequence packets.

7. The method for managing network data transmission based on artificial intelligence according to claim 6, characterized in that: The step of traversing all target network nodes having upstream nodes along a predetermined transmission path and sequentially adjusting the data packet characteristics of corresponding neutralization sequence packets according to the secondary transmission stress memory state comprises the following steps: Number all target network nodes with upstream nodes on the predetermined transmission path in ascending order, and select the node with the smallest node number as the current target node; Obtain target data packet characteristics of a target neutralization sequence packet sent to an upstream node of a current target node; Input the target data packet characteristics into the artificial intelligence model, and output the secondary transmission stress memory state generated by the target neutralization sequence packet on the current target node through the artificial intelligence model; Retrieve the original transmission stress memory state of the current target node from the path stress map; Performing vector superposition of the original transmission stress memory state's feature vector and the secondary transmission stress memory state's feature vector to generate a composite transmission stress memory state vector; Regenerate the data packet characteristics of the target neutralization sequence packet based on the state vector of the composite transmission stress memory state and using the reverse resonance generation algorithm; The next node is selected as the current target node according to the node number, and the above data packet feature adjustment steps are repeated until all nodes with node numbers have been selected as the current target nodes.

8. The method for managing network data transmission based on artificial intelligence according to claim 7, characterized in that: The step of superimposing the characteristic vector of the original transmission stress memory state with the characteristic vector of the secondary transmission stress memory state to generate a composite transmission stress memory state vector comprises the following steps: Determining a first confidence weight for a characteristic vector of an original transmission stress memory state according to characteristic stability of the service data stream; determining a second confidence weight for a feature vector of the secondary transmission stress memory state based on the structural complexity of the target neutralization sequence package; A weighted sum operation is performed on the feature vector of the original transmission stress memory state and the feature vector of the secondary transmission stress memory state in combination with the first confidence weight and the second confidence weight, and a result of the weighted sum operation is used as a composite transmission stress memory state vector.

9. An artificial intelligence-based network data transmission management system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the network data transmission management method based on artificial intelligence as described in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instruction is executed by a processor, the processor is configured to execute the network data transmission management method based on artificial intelligence according to any one of claims 1 to 8.

Images