An AI-based network communication link adaptive optimization method in a complex environment

Abstract

The application discloses an AI-based network communication link adaptive optimization method in a complex environment and belongs to the technical field of network communication. In view of the poor adaptability, single optimization target, response lag and lack of cross-layer cooperation of the network communication link optimization method in the prior art in a complex environment, the application collects multidimensional data of a physical layer, a data link layer and an environment layer through a multimodal link state perception module, constructs a space-time joint link state prediction model, combines a multi-target deep reinforcement learning decision engine and realizes dynamic adaptive adjustment of cross-layer parameters. The application simultaneously collects multidimensional data of the physical layer, the data link layer, the environment layer and a service layer and can comprehensively and accurately reflect the link state in the complex environment. Furthermore, the application combines LSTM and GAT to construct a space-time joint prediction model, can simultaneously capture the time variation law and the spatial topological relationship of the link state and significantly improves the precision and reliability of the link state prediction.

Description

Technical Field

[0001] This invention belongs to the field of network communication technology, specifically relating to an AI-based adaptive optimization method for network communication links in complex environments. Background Technology

[0002] With the rapid development of 5G / 6G technology and the widespread adoption of IoT applications, network communication systems are facing increasingly complex application environments. In scenarios such as the Industrial Internet, the Internet of Vehicles, and emergency communications, the communication environment is typically characterized by severe electromagnetic interference, dynamic changes in network topology, diverse service requirements, and high real-time demands. Traditional network communication link optimization methods are mainly based on fixed rules and thresholds, using pre-set parameter adjustment strategies to cope with changes in link status.

[0003] However, traditional methods have the following obvious drawbacks:

[0004] 1. Poor adaptability: Traditional methods can only handle a limited set of preset scenarios and cannot cope with complex and ever-changing real-world environments. In particular, the optimization effect drops sharply when there are unknown interference sources and sudden topology changes.

[0005] 2. Single optimization objective: Traditional methods usually only focus on a single performance metric, such as throughput or latency, and cannot balance multiple conflicting optimization objectives at the same time.

[0006] 3. Delayed response: Traditional methods adjust based on the current link status and cannot predict future link changes, resulting in adjustments always lagging behind actual needs.

[0007] 4. Lack of cross-layer collaboration: Traditional methods usually optimize at a single protocol layer, which cannot achieve collaborative optimization of the physical layer, data link layer and network layer, thus limiting the improvement of overall performance.

[0008] In recent years, the development of artificial intelligence technology has provided new ideas for solving the above problems. AI-based link optimization methods can automatically discover complex patterns and rules by learning from historical data, achieving more intelligent and adaptive link adjustments. However, existing AI-driven link optimization methods still have some shortcomings: most methods only utilize single-dimensional link state data, ignoring the influence of environmental factors and spatial topology; it is difficult to balance the accuracy and real-time performance of prediction models; and the reward function design of reinforcement learning algorithms is unreasonable, failing to effectively balance multiple optimization objectives.

[0009] Therefore, there is a need for an AI-based adaptive optimization method for network communication links in complex environments to address the problems of poor adaptability, single optimization objective, slow response, and lack of cross-layer collaboration in existing network communication link optimization methods under complex environments. Summary of the Invention

[0010] The purpose of this invention is to provide an AI-based adaptive optimization method for network communication links in complex environments, in order to solve the problems mentioned in the background art.

[0011] To achieve the above objectives, the present invention provides the following technical solution: an AI-based adaptive optimization method for network communication links in complex environments, comprising the following steps:

[0012] S1: Multimodal link status awareness, real-time collection of physical layer parameters, data link layer parameters, environmental electromagnetic parameters and service traffic characteristic data of communication nodes;

[0013] S2: Data preprocessing and feature engineering, which cleans, normalizes and extracts features from the collected raw data to generate standardized link state feature vectors;

[0014] S3: Spatiotemporal joint link state prediction, which inputs the standardized feature vector into the pre-trained spatiotemporal joint prediction model and outputs the link state prediction results within the next T time steps;

[0015] S4: Multi-objective reinforcement learning decision-making, inputting the link state prediction results and current service QoS requirements into the multi-objective deep reinforcement learning decision-making engine to generate the optimal cross-layer parameter adjustment strategy;

[0016] S5: Link Adaptive Execution and Feedback. Based on the generated adjustment strategy, the physical layer, data link layer, and network layer parameters of the communication link are adjusted synchronously, and the adjusted link performance data is fed back to the decision engine and prediction model for online updates.

[0017] It should be noted in the scheme that the physical layer parameters collected in step S1 include signal strength, signal-to-noise ratio, channel gain, Doppler shift, and bit error rate; the data link layer parameters include bandwidth utilization, frame loss rate, retransmission count, and queue length; the environmental electromagnetic parameters include interference signal strength, number of interference sources, and spectrum occupancy; and the service traffic characteristic data include data packet size, transmission rate, burstiness, and latency sensitivity.

[0018] It is worth noting that the data preprocessing in step S2 includes outlier removal, missing value filling, and data normalization. The data normalization adopts the max-min normalization method to map all feature values ​​to the [0,1] interval. The feature engineering adopts the principal component analysis (PCA) method to reduce the dimensionality of high-dimensional features and retains the principal components with a cumulative contribution rate greater than 95% as the final link state feature vector.

[0019] Furthermore, it should be noted that the spatiotemporal joint prediction model in step S3 is composed of a cascaded Long Short-Term Memory (LSTM) network and a Graph Attention (GAT) network, where LSTM is used to extract the time-series features of link states and GAT is used to extract the spatial correlation features of the communication network topology.

[0020] As a preferred implementation, the multi-objective deep reinforcement learning decision engine in step S4 adopts the Deep Deterministic Policy Gradient (DDPG) algorithm, whose reward function simultaneously considers four optimization objectives: throughput, transmission delay, packet loss rate, and device energy consumption.

[0021] In a preferred embodiment, the cross-layer parameters adjusted in step S5 include: modulation and coding scheme, transmit power and channel selection of the physical layer; frame length, retransmission mechanism and media access control (MAC) protocol parameters of the data link layer; and routing and traffic allocation strategies of the network layer.

[0022] As a preferred implementation, the method also includes an edge-cloud collaborative training step: performing real-time link status awareness, prediction, and decision-making at the edge nodes, collecting historical data from multiple edge nodes in the cloud for offline training and global updates of the model, and distributing the updated model parameters to the edge nodes.

[0023] As a preferred implementation, the reward function of the multi-objective deep reinforcement learning decision engine The specific calculation formula is as follows:

[0024]

[0025] in, for Network throughput at any given time This represents the theoretical maximum throughput.

[0026] for Average transmission delay at time, The maximum allowed latency for the business;

[0027] for Packet loss rate at any given moment;

[0028] for Real-time device energy consumption This represents the maximum allowable energy consumption of the equipment.

[0029] , , , These are the weighting coefficients for throughput, latency, packet loss rate, and energy consumption, respectively, and they satisfy... Each weighting coefficient can be dynamically adjusted according to the QoS requirements of different services.

[0030] Compared with existing technologies, the AI-based adaptive optimization method for network communication links in complex environments provided by this invention has at least the following beneficial effects:

[0031] (1) This invention simultaneously collects multi-dimensional data from the physical layer, data link layer, environment layer and business layer, which can comprehensively and accurately reflect the link status under complex environment, providing a rich information foundation for subsequent prediction and decision-making; and this invention combines LSTM and GAT to construct a spatiotemporal joint prediction model, which can simultaneously capture the temporal change law and spatial topological relationship of the link status, significantly improving the accuracy and reliability of link status prediction.

[0032] (2) The present invention adopts a multi-objective deep reinforcement learning algorithm, which can simultaneously optimize multiple conflicting objectives such as throughput, latency, packet loss rate and energy consumption, and realize cross-layer parameter coordination adjustment of physical layer, data link layer and network layer to maximize the overall network performance. At the same time, the present invention makes decisions based on the prediction results of future link status, which can take adjustment measures in advance, avoid the problem of response lag of traditional methods, and significantly improve the real-time performance and stability of the system.

[0033] (3) The present invention adopts an edge-cloud collaborative training and inference architecture, which not only ensures the low latency requirement of real-time decision-making, but also utilizes the powerful computing power of the cloud to perform global training and updating of the model, thereby improving the generalization ability and adaptability of the model. Attached Figure Description

[0034] Figure 1 This is a flowchart of an AI-based adaptive optimization method for network communication links in complex environments, according to the present invention. Detailed Implementation

[0035] The present invention will be further described below with reference to embodiments.

[0036] Please see Figure 1 This invention provides an AI-based adaptive optimization method for network communication links in complex environments, comprising the following steps:

[0037] S1: Multimodal link status awareness, real-time collection of physical layer parameters, data link layer parameters, environmental electromagnetic parameters and service traffic characteristic data of communication nodes.

[0038] Specifically, the physical layer parameters collected include signal strength, signal-to-noise ratio, channel gain, Doppler shift, and bit error rate;

[0039] Data link layer parameters include bandwidth utilization, frame loss rate, retransmission count, and queue length; environmental electromagnetic parameters include interference signal strength, number of interference sources, and spectrum occupancy.

[0040] Business traffic characteristic data includes packet size, transmission rate, burstiness, and latency sensitivity.

[0041] Furthermore, physical layer parameters: signal strength (RSSI) and signal-to-noise ratio (SNR) are collected every 10ms; channel gain, Doppler shift, and bit error rate are collected every 100ms.

[0042] Data link layer parameters: bandwidth utilization, frame loss rate, retransmission count, and MAC layer queue length are collected every 50ms;

[0043] Environmental electromagnetic parameters: interference signal strength, number of interference sources, and spectrum occupancy rate are collected every 200ms;

[0044] Business traffic characteristic data: data packet size distribution, average transmission rate, traffic burstiness, and business latency sensitivity are collected every 1 second;

[0045] All collected data are appended with a unified timestamp with an accuracy of 1ms and synchronized via the node's local NTP clock, with a synchronization error of ≤1ms.

[0046] Physical layer parameters are collected through the radio frequency transceiver module of the communication node; data link layer parameters are obtained through the driver layer of the network interface card; environmental electromagnetic parameters are collected through an external portable spectrum analyzer; and service traffic characteristic data are collected through kernel-mode traffic monitoring tools.

[0047] S2: Data preprocessing and feature engineering, which cleans, normalizes and extracts features from the collected raw data to generate standardized link state feature vectors.

[0048] Specifically, data preprocessing includes outlier removal, missing value imputation, and data normalization. Data normalization uses the max-min normalization method to map all feature values ​​to the [0,1] interval. Feature engineering uses principal component analysis (PCA) to reduce the dimensionality of high-dimensional features and retains the principal components with a cumulative contribution rate greater than 95% as the final link state feature vector.

[0049] The data normalization method, using the max-min normalization approach, is calculated using the following formula:

[0050]

[0051] In the formula, These are the original eigenvalues. and These are the minimum and maximum values ​​of this feature in historical data, respectively.

[0052] Furthermore, outlier removal adopts the 3σ principle. Data points that exceed the range of ±3 times the historical mean of the feature are identified as outliers and deleted directly.

[0053] S3: Spatiotemporal Joint Link State Prediction. The standardized feature vector is input into the pre-trained spatiotemporal joint prediction model, and the link state prediction results within the next T time steps are output.

[0054] Specifically, the spatiotemporal joint prediction model consists of a cascaded Long Short-Term Memory (LSTM) network and a Graph Attention (GAT) network. The LSTM is used to extract the temporal series features of the link states, while the GAT is used to extract the spatial correlation features of the communication network topology.

[0055] S4: Multi-objective reinforcement learning decision-making, inputs the link state prediction results and current service QoS requirements into the multi-objective deep reinforcement learning decision engine to generate the optimal cross-layer parameter adjustment strategy.

[0056] Specifically, the multi-objective deep reinforcement learning decision engine adopts the Deep Deterministic Policy Gradient (DDPG) algorithm, whose reward function simultaneously considers four optimization objectives: throughput, transmission delay, packet loss rate, and device energy consumption.

[0057] Furthermore, the reward function of the multi-objective deep reinforcement learning decision engine The specific calculation formula is as follows:

[0058]

[0059] in, for Network throughput at any given time This represents the theoretical maximum throughput.

[0060] for Average transmission delay at time, The maximum allowed latency for the business;

[0061] for Packet loss rate at any given moment;

[0062] for Real-time device energy consumption This represents the maximum allowable energy consumption of the equipment.

[0063] , , , These are the weighting coefficients for throughput, latency, packet loss rate, and energy consumption, respectively, and they satisfy... Each weighting coefficient can be dynamically adjusted according to the QoS requirements of different services.

[0064] S5: Link Adaptive Execution and Feedback. Based on the generated adjustment strategy, the physical layer, data link layer, and network layer parameters of the communication link are adjusted synchronously, and the adjusted link performance data is fed back to the decision engine and prediction model for online updates.

[0065] Specifically, the cross-layer parameters to be adjusted include: modulation and coding schemes, transmit power, and channel selection at the physical layer; frame length, retransmission mechanism, and Media Access Control (MAC) protocol parameters at the data link layer; and routing and traffic allocation strategies at the network layer.

[0066] In addition, the method of the present invention also includes an edge-cloud collaborative training step: performing real-time link status perception, prediction and decision-making at the edge nodes, collecting historical data from multiple edge nodes in the cloud for offline training and global updating of the model, and sending the updated model parameters to the edge nodes.

[0067] In summary, this invention addresses the industry pain points of link optimization in complex communication scenarios such as the Industrial Internet, the Internet of Vehicles, and emergency communications. It constructs a complete adaptive optimization system of "multimodal perception, spatiotemporal joint prediction, multi-objective cross-layer decision-making, and closed-loop feedback update." Through the organic integration of four-dimensional multimodal link state perception technology, LSTM-GAT cascaded spatiotemporal joint prediction technology, multi-objective deep reinforcement learning cross-layer decision-making technology, and edge-cloud collaborative training technology, it fundamentally solves the technical problems of poor adaptability, single optimization target, slow response, and lack of cross-layer collaboration in traditional link optimization methods.

[0068] This invention represents a shift from passive, reactive optimization to proactive, predictive optimization. It can simultaneously address multiple conflicting optimization objectives—network throughput, transmission latency, packet loss rate, and device power consumption—in complex environments characterized by severe electromagnetic interference, dynamic topology changes, and diverse service requirements, achieving global optimization of communication link performance. The invention's technical solution is highly versatile, compatible with existing mainstream communication protocols and hardware devices such as 5G / 6G, WiFi, and LoRa. It can be rapidly deployed without large-scale hardware modifications, demonstrating significant engineering application value and market potential. Furthermore, the invention's technical architecture possesses excellent scalability, allowing for further enhancement of optimization effects through the introduction of more advanced AI algorithms and communication technologies, thus adapting to the development needs of future next-generation communication networks.

[0069] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An AI-based adaptive optimization method for network communication links in complex environments, characterized in that: Includes the following steps: S1: Multimodal link status awareness, real-time collection of physical layer parameters, data link layer parameters, environmental electromagnetic parameters and service traffic characteristic data of communication nodes; S2: Data preprocessing and feature engineering, which cleans, normalizes and extracts features from the collected raw data to generate standardized link state feature vectors; S3: Spatiotemporal joint link state prediction, which inputs the standardized feature vector into the pre-trained spatiotemporal joint prediction model and outputs the link state prediction results within the next T time steps; S4: Multi-objective reinforcement learning decision-making, inputting the link state prediction results and current service QoS requirements into the multi-objective deep reinforcement learning decision-making engine to generate the optimal cross-layer parameter adjustment strategy; S5: Link Adaptive Execution and Feedback. Based on the generated adjustment strategy, the physical layer, data link layer, and network layer parameters of the communication link are adjusted synchronously, and the adjusted link performance data is fed back to the decision engine and prediction model for online updates.

2. The AI-based adaptive optimization method for network communication links in complex environments according to claim 1, characterized in that: The physical layer parameters collected in step S1 include signal strength, signal-to-noise ratio, channel gain, Doppler shift, and bit error rate; the data link layer parameters include bandwidth utilization, frame loss rate, retransmission count, and queue length; the environmental electromagnetic parameters include interference signal strength, number of interference sources, and spectrum occupancy; and the service traffic characteristic data include data packet size, transmission rate, burstiness, and latency sensitivity.

3. The AI-based adaptive optimization method for network communication links in complex environments according to claim 1, characterized in that: The data preprocessing in step S2 includes outlier removal, missing value filling and data normalization. The data normalization adopts the max-min normalization method to map all feature values ​​to the [0,1] interval. Feature engineering uses principal component analysis to reduce the dimensionality of high-dimensional features, retaining principal components with a cumulative contribution rate greater than 95% as the final link state feature vector.

4. The AI-based adaptive optimization method for network communication links in complex environments according to claim 1, characterized in that: The spatiotemporal joint prediction model in step S3 consists of a cascaded Long Short-Term Memory network and a Graph Attention network, where LSTM is used to extract the time-series features of link states and GAT is used to extract the spatial correlation features of the communication network topology.

5. The AI-based adaptive optimization method for network communication links in complex environments according to claim 1, characterized in that: The multi-objective deep reinforcement learning decision engine in step S4 adopts a deep deterministic policy gradient algorithm, whose reward function simultaneously considers four optimization objectives: throughput, transmission delay, packet loss rate, and device energy consumption.

6. The AI-based adaptive optimization method for network communication links in complex environments according to claim 1, characterized in that: The cross-layer parameters adjusted in step S5 include: modulation and coding scheme, transmit power, and channel selection at the physical layer; frame length, retransmission mechanism, and media access control protocol parameters at the data link layer; and routing and traffic allocation strategies at the network layer.

7. The AI-based adaptive optimization method for network communication links in complex environments according to claim 1, characterized in that: It also includes edge-cloud collaborative training steps: performing real-time link status awareness, prediction and decision-making at edge nodes, collecting historical data from multiple edge nodes in the cloud for offline training and global updates of the model, and distributing the updated model parameters to the edge nodes.

8. The AI-based adaptive optimization method for network communication links in complex environments according to claim 5, characterized in that: The reward function of the multi-objective deep reinforcement learning decision engine The specific calculation formula is as follows:

9. Among them, for Network throughput at any given time This represents the theoretical maximum throughput. for Average transmission delay at time, The maximum allowed latency for the business; for Packet loss rate at any given moment; for Real-time device energy consumption This represents the equipment's maximum permissible energy consumption. , , , These are the weighting coefficients for throughput, latency, packet loss rate, and energy consumption, respectively, and they satisfy... Each weighting coefficient can be dynamically adjusted according to the QoS requirements of different services.

Images