A method for protocol optimization and dynamic routing selection for network communication systems
By optimizing the communication protocol through deep learning models and dynamic routing selection algorithms, the problems of excessive bandwidth resource occupation and energy consumption of traditional methods in complex network environments are solved, and efficient and stable network communication is achieved.
Patent Information
- Application Number
- CN202510329864.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-20
- Publication Date
- 2025-09-30
- Estimated Expiration
- 2045-03-20
AI Technical Summary
Traditional communication protocols and routing methods are difficult to meet the requirements of efficiency, stability and security in large-scale, highly concurrent and changing network scenarios, leading to problems such as congestion, delay and packet loss, especially in complex network environments where bandwidth resources are excessively occupied and energy consumption increases.
A deep learning model (GCN combined with TCN) is used to predict future traffic trends, and a dynamic routing algorithm (distance vector and link state routing algorithm) is combined to optimize the communication protocol. The deep learning model is used to predict future traffic trends and dynamically adjust parameters. Distributed computing and broadcast mechanisms are combined to optimize routing selection, realizing dynamic routing selection and parameter adjustment.
It improves the transmission efficiency of network communications, reduces latency and packet loss, enhances network stability and resource utilization, and provides a better network experience.
Smart Images
Figure CN120110977B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of communications, and in particular to a method for performing protocol optimization and dynamic routing selection on a network communication system. Background Art
[0002] With the rapid development of network technology, network communications have become an integral part of modern society, and the security and efficiency of data transmission have become increasingly important research topics. However, in complex network environments, traditional communication protocols and routing methods often struggle to meet the requirements of efficiency, stability, and security. Particularly in large-scale, highly concurrent, and highly dynamic network scenarios, traditional communication protocols and routing methods are prone to congestion, latency, and packet loss, severely impacting the quality and efficiency of network communications. In particular, existing dynamic routing protocols periodically broadcast HELLO and TC messages to maintain neighbor relationships and update routing tables. However, in large-scale networks, this broadcast mechanism consumes significant bandwidth resources and increases energy consumption. Summary of the Invention
[0003] In order to solve the above technical problems, the present invention adopts a technical solution: a method for performing protocol optimization and dynamic routing selection on a network communication system, comprising the following steps:
[0004] S01. Detect network status, including bandwidth usage, packet loss rate, round-trip delay, queue length, transmission control window, traffic direction, and network topology.
[0005] S02: Determine whether protocol optimization is required. If so, perform protocol optimization. If not, execute step S03.
[0006] Protocol optimization is accomplished through a deep learning model that uses a GCN combined with a TCN architecture. The first layer is GCN, and the second layer is TCN. All network nodes and their connections are first input into the GCN, which then represents them as a graph structure. This aggregates each node's adjacency information and generates node features containing local topological characteristics. The GCN-generated node features are then input into the TCN as time series data. The TCN's dilated convolution operation is used to capture long-term temporal dependencies, thereby predicting future traffic trends.
[0007] S03. Determine whether dynamic routing is required. If so, perform dynamic routing. If not, update network status information.
[0008] Dynamic routing is achieved through distance vector routing or link state routing. In the distance vector routing algorithm, each router maintains a vector table that records the best exit and distance to each destination. The router periodically sends update information to its neighbors and updates its own vector table based on the received information. The link state routing algorithm uses open shortest path first. Each router discovers its neighbors and measures the distance to them, then broadcasts link state information. All routers build a complete network topology based on the collected link state information and use the shortest path algorithm to calculate the shortest path to all destinations.
[0009] S04. Continue to monitor the network status until the system is terminated or the exception processing is completed.
[0010] Furthermore, step S02 also includes dynamic parameter adjustment, setting a hierarchical parameter adjustment mechanism, including first-level parameter adjustment and second-level parameter adjustment. When the deep learning model predicts that the future traffic trend is about to be bandwidth saturated or delay exceeded, the first-level parameter adjustment is performed first, including dynamically reducing the TCP window size or enabling adaptive retransmission control, and the adaptive retransmission control is to adjust the retransmission time interval; when the first-level parameter adjustment does not meet the requirements, the second-level parameter adjustment is started, including adaptive switching of the compression algorithm or segmentation management of the data packet. The adaptive switching of the compression algorithm is realized according to the bandwidth and traffic thresholds. The segmentation management of the data packet is to set multiple data packet size thresholds, and the data packet is split when the data packet size threshold is reached.
[0011] Furthermore, a compression algorithm library is set up, and multiple compression algorithms are preset, including zlib, LZ4, and Brotli. Adaptive switching of compression algorithms is achieved according to the thresholds of bandwidth and traffic. When the bandwidth is less than the set threshold but the processing speed is greater than the set threshold, LZ4 is selected. When the bandwidth is greater than the set threshold and the traffic is greater than the set threshold, Brotli is selected.
[0012] Furthermore, in order to achieve dynamic parameter adjustment, the real-time network status is collected after each parameter adjustment, and the parameter adjustment strategy is optimized based on the feedback reinforcement learning algorithm. The implementation process of the reinforcement learning algorithm is as follows: define the network status, that is, the initial collection data type, including bandwidth and packet loss rate. The action set includes the selection of compression algorithm, adjustment of TCP window size, and segmented sending of data packets. After the network communication system takes action, the reward value is calculated based on the improvement of the network status. If the delay or packet loss rate is reduced, a positive reward is obtained, otherwise a negative reward is obtained. After each action is executed, the Q value table is updated through reinforcement learning. When encountering similar states subsequently, the optimal adjustment strategy is automatically selected.
[0013] Furthermore, both GCN and TCN are 2-layer.
[0014] Furthermore, the training data for the deep learning model comes from the network traffic analysis tool NetFlow and the router log system.
[0015] Furthermore, the method also optimizes the broadcast mechanism according to the node status, including the following steps:
[0016] S21. Each node continuously monitors its own status and receives the status of neighboring nodes, including power and traffic load. When the node status changes or receives specific trigger information, it determines whether the broadcast trigger condition is met;
[0017] S22. When the broadcast trigger condition is met, the node will generate and broadcast a HELLO or TC message containing key status information, including node ID, power status, load information, neighbor node list and update timestamp;
[0018] S23. Dynamically adjust the broadcast cycle according to the network change frequency and node status. When the network topology change frequency is less than the set value, extend the broadcast cycle. When the network topology change frequency is greater than the set value, shorten the broadcast cycle.
[0019] Furthermore, the fast convergence mechanism adopts a distributed computing approach, where each node independently calculates the routing table based on the information provided by neighboring nodes, and transmits the updated routing information to other nodes through broadcasting.
[0020] Furthermore, when calculating the routing table, each node not only selects the optimal path as the main path, but also selects several suboptimal paths as alternative paths based on the objective function value. When the main path fails or is congested, the node switches to the alternative path for data transmission.
[0021] Furthermore, the node continuously monitors the network status and routing performance indicators and adjusts the routing strategy and weight coefficients in real time based on the monitoring results. When the delay, packet loss rate or bandwidth utilization rate exceeds the set threshold, the routing adjustment operation is triggered. First, the weight coefficient of the affected path is updated through the dynamic weighting algorithm. The adjusted weight coefficient will be input into the weighted shortest path algorithm, and the optimal path will be recalculated according to the new weight. For multiple paths with similar weights, the system adopts an equal-cost load sharing strategy to evenly distribute traffic to avoid congestion on a single path.
[0022] Beneficial effects of the present invention: The present invention integrates a network status monitoring module inside the communication device to obtain real-time mobile network status information. The communication device integrates a routing calculation and selection module inside to calculate candidate paths and perform performance evaluation based on network status information, and select the optimal path for data transmission. Through actual application tests, the present invention can effectively improve the transmission efficiency of network communications, reduce latency and packet loss rate, and improve network stability and reliability. At the same time, the dynamic routing selection mechanism can adjust the route in real time according to the network status, further improving the utilization rate of network resources and overall performance. It is expected that the implementation of the present invention will bring users a better network usage experience and contribute to the technological progress and development in the field of network communications. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] Figure 1 The following is a flow chart of the method described in the embodiment. DETAILED DESCRIPTION
[0024] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.
[0025] Example 1
[0026] This embodiment discloses a method for performing protocol optimization and dynamic routing selection on a network communication system. Figure 1 As shown, the following steps are included:
[0027] S01, start, system initialization, check network status;
[0028] S02. Determine whether protocol optimization is required based on the detected network status. If so, perform protocol optimization. If not, execute step S03.
[0029] S03. Determine whether dynamic routing is required. If so, perform dynamic routing. If not, update network status information.
[0030] S04. Continue to monitor the network status until the system is terminated or the exception processing is completed.
[0031] In this embodiment, the network status detected in step S01 includes collecting traffic patterns and status characterization data from network devices. The traffic patterns and status characterization data include bandwidth usage, packet loss rate, round-trip delay, queue length, transmission control window, and traffic direction. Communication protocol optimization is performed based on these data. The optimization process is as follows:
[0032] 1. Deep Learning Model Construction and Training. The deep learning model uses a GCN combined with TCN architecture, with the GCN as the first layer and the TCN as the second layer. Both GCN and TCN are two layers. First, all network nodes and their connections are input into the GCN. The GCN represents all network nodes and their connections as a graph structure, aggregating each node's adjacency information and generating node features containing local topological characteristics. The node features generated by the GCN are then input into the TCN as time series data. The TCN's dilated convolution operation is used to capture long-term temporal dependencies, thereby predicting future traffic trends.
[0033] 2. Dynamic parameter adjustment. A hierarchical parameter adjustment mechanism is set up, including first-level parameter adjustment and second-level parameter adjustment. When the deep learning model predicts that the future traffic trend is about to be bandwidth saturated or the delay exceeds the limit, the first-level parameter adjustment is performed first, including dynamically reducing the TCP window size or enabling adaptive retransmission control. The adaptive retransmission control is to adjust the retransmission time interval; when the first-level parameter adjustment does not meet the requirements, the second-level parameter adjustment is started, including adaptive switching of the compression algorithm or segmentation management of the data packet. The adaptive switching of the compression algorithm is realized according to the bandwidth and traffic thresholds. The segmentation management of the data packet is to set multiple data packet size thresholds, and the data packet is split when the size threshold is reached.
[0034] In this embodiment, a compression algorithm library is set up, and multiple compression algorithms are preset, including zlib, LZ4, and Brotli. Adaptive switching of the compression algorithm is achieved according to the thresholds of bandwidth and traffic. When the bandwidth is less than the set threshold but the processing speed is greater than the set threshold, LZ4 is selected. When the bandwidth is greater than the set threshold and the traffic is greater than the set threshold, Brotli is selected.
[0035] In order to achieve dynamic parameter adjustment, the real-time network status is collected after each parameter adjustment, and the parameter adjustment strategy is optimized based on the feedback reinforcement learning algorithm. The implementation process of the reinforcement learning algorithm is as follows: define the network status, that is, the initial collection data type, including bandwidth and packet loss rate. The action set includes the selection of compression algorithm, adjustment of TCP window size, and segmented sending of data packets. After the network communication system takes action, the reward value is calculated based on the improvement of the network status. If the delay or packet loss rate is reduced, a positive reward is obtained, otherwise a negative reward is obtained. After each action is executed, the Q value table is updated through reinforcement learning. When encountering similar states subsequently, the optimal adjustment strategy is automatically selected.
[0036] In this embodiment, the training data comes from the network traffic analysis tool NetFlow and the router log system.
[0037] In this embodiment, detecting network status also involves each node continuously monitoring its own status (such as battery level and traffic load) and receiving the status of neighboring nodes. The node's own status and neighboring node status include battery level and traffic load. When a node's status changes (such as battery level falling below a threshold or traffic load increasing) or receives specific trigger information (such as a neighboring node status change notification), it determines whether a broadcast trigger condition has been met. When the broadcast trigger condition is met, the node generates and broadcasts a HELLO or TC message containing key status information, including the node ID, battery level, load information, neighbor node list, and update timestamp, to reduce the amount of transmitted data. The broadcast period is dynamically adjusted based on the frequency of network changes and node status. The broadcast period is extended when the frequency of network topology changes is less than a set value, and shortened when the frequency of network topology changes is greater than a set value.
[0038] In this embodiment, each node continuously monitors its own status through the following implementations: power consumption, CPU / memory monitoring modules, traffic monitoring modules, and traffic monitoring sensors (NetFlow). The power consumption sensor monitors the node's power status; network devices have built-in CPU and memory monitoring capabilities, and the system monitoring module obtains real-time usage of these key resources; NetFlow collects real-time traffic load information. Because sensor standards vary across devices, and sensor accuracy and real-time performance may affect monitoring effectiveness, the system can adopt an architecture based on virtual sensors and software-defined monitoring (SDM) to enhance feasibility and adaptability. Virtual sensors simulate power consumption, load, and temperature status by collecting existing device data (such as network traffic records), reducing reliance on physical sensors. SDM, on the other hand, centrally controls and dynamically configures monitoring policies for different devices to achieve cross-device data unification and management.
[0039] The key status information in the HELLO or TC message is interpreted as follows:
[0040] Node ID, used by neighboring nodes to accurately identify the source of information;
[0041] Battery status, showing the current battery percentage or low battery warning, so that neighboring nodes can consider the node's battery life when selecting routes;
[0042] Load information (CPU / memory usage) helps neighboring nodes evaluate the processing capacity of the node;
[0043] Bandwidth usage, which indicates the node's current bandwidth usage to avoid data loss or delay caused by congestion;
[0044] Neighbor node list, which provides the current connection relationship of nodes and helps optimize routing calculation of network topology;
[0045] Update timestamps to ensure that neighboring nodes can recognize the real-time nature of the information.
[0046] In this embodiment, detecting the network status also includes detecting the network topology. When the network topology changes, the routing table is updated through a fast convergence mechanism, thereby achieving dynamic routing selection.
[0047] In this embodiment, an objective function is constructed to evaluate the pros and cons of different paths. The design of the objective function should be able to reflect the goals of network optimization, such as minimizing delay, maximizing bandwidth utilization, and improving network reliability. This embodiment constructs the objective function based on round-trip delay, bandwidth, and packet loss rate:
[0048] ;
[0049] Where n is the number of total packets or transmission paths, representing the total number of all delay terms that need to be considered.
[0050] Represents the delay value of the i-th path or the i-th data packet. These are the delay times for a specific path or packet, measured in milliseconds (ms). α is the weight for the delay term; β is the weight for the bandwidth utilization term, which represents the ratio of used bandwidth to total available bandwidth; and γ is the weight for the packet loss rate term, which represents the ratio of packets lost during transmission to the total number of packets sent. Indicates the bandwidth used. is the total available bandwidth, Indicates the packet loss rate.
[0051] After determining the objective function, it is necessary to select an appropriate algorithm to implement dynamic routing. This embodiment implements dynamic routing through a distance vector routing algorithm or a link state routing algorithm. The distance vector routing algorithm uses a routing information protocol. Each router maintains a vector table that records the best exit and distance to each destination. The router periodically sends update information to its neighbors and updates its own vector table based on the received information. The link state routing algorithm uses an open shortest path first method. Each router discovers its neighbors and measures the distance to them, then broadcasts link state information. All routers construct a complete network topology based on the collected link state information and use the shortest path algorithm to calculate the shortest path to all destinations. The fast convergence mechanism uses a distributed computing method. Each node independently calculates the routing table based on the information provided by neighboring nodes and transmits the updated routing information to other nodes through broadcasting.
[0052] In this embodiment, a fast convergence mechanism rapidly updates routing tables when network topology changes, reducing route oscillation and packet loss. This fast convergence mechanism is implemented through event-driven triggering. When a topology change (such as a link disconnection or a new node joining) is detected using Link State Advertisements (LSAs), routing updates are initiated immediately, rather than relying on periodic updates at fixed intervals. This event-driven mechanism enables instant propagation of routing changes, avoiding the latency associated with traditional periodic updates. Each node periodically broadcasts Link State Advertisements (LSAs), which contain its own link information and information about connected neighbors. When a link state changes, the node immediately generates a new LSA and broadcasts it to other nodes. The convergence mechanism utilizes a distributed computing approach, with each node independently calculating its routing table based on information provided by neighboring nodes and broadcasting the updated routing information to other nodes. In a distributed environment, each node independently performs routing calculations and table updates, independent of a central control node. Each node independently generates a topology map and uses a distributed Dijkstra algorithm to parallelly compute the shortest paths to other nodes. In distance vector routing algorithms (such as RIP), nodes iteratively calculate the optimal path based on updates from neighboring nodes using the distributed Bellman-Ford algorithm. When topology changes occur, only the changed information is broadcast, rather than the entire routing table. Using differential updates significantly reduces the amount of broadcast data and speeds up information dissemination. For example, if a link is disconnected, a node only needs to notify its neighbors that the link is unreachable, eliminating the need to rebroadcast all path information. Furthermore, using compressed broadcast technology, only critical information (such as the changed node ID and the delay of the changed path) is included in the broadcast, reducing data volume and improving dissemination efficiency.
[0053] To address network failures and congestion, the present invention introduces an alternative path management mechanism. When calculating the routing table, each node not only selects the optimal path as the primary path but also selects several suboptimal paths as alternative paths based on the objective function value. When the primary path fails or becomes congested, the node switches to the alternative path for data transmission.
[0054] Nodes continuously monitor network status and routing performance metrics and adjust routing strategies and weights in real time based on these results. When latency, packet loss, or bandwidth utilization exceed set thresholds (e.g., latency exceeding 200ms, packet loss exceeding 5%, or bandwidth utilization exceeding 90%), routing adjustments are triggered. A dynamic weighting algorithm is first used to update the weights of affected paths. For example, paths with high latency or high packet loss are weighted higher, thereby reducing their priority in routing. The adjusted weights are then fed into a weighted shortest path algorithm, which recalculates the optimal path based on the new weights. For multiple paths with similar weights, the system employs an equal-cost load balancing (ECMP) strategy to evenly distribute traffic and avoid congestion on a single path. Furthermore, the system can adaptively adjust trigger thresholds based on historical data to adapt to changes in network traffic, further improving the accuracy and flexibility of these adjustments.
[0055] This invention uses deep learning and big data analysis technologies to intelligently optimize protocols used in network communications. By analyzing historical communication data, it learns and predicts information such as traffic patterns and congestion conditions in network communications, thereby dynamically adjusting communication protocol parameters and improving the efficiency and stability of network communications.
[0056] Dynamic routing is achieved through the use of multi-objective optimization algorithms and real-time network monitoring technology. By monitoring network status in real time, such as bandwidth, latency, and packet loss rate, and combining objective functions (such as minimizing transmission time and maximizing transmission efficiency), the optimal routing path is calculated and packets are transmitted along that path. By optimizing the protocol's packet structure and transmission mechanism, redundant data is reduced and data transmission efficiency is improved.
[0057] The above description is only the basic principle and preferred embodiments of the present invention. Improvements and substitutions made by those skilled in the art based on the present invention fall within the protection scope of the present invention.
Claims
1. A method for performing protocol optimization and dynamic routing selection on a network communication system, characterized in that: The following steps are involved: S01. Detect network status, including bandwidth usage, packet loss rate, round-trip delay, queue length, transmission control window, traffic direction, and network topology. S02: Determine whether protocol optimization is required. If so, perform protocol optimization. If not, execute step S03. Protocol optimization is accomplished through a deep learning model that uses a GCN combined with a TCN architecture. The first layer is GCN, and the second layer is TCN. All network nodes and their connections are first input into the GCN, which then represents them as a graph structure. This aggregates each node's adjacency information and generates node features containing local topological characteristics. The GCN-generated node features are then input into the TCN as time series data. The TCN's dilated convolution operation is used to capture long-term temporal dependencies, thereby predicting future traffic trends. S03. Determine whether dynamic routing is required. If so, perform dynamic routing. If not, update network status information. Dynamic routing is achieved through distance vector routing or link state routing. In the distance vector routing algorithm, each router maintains a vector table that records the best exit and distance to each destination. The router periodically sends update information to its neighbors and updates its own vector table based on the received information. The link state routing algorithm uses open shortest path first. Each router discovers its neighbors and measures the distance to them, then broadcasts link state information. All routers build a complete network topology based on the collected link state information and use the shortest path algorithm to calculate the shortest path to all destinations. S04. Continue to monitor the network status until the end.
2. The method for performing protocol optimization and dynamic routing selection for a network communication system according to claim 1, wherein: Step S02 also includes dynamic parameter adjustment, setting a hierarchical parameter adjustment mechanism, including first-level parameter adjustment and second-level parameter adjustment. When the deep learning model predicts that the future traffic trend is about to be bandwidth saturated or delay exceeded, the first-level parameter adjustment is performed first, including dynamically reducing the TCP window size or enabling adaptive retransmission control, and the adaptive retransmission control is to adjust the retransmission time interval; when the first-level parameter adjustment does not meet the requirements, the second-level parameter adjustment is started, including adaptive switching of the compression algorithm or segmentation management of the data packet. The adaptive switching of the compression algorithm is realized according to the bandwidth and traffic thresholds. The segmentation management of the data packet is to set multiple data packet size thresholds, and the data packet is split when the data packet size threshold is reached.
3. The method for performing protocol optimization and dynamic routing selection for a network communication system according to claim 1, wherein: Set the compression algorithm library and preset multiple compression algorithms, including zlib, LZ4, and Brotli. Adaptive switching of compression algorithms is implemented based on bandwidth and traffic thresholds. When the bandwidth is lower than the set threshold but the processing speed is higher than the set threshold, LZ4 is selected. When the bandwidth and traffic are both higher than the set threshold, Brotli is selected.
4. The method for performing protocol optimization and dynamic routing selection for a network communication system according to claim 1, wherein: In order to achieve dynamic parameter adjustment, the real-time network status is collected after each parameter adjustment, and the parameter adjustment strategy is optimized based on the feedback reinforcement learning algorithm. The implementation process of the reinforcement learning algorithm is as follows: define the network status, that is, the initial collection data type, including bandwidth and packet loss rate. The action set includes the selection of compression algorithm, adjustment of TCP window size, and segmented sending of data packets. After the network communication system takes action, the reward value is calculated based on the improvement of the network status. If the delay or packet loss rate is reduced, a positive reward is obtained, otherwise a negative reward is obtained. After each action is executed, the Q value table is updated through reinforcement learning. When encountering similar states subsequently, the optimal adjustment strategy is automatically selected.
5. The method for performing protocol optimization and dynamic routing selection for a network communication system according to claim 1, wherein: Both GCN and TCN have 2 layers.
6. The method for performing protocol optimization and dynamic routing selection on a network communication system according to claim 1, wherein: The training data for the deep learning model comes from the network traffic analysis tool NetFlow and the router log system.
7. The method for performing protocol optimization and dynamic routing selection on a network communication system according to claim 1, wherein: This method also optimizes the broadcast mechanism according to the node status, including the following steps: S21. Each node continuously monitors its own status and receives the status of neighboring nodes, including power and traffic load. When the node status changes or receives specific trigger information, it determines whether the broadcast trigger condition is met; S22. When the broadcast trigger condition is met, the node will generate and broadcast a HELLO or TC message containing key status information, including node ID, power status, load information, neighbor node list and update timestamp; S23. Dynamically adjust the broadcast cycle according to the network change frequency and node status. When the network topology change frequency is less than the set value, extend the broadcast cycle. When the network topology change frequency is greater than the set value, shorten the broadcast cycle.
8. The method for performing protocol optimization and dynamic routing selection on a network communication system according to claim 1, wherein: The fast convergence mechanism adopts a distributed computing method. Each node independently calculates the routing table based on the information provided by neighboring nodes, and transmits the updated routing information to other nodes through broadcasting.
9. The method for performing protocol optimization and dynamic routing selection on a network communication system according to claim 1, wherein: When calculating the routing table, each node not only selects the optimal path as the main path, but also selects several suboptimal paths as alternative paths based on the objective function value. When the main path fails or is congested, the node switches to the alternative path for data transmission.
10. The method for performing protocol optimization and dynamic routing selection on a network communication system according to claim 1, wherein: The node continuously monitors network status and routing performance indicators and adjusts routing strategies and weight coefficients in real time based on the monitoring results. When the delay, packet loss rate or bandwidth utilization rate exceeds the set threshold, the routing adjustment operation is triggered. First, the weight coefficient of the affected path is updated through a dynamic weighting algorithm. The adjusted weight coefficient will be input into the weighted shortest path algorithm, and the optimal path will be recalculated based on the new weight. For multiple paths with similar weights, the system adopts an equal-cost load sharing strategy to evenly distribute traffic to avoid congestion on a single path.