A communication system based on high-speed signal connection switch dividing VLAN into network cards

Abstract

The present invention relates to the technical field of virtual local area networks, and specifically to a communication system for dividing VLANs into network cards based on high-speed signal connection switches. The system includes a network status monitoring module, a traffic prediction module, a resource allocation decision module, a traffic control execution module, a behavior pattern analysis module, a VLAN configuration adjustment module, a resource allocation optimization module, and a network configuration adaptive adjustment module. In the present invention, software-defined network technology and time series analysis are combined, anomaly detection and long short-term memory network convolutional neural network are used to predict traffic, bandwidth, priority and routing are adjusted through optimization theory and reinforcement learning, network flexibility is enhanced, software-defined network technology improves policy deployment efficiency, recurrent neural network and long short-term memory network deepen behavior recognition, and the minimum cut maximum flow algorithm is used to optimize VLAN configuration and improve resource utilization. The network adaptive adjustment adopts genetic algorithm and simulated annealing to optimize routing and automatically adapt to QoS strategy and enhance adaptive capability.

Description

Technical Field

[0001] The invention relates to the technical field of virtual local area network, and in particular to a communication system for dividing VLAN into network cards based on a high-speed signal connection switch. Background Art

[0002] Virtual Local Area Network (VLAN) technology belongs to the field of network engineering and management, especially in the design and application of Local Area Network (LAN). VLAN allows network administrators to create multiple logical segments on a physical network, each segment operates as an independent network domain, increasing the flexibility and security of the network. This technology is implemented by configuring VLAN identifiers on network devices such as switches, so that data packets from different VLANs can be transmitted in isolation on the same physical network. In this way, VLAN technology can reduce broadcast domains, improve network efficiency, and enhance data security and isolation. The configuration and management of VLAN requires network devices that support VLAN functions and corresponding network management software. Through these devices and software, the logical structure of the network can be flexibly designed and adjusted to meet different network application requirements.

[0003] Among them, the communication system based on high-speed signal connection switch to divide VLAN into network cards is an implementation method of creating multiple logical networks within a physical network using VLAN technology. The purpose of this system is to divide the high-speed signal connection into different virtual networks (VLANs) by configuring the switch, thereby providing an independent network environment for each network card connected to the switch. The effects of doing so include improving network security because data is isolated between different VLANs; increasing network flexibility because the network can be logically divided as needed without changing the physical connection; and optimizing network performance, alleviating network congestion by reducing unnecessary data broadcasts.

[0004] In actual operation, traditional systems often lack real-time and accurate network status monitoring, making it difficult to effectively predict future traffic patterns and demands, resulting in inflexible or inaccurate responses to resource allocation and traffic control strategies. In addition, traditional network configuration adjustments often rely on manual intervention and lack efficient automation tools, making network optimization responses slow and unable to adapt to traffic changes and demand fluctuations in a timely manner. The lack of in-depth behavioral pattern analysis makes network management and optimization lack of pertinence, resulting in unreasonable allocation of network resources, affecting the overall performance of the network and data transmission efficiency. These deficiencies ultimately lead to a decline in network service quality, impaired user experience, and increased network maintenance and management costs. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings in the prior art and to propose a communication system based on high-speed signal connection switch to divide VLAN into network cards.

[0006] In order to achieve the above object, the present invention adopts the following technical scheme: a communication system based on high-speed signal connection switch dividing VLAN into network cards includes a network status monitoring module, a traffic prediction module, a resource allocation decision module, a traffic control execution module, a behavior pattern analysis module, a VLAN configuration adjustment module, a resource allocation optimization module, and a network configuration adaptive adjustment module;

[0007] The network status monitoring module is based on software-defined network technology, uses time series analysis and anomaly detection algorithms to monitor network bandwidth utilization, delay and packet loss rate indicators in real time, analyzes network behavior, and generates network performance indicators;

[0008] The traffic prediction module uses a long short-term memory network model or a convolutional neural network model to perform forecast analysis on future traffic patterns and demands based on network performance indicators, and generates traffic trend forecasts;

[0009] The resource allocation decision module uses optimization theory and reinforcement learning methods based on traffic trend prediction to dynamically adjust bandwidth allocation, priority and routing strategies through PID controller or deep reinforcement learning model to generate dynamic adjustment strategies;

[0010] The flow control execution module deploys and adjusts the flow control strategy based on the dynamic adjustment strategy and uses software defined network technology, automatically adjusts the network configuration according to the real-time data and prediction results, and generates flow scheduling instructions;

[0011] The behavior pattern analysis module uses a recurrent neural network model or a long short-term memory network model based on network performance indicators to identify and analyze behavior patterns in network traffic and generate behavior pattern insights;

[0012] The VLAN configuration adjustment module uses the minimum cut maximum flow algorithm based on behavioral pattern insights to dynamically optimize VLAN division and network partitioning, improve the utilization efficiency of network resources and data transmission rate, and generate a network partition optimization solution;

[0013] The resource allocation optimization module uses Q learning or deep Q network model based on the network partition optimization solution to adjust the network resource allocation strategy in real time and generate resource optimization configuration;

[0014] The network configuration adaptive adjustment module is based on resource optimization configuration, adopts genetic algorithm and simulated annealing algorithm to optimize the flow distribution and route selection in the network, automatically adjusts the QoS strategy, matches the changes in network conditions, and generates an adaptive configuration strategy.

[0015] As a further solution of the present invention, the network performance indicators are specifically real-time data and trend analysis of network bandwidth utilization, delay indicators, and packet loss rate performance indicators; the traffic trend prediction includes future traffic trends, peak predictions, and demand change analysis; the dynamic adjustment strategy is specifically a dynamic bandwidth allocation scheme, priority adjustment rules, and routing selection strategies; the traffic scheduling instructions are specifically network configuration adjustment instructions and traffic scheduling instructions; the behavior pattern insights include periodic changes in network usage and abnormal traffic behavior pattern analysis; the network partition optimization scheme is specifically a VLAN division scheme and a network partition optimization strategy; the resource optimization configuration is specifically bandwidth allocation optimization and routing selection optimization strategies; the adaptive configuration strategy is specifically traffic engineering optimization and dynamic QoS adjustment rules.

[0016] As a further solution of the present invention, the network status monitoring module includes a time series analysis submodule, an anomaly detection submodule, and a real-time monitoring submodule;

[0017] The time series analysis submodule is based on software defined network technology, collects network bandwidth utilization bandwidth_usage, delay latency, and packet loss rate packet_loss indicators, adopts an autoregressive moving average model, calls the ARIMA function through the Python statsmodels library, sets the model parameter order=(2,1,2), where the first term 2 represents the order of the autoregressive term AR, 1 represents the difference order I, and the second term 2 represents the order of the moving average term MA, performs the fitting operation of the time series data, and generates a trend analysis of network indicators;

[0018] The anomaly detection submodule uses the isolation forest algorithm based on network indicator trend analysis, and uses the IsolationForest function of sklearn.ensemble to set the parameter n_estimators=100 to represent the number of trees in the forest, and contamination=auto to automatically estimate the proportion of abnormal points in the data set, perform the training operation of the isolation forest model, and generate abnormal behavior recognition;

[0019] The real-time monitoring submodule is based on abnormal behavior identification and uses software-defined network technology to set the data collection frequency parameter to frequency=5, which means that network performance data is collected every 5 seconds. The monitor function of the Traffic Monitor tool is used to set the monitoring parameter interval=5, and network performance data is collected every 5 seconds, thereby performing real-time data monitoring and generating network performance indicators.

[0020] As a further solution of the present invention, the traffic prediction module includes an LSTM prediction submodule, a CNN prediction submodule, and a traffic pattern analysis submodule;

[0021] The LSTM prediction submodule uses a long short-term memory network model based on network performance indicator data, and uses the TensorFlow library to build an LSTM network structure, including setting the number of hidden layer units to 128 to capture the long-term dependencies of the time series, setting the learning rate to 0.001 to control the speed of the learning process, setting the batch size to 32 to balance the training efficiency and network performance, using the Adam optimizer for parameter optimization, and using the mean_squared_error loss function to evaluate the difference between the model prediction value and the true value to generate a time series prediction analysis;

[0022] The CNN prediction submodule adopts a convolutional neural network model based on the time series prediction analysis results, and uses the PyTorch library to define the CNN model structure, including adding a Conv1D layer, setting the number of filters to 64 to extract local features of traffic data, setting the kernel size to 3 to capture the relationship between adjacent data points, selecting ReLU as the activation function to introduce nonlinearity, applying the MaxPooling1D layer for feature downsampling, and using Adam as the optimizer. The loss function selects cross_entropy to handle classification problems and generate local feature analysis results;

[0023] The traffic pattern analysis submodule is based on the time series prediction analysis results and the local feature analysis results, and comprehensively applies statistical analysis techniques. It performs data operations through Python's NumPy library, calculates the moving average to determine the overall trend of the traffic, sets the window size to 7 days to capture traffic fluctuations within a week, and calculates the standard deviation to identify abnormal traffic fluctuations. It generates a traffic trend forecast by calling the mean and std functions of the NumPy library.

[0024] As a further solution of the present invention, the resource allocation decision module includes an optimization theory submodule, a PID control submodule, and a first reinforcement learning submodule;

[0025] The optimization theory submodule is based on traffic trend prediction, adopts optimization theory, uses linear programming method to build a resource allocation model, defines the objective function as minimize(total_cost) to minimize the total cost, sets constraints = [bandwidth> = 100Mbps, priority < = 10], captures the optimal resource allocation plan through the Simplex algorithm, and generates a resource optimization plan;

[0026] The PID control submodule is based on a resource optimization solution and adopts PID control theory. It uses a proportional-integral-differential controller model to adjust the parameters Kp=0.1, Ki=0.01, Kd=0.001 to feedback the control system performance, adjusts the target value setpoint=desired_performance according to real-time performance data, and automatically corrects the deviation of bandwidth allocation, priority and routing strategy through an adjustment algorithm to generate dynamic adjustment parameters;

[0027] The first reinforcement learning submodule is based on dynamically adjusted parameters, adopts deep reinforcement learning, uses the Q learning framework for self-learning and matching, sets the environment state, action and reward mechanism, defines the learning rate learning_rate=0.01, the reward decay rate discount_rate=0.9, and the exploration rate epsilon=0.1 as the key parameters for model training, and adjusts the bandwidth allocation, priority and routing strategy through iterative training to match the changes in the network environment to generate a dynamic adjustment strategy.

[0028] As a further solution of the present invention, the traffic control execution module includes an SDN deployment submodule, a policy adjustment submodule, and a configuration automation submodule;

[0029] The SDN deployment submodule uses software defined network technology to communicate with the network switch through the OpenFlow protocol based on a dynamic adjustment strategy, and executes the command item add-flow to add a new flow table entry to the switch, wherein the parameter match is set to ip, ip_src=192.168.1.1, ip_dst=192.168.2.1 to match the target data flow, and the action parameter action is set to output:1 to specify the port to which the data packet should be forwarded, and is completed by interacting with the REST API interface of the SDN controller OpenDaylight to generate an initial flow control configuration;

[0030] The policy adjustment submodule is based on the initial flow control configuration, according to the real-time performance data of the network and the flow trend prediction, and calls the REST API to dynamically adjust the flow table items in the SDN controller, including sending a command to modify the flow table items to the controller through an HTTP PUT request, and the request body carries the new configuration information priority to adjust the processing priority of the flow table items, action=output:2 to change the forwarding port of the data packet, and establishes an updated flow control policy;

[0031] The configuration automation submodule uses the Ansible network automation tool to define tasks to automatically deploy new network configurations based on the updated traffic control strategy. The task parameters specify the address of the target device hosts=switch1 and the required operations. The task-name:Update flow table in the Ansible playbook is used to define the operations of adding add-flow, modifying mod-flow and deleting del-flow of flow table items. The submodule interacts with the network device through the SSH protocol, automatically applies the single-pipe traffic control strategy, and generates traffic scheduling instructions.

[0032] As a further solution of the present invention, the behavior pattern analysis module includes an RNN analysis submodule, an LSTM analysis submodule, and a pattern recognition submodule;

[0033] The RNN analysis submodule executes the construction of the recurrent neural network model based on the network performance indicators, and uses the TensorFlow library to define the model, including the utilization rate, delay, and packet loss rate indicator data of the input layer receiving bandwidth. The hidden layer captures the patterns and dependencies in the time series by setting the number of recurrent units to 128. The activation function selects relu to improve the model's ability to capture nonlinear features. The optimizer uses Adam to optimize the training process, and the loss function uses mean_squared_error. The model is trained for network performance indicators to generate time-dependent characteristic analysis.

[0034] The LSTM analysis submodule is based on time-dependent characteristic analysis, adopts a long short-term memory network model, and is implemented through the Keras library. The model structure design includes an input layer processing the results of the time-dependent characteristic analysis, and the LSTM layer is configured with 256 units to analyze the long-term dependency in the network behavior. The activation function uses sigmoid, the optimizer is set to Adam in the model training, and the loss function uses categorical_crossentropy. It analyzes the periodic changes and abnormal behaviors in the network behavior pattern and generates the network behavior pattern recognition result;

[0035] The pattern recognition submodule implements pattern recognition technology based on the network behavior pattern recognition results, applies the support vector machine algorithm in the scikit-learn library, and the SVM model parameter configuration includes the penalty coefficient C=1.0 for the error term and the kernel function kernel=rbf to handle the nonlinear problems in the network behavior pattern recognition results. The SVM model is trained to distinguish normal and abnormal behaviors in network traffic and generate insights into network traffic behavior.

[0036] As a further solution of the present invention, the VLAN configuration adjustment module includes a VLAN automatic division submodule, a graph theory optimization submodule, and a network partition adjustment submodule;

[0037] The VLAN automatic division submodule is based on the insight of network traffic behavior, adopts the minimum cut maximum flow algorithm, and constructs a network flow graph through the Python library NetworkX. The nodes in the graph represent multiple switches in the network, and the edges represent the connections between switches. The capacity of the edges constrains the maximum data transmission capacity range of the connection. The maximum_flow function in NetworkX is used to specify the source node and sink node parameters, calculate the estimated maximum flow from the source node to the sink node in the network, and generate a network flow optimization plan;

[0038] The graph theory optimization submodule is based on the network flow optimization scheme, and iteratively applies the Kernighan-Lin algorithm to perform refinement optimization of network partitions. By iteratively exchanging nodes in the network partitions, the sum of edge weights between partitions is reduced, and the traffic and delay across VLANs are reduced. A Python script is used to define the number and conditions of partition exchanges, as well as preset items for evaluating the efficiency of partition configuration before and after the exchange. By calculating the reduction in the sum of edge weights after the exchange, it is determined whether to perform the exchange, and a refined VLAN optimization configuration is generated;

[0039] The network partition adjustment submodule is based on the refined VLAN optimization configuration, uses the network automation tool Ansible to deploy new VLAN configuration to network devices, defines Ansible playbook, which includes VLAN creation, modification and port allocation tasks for target network devices, and parameters including VLAN ID and port list are passed to network devices through playbook. By utilizing Ansible's automation capabilities, it communicates with network devices through the SSH protocol, executes configuration update tasks, and generates a network partition optimization plan.

[0040] As a further solution of the present invention, the resource allocation optimization module includes a second reinforcement learning submodule, a resource optimization submodule, and a delay reduction submodule;

[0041] The second reinforcement learning submodule executes a deep Q learning algorithm based on the network partition optimization scheme, and uses the TensorFlow and Keras libraries in the Python environment to build a deep Q network. The state space is defined by the current resource usage of the network partition, and the action space is composed of decisions to increase or decrease the bandwidth allocation of multiple partitions, including reducing latency and improving bandwidth utilization. The implementation of the deep Q learning algorithm includes setting the learning rate to 0.01 and the discount rate to 0.9, using the model.fit method to train the model, cyclically updating the Q value to capture the optimal network resource allocation strategy, and generating a preliminary adjustment strategy for network resources;

[0042] The resource optimization submodule performs resource allocation optimization based on the preliminary adjustment strategy of network resources. It adopts genetic algorithm, uses Python's DEAP library, sets the initial population size to 50, the crossover rate to 0.8, and the mutation rate to 0.2, and defines the fitness function to maximize the overall performance of the network, including bandwidth utilization and network delay. The genetic algorithm optimizes resource allocation in the iterative process through selection, crossover, and mutation operations. During the execution process, the quality of each generation of population is evaluated according to the fitness function, and the optimal individual is selected as the resource allocation solution to generate a refined network resource adjustment strategy.

[0043] The delay reduction submodule executes the simulated annealing algorithm based on the refined network resource adjustment strategy, sets the initial temperature to 1000, the cooling rate to 0.95, and the objective function to calculate the total network delay. Through the simanneal library of Python, during the simulated annealing process, resource allocation decisions are randomly selected for analysis. If the new resource allocation decision can reduce the network delay, the new resource allocation decision is accepted. After each iteration, the temperature is lowered until the cooling end is reached. The simulated annealing algorithm is executed to cyclically optimize the network resource allocation and generate the resource optimization configuration.

[0044] As a further solution of the present invention, the network configuration adaptive adjustment module includes a genetic algorithm submodule, a simulated annealing submodule, and a QoS policy adjustment submodule;

[0045] The genetic algorithm submodule is based on resource optimization configuration, executes genetic algorithm to perform initial optimization of network traffic distribution and route selection, initializes a population, each individual represents a set of network configuration schemes, including traffic distribution ratio and route selection, and the algorithm parameter settings include population size of 100, crossover rate of 0.7, mutation rate of 0.01, fitness function based on network delay and bandwidth utilization calculation, and uses Python's DEAP library to create algorithm logic, including selection operation using roulette selection method, crossover operation using single point crossover, mutation operation using random bit mutation, and the number of iterations is set to 50 generations to generate genetic algorithm optimization results;

[0046] The simulated annealing submodule is based on the optimization results of the genetic algorithm, and the simulated annealing algorithm is used to refine and adjust the above results. The initial temperature is set to 1000, the cooling rate is 0.95, the number of iterations for each temperature is 100, and the objective function is also based on network delay and bandwidth utilization. The algorithm is established through a Python script, in which the temperature reduction process adopts an exponential cooling scheme, and fine-tuning is performed by randomly selecting network configuration parameters, including adjusting routing weights or traffic distribution ratios, capturing the optimal network configuration, and generating simulated annealing adjustment results;

[0047] The QoS policy adjustment submodule automatically adjusts the network QoS policy to match the changes in network conditions based on the simulated annealing adjustment results, converts the routing selection and traffic distribution policies in the adjustment results into QoS rules by analyzing them, performs priority control by setting DSCP marks, reserves the bandwidth at 20Mbps, adopts a fair queue algorithm for queue management policy, uses the Cisco IOS command ip qos dscp af31 for configuration, and automatically sends them to routers and switches through scripts to generate an adaptive configuration policy.

[0048] Compared with the prior art, the advantages and positive effects of the present invention are:

[0049] In the present invention, software-defined network technology and time series analysis are used in combination with anomaly detection algorithms to improve network monitoring capabilities, including real-time monitoring of bandwidth utilization, delay and packet loss rate, to ensure the accuracy of performance indicators. Long short-term memory networks or convolutional neural networks are applied to traffic prediction, and bandwidth allocation, priority and routing strategies are dynamically adjusted through optimization theory and reinforcement learning, especially PID controllers or deep reinforcement learning models, to enhance network flexibility. Software-defined network technology improves the efficiency of policy deployment in traffic control execution, and recurrent neural networks or long short-term memory networks deepen behavioral pattern recognition. VLAN configuration is optimized by the minimum cut maximum flow algorithm to improve resource utilization. Resource allocation optimization uses Q learning or deep Q networks, and the network configuration adaptive adjustment module uses genetic algorithms and simulated annealing algorithms to optimize routing selection, automatically adapt to QoS strategies, and improve adaptive capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a system flow chart of the present invention;

[0051] Figure 2 It is a schematic diagram of the system framework of the present invention;

[0052] Figure 3 It is a flow chart of the network status monitoring module of the present invention;

[0053] Figure 4 It is a flow chart of the flow prediction module of the present invention;

[0054] Figure 5 It is a flow chart of the resource allocation decision module of the present invention;

[0055] Figure 6 It is a flow chart of the flow control execution module of the present invention;

[0056] Figure 7 It is a flow chart of the behavior pattern analysis module of the present invention;

[0057] Figure 8 It is a flow chart of the VLAN configuration adjustment module of the present invention;

[0058] Fig. 9 It is a flow chart of the resource allocation optimization module of the present invention;

[0059] Fig.10 This is a flow chart of the network configuration adaptive adjustment module of the present invention. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0061] In the description of the present invention, it should be understood that the terms length, width, up, down, front, back, left, right, vertical, horizontal, top, bottom, inside, outside, etc. indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, which are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific position, be constructed and operated in a specific position, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of multiple is two or more, unless otherwise clearly and specifically defined.

[0062] Embodiment 1

[0063] See also Figure 1 to Figure 2 , a communication system based on high-speed signal connection switch dividing VLAN into network cards includes a network status monitoring module, a traffic prediction module, a resource allocation decision module, a traffic control execution module, a behavior pattern analysis module, a VLAN configuration adjustment module, a resource allocation optimization module, and a network configuration adaptive adjustment module;

[0064] The network status monitoring module is based on software-defined network technology and uses time series analysis and anomaly detection algorithms to monitor network bandwidth utilization, latency, and packet loss rate in real time, analyze network behavior, and generate network performance indicators.

[0065] The traffic prediction module uses long short-term memory network models or convolutional neural network models to predict and analyze future traffic patterns and demands based on network performance indicators and generate traffic trend forecasts;

[0066] The resource allocation decision module uses optimization theory and reinforcement learning methods based on traffic trend prediction to dynamically adjust bandwidth allocation, priority and routing strategies through PID controller or deep reinforcement learning model to generate dynamic adjustment strategies.

[0067] The traffic control execution module deploys and adjusts the traffic control strategy based on dynamic adjustment strategies and software-defined network technology, automatically adjusts the network configuration according to real-time data and prediction results, and generates traffic scheduling instructions;

[0068] The behavior pattern analysis module uses a recurrent neural network model or a long short-term memory network model based on network performance indicators to identify and analyze behavior patterns in network traffic and generate behavior pattern insights;

[0069] The VLAN configuration adjustment module uses the minimum cut maximum flow algorithm based on behavioral pattern insights to dynamically optimize VLAN division and network partitioning, improve the efficiency of network resource utilization and data transmission rate, and generate network partition optimization solutions;

[0070] The resource allocation optimization module uses Q learning or deep Q network model based on the network partition optimization solution to adjust the network resource allocation strategy in real time and generate resource optimization configuration;

[0071] The network configuration adaptive adjustment module is based on resource optimization configuration, and uses genetic algorithms and simulated annealing algorithms to optimize traffic distribution and routing selection in the network, automatically adjust QoS policies, match changes in network conditions, and generate adaptive configuration policies.

[0072] Network performance indicators specifically include real-time data and trend analysis of network bandwidth utilization, latency indicators, and packet loss rate performance indicators. Traffic trend prediction includes future traffic trends, peak predictions, and demand change analysis. Dynamic adjustment strategies specifically include dynamic bandwidth allocation plans, priority adjustment rules, and routing selection strategies. Traffic scheduling instructions specifically include network configuration adjustment instructions and traffic scheduling instructions. Behavioral pattern insights include periodic changes in network usage and abnormal traffic behavior pattern analysis. Network partition optimization plans specifically include VLAN division plans and network partition optimization strategies. Resource optimization configuration specifically includes bandwidth allocation optimization and routing selection optimization strategies. Adaptive configuration strategies specifically include traffic engineering optimization and dynamic QoS adjustment rules.

[0073] In the network status monitoring module, the network bandwidth utilization, latency, and packet loss rate and other indicator data are collected through software-defined network technology, and these data are stored in the format of time series. Anomaly detection algorithms, such as the statistical isolation forest algorithm, are used to analyze the time series data to identify anomalies that are significantly different from the normal network behavior pattern. The process calculates the anomaly score for each data point, and a high score indicates potential abnormal behavior. At the same time, the time series is preprocessed using moving average or exponential smoothing methods to reduce the impact of noise on anomaly detection. In this way, the network status is monitored in real time and a network performance indicator report is generated. The report records the real-time operation status of the network in detail, providing an accurate data basis for subsequent modules.

[0074] In the traffic forecasting module, long short-term memory networks or convolutional neural network models are used to predict future traffic patterns and demands based on network performance indicators. These models can accurately predict network traffic trends by learning long-term dependencies and short-term changes in time series data. The training process includes the selection of time windows, feature standardization, and model parameter tuning to minimize prediction errors. Long short-term memory networks handle long-term dependency issues through their unique gating mechanism, while convolutional neural networks excel in capturing local patterns in time series. The generated traffic trend forecast report describes in detail the traffic changes in the future period, providing a scientific basis for resource allocation decisions.

[0075] In the resource allocation decision module, based on traffic trend prediction, optimization theory and reinforcement learning methods, such as proportional-integral-derivative (PID) controller and deep reinforcement learning model, are used to dynamically adjust bandwidth allocation, priority and routing strategies. The deep reinforcement learning model learns the optimal strategy through interaction with the environment, while the PID controller adjusts the system output according to the error value to achieve the expected performance target. In this process, the model receives network performance indicators in real time and optimizes the decision strategy through the reward mechanism to ensure the efficiency and fairness of resource allocation. The generated dynamic adjustment strategy report guides the traffic control execution module to ensure the optimal utilization of network resources.

[0076] In the traffic control execution module, software-defined network technology is used to deploy and adjust traffic control strategies based on dynamic adjustment strategies. New routing strategies, bandwidth allocation, and priority settings are quickly deployed through programmatic definition, while network configuration is automatically adjusted based on real-time data and prediction results. This process achieves a high degree of automation and flexibility in network configuration, can quickly respond to changes in network status, optimize data flow, reduce congestion and delay, and generate traffic scheduling instructions to ensure that the network operates in the best state.

[0077] In the behavior pattern analysis module, the behavior pattern recognition and analysis in network traffic is performed based on network performance indicators through recurrent neural network or long short-term memory network models. The model learns the time-dependent features in network traffic data and identifies normal and abnormal traffic patterns. This analysis process not only relies on the deep learning ability of the model, but also includes the detailed operations of data preprocessing, feature extraction and model training. The generated behavior pattern insight report provides important information for network security and management, helping network administrators identify potential security threats and network failures and optimize network performance.

[0078] In the VLAN configuration adjustment module, based on behavioral pattern insights, the minimum cut maximum flow algorithm is used to dynamically optimize VLAN division and network partitioning. The algorithm analyzes the demand and capacity of network traffic and calculates the optimal network segmentation plan to improve the efficiency of data transmission and the utilization of network resources. This process includes building a network flow graph, calculating the maximum flow, and identifying the minimum cut set to determine the optimal VLAN configuration. The generated network partition optimization plan report details how to reconfigure the network to meet current traffic demand and performance goals, and improve the scalability and flexibility of the network.

[0079] In the resource allocation optimization module, Q learning or deep Q network models are used to adjust the network resource allocation strategy in real time based on the network partition optimization scheme. These models learn how to make the best resource allocation decisions under given network conditions by exploring and utilizing strategies. The process includes defining the state space, action space, and reward function. The model adjusts its strategy through repeated trial and error learning to maximize long-term rewards. The generated resource optimization configuration report provides a set of optimized network resource allocation solutions to ensure that the network can operate efficiently under various conditions.

[0080] In the network configuration adaptive adjustment module, based on resource optimization configuration, genetic algorithms and simulated annealing algorithms are used to optimize the traffic distribution and routing selection in the network. Genetic algorithms simulate natural selection and genetic mechanisms to continuously iterate and find the optimal solution, while simulated annealing algorithms simulate the physical annealing process to effectively avoid local optimal solutions and find global optimal solutions. These algorithms combine the real-time performance data of the network to dynamically adjust the network configuration to automatically adapt to changes in network conditions. The generated adaptive configuration strategy report describes in detail how to automatically adjust network settings through intelligent algorithms to ensure network performance and service quality.

[0081] See also Figure 3 ,The network status monitoring module includes a time series analysis submodule, anomaly detection submodule, and real-time monitoring submodule;

[0082] The time series analysis submodule is based on software-defined network technology, collects network bandwidth utilization bandwidth_usage, delay latency, and packet loss rate packet_loss indicators, adopts the autoregressive moving average model, calls the ARIMA function through the Python statsmodels library, sets the model parameter order = (2,1,2), where the first term 2 represents the order of the autoregressive term AR, 1 represents the difference order I, and the second term 2 represents the order of the moving average term MA, performs the fitting operation of the time series data, and generates the trend analysis of the network indicators;

[0083] The anomaly detection submodule is based on the trend analysis of network indicators and adopts the isolation forest algorithm. Through the IsolationForest function of sklearn.ensemble, the parameter n_estimators = 100 is set to indicate the number of trees in the forest, and contamination = auto is used to automatically estimate the proportion of abnormal points in the data set, and the training operation of the isolation forest model is performed to generate abnormal behavior recognition;

[0084] The real-time monitoring submodule is based on abnormal behavior identification and uses software-defined network technology to set the data collection frequency parameter to frequency = 5, which means that network performance data is collected every 5 seconds. The monitor function of the Traffic Monitor tool is used to set the monitoring parameter interval = 5, and network performance data is collected every 5 seconds. Real-time data monitoring is performed to generate network performance indicators.

[0085] In the time series analysis submodule, based on the application background of software defined network (SDN) technology, the first thing to do is to collect network performance data, including network bandwidth utilization (bandwidth_usage), delay (latency) and packet loss rate (packet_loss). These data are stored in time series format, and each indicator changes over time, which can be formally represented as a sequence of timestamps and indicator values. For example, the data of network bandwidth utilization is recorded in the form of (timestamp, bandwidth_usage), where timestamp represents the time point of data collection, and bandwidth_usage represents the network bandwidth utilization at that time point. After the data is collected, the submodule uses the autoregressive moving average (ARIMA) model to analyze the time series data. The ARIMA model is a commonly used method in time series prediction. It models the autoregressive (AR) and moving average (MA) parts of historical data, and adds a difference (I) operation to stabilize the series, so that future data points can be predicted. In this submodule, the ARIMA function is called through the Python statsmodels library, and the model parameter order = (2,1,2) is set for implementation. The parameter order consists of three parts, corresponding to the three key parts of the ARIMA model: 2 represents the order of the autoregressive term (AR), indicating that the values ​​of the previous two time points will be taken into account when predicting the current value; 1 represents the difference order (I), indicating that the original data needs to be differentiated once before the model is fitted to eliminate the non-stationarity of the data; 2 represents the order of the moving average term (MA), which means that the error terms of the previous two time points will be used in the model to adjust the current prediction. With the set ARIMA model parameters, the model will perform fitting operations on the time series data of each network performance indicator. This process involves using relevant information in the historical data to estimate the model parameters, including the autoregressive coefficient, the moving average coefficient, and the overall average level of the series. After completing these calculations, the model can predict the future network performance indicator values ​​based on the existing data, such as the trends of bandwidth utilization, delay, and packet loss rate. Finally, the time series analysis submodule generates an analysis report on the future trend of network performance indicators. This report not only contains the forecast value, but also includes quantitative analysis of the trend of network performance changes, such as confidence intervals and other statistical information, to provide decision support for network managers and help identify emerging performance bottlenecks or areas that need to be optimized in the future.

[0086] In the anomaly detection submodule, based on the network indicator trend analysis report generated by the previous submodule, the isolation forest algorithm is used to identify abnormal behaviors. Isolation forest is a tree-based anomaly detection algorithm. Its core idea is to randomly select a feature and its split value to split the data set into two parts. This process is repeated on each tree in the forest. Since the anomalies are sparsely distributed, they are easier to be isolated, that is, the number of splits required to reach the leaf node in the tree structure is relatively small. In actual operation, the isolation forest algorithm is implemented through the IsolationForest function of sklearn.ensemble. The parameter n_estimators = 100 is set to define the number of trees in the forest, and contamination = 'auto' automatically estimates the proportion of anomalies in the data set. The isolation forest model first randomly selects a feature and a split value, and then splits the data into two parts according to this split value. This process is recursively repeated on each tree until the data can no longer be split, that is, each data point is isolated on a leaf node. By calculating the average path length required for a data point to reach a leaf node, the isolation forest can evaluate an anomaly score for each point. The shorter the path required for a data point to reach a leaf node, the higher its anomaly score, and thus it is judged to be anomaly. After the isolation forest model is trained, it can identify abnormal behaviors such as sudden reduction in bandwidth utilization, increased latency, or increased packet loss based on historical and current data of network performance indicators. The results of anomaly detection are presented in the form of a report, which lists the detected abnormal events and their related attributes, such as the time of occurrence, the affected network performance indicators, and the degree of anomaly score. This information is crucial for network managers because they can quickly identify and locate network performance problems and take corresponding preventive or repair measures to ensure the stable operation of the network.

[0087] In the real-time monitoring submodule, based on the results of abnormal behavior identification, the network performance indicators are continuously monitored through software-defined network technology. In this submodule, the data collection frequency parameter frequency = 5 is set, which means that network performance data is collected every 5 seconds. This is achieved by using the monitor function of the Traffic Monitor tool, where the monitoring parameter interval = 5 sets the monitoring time interval, that is, network performance data is collected every 5 seconds to generate network performance indicators.

[0088] See also Figure 4 ,The traffic prediction module includes LSTM prediction submodule, CNN prediction submodule, and traffic pattern analysis submodule;

[0089] The LSTM prediction submodule uses the long short-term memory network model based on network performance indicator data and uses the TensorFlow library to build the LSTM network structure, including setting the number of hidden layer units to 128 to capture the long-term dependencies of the time series, setting the learning rate to 0.001 to control the speed of the learning process, setting the batch size to 32 to balance the training efficiency and network performance, using the Adam optimizer for parameter optimization, and using the mean_squared_error loss function to evaluate the difference between the model prediction value and the true value to generate time series prediction analysis;

[0090] The CNN prediction submodule adopts the convolutional neural network model based on the time series prediction analysis results. The PyTorch library is used to define the CNN model structure, including adding the Conv1D layer, setting the number of filters to 64 to extract the local features of the traffic data, setting the kernel size to 3 to capture the relationship between adjacent data points, selecting ReLU as the activation function to introduce nonlinearity, applying the MaxPooling1D layer for feature downsampling, and using Adam as the optimizer. The loss function selects cross_entropy to handle the classification problem and generate local feature analysis results;

[0091] The traffic pattern analysis submodule is based on the results of time series prediction analysis and local feature analysis, and comprehensively applies statistical analysis techniques. It performs data operations through Python's NumPy library, calculates the moving average to determine the overall trend of traffic, sets the window size to 7 days to capture traffic fluctuations within a week, and calculates the standard deviation to identify abnormal traffic fluctuations. It generates traffic trend forecasts by calling the mean and std functions of the NumPy library.

[0092] In the LSTM prediction submodule, the long short-term memory network (LSTM) is implemented using the TensorFlow library to process time-based series data. The data format is time series data, which contains network performance indicators such as latency, throughput, etc., formatted in a form suitable for input to the LSTM network. This includes standardization or normalization of the data, which converts the data into the form of one or more time steps. Specifically, an LSTM network is initialized, in which the number of hidden layer units is set to 128, which is to capture the long-term dependencies in the time series data. The learning rate is set to 0.001 to control the speed of the learning process to optimize the efficiency and accuracy of the network weight update. The batch size is set to 32, which considers the balance between maintaining training efficiency and reducing memory consumption. The Adam optimizer is used because it exhibits good convergence under various conditions and is suitable for most network training tasks. The loss function uses mean_squared_error, which directly reflects the difference between the model prediction value and the true value. The optimization goal is to reduce this error and improve the accuracy of the prediction. With this setting, the LSTM is able to update the weights in each training iteration, gradually improving the model to better predict future network performance indicators.

[0093] In the CNN prediction submodule, implemented with the help of the PyTorch library, we focus on extracting local features of time series data. The data input used in this part is the output of the LSTM module, which is the result of the time series prediction analysis. The CNN model adds a Conv1D layer with 64 filters and a kernel size of 3. This is designed to effectively capture local patterns and features in the sequence data. The activation function uses ReLU because it can introduce nonlinearity while maintaining computational simplicity, which helps to handle complex data relationships. The MaxPooling1D layer is used for feature downsampling, reducing the number of parameters and computational complexity while retaining important feature information. The Adam optimizer is also used because its adaptive learning rate makes the training process more stable. The loss function uses cross_entropy, which is particularly suitable for handling classification problems and measures the difference between the model output and the true label. Through these operations, the CNN model strengthens its ability to learn local features of time series data and improves the accuracy and generalization of predictions.

[0094] In the traffic pattern analysis submodule, statistical analysis techniques are comprehensively applied, mainly based on the NumPy library to process data. The input data comes from the prediction results of the first two submodules, which include time series prediction and local feature analysis results. By calculating the moving average, the window size is set to 7 days. This operation is designed to capture the traffic fluctuation trend within a week and help understand how the traffic changes over time. The calculation of the standard deviation further identifies abnormal traffic fluctuations, which is achieved by evaluating the degree of discreteness of the traffic data. The application of these statistical methods not only provides an in-depth understanding of network traffic patterns, but also provides strong support for network management and planning. Through these analyses, it is possible to predict traffic trends, identify problem areas, and provide decision support for network optimization and expansion.

[0095] See also Figure 5 ,The resource allocation decision module includes an optimization theory submodule, a PID control submodule, and a first reinforcement learning submodule;

[0096] The optimization theory submodule is based on traffic trend prediction, adopts optimization theory, uses linear programming method to build resource allocation model, defines the objective function as minimize(total_cost) to minimize the total cost, and sets constraints

[0097] constraints = [bandwidth> = 100Mbps, priority < = 10], the optimal resource allocation solution is captured through the Simplex algorithm to generate a resource optimization solution;

[0098] The PID control submodule is based on the resource optimization scheme and adopts the PID control theory. It uses the proportional-integral-differential controller model to adjust the parameters Kp=0.1, Ki=0.01, Kd=0.001 to feedback the control system performance. It adjusts the target value setpoint=desired_performance according to the real-time performance data. It automatically corrects the deviation of bandwidth allocation, priority and routing strategy through the adjustment algorithm to generate dynamic adjustment parameters.

[0099] The first reinforcement learning submodule is based on dynamic adjustment parameters, adopts deep reinforcement learning, uses the Q learning framework for self-learning and matching, sets the environment state, action and reward mechanism, defines the learning rate learning_rate = 0.01, the reward decay rate discount_rate = 0.9, and the exploration rate epsilon = 0.1 as the key parameters for model training, and adjusts the bandwidth allocation, priority and routing strategy through iterative training to match the changes in the network environment, and generates a dynamic adjustment strategy.

[0100] In the optimization theory submodule, the key step is to use the linear programming method to build a resource allocation model through traffic trend prediction data. This process starts with defining a clear objective function minimize(total_cost), which aims to minimize the total cost through resource optimization. In this model, the constraints are set to bandwidth>=100Mbps and priority<=10 to ensure that the basic operation requirements of the network are met. The Simplex algorithm, as an efficient optimization algorithm, is used to deal with these constraints and find the optimal solution for resource allocation under given conditions. This algorithm iteratively improves the estimated value of the solution until the optimal point that minimizes the total cost is found. During the operation, each step carefully considers the availability of resources and the priority of demand to ensure that the generated resource optimization solution is both economical and efficient, providing a solid foundation for network operations.

[0101] In the PID control submodule, parameter adjustment through the proportional-integral-differential (PID) controller model based on the resource optimization scheme generated previously is the key link to achieve detailed control. Here, the parameter settings of Kp=0.1, Ki=0.01, and Kd=0.001 are optimized based on real-time feedback of system performance. This method can automatically adjust the control intensity according to the current performance of the system, correct the deviation in real time, and stabilize the system performance in an ideal state. The PID controller ensures the flexibility and responsiveness of network resource allocation by continuously monitoring the difference between system performance and expected performance and adjusting the input to narrow this gap. In this way, the dynamic adjustment parameters are continuously optimized, so that the network resource allocation can sensitively respond to the changes in actual traffic and load, and improve the overall operation efficiency of the network.

[0102] In the first reinforcement learning submodule, resource allocation is further optimized through deep reinforcement learning based on dynamically adjusted parameters. Here, the Q learning framework is adopted. The key is to guide the learning process through environmental states, actions and reward mechanisms. The settings of parameters such as learning rate learning_rate=0.01, reward decay rate discount_rate=0.9, exploration rate epsilon=0.1 are intended to optimize the learning process of the model and ensure that the model can effectively learn from past experience and adjust strategies. In this way, the model can find a balance between exploration and utilization, and gradually learn how to make the best resource allocation decisions in a changing network environment. In this process, the model continuously learns through trial and error, evaluates the effects of different resource allocation strategies, and finally can autonomously generate dynamic adjustment strategies that adapt to changes in the network environment, thereby realizing adaptive optimization of network resource management.

[0103] See also Figure 6,The traffic control execution module includes the SDN deployment submodule, the policy adjustment submodule, and the configuration automation submodule;

[0104] The SDN deployment submodule uses software-defined network technology to communicate with network switches through the OpenFlow protocol based on dynamic adjustment strategies, and executes the command item add-flow to add new flow table items to the switch, where the parameter match is set to ip, ip_src = 192.168.1.1, ip_dst = 192.168.2.1 to match the target data flow, and the action parameter action is set to output:1 to specify the port to which the data packet should be forwarded. This is completed by interacting with the REST API interface of the SDN controller OpenDaylight to generate the initial flow control configuration;

[0105] Based on the initial flow control configuration, the policy adjustment submodule calls the REST API to dynamically adjust the flow table items in the SDN controller according to the real-time performance data and traffic trend prediction of the network, including sending a command to modify the flow table items to the controller through an HTTP PUT request. The request body carries the new configuration information priority to adjust the processing priority of the flow table items, action=output:2 to change the forwarding port of the data packet, and establishes an updated flow control policy.

[0106] The configuration automation submodule uses the Ansible network automation tool to define tasks to automatically deploy new network configurations based on the updated traffic control policy. The task parameters specify the target device address hosts=switch1 and the required operations. The task -name:Update flow table in the Ansible playbook is used to define the operations of adding add-flow, modifying mod-flow, and deleting del-flow flow table items. The SSH protocol is used to interact with network devices, automatically apply single-pipe traffic control policies, and generate traffic scheduling instructions.

[0107] In the SDN deployment submodule, software-defined network technology communicates with network switches through the OpenFlow protocol, which is the core of network traffic control and management. Specifically, this submodule adds new flow table entries to the switch by executing the add-flow command item to match and control the target data flow. The parameter match in the command item is set to ip, ip_src = 192.168.1.1, ip_dst = 192.168.2.1, which accurately specifies the data flow with source IP address 192.168.1.1 and destination IP address 192.168.2.1, and the action parameter action is set to output:1, indicating that these data packets should be forwarded to port 1. These operations are completed by interacting with the REST API interface of the SDN controller OpenDaylight. The interface uses HTTP request and response to allow users to programmatically control network devices. During the interaction, the constructed HTTP request carries the necessary JSON format data, including detailed configuration information of the flow table items. After receiving these requests, the SDN controller parses the data in the request body and sends the corresponding flow table item rules to the specified network device. The initial flow control configuration file generated by this operation provides a basis for the management and scheduling of network traffic, ensuring the efficient use of network resources and the smooth operation of network traffic.

[0108] In the policy adjustment submodule, based on the initial flow control configuration, the dynamic adjustment of flow table items is performed according to the real-time performance data and traffic trend prediction of the network. This submodule sends an HTTP PUT request to the SDN controller by calling the REST API. The request body carries new configuration information, such as priority is used to adjust the processing priority of the flow table item, and action=output:2 changes the forwarding port of the data packet. This dynamic adjustment strategy allows the network to quickly adapt to changes according to the current status and needs and optimize the allocation of network resources. For example, when the bandwidth demand of a certain data flow increases, the transmission quality of the key data flow is ensured by increasing the priority of its flow table item; or the forwarding path of the data flow is adjusted according to the network congestion situation to avoid potential bottlenecks. By precisely controlling the behavior of the data flow, the policy adjustment submodule not only improves the performance and stability of the network, but also enhances the flexibility and scalability of the network. The generated updated flow control policy file records all adjusted flow table item configurations, providing important reference information for network operation and maintenance personnel.

[0109] In the configuration automation submodule, the new network configuration is automatically deployed with the help of the Ansible network automation tool, further improving the efficiency and accuracy of network management. In this submodule, the defined task automation process includes specifying the address of the target device and the operations to be performed, such as adding, modifying or deleting flow table items. The tasks in the Ansible playbook are defined in YAML format, such as -name:Update flow table, which clearly indicates the name of the task to be performed and the related operation instructions, including add-flow, mod-flow and del-flow. These tasks interact with network devices through the SSH protocol to automatically update the flow control policy without manual intervention, greatly reducing the risk of configuration errors and improving the speed and reliability of operations. The successful implementation of the configuration automation submodule not only ensures the consistency and accuracy of network configuration, but also achieves a high degree of network operation automation, making network management more efficient and flexible. The generated traffic scheduling instruction file records the execution status and results of each operation in detail, providing reliable data support for the continuous optimization and management of the network.

[0110] See also Figure 7 ,The behavior pattern analysis module includes RNN analysis submodule, LSTM analysis submodule, and pattern recognition submodule;

[0111] The RNN analysis submodule builds a recurrent neural network model based on network performance indicators. The model is defined using the TensorFlow library, including the input layer receiving bandwidth utilization, latency, and packet loss rate indicator data. The hidden layer captures the patterns and dependencies in the time series by setting the number of recurrent units to 128. The activation function selects relu to improve the model's ability to capture nonlinear features. The optimizer uses Adam to optimize the training process, and the loss function uses mean_squared_error. The model is trained based on network performance indicators to generate time-dependent characteristic analysis.

[0112] The LSTM analysis submodule is based on time-dependent characteristic analysis and adopts the long short-term memory network model. It is implemented through the Keras library. The model structure design includes the input layer processing the results of time-dependent characteristic analysis. The LSTM layer is configured with 256 units to analyze the long-term dependency in network behavior. The activation function uses sigmoid. The optimizer is set to Adam in model training. The loss function uses categorical_crossentropy. It analyzes the periodic changes and abnormal behaviors in the network behavior pattern and generates the network behavior pattern recognition results.

[0113] The pattern recognition submodule implements pattern recognition technology based on the network behavior pattern recognition results, applies the support vector machine algorithm in the scikit-learn library, and the SVM model parameter configuration includes the penalty coefficient C=1.0 for the error term and the kernel function kernel=rbf to deal with the nonlinear problems in the network behavior pattern recognition results. The SVM model is trained to distinguish between normal and abnormal behaviors in network traffic and generate insights into network traffic behavior.

[0114] In the RNN analysis submodule, the recurrent neural network (RNN) is specifically used to process sequence data, enabling it to capture patterns and dependencies in time series data. Network performance indicators such as bandwidth utilization, latency, and packet loss rate are used as input data. These data need to be preprocessed before entering the model, such as normalization, to ensure that the model can learn effectively. The RNN model is built using the TensorFlow library, which includes an input layer and one or more hidden layers. The number of recurrent units in the hidden layer is set to 128 in order to capture the complex dependencies in time series data. The use of the ReLU activation function helps the model introduce nonlinearity in the learning process and improves the ability to recognize complex patterns. The Adam optimizer is used for model training because the Adam optimizer can show good convergence under a variety of different conditions, making the training process more efficient. As a loss function, mean_squared_error is directly optimized for the difference between the model's predicted value and the actual value, with the goal of minimizing this error. Through the training and optimization of the RNN model, time-dependent characteristic analysis results can be generated, which can reveal the inherent laws of network performance indicators changing over time and provide a basis for subsequent network behavior pattern recognition.

[0115] In the LSTM analysis submodule, a long short-term memory network (LSTM) model is used to process the time-dependent feature analysis results generated by the RNN analysis submodule. LSTM is particularly suitable for analyzing and capturing long-term dependencies, making it an ideal choice for processing time series data. The model is implemented using the Keras library, where the LSTM layer is configured with 256 units to enhance the model's ability to analyze long-term dependencies in network behavior. The sigmoid activation function is used in the LSTM layer to help the model perform in classification tasks, as its output range is from 0 to 1, which is suitable for handling binary classification problems. The Adam optimizer and categorical_crossentropy are also used as the loss function during model training. This setting is aimed at optimizing the performance of the model when analyzing periodic changes and abnormal behavior in network behavior patterns. Through in-depth training of the LSTM model, network behavior pattern recognition results can be generated, which provides an accurate basis for identifying normal and abnormal behaviors in network traffic.

[0116] In the pattern recognition submodule, the support vector machine (SVM) algorithm is used, implemented through the scikit-learn library, to specifically process the network behavior pattern recognition results generated by the LSTM analysis submodule. The SVM model effectively handles nonlinear problems by configuring the penalty coefficient C of the error term to 1.0 and using the radial basis function (rbf) kernel. The training of the SVM model aims to distinguish between normal and abnormal behaviors in network traffic, which is critical for network security and performance management. Through accurate pattern recognition, the SVM model can help the operation and maintenance team to promptly discover and solve network problems, thereby improving the stability and security of the network.

[0117] See also Figure 8 ,The VLAN configuration adjustment module includes a VLAN automatic division submodule, a graph theory optimization submodule, and a network partition adjustment submodule;

[0118] The VLAN automatic division submodule is based on the insight of network traffic behavior and adopts the minimum cut maximum flow algorithm. It constructs a network flow graph through the Python library NetworkX. The nodes in the graph represent multiple switches in the network, and the edges represent the connections between switches. The capacity of the edges constrains the maximum data transmission capacity range of the connection. It uses the maximum_flow function in NetworkX to specify the source node and sink node parameters, calculate the estimated maximum flow from the source node to the sink node in the network, and generate a network flow optimization plan.

[0119] The graph optimization submodule is based on the network flow optimization solution and iteratively applies the Kernighan-Lin algorithm to refine the network partitions. It iteratively exchanges nodes in the network partitions to reduce the sum of edge weights between partitions, reduce cross-VLAN traffic and latency, and use Python scripts to define the number and conditions of partition exchanges, as well as preset items for evaluating the efficiency of partition configuration before and after exchanges. It determines whether to perform the exchange by calculating the reduction in the sum of edge weights after the exchange, and generates a refined VLAN optimization configuration.

[0120] The network partition adjustment submodule is based on the refined VLAN optimization configuration. It uses the network automation tool Ansible to deploy new VLAN configurations to network devices and define Ansible playbooks, which include VLAN creation, modification, and port allocation tasks for target network devices. Parameters including VLAN ID and port list are passed to network devices through playbooks. By leveraging Ansible's automation capabilities, it communicates with network devices through the SSH protocol, executes configuration update tasks, and generates network partition optimization solutions.

[0121] In the VLAN automatic division submodule, the minimum cut maximum flow algorithm is used to construct a network flow graph with the Python library NetworkX. In the graph, nodes represent switches in the network, edges represent connections between switches, and the capacity constraints of edges represent the maximum data transmission capacity of the connection. By specifying the source node and sink node parameters through the maximum_flow function in NetworkX, the estimated maximum flow from the source node to the sink node is calculated, and this process generates a network flow optimization plan. The minimum cut maximum flow algorithm ensures that the data flow in the network can reach the optimal state by finding the path that maximizes the flow from the source point to the sink point. The calculation of the maximum flow between the source node and the sink node is based on the capacity constraint of each edge, reflecting the transmission capacity of the network. The optimization plan includes suggestions for adjusting the network flow, such as increasing or decreasing the flow on a specific path to balance the network load and reduce congestion. The generated network flow optimization plan document describes in detail the specific measures to improve network performance and enhance data transmission efficiency, providing a scientific basis for network administrators to adjust and optimize the network structure.

[0122] In the graph optimization submodule, the Kernighan-Lin algorithm is applied iteratively to refine the network partitions, aiming to reduce cross-VLAN traffic and latency. The algorithm defines the number and conditions of partition exchanges through Python scripts, as well as preset items to evaluate the efficiency of partition configuration before and after the exchange. The algorithm reduces the sum of edge weights between partitions by iteratively exchanging nodes in the network partitions. The reduction in the sum of edge weights after the exchange is calculated for each iteration, which is used as the basis for whether to perform the exchange.

[0123] The core of the Kernighan-Lin algorithm is to optimize the segmentation of the network, reduce the data transmission requirements between different partitions, thereby reducing the overall burden of the network and improving the efficiency and speed of data transmission. The refined VLAN optimization configuration file records all the optimized network partition information, including the division of each VLAN, which VLAN each switch belongs to, etc., providing strong support for the efficient operation and management of the network.

[0124] In the network partition adjustment submodule, the network automation tool Ansible is used to deploy new VLAN configurations to network devices. By defining Ansible playbooks, including tasks for VLAN creation, modification, and port allocation, parameters such as VLAN IDs and port lists are passed to network devices through playbooks. Ansible's automation capabilities communicate with network devices through the SSH protocol, perform configuration update tasks, and generate network partition optimization plans. This process enables rapid deployment and updating of network configurations, ensuring that the network configuration is consistent with the optimization plan. The network partition optimization plan file provides detailed deployment results and network configuration information, enabling network administrators to clearly understand the current status of the network, facilitating subsequent monitoring and maintenance work, thereby ensuring continuous optimization and improvement of the network architecture. By automating the deployment of new VLAN configurations, the reliability and performance of the network have been significantly improved, while also reducing human errors and improving work efficiency.

[0125] See also Fig. 9 ,The resource allocation optimization module includes a second reinforcement learning submodule, a resource optimization submodule, and a delay reduction submodule;

[0126] The second reinforcement learning submodule executes the deep Q learning algorithm based on the network partition optimization scheme, and uses the TensorFlow and Keras libraries in the Python environment to build a deep Q network. The state space is defined by the current resource usage of the network partition, and the action space is composed of decisions to increase or decrease the bandwidth allocation of multiple partitions, including reducing latency and improving bandwidth utilization. The implementation of the deep Q learning algorithm includes setting the learning rate to 0.01 and the discount rate to 0.9, using the model.fit method to train the model, cyclically updating the Q value to capture the optimal network resource allocation strategy, and generating a preliminary adjustment strategy for network resources;

[0127] The resource optimization submodule performs resource allocation optimization based on the preliminary adjustment strategy of network resources. It uses a genetic algorithm and the DEAP library of Python to set the initial population size to 50, the crossover rate to 0.8, and the mutation rate to 0.2. It defines the fitness function to maximize the overall performance of the network, including bandwidth utilization and network latency. The genetic algorithm optimizes resource allocation in an iterative process through selection, crossover, and mutation operations. During the execution process, the fitness function is used to evaluate the quality of each generation of population, select the best individual as the resource allocation solution, and generate a refined network resource adjustment strategy.

[0128] The delay reduction submodule executes the simulated annealing algorithm based on the refined network resource adjustment strategy, sets the initial temperature to 1000, the cooling rate to 0.95, and the objective function to calculate the total network delay. Through Python's simanneal library, during the simulated annealing process, resource allocation decisions are randomly selected for analysis. If the new resource allocation decision can reduce network delay, the new resource allocation decision is accepted. After each iteration, the temperature is lowered until the cooling end is reached. The simulated annealing algorithm is executed to cyclically optimize network resource allocation and generate resource optimization configuration.

[0129] In the second reinforcement learning submodule, the deep Q-learning algorithm is applied to the network partition optimization scheme, aiming to dynamically adjust network resources to maximize performance indicators. The deep Q-network (DQN) model built using TensorFlow and Keras libraries accepts the current resource usage of the network as the state space, and the action space is defined as the modification of bandwidth allocation decisions for multiple network partitions, including increasing or decreasing allocations to optimize latency and bandwidth utilization. During model training, the learning rate is set to 0.01 to control the speed of the learning steps, while the discount rate is 0.9 to balance the importance of immediate rewards and future rewards. Through the model.fit method, the model cyclically updates the Q value during the training process to gradually approach the optimal network resource allocation strategy. The preliminary network resource adjustment strategy generated by this process not only provides a basis for resource optimization, but also can make adjustments based on real-time performance feedback, thereby achieving self-optimization of network performance.

[0130] In the resource optimization submodule, a genetic algorithm is used to perform resource allocation optimization based on the preliminary adjustment strategy of network resources generated by the second reinforcement learning submodule. Using the DEAP library of Python, this process starts with initializing the population, and the population size is set to 50, representing the resource allocation solution. The crossover rate and mutation rate are set to 0.8 and 0.2 respectively to promote the diversity of the population and the innovation of solutions. The fitness function is defined to maximize the overall performance of the network by considering bandwidth utilization and network latency. The core operations of the genetic algorithm include selection, crossover and mutation, through which resource allocation is optimized in each iteration. Each generation of the population is evaluated according to the fitness function, and the individuals with the best performance are selected as the parents of the next generation. The refined network resource adjustment strategy generated by this process provides an optimized resource allocation solution for the network, thereby improving network performance and service quality.

[0131] In the delay reduction submodule, the simulated annealing algorithm is applied to further reduce network delay based on the refined network resource adjustment strategy generated by the resource optimization submodule. By setting the initial temperature and cooling rate, the simulated annealing algorithm simulates the physical annealing process during the optimization process to find the global optimal solution and avoid falling into the local optimum. The objective function is used to calculate the total network delay, and the randomly selected resource allocation decision in each iteration is accepted if it can reduce the network delay. As the temperature gradually decreases, the algorithm gradually reduces the probability of accepting new solutions until the cooling end is reached. The resource optimization configuration generated by this process aims to minimize network delay by fine-tuning the network resource allocation, thereby improving the overall performance and responsiveness of the network.

[0132] See also Fig.10 ,The network configuration adaptive adjustment module includes a genetic algorithm submodule, a simulated annealing submodule, and a QoS policy adjustment submodule;

[0133] The genetic algorithm submodule is based on resource optimization configuration, executes genetic algorithm to perform initial optimization of network traffic distribution and route selection, initializes a population, and each individual represents a set of network configuration schemes, including traffic distribution ratio and route selection. The algorithm parameter settings include population size of 100, crossover rate of 0.7, mutation rate of 0.01, fitness function based on network delay and bandwidth utilization calculation, and uses Python's DEAP library to create algorithm logic, including roulette selection method for selection operation, single-point crossover for crossover operation, random bit mutation for mutation operation, and the number of iterations is set to 50 generations to generate genetic algorithm optimization results;

[0134] The simulated annealing submodule uses the simulated annealing algorithm to refine and adjust the above results based on the optimization results of the genetic algorithm. The initial temperature is set to 1000, the cooling rate is 0.95, the number of iterations for each temperature is 100, and the objective function is also based on network delay and bandwidth utilization. The algorithm is established through Python scripts. The temperature reduction process adopts an exponential cooling scheme. The network configuration parameters are randomly selected for fine-tuning, including adjusting the routing weight or traffic distribution ratio, to capture the optimal network configuration and generate simulated annealing adjustment results.

[0135] The QoS policy adjustment submodule automatically adjusts the network QoS policy to match the changes in network conditions based on the simulated annealing adjustment results. It converts the routing selection and traffic distribution policies in the adjustment results into QoS rules by analyzing them. It performs priority control by setting DSCP marks, and reserves the bandwidth at 20Mbps. The queue management policy adopts the fair queue algorithm and is configured using the Cisco IOS command ip qos dscp af31. It is automatically sent to routers and switches through scripts to generate an adaptive configuration policy.

[0136] In the genetic algorithm submodule, the initial optimization of network traffic distribution and routing selection is performed, and iterative evolution is performed by constructing a population. Each individual represents a set of network configuration solutions, covering traffic distribution ratio and routing selection. The algorithm logic is implemented using Python's DEAP library. The population size is set to 100, the crossover rate and mutation rate are 0.7 and 0.01 respectively, and the fitness function is calculated based on network delay and bandwidth utilization. The algorithm selection operation uses the roulette selection method to ensure that excellent individuals are selected with a higher probability. The crossover operation uses single-point crossover to ensure the diversity of genes. The mutation operation introduces new gene mutations through random bit mutation to prevent the algorithm from premature convergence. The number of iterations is set to 50 generations. Each generation generates a new population through selection, crossover, and mutation operations, gradually improving the fitness of the entire population, and finally generating the genetic algorithm optimization result. The network configuration solution generated by this process aims to balance network delay and bandwidth utilization. The fitness function guides the search process to evolve in the direction of optimizing network performance. The optimization results are saved in the form of files, including the optimal distribution ratio of network traffic and routing selection strategy, providing a scientific decision-making basis for network operation and maintenance.

[0137] In the simulated annealing submodule, the simulated annealing algorithm is applied to refine the optimization results of the genetic algorithm. When the algorithm starts, the initial temperature is set to 1000, and the exponential cooling scheme is used to reduce the temperature at a cooling rate of 0.95. 100 iterations are performed at each temperature. In each iteration, the network configuration parameters are randomly selected for fine-tuning, the routing weights or traffic distribution ratios are adjusted, and the fitness of the new network configuration scheme is calculated. If the fitness of the new scheme is better than the current scheme or a poor scheme is accepted with a certain probability, the local optimal solution is escaped until the system is cooled. The simulated annealing algorithm effectively balances exploration and utilization by allowing the acceptance of inferior solutions at high temperatures, and gradually reducing the probability of accepting inferior solutions by decreasing the temperature, and finally captures the optimal network configuration. The adjustment results record in detail the finely adjusted network configuration scheme, including the finalized routing weights and traffic distribution ratios, providing a more accurate and efficient configuration strategy for network management.

[0138] In the QoS policy adjustment submodule, based on the simulated annealing adjustment results, the optimized routing and traffic distribution strategies are converted into QoS rules to achieve automatic adjustment of network QoS strategies. By analyzing the routing and traffic distribution in the optimization results, the corresponding DSCP tags are set for priority control, the bandwidth reservation is set to 20Mbps, and the fair queue algorithm is used as the queue management strategy to ensure fair and efficient transmission of network traffic. The QoS rules are configured through the Cisco IOS command ip qos dscpaf31, and the configuration is sent to routers and switches in combination with script automation technology to achieve automatic deployment of network configuration. The adaptive configuration strategy generated by this process effectively matches the changes in network conditions, ensures the quality of network services, improves the utilization efficiency of network resources and the processing capacity of network traffic, ensures that key business traffic can obtain high-priority transmission guarantees in the network, and provides users with more stable and reliable network services.

[0139] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A communication system based on high-speed signal connection switch to divide VLAN into network cards, characterized in that: The system includes a network status monitoring module, a traffic prediction module, a resource allocation decision module, a traffic control execution module, a behavior pattern analysis module, a VLAN configuration adjustment module, a resource allocation optimization module and a network configuration adaptive adjustment module; The network status monitoring module is based on software-defined network technology, uses time series analysis and anomaly detection algorithms to monitor network bandwidth utilization, delay and packet loss rate indicators in real time, analyzes network behavior, and generates network performance indicators; The traffic prediction module uses a long short-term memory network model or a convolutional neural network model to perform forecast analysis on future traffic patterns and demands based on network performance indicators, and generates traffic trend forecasts; The resource allocation decision module uses optimization theory and reinforcement learning methods based on traffic trend prediction to dynamically adjust bandwidth allocation, priority and routing strategies through PID controller or deep reinforcement learning model to generate dynamic adjustment strategies; The flow control execution module deploys and adjusts the flow control strategy based on the dynamic adjustment strategy and uses software-defined network technology, automatically adjusts the network configuration according to real-time data and prediction results, and generates flow scheduling instructions; The behavior pattern analysis module uses a recurrent neural network model or a long short-term memory network model based on network performance indicators to identify and analyze behavior patterns in network traffic and generate behavior pattern insights; The behavioral pattern insights include periodic changes in network usage and abnormal traffic behavior pattern analysis; The behavior pattern analysis module includes an RNN analysis submodule, an LSTM analysis submodule and a pattern recognition submodule; The RNN analysis submodule executes the construction of the recurrent neural network model based on the network performance indicators, and uses the TensorFlow library to define the model, including the utilization rate, delay and packet loss rate indicator data of the input layer receiving bandwidth. The hidden layer captures the patterns and dependencies in the time series by setting the number of recurrent units to 128. The activation function selects relu to improve the model's ability to capture nonlinear features. The optimizer uses Adam to optimize the training process, and the loss function uses mean_squared_error. The model is trained for the network performance indicators to generate time-dependent characteristic analysis. The LSTM analysis submodule is based on time-dependent characteristic analysis, adopts a long short-term memory network model, and is implemented through the Keras library. The model structure design includes an input layer processing the results of the time-dependent characteristic analysis, and the LSTM layer is configured with 256 units to analyze the long-term dependency in the network behavior. The activation function uses sigmoid, the optimizer is set to Adam in the model training, and the loss function uses categorical_crossentropy. It analyzes the periodic changes and abnormal behaviors in the network behavior pattern and generates the network behavior pattern recognition result; The pattern recognition submodule implements pattern recognition technology based on the network behavior pattern recognition results, applies the support vector machine algorithm in the scikit-learn library, and configures the SVM model parameters including the penalty coefficient C=1.0 for the error term and the kernel function kernel=rbf to handle the nonlinear problem in the network behavior pattern recognition results. The SVM model is trained to distinguish between normal and abnormal behaviors in network traffic and generate insights into network traffic behavior. The VLAN configuration adjustment module uses the minimum cut maximum flow algorithm based on behavioral pattern insights to dynamically optimize VLAN division and network partitioning, improve the utilization efficiency of network resources and data transmission rate, and generate a network partition optimization solution; The resource allocation optimization module uses Q learning or deep Q network model based on the network partition optimization solution to adjust the network resource allocation strategy in real time and generate resource optimization configuration; The network configuration adaptive adjustment module is based on resource optimization configuration, adopts genetic algorithm and simulated annealing algorithm to optimize the flow distribution and route selection in the network, automatically adjusts the QoS strategy, matches the changes in network conditions, and generates an adaptive configuration strategy.

2. The communication system according to claim 1, wherein the VLAN is divided into network cards based on a high-speed signal connection switch, characterized in that: The network performance indicators specifically include real-time data and trend analysis of network bandwidth utilization, delay indicators and packet loss rate performance indicators; the traffic trend prediction includes future traffic trends, peak predictions and demand change analysis; the dynamic adjustment strategy specifically includes dynamic bandwidth allocation schemes, priority adjustment rules and routing selection strategies; the traffic scheduling instructions specifically include network configuration adjustment instructions and traffic scheduling instructions; the network partition optimization scheme specifically includes VLAN division schemes and network partition optimization strategies; the resource optimization configuration specifically includes bandwidth allocation optimization and routing selection optimization strategies; the adaptive configuration strategy specifically includes traffic engineering optimization and dynamic QoS adjustment rules.

3. The communication system according to claim 1, wherein the VLAN is divided into network cards based on a high-speed signal connection switch, characterized in that: The network status monitoring module includes a time series analysis submodule, an anomaly detection submodule and a real-time monitoring submodule; The time series analysis submodule is based on software defined network technology, collects network bandwidth utilization bandwidth_usage, delay latency and packet loss rate packet_loss indicators, adopts autoregressive moving average model, calls ARIMA function through Python's statsmodels library, sets model parameter order=(2,1,2), where the first term 2 represents the order of autoregressive term AR, 1 represents the difference order I, and the second term 2 represents the order of moving average term MA, performs fitting operation of time series data, and generates network indicator trend analysis; The anomaly detection submodule uses the isolation forest algorithm based on network indicator trend analysis, through the IsolationForest function of sklearn.ensemble, sets the parameter n_estimators=100 to represent the number of trees in the forest, and contamination=auto to automatically estimate the proportion of abnormal points in the data set, executes the training operation of the isolation forest model, and generates abnormal behavior recognition; The real-time monitoring submodule is based on abnormal behavior identification and uses software-defined network technology to set the data collection frequency parameter to frequency=5, which means that network performance data is collected every 5 seconds. The monitor function of the Traffic Monitor tool is used to set the monitoring parameter interval=5, and network performance data is collected every 5 seconds. Real-time data monitoring is performed to generate network performance indicators.

4. The communication system according to claim 1, wherein the VLAN is divided into network cards based on a high-speed signal connection switch, characterized in that: The traffic prediction module includes an LSTM prediction submodule, a CNN prediction submodule and a traffic pattern analysis submodule; The LSTM prediction submodule uses a long short-term memory network model based on network performance indicator data, and uses the TensorFlow library to build an LSTM network structure, including setting the number of hidden layer units to 128 to capture the long-term dependencies of the time series, setting the learning rate to 0.001 to control the speed of the learning process, setting the batch size to 32 to balance the training efficiency and network performance, using the Adam optimizer for parameter optimization, and using the mean_squared_error loss function to evaluate the difference between the model prediction value and the true value to generate a time series prediction analysis; The CNN prediction submodule adopts a convolutional neural network model based on the time series prediction analysis results, and uses the PyTorch library to define the CNN model structure, including adding a Conv1D layer, setting the number of filters to 64 to extract local features of traffic data, setting the kernel size to 3 to capture the relationship between adjacent data points, selecting ReLU as the activation function to introduce nonlinearity, applying the MaxPooling1D layer for feature downsampling, and using Adam as the optimizer. The loss function selects cross_entropy to handle classification problems and generate local feature analysis results; The traffic pattern analysis submodule is based on the time series prediction analysis results and the local feature analysis results, and comprehensively applies statistical analysis techniques. It performs data operations through Python's NumPy library, calculates the moving average to determine the overall trend of the traffic, sets the window size to 7 days to capture traffic fluctuations within a week, and calculates the standard deviation to identify abnormal traffic fluctuations. It generates a traffic trend forecast by calling the mean and std functions of the NumPy library.

5. The communication system based on high-speed signal connection switch dividing VLAN into network cards according to claim 1, characterized in that: The resource allocation decision module includes an optimization theory submodule, a PID control submodule and a first reinforcement learning submodule; The optimization theory submodule is based on traffic trend prediction, adopts optimization theory, uses linear programming method to build a resource allocation model, defines the objective function as minimize(total_cost) to minimize the total cost, sets constraints=[bandwidth >= 100Mbps, priority <= 10], captures the optimal resource allocation plan through the Simplex algorithm, and generates a resource optimization plan; The PID control submodule is based on the resource optimization scheme, adopts PID control theory, uses the proportional-integral-differential controller model to adjust the parameters Kp=0.1, Ki=0.01, Kd=0.001 to feedback the control system performance, adjusts the target value setpoint=desired_performance according to the real-time performance data, and automatically corrects the deviation of bandwidth allocation, priority and routing strategy through the adjustment algorithm to generate dynamic adjustment parameters; The first reinforcement learning submodule is based on dynamic adjustment parameters, adopts deep reinforcement learning, uses the Q learning framework for self-learning and matching, sets the environment state, action and reward mechanism, defines the learning rate learning_rate=0.01, the reward decay rate discount_rate=0.9, and the exploration rate epsilon=0.1 as the key parameters for model training, and adjusts the bandwidth allocation, priority and routing strategy through iterative training to match the changes in the network environment to generate a dynamic adjustment strategy.

6. The communication system based on high-speed signal connection switch dividing VLAN into network cards according to claim 1, characterized in that: The traffic control execution module includes an SDN deployment submodule, a policy adjustment submodule and a configuration automation submodule; The SDN deployment submodule uses software defined network technology to communicate with the network switch through the OpenFlow protocol based on a dynamic adjustment strategy, and executes the command item add-flow to add a new flow table entry to the switch, wherein the parameter match is set to ip, ip_src=192.168.1.1, ip_dst=192.168.2.1 to match the target data flow, and the action parameter action is set to output:1 to specify the port to which the data packet should be forwarded, and is completed by interacting with the REST API interface of the SDN controller OpenDaylight to generate an initial flow control configuration; The policy adjustment submodule is based on the initial flow control configuration, according to the real-time performance data of the network and the flow trend prediction, and calls the REST API to dynamically adjust the flow table items in the SDN controller, including sending a command to modify the flow table items to the controller through an HTTP PUT request, and the request body carries the new configuration information priority to adjust the processing priority of the flow table items, action=output:2 to change the forwarding port of the data packet, and establishes an updated flow control policy; The configuration automation submodule uses the Ansible network automation tool to define tasks to automatically deploy new network configurations based on the updated traffic control strategy. The task parameters specify the address of the target device hosts=switch1 and the required operations. The task - name: Update flow table in the Ansible playbook is used to define the operations of adding add-flow, modifying mod-flow and deleting del-flow flow table items. The SSH protocol is used to interact with network devices, automatically apply the single-pipe traffic control strategy, and generate traffic scheduling instructions.

7. The communication system based on high-speed signal connection switch dividing VLAN into network cards according to claim 1, characterized in that: The VLAN configuration adjustment module includes a VLAN automatic division submodule, a graph theory optimization submodule and a network partition adjustment submodule; The VLAN automatic division submodule is based on the insight of network traffic behavior, adopts the minimum cut maximum flow algorithm, and constructs a network flow graph through the Python library NetworkX. The nodes in the graph represent multiple switches in the network, and the edges represent the connections between switches. The capacity of the edges constrains the maximum data transmission capacity range of the connection. The maximum_flow function in NetworkX is used to specify the source node and sink node parameters, calculate the estimated maximum flow from the source node to the sink node in the network, and generate a network flow optimization plan; The graph theory optimization submodule is based on the network flow optimization scheme, and iteratively applies the Kernighan-Lin algorithm to perform refinement optimization of network partitions. By iteratively exchanging nodes in the network partitions, the sum of edge weights between partitions is reduced, and the traffic and delay across VLANs are reduced. A Python script is used to define the number and conditions of partition exchanges, as well as preset items for evaluating the efficiency of partition configuration before and after the exchange. By calculating the reduction in the sum of edge weights after the exchange, it is determined whether to perform the exchange, and a refined VLAN optimization configuration is generated; The network partition adjustment submodule is based on the refined VLAN optimization configuration, uses the network automation tool Ansible to deploy new VLAN configuration to network devices, defines Ansible playbook, which includes VLAN creation, modification and port allocation tasks for target network devices, and parameters including VLAN ID and port list are passed to network devices through playbook. By utilizing Ansible's automation capabilities, it communicates with network devices through the SSH protocol, executes configuration update tasks, and generates a network partition optimization plan.

8. The communication system based on high-speed signal connection switch dividing VLAN into network cards according to claim 1, characterized in that: The resource allocation optimization module includes a second reinforcement learning submodule, a resource optimization submodule and a delay reduction submodule; The second reinforcement learning submodule executes a deep Q learning algorithm based on the network partition optimization scheme, and uses the TensorFlow and Keras libraries in the Python environment to build a deep Q network. The state space is defined by the current resource usage of the network partition, and the action space is composed of decisions to increase or decrease the bandwidth allocation of multiple partitions, including reducing latency and improving bandwidth utilization. The implementation of the deep Q learning algorithm includes setting the learning rate to 0.01 and the discount rate to 0.9, using the model.fit method to train the model, cyclically updating the Q value to capture the optimal network resource allocation strategy, and generating a preliminary adjustment strategy for network resources; The resource optimization submodule performs resource allocation optimization based on the preliminary adjustment strategy of network resources. It adopts genetic algorithm, uses Python's DEAP library, sets the initial population size to 50, the crossover rate to 0.8, and the mutation rate to 0.2, and defines the fitness function to maximize the overall performance of the network, including bandwidth utilization and network delay. The genetic algorithm optimizes resource allocation in the iterative process through selection, crossover and mutation operations. During the execution process, the quality of each generation of population is evaluated according to the fitness function, and the optimal individual is selected as the resource allocation solution to generate a refined network resource adjustment strategy. The delay reduction submodule executes the simulated annealing algorithm based on the refined network resource adjustment strategy, sets the initial temperature to 1000, the cooling rate to 0.95, and the objective function to calculate the total network delay. Through the simanneal library of Python, during the simulated annealing process, resource allocation decisions are randomly selected for analysis. If the new resource allocation decision can reduce the network delay, the new resource allocation decision is accepted. After each iteration, the temperature is lowered until the cooling end is reached. The simulated annealing algorithm is executed to cyclically optimize the network resource allocation and generate the resource optimization configuration.

9. The communication system based on high-speed signal connection switch dividing VLAN into network cards according to claim 1, characterized in that: The network configuration adaptive adjustment module includes a genetic algorithm submodule, a simulated annealing submodule and a QoS policy adjustment submodule; The genetic algorithm submodule is based on resource optimization configuration, executes genetic algorithm to perform initial optimization of network traffic distribution and route selection, initializes a population, each individual represents a set of network configuration schemes, including traffic distribution ratio and route selection, and the algorithm parameter settings include population size of 100, crossover rate of 0.7, mutation rate of 0.01, fitness function based on network delay and bandwidth utilization calculation, and uses Python's DEAP library to create algorithm logic, including selection operation using roulette selection method, crossover operation using single point crossover, mutation operation using random bit mutation, and the number of iterations is set to 50 generations to generate genetic algorithm optimization results; The simulated annealing submodule is based on the optimization results of the genetic algorithm, and the simulated annealing algorithm is used to refine and adjust the above results. The initial temperature is set to 1000, the cooling rate is 0.95, the number of iterations for each temperature is 100, and the objective function is also based on network delay and bandwidth utilization. The algorithm is established through a Python script, in which the temperature reduction process adopts an exponential cooling scheme, and fine-tuning is performed by randomly selecting network configuration parameters, including adjusting routing weights or traffic distribution ratios, capturing the optimal network configuration, and generating simulated annealing adjustment results; The QoS policy adjustment submodule automatically adjusts the network QoS policy to match the changes in network conditions based on the simulated annealing adjustment results, converts the routing selection and traffic distribution policies in the adjustment results into QoS rules by analyzing them, performs priority control by setting DSCP marks, reserves the bandwidth at 20Mbps, adopts the fair queue algorithm for queue management policy, uses the CiscoIOS command ip qos dscp af31 for configuration, and automatically sends them to routers and switches through scripts to generate an adaptive configuration policy.

Images