Flow control and load balancing method in network communication

Abstract

The invention relates to the technical field of network communication, and discloses a flow control and load balancing method in network communication, which comprises the following steps of: collecting and preprocessing historical network communication data by using a network monitoring tool to form a basic data set; thirdly, performing flow characteristic analysis on the basic data set, including a flow peak time period, a user behavior pattern, a network topology structure and the like, so as to construct a network communication training data set; and then, a traffic prediction model and a resource allocation model are respectively constructed by adopting a dual machine learning model design, and a residual error fitting model is constructed by utilizing residual errors of the two models, so that accurate traffic control and load balancing are realized. By training and optimizing the model, the real-time network communication data can be predicted, the potential risk can be identified, and the targeted strategy can be formulated. According to the invention, the accuracy and flexibility of flow control and load balancing are improved, and reasonable utilization and efficient operation of network resources are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of network communications, and in particular to a method for flow control and load balancing in network communications. Background Art

[0002] In large-scale network systems, how to effectively control traffic and load balance to ensure the rational use and efficient operation of network resources has become a technical problem that needs to be solved urgently. Although the existing traffic control and load balancing methods in network communications can alleviate the problems of network congestion and uneven resource allocation to a certain extent, they still have many defects and are difficult to meet the growing network needs.

[0003] At present, common traffic control and load balancing methods mainly rely on static rules or simple dynamic adjustment strategies. For example, some methods trigger the traffic control mechanism by setting a fixed threshold. When the network traffic exceeds the threshold, measures such as current limiting and redirection are taken. However, this method lacks flexibility and cannot adapt to the dynamic changes of network traffic, which easily leads to waste of resources or excessive control.

[0004] Existing methods often use a single model or algorithm when performing traffic prediction and resource allocation, ignoring the complexity and diversity of network traffic. For example, some methods only build prediction models based on historical traffic data, without considering the impact of factors such as user behavior patterns and network topology on traffic. This results in limited accuracy of prediction results, making it difficult to accurately guide traffic control and resource allocation.

[0005] Taking a large e-commerce platform as an example, during promotional activities, network traffic will increase dramatically, but existing technologies are often unable to accurately predict this traffic peak and allocate resources in advance. Summary of the invention

[0006] The purpose of the present invention is to provide a method for flow control and load balancing in network communication to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solution: a flow control and load balancing method in network communication, the method comprising:

[0008] Use network monitoring tools to collect historical network communication data and pre-process the collected data to form a basic network communication data set;

[0009] Conduct traffic feature analysis on the basic data set, including traffic peak period analysis, user behavior pattern analysis, and network topology structure impact analysis, and merge the analysis result data with the basic data set to form a network communication training data set;

[0010] The dual machine learning model design is adopted to construct a traffic prediction model and a resource allocation model respectively. The machine learning algorithm is used to fit the network traffic and resource allocation. The residuals of the two models are used to construct a residual fitting model to estimate the causal effect of traffic prediction on resource allocation and achieve accurate traffic control and load balancing. The traffic prediction model is used to predict the traffic distribution and change trend in the network, and the resource allocation model is used to determine the resource allocation strategy of each network node or server.

[0011] The dual machine learning model is trained using the network communication training dataset and the model parameters are optimized using the back-propagation algorithm.

[0012] Use the trained model to predict real-time network communication data and identify potential traffic congestion and resource imbalance risks; and formulate targeted traffic control strategies and resource allocation strategies based on the prediction results;

[0013] Apply the developed traffic control strategy and load balancing solution to the actual network communication system, monitor the effective execution of the strategy, and make continuous fine-tuning and optimization of the strategy with the help of real-time monitoring system and feedback adjustment mechanism.

[0014] Preferably, the steps for implementing the traffic peak period analysis include:

[0015] S101: using a network monitoring tool to collect network communication data, including packet size, transmission time, source address, destination address, protocol type, and timestamp information;

[0016] S102: Based on the collected data set, the network communication data volume is arranged in chronological order with the timestamp as the reference to construct a traffic time series; for requirements of different time granularities, the traffic time series of corresponding granularities are constructed respectively;

[0017] S103: Use a sliding window algorithm combined with a threshold determination method to perform peak detection on the traffic time series; set a sliding window of a fixed size, slide on the time series, and calculate the average or median of the traffic in the window as a benchmark value; at the same time, set a dynamic threshold, which can be adjusted according to historical data or business needs and is a certain multiple of the benchmark value; when the traffic value in the window exceeds the threshold, it is determined to be a traffic peak period; the formula for calculating the sliding window average value is:

[0018]

[0019] Among them, F(i) is the flow value at time i, and w is the window size;

[0020] S104: For each traffic peak period detected, extract its characteristic parameters, including peak time, peak duration, peak traffic size, and traffic change trend before and after the peak; use these characteristic parameters to further analyze the periodicity of the peak.

[0021] Preferably, the data preprocessing method includes data cleaning, which specifically includes:

[0022] For missing values ​​in the data set, linear interpolation algorithm is used to fill them;

[0023] For the noisy data in the data set, the sliding average filtering algorithm is used to denoise the data.

[0024] Preferably, the data preprocessing method also includes:

[0025] Standardize the continuous data to eliminate the dimension difference. Use the Z-score method to standardize the data. Subtract the mean of each numerical data and divide it by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for standardization is: Z = (X-μ) / σ; where X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data;

[0026] Perform one-hot encoding conversion on categorical data.

[0027] Preferably, the steps for implementing the user behavior pattern analysis include:

[0028] S201: Collect user network communication data, including user ID, access time, access duration, access service, data transmission volume and operation type;

[0029] S202: Based on the collected data set, the network communication behavior of each user is arranged in chronological order with the user ID as an identifier to construct a user behavior sequence; the sequence includes the user's access behavior records at different time points, which is used to analyze the user's behavior pattern;

[0030] S203: using a clustering algorithm K-means to perform pattern recognition on user behavior sequences; first, defining a feature vector according to user behavior characteristics; then, using a clustering algorithm to divide users into different behavior pattern groups, including high-frequency users, low-frequency users, and nighttime active users; the user behavior characteristics include access frequency, access duration, and preferred services;

[0031] S204: For each identified user behavior pattern group, extract its characteristic parameters, including group size, behavior pattern distribution, typical behavior path and pattern conversion law;

[0032] S205: Using the characteristic parameters extracted in step S204, further analyze the stability and periodicity of the user behavior pattern and its relationship with the traffic peak.

[0033] Preferably, the steps for implementing the network topology impact analysis include:

[0034] S301: Collect network topology data, where the network topology data includes node information, link information, and network hierarchy;

[0035] S302: Based on the collected data set, a network topology model is constructed; using graph theory, the network topology is represented as a directed graph, where nodes represent network devices, edges represent the connection relationship between devices, and the weight of the edge represents the bandwidth or delay attribute of the link; the graph theory representation is: suppose the network topology is a graph G = (V, E), where V is a node set and E is an edge set; node v i ∈V, edge e ij =(v i ,v j )∈E;

[0036] S303: Extract features of the network topology model, including node degree distribution, path length, and clustering coefficient; these features are used to describe the connectivity, compactness, and location of key nodes of the network, and then analyze the impact of the network topology on traffic distribution and transmission performance.

[0037] Preferably, the dual machine learning model is designed as follows:

[0038] Build a traffic prediction model: Use the long short-term memory network (LSTM) algorithm to build a traffic prediction model. This model predicts the traffic distribution and change trend in the network by capturing the long-term dependencies in time series data.

[0039] Constructing a resource allocation model: A decision tree algorithm is used to construct a resource allocation model, which determines the resource allocation strategy of each network node or server based on traffic prediction results and network status;

[0040] Constructing residual fitting model: Using the residuals of the traffic prediction model and the resource allocation model, a linear regression model is used for fitting to estimate the causal effect of traffic prediction on resource allocation; the formula of the linear regression model is:

[0041]

[0042] in, is the predicted value of the residual fitting model, β0 and β1 are the regression coefficients, and y pred is the predicted value of the traffic prediction model, y actual is the actual flow value, ∈ is the error term;

[0043] The traffic prediction model, resource allocation model and residual fitting model are integrated together to form a dual machine learning model. The model parameters are optimized through the back propagation algorithm, so that the model can accurately predict network traffic and formulate corresponding resource allocation strategies to achieve traffic control and load balancing.

[0044] Preferably, by setting a resource utilization threshold, when the predicted resource utilization exceeds the threshold, it is determined that there is a risk of resource imbalance, and nodes or servers with resource shortage or surplus are determined.

[0045] Preferably, the flow control strategy includes: using a flow rate limiting method to perform flow control on identified flow congestion points, and dynamically adjusting the allocation of network bandwidth according to flow prediction results.

[0046] Preferably, the resource allocation strategy includes: supplementing resources for nodes or servers with resource shortages, including increasing computing resources and storage resources, according to resource imbalance risk assessment results; and releasing or reallocating resources for nodes or servers with excess resources.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] The present invention uses network monitoring tools to collect comprehensive network communication data and conducts in-depth traffic feature analysis, including multiple dimensions such as traffic peak time periods, user behavior patterns, and network topology structures, so as to more accurately predict the distribution and change trends of network traffic. Based on this, the constructed dual machine learning model can accurately guide traffic control and resource allocation, effectively avoid waste or excessive control of resources, and improve the utilization of network resources.

[0049] Compared with the existing methods that rely on static rules or simple dynamic adjustment strategies, the dual machine learning model used in the present invention can dynamically predict and adjust according to real-time network communication data. This enables the flow control and load balancing strategies to flexibly adapt to the dynamic changes of network traffic, ensuring the stable operation and efficient performance of the network system. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A processing step diagram of the flow control and load balancing method in network communication according to the present invention;

[0051] Figure 2 A diagram of the steps for analyzing peak traffic periods;

[0052] Figure 3 Schematic diagram of the design of the dual machine learning model. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] See also Figure 1-3 The present invention provides a technical solution: a flow control and load balancing method in network communication, the method comprising:

[0055] Use network monitoring tools, such as network traffic analyzers and log collection systems, to collect historical network communication data. The collected data includes but is not limited to traffic size, flow direction, timestamp, user identity, network node information, etc. Then, preprocess the collected data. The data preprocessing method includes data cleaning, which includes:

[0056] ① For missing values ​​in the data set, linear interpolation algorithm is used to fill them.

[0057] ② For the noisy data in the data set, the sliding average filtering algorithm is used to denoise the data.

[0058] ③ Standardize the continuous data to eliminate dimensional differences. Use the Z-score method to standardize the data. Subtract the mean of each numerical data and divide it by its standard deviation so that the processed data conforms to the standard normal distribution. The standardization formula is: Z = (X-μ) / σ; where X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data.

[0059] ④Perform one-hot encoding conversion on categorical data.

[0060] Conduct in-depth traffic feature analysis on the basic data set. This includes: conducting traffic peak period analysis to identify the peak and valley periods of network traffic; conducting user behavior pattern analysis to explore the habits and patterns of users accessing the network; conducting network topology structure impact analysis to evaluate the impact of network structure on traffic distribution. Merge the analysis result data with the basic data set to form a network communication training data set.

[0061] The dual machine learning model design is adopted to build a traffic prediction model and a resource allocation model respectively. The traffic prediction model uses machine learning algorithms such as time series analysis and neural networks to predict the traffic distribution and change trend in the network. The resource allocation model determines the resource allocation strategy of each network node or server based on the prediction results and network conditions. At the same time, the residuals of the two models are used to build a residual fitting model to estimate the causal effect of traffic prediction on resource allocation, so as to achieve accurate traffic control and load balancing.

[0062] The dual machine learning model is trained using the network communication training data set. The model parameters are continuously adjusted through optimization methods such as the back propagation algorithm to make the model's prediction results more accurate.

[0063] Use the trained model to predict real-time network communication data. Identify potential traffic congestion and resource imbalance risks through prediction results. Develop targeted traffic control strategies and resource allocation strategies based on prediction results and network conditions. The formulation of strategies needs to comprehensively consider multiple factors such as network performance, user experience, and resource utilization.

[0064] Apply the developed traffic control strategy and load balancing solution to the actual network communication system. Monitor and evaluate the implementation of the strategy through the real-time monitoring system. At the same time, with the help of the feedback adjustment mechanism, the strategy is continuously fine-tuned and optimized according to the real-time network status and user needs to ensure the stable operation and efficient performance of the network system.

[0065] This outbreak is further described below in conjunction with Examples 1 to 3:

[0066] Embodiment 1:

[0067] The specific implementation method of traffic peak period analysis is as follows:

[0068] Use the monitoring tool Wireshark deployed in the network to continuously collect network communication data. The collected data includes but is not limited to packet size, transmission time, source address, destination address, protocol type and timestamp information. Clean the collected raw data to remove duplication, missing and outliers to ensure the accuracy and completeness of the data.

[0069] Based on the preprocessed data set, the network communication data volume is arranged in chronological order with the timestamp as the benchmark to construct the traffic time series. In order to meet the analysis requirements of different time granularities, the traffic time series at the minute level, hour level, and day level are constructed respectively. For the hourly time series, the network communication data volume per hour is used as the data point of that hour.

[0070] The sliding window algorithm combined with the threshold judgment method is used to detect the peak value of the traffic time series. Set a sliding window of fixed size (such as 3 hours), slide it on the time series, and calculate the average value of the traffic in the window as the benchmark value. At the same time, set a dynamic threshold, which is a certain multiple of the benchmark value (such as 1.5 times). When the traffic value in the window exceeds the threshold, it is determined to be a traffic peak period. For example, during the 14:00-17:00 period of a certain day, the traffic continues to be higher than the threshold, so this period is determined to be a traffic peak period. The formula for calculating the sliding window average value is:

[0071]

[0072] Among them, F(i) is the flow value at time i, and w is the window size.

[0073] For each traffic peak period detected, its characteristic parameters are extracted. The characteristic parameters include peak time, peak duration, peak traffic size, traffic change trend before and after the peak, etc. For example, for the peak period of 14:00-17:00 mentioned above, characteristic parameters such as peak time of 15:30, peak duration of 3 hours, peak traffic size of 500Mbps, rising trend of traffic before the peak, and falling trend of traffic after the peak are extracted.

[0074] The specific implementation of user behavior pattern analysis is as follows:

[0075] Using monitoring tools and user log systems deployed in the network, users' network communication data is continuously collected. For example, in a large online shopping platform, users' network communication data in the past month is collected, including user ID, access time, access duration, access pages (such as product details page, shopping cart page, payment page, etc.), data transmission volume and operation type (such as click, browse, purchase, etc.).

[0076] In the data preprocessing stage, we first removed invalid, missing, and abnormal records, such as empty user IDs, negative access times, or access times that were outside a reasonable range, etc. Then, we performed a consistency check on the data to ensure that the access records corresponding to each user ID were arranged in chronological order, forming a basic user behavior data set.

[0077] Based on the preprocessed data set, the network communication behavior sequence of each user is constructed with the user ID as the identifier. For example, for a user with the user ID "U12345", the behavior sequence may be as follows: U12345: [(timestamp 1, visit product details page, browse, 5 minutes), (timestamp 2, visit shopping cart page, click, 2 minutes), ...]. Such a sequence can reflect the user's access behavior and its changes at different time points.

[0078] Select the K-means clustering algorithm to perform pattern recognition on user behavior sequences. Define feature vectors based on user behavior characteristics (such as access frequency, access duration, preferred pages, etc.). For example, feature vectors include:

[0079] Use the K-means clustering algorithm to divide users into different behavior pattern groups. In practical applications, it is necessary to select an appropriate number of clusters K based on business needs and data distribution. For example, after multiple experiments and evaluations, K = 5 was finally selected to divide users into five behavior pattern groups: high-frequency purchasing users, low-frequency browsing users, nighttime active users, weekend shopping users, and promotion-sensitive users.

[0080] For each identified user behavior pattern group, its characteristic parameters were extracted and analyzed in depth. For example, it was found that high-frequency purchasing users are usually active during specific time periods during the day and evening on weekdays, and prefer to visit product detail pages and payment pages; while night-active users mainly visit from late at night to early morning, and prefer to browse and add to shopping carts.

[0081] These characteristic parameters provide important inputs for subsequent traffic prediction and resource allocation models. Based on the periodicity of user behavior patterns, the network traffic distribution in a certain period of time in the future can be predicted, and targeted traffic control strategies and resource allocation strategies can be formulated accordingly.

[0082] The specific implementation method of network topology impact analysis is as follows:

[0083] Based on the preprocessed data set, the network topology model is constructed using graph theory. The network topology is represented as a directed graph, where nodes represent network devices (such as routers, switches, servers, etc.) and edges represent the connection relationship between devices. The weight of the edge can represent the bandwidth, delay and other attributes of the link to reflect the actual transmission capacity of the network. The graph theory can be represented as follows: Let the network topology be a graph G = (V, E), where V is the node set and E is the edge set; node v i ∈V, edge e ij =(v i ,v j )∈E. For example, in a network with 10 nodes and 15 links, a graph model with 10 vertices and 15 edges can be constructed. Each vertex represents a network device, and each edge represents the connection relationship between devices, and is annotated with attributes such as bandwidth and latency.

[0084] Feature extraction is performed on the network topology model to describe the connectivity, compactness and location of key nodes of the network. Common features include node degree distribution, path length, clustering coefficient, betweenness centrality, etc. For example, node degree represents the number of edges connected to a node. In a specific network, some nodes may be found to have a higher degree, which means that these nodes are key nodes or hubs in the network. Path length reflects the shortest distance between any two points in the network, which helps to identify bottleneck areas in the network. The clustering coefficient describes the degree of clustering of nodes in the network, while betweenness centrality measures the importance of a node in the network, that is, how many shortest paths pass through the node.

[0085] Use traffic data and topology features to perform correlation analysis to reveal the impact of network topology on traffic distribution and transmission performance. Statistical methods (such as correlation analysis) can be used for analysis. In a specific network, it may be found that traffic is positively correlated with the degree of certain nodes, that is, nodes with higher degrees tend to carry more traffic. It may also be found that traffic is negatively correlated with path length, that is, links with shorter paths tend to transmit more traffic. These correlations help identify key nodes and links that affect traffic transmission, as well as possible bottleneck areas.

[0086] Embodiment 2:

[0087] The dual machine learning model design approach includes:

[0088] Constructing a traffic prediction model: Long short-term memory network (LSTM) is used as the core algorithm of the traffic prediction model. In the specific implementation, the traffic data in the network communication training data set is used as the input of the LSTM model, and the model is trained to predict the traffic distribution and change trend in the future. The parameter setting and training process of the LSTM model are adjusted according to actual needs. Assuming there is a data set containing daily network traffic in the past year, the first 90% of the data can be used to train the LSTM model, and the last 10% of the data can be used to test the model performance. By adjusting the number of layers, number of hidden units, learning rate and other parameters of the LSTM model, the prediction accuracy of the model on the test set is maximized.

[0089] Constructing a resource allocation model: The decision tree algorithm is used as the core of the resource allocation model. In specific implementation, the traffic prediction results and network status (such as node load and link bandwidth) are used as the input of the decision tree model. Through training, the model can output the resource allocation strategy of each network node or server. The decision tree construction process includes feature selection (selecting the features that have the greatest impact on resource allocation), decision tree generation (dividing the data set according to the feature value and constructing the decision tree) and pruning (removing redundant branches and nodes).

[0090] According to the prediction results of the LSTM model and the current network status, a decision tree model is used to allocate appropriate resources (such as CPU, memory, bandwidth, etc.) to each node or server. By adjusting the depth of the decision tree, pruning strategy and other parameters, the resource allocation strategy can minimize resource costs while meeting network performance requirements.

[0091] Constructing residual fitting model: Using the residuals of the traffic prediction model and the resource allocation model (i.e. the difference between the predicted value and the actual value), a linear regression model is used for fitting. The formula of the linear regression model is:

[0092]

[0093] in, is the predicted value of the residual fitting model, β0 and β1 are the regression coefficients, and y pred is the predicted value of the traffic prediction model, y actual is the actual flow value, ∈ is the error term.

[0094] In the specific implementation, the residual is used as the input of the linear regression model, and the model is trained to estimate the causal effect of traffic prediction on resource allocation. The parameters of the linear regression model include the regression coefficient and the intercept, which are solved by optimization algorithms such as the least squares method.

[0095] If there is a certain deviation between the predicted traffic of the LSTM model and the actual traffic, the linear regression model can be used to fit the deviation to obtain the parameters of the residual fitting model. Then, the residual fitting model can be used to correct the prediction results of the LSTM model to improve the accuracy of the resource allocation strategy.

[0096] The dual machine learning model is trained using the network communication training dataset. The training dataset is divided into multiple batches, each containing a certain number of samples. Then, the back propagation algorithm is used to optimize the model parameters so that the loss function (such as mean square error) of the model on the training set is minimized.

[0097] The performance of the model is evaluated through methods such as cross-validation, and the model is fine-tuned and optimized based on the evaluation results. The training dataset is divided into multiple subsets, each of which is used to train and validate the model. Then, the model is adjusted based on the performance indicators on the validation set (such as accuracy, recall, and F1 score) to improve the generalization ability of the model.

[0098] Here is a Python code example for building a dual machine learning model:

[0099]

[0100] Use the trained model to predict real-time network communication data. Use the real-time network communication data as the input of the LSTM model to obtain the traffic prediction results for a period of time in the future. Then, use the prediction results as the input of the decision tree model to obtain the resource allocation strategy of each network node or server.

[0101] Develop targeted traffic control strategies and resource allocation strategies based on the prediction results and resource allocation strategies. Develop corresponding traffic control strategies (such as flow control or caching) based on the traffic peak periods and user behavior patterns in the prediction results. Develop corresponding resource allocation strategies (such as load balancing or resource scheduling) based on the resource allocation results and network status in the resource allocation strategy.

[0102] Embodiment 3:

[0103] Based on the results of traffic congestion identification and resource imbalance risk assessment, formulate targeted traffic control strategies and resource allocation strategies.

[0104] Traffic control strategy:

[0105] For identified traffic congestion points (such as the 9:00-10:00 am period), traffic shaping technology is used to limit or delay the traffic of some non-critical businesses to alleviate network congestion.

[0106] Dynamically adjust the allocation of network bandwidth. For example, increase bandwidth resources during the 9:00-10:00 am period to ensure that the communication quality of key businesses is not affected.

[0107] Resource allocation strategy:

[0108] For servers with insufficient resources (such as the servers predicted above), resources are supplemented, such as adding hardware resources such as CPU and memory, or optimizing server configuration to improve resource utilization efficiency.

[0109] Release or reallocate resources for nodes or servers with excess resources, such as allocating some idle resources to other nodes or servers that need resources, to improve overall resource utilization efficiency.

[0110] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0111] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for flow control and load balancing in network communication, characterized in that: The method comprises: Use network monitoring tools to collect historical network communication data and pre-process the collected data to form a basic network communication data set; Conduct traffic feature analysis on the basic data set, including traffic peak period analysis, user behavior pattern analysis, and network topology structure impact analysis, and merge the analysis result data with the basic data set to form a network communication training data set; The dual machine learning model design is adopted to construct a traffic prediction model and a resource allocation model respectively. The machine learning algorithm is used to fit the network traffic and resource allocation. The residuals of the two models are used to construct a residual fitting model to estimate the causal effect of traffic prediction on resource allocation and achieve accurate traffic control and load balancing. The traffic prediction model is used to predict the traffic distribution and change trend in the network, and the resource allocation model is used to determine the resource allocation strategy of each network node or server. The dual machine learning model is trained using the network communication training dataset and the model parameters are optimized using the back-propagation algorithm. Use the trained model to predict real-time network communication data and identify potential traffic congestion and resource imbalance risks; and formulate targeted traffic control strategies and resource allocation strategies based on the prediction results; Apply the developed traffic control strategy and load balancing solution to the actual network communication system, monitor the effective execution of the strategy, and make continuous fine-tuning and optimization of the strategy with the help of real-time monitoring system and feedback adjustment mechanism.

2. The method for flow control and load balancing in network communication according to claim 1, characterized in that: The steps to implement traffic peak period analysis include: S101: using a network monitoring tool to collect network communication data, including packet size, transmission time, source address, destination address, protocol type, and timestamp information; S102: Based on the collected data set, the network communication data volume is arranged in chronological order with the timestamp as the reference to construct a traffic time series; for requirements of different time granularities, the traffic time series of corresponding granularities are constructed respectively; S103: Use a sliding window algorithm combined with a threshold determination method to perform peak detection on the traffic time series; set a sliding window of a fixed size, slide on the time series, and calculate the average or median of the traffic in the window as a benchmark value; at the same time, set a dynamic threshold, which can be adjusted according to historical data or business needs and is a certain multiple of the benchmark value; when the traffic value in the window exceeds the threshold, it is determined to be a traffic peak period; the formula for calculating the sliding window average value is: Among them, F(i) is the flow value at time i, and w is the window size; S104: For each traffic peak period detected, extract its characteristic parameters, including peak time, peak duration, peak traffic size, and traffic change trend before and after the peak; use these characteristic parameters to further analyze the periodicity of the peak.

3. The method for flow control and load balancing in network communication according to claim 1, characterized in that: Data preprocessing methods include data cleaning, including: For missing values ​​in the data set, linear interpolation algorithm is used to fill them; For the noisy data in the data set, the sliding average filtering algorithm is used to denoise the data.

4. The method for flow control and load balancing in network communication according to claim 3, characterized in that: Data preprocessing methods also include: Standardize the continuous data to eliminate the dimension difference. Use the Z-score method to standardize the data. Subtract the mean of each numerical data and divide it by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for standardization is: Z = (X-μ) / σ; where X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data; Perform one-hot encoding conversion on categorical data.

5. The method for flow control and load balancing in network communication according to claim 1, characterized in that: The implementation steps of user behavior pattern analysis include: S201: Collect user network communication data, including user ID, access time, access duration, access service, data transmission volume and operation type; S202: Based on the collected data set, the network communication behavior of each user is arranged in chronological order with the user ID as an identifier to construct a user behavior sequence; the sequence includes the user's access behavior records at different time points, which is used to analyze the user's behavior pattern; S203: using a clustering algorithm K-means to perform pattern recognition on user behavior sequences; first, defining a feature vector according to user behavior characteristics; then, using a clustering algorithm to divide users into different behavior pattern groups, including high-frequency users, low-frequency users, and nighttime active users; the user behavior characteristics include access frequency, access duration, and preferred services; S204: For each identified user behavior pattern group, extract its characteristic parameters, including group size, behavior pattern distribution, typical behavior path and pattern conversion law; S205: Using the characteristic parameters extracted in step S204, further analyze the stability and periodicity of the user behavior pattern and its relationship with the traffic peak.

6. The method for flow control and load balancing in network communication according to claim 1, characterized in that: The implementation steps of network topology impact analysis include: S301: Collect network topology data, where the network topology data includes node information, link information, and network hierarchy; S302: Based on the collected data set, a network topology model is constructed; using graph theory, the network topology is represented as a directed graph, where nodes represent network devices, edges represent the connection relationship between devices, and the weight of the edge represents the bandwidth or delay attribute of the link; the graph theory representation is: suppose the network topology is a graph G = (V, E), where V is a node set and E is an edge set; node v i ∈V, edge e ij =(v i ,v j )∈E; S303: Extract features of the network topology model, including node degree distribution, path length, and clustering coefficient; these features are used to describe the connectivity, compactness, and location of key nodes of the network, and then analyze the impact of the network topology on traffic distribution and transmission performance.

7. The method for flow control and load balancing in network communication according to claim 1, characterized in that: The dual machine learning model is designed in the following way: Build a traffic prediction model: Use the long short-term memory network (LSTM) algorithm to build a traffic prediction model. This model predicts the traffic distribution and change trend in the network by capturing the long-term dependencies in time series data. Constructing a resource allocation model: A decision tree algorithm is used to construct a resource allocation model, which determines the resource allocation strategy of each network node or server based on traffic prediction results and network status; Construct residual fitting model: Use the residuals of traffic forecast model and resource allocation model to fit the linear regression model to estimate the causal effect of traffic forecast on resource allocation; The formula for the linear regression model is: in, is the predicted value of the residual fitting model, β0 and β1 are the regression coefficients, and y pred is the predicted value of the traffic prediction model, y actual is the actual flow value, ∈ is the error term; The traffic prediction model, resource allocation model and residual fitting model are integrated together to form a dual machine learning model. The model parameters are optimized through the back propagation algorithm, so that the model can accurately predict network traffic and formulate corresponding resource allocation strategies to achieve traffic control and load balancing.

8. The method for flow control and load balancing in network communication according to claim 1, characterized in that: By setting a resource utilization threshold, when the predicted resource utilization exceeds the threshold, it is determined that there is a risk of resource imbalance, and the nodes or servers with resource shortage or surplus are identified.

9. The method for flow control and load balancing in network communication according to claim 8, characterized in that: The flow control strategy includes: using a flow rate limiting method to perform flow control on identified flow congestion points, and dynamically adjusting the allocation of network bandwidth according to flow prediction results.

10. The method for flow control and load balancing in network communication according to claim 9, characterized in that: The resource allocation strategy includes: supplementing resources for nodes or servers with resource shortages, including increasing computing resources and storage resources, according to the resource imbalance risk assessment results; and releasing or reallocating resources for nodes or servers with excess resources.

Images