A method, medium and system for optimizing operator traffic scheduling

Abstract

The present invention provides an operator traffic scheduling optimization method, medium and system, which belongs to the technical field of traffic prediction model, and includes: firstly, dividing the bandwidth nodes on the operator side into fixed time windows of 5 minutes, and arranging the threshold of each node and the user side traffic rate in descending order. Then, the user side traffic is allocated to the operator side maximum threshold node according to the sorting order, and the node threshold sorting is updated in real time. When the user traffic is greater than the current maximum threshold, a suitable node is selected for peak cutting according to the remaining peak cutting capacity of the node. In order to improve the traffic distribution accuracy within the time window, each window is further divided into 15 20-second sub-windows, and the average traffic value of each sub-window is dynamically calculated. At the same time, the actual average value of the peak cutting node is recorded, and a new peak cutting node is added when it is lower than the current threshold. Finally, the pre-trained user side traffic prediction model is used to adjust the node threshold and user traffic sorting in advance to make good traffic distribution.

Description

Technical Field

[0001] The present invention belongs to the technical field of traffic prediction models, and in particular, relates to an operator traffic scheduling optimization method, medium and system. Background Art

[0002] In recent years, with the rapid development of information technology, various digital applications and services have become popular, and people's demand for network bandwidth has continued to grow. In order to meet users' growing demand for bandwidth, operators have continuously increased their investment in network infrastructure and expanded network bandwidth resources. However, due to the uncertainty of user behavior and the dynamic nature of network traffic, there is often a mismatch between the bandwidth resource supply on the operator side and the bandwidth demand on the user side, resulting in low bandwidth resource utilization efficiency.

[0003] At present, some bandwidth resource scheduling optimization methods adopted by the industry mainly include: 1) predicting future bandwidth demand based on user historical usage and allocating network resources in advance; 2) implementing dynamic pricing strategies when allocating bandwidth resources to different users to guide users to use bandwidth reasonably; 3) applying network function virtualization technology to dynamically adjust network topology and optimize resource allocation. However, these methods still have the problem of mismatch between bandwidth supply on the operator side and bandwidth demand on the user side in actual applications. Therefore, how to further improve the utilization efficiency of bandwidth resources on the operator side and optimize the bandwidth experience on the user side has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] In view of this, the present invention provides an operator traffic scheduling optimization method, medium and system, which can solve the problem of mismatch between operator-side bandwidth supply and user-side bandwidth demand in the prior art.

[0005] The present invention is achieved in that:

[0006] A first aspect of the present invention provides an operator traffic scheduling optimization method, comprising the following steps:

[0007] S10, dividing the time dimension of the bandwidth node on the operator side into time windows of fixed length, each time window is set to 5 minutes;

[0008] S20, arranging the thresholds of each bandwidth node on the operator side in descending order as threshold sorting, and arranging the bandwidth flow rate on the user side in descending order;

[0009] S30, allocating the user-side traffic to the maximum threshold node on the operator side in descending order, and updating the threshold ranking of the operator-side node after processing each user-side node;

[0010] S40, when the traffic of the user-side node is greater than the current maximum threshold of the operator-side, select a peak-cutting node to process the user-side node according to the remaining number of peak-cutting nodes of each operator-side bandwidth node;

[0011] S50, for each operator, further divide each time window into 15 sub-windows of 20 seconds, dynamically calculate the flow value before the start of each sub-window in each time window, and update the flow average value of the remaining sub-windows in the time window in real time;

[0012] S60, recording the actual average value of the peak shaving node when the peak shaving node quota is used, and comparing it with the threshold value of the current operator-side node. If the actual average value of the peak shaving node is lower than the threshold value of the current node, adding a new peak shaving node;

[0013] S70, using a pre-trained user-side traffic prediction model to predict the user-side traffic change trend in a future time window based on historical user-side bandwidth usage data to obtain predicted user traffic;

[0014] S80. Adjust the operator-side bandwidth node threshold sorting and the user-side bandwidth traffic rate sorting in S20 in advance according to the predicted user traffic to prepare for traffic allocation in the next time window, and then repeat steps S30 to S70.

[0015] On the basis of the above technical solution, an operator traffic scheduling optimization method of the present invention can also be improved as follows:

[0016] The threshold of each bandwidth node on the operator side is specifically the maximum flow value that can be processed by each bandwidth node on the operator side within the current time window.

[0017] Furthermore, the user-side bandwidth flow rate is specifically an average bandwidth usage of each user in the current time window.

[0018] Furthermore, the peak shaving nodes are specifically the top 5% nodes with the highest bandwidth values ​​that are not included in the billing under the 95 billing mode.

[0019] Furthermore, the remaining number of peak shaving nodes is specifically the number of nodes currently used for peak shaving by each operator-side bandwidth node; and selecting the peak shaving node to process the user-side node is specifically to distribute the user-side traffic that exceeds the current operator-side maximum threshold to the available peak shaving nodes, so that the actual traffic of the operator-side node does not exceed its threshold.

[0020] Furthermore, the dynamic calculation of the flow value before the start of each sub-window within each time window is specifically calculated based on the actual flow data of the sub-window that has passed and the predicted flow of the remaining sub-windows.

[0021] Furthermore, the user-side traffic prediction model adopts a long short-term memory network LSTM structure, specifically including an input layer, an LSTM layer, a fully connected layer and an output layer; the input layer receives historical time series data, the LSTM layer is used to capture the long-term dependencies of the time series, the fully connected layer is used for feature mapping, and the output layer generates prediction results.

[0022] The step of constructing a training data set for the user-side traffic prediction model specifically includes:

[0023] 1) Collect historical user-side bandwidth usage data;

[0024] 2) preprocessing the historical user-side bandwidth usage data, including denoising, normalization, and missing value processing;

[0025] 3) Divide the preprocessed data into sequence samples of fixed length in chronological order;

[0026] 4) For each sequence sample, use the sliding window method to generate input features and corresponding target values;

[0027] 5) Randomly divide the generated input feature and target value pairs into training set, validation set and test set.

[0028] The steps of training the user-side traffic prediction model specifically include:

[0029] 1) Initialize LSTM network parameters;

[0030] 2) Use the training set data for forward propagation and calculate the predicted value;

[0031] 3) Calculate the mean square error loss between the predicted value and the actual value;

[0032] 4) Update network parameters using back-propagation algorithm and optimizer;

[0033] 5) Evaluate the model performance on the validation set and save the model if the performance improves;

[0034] 6) Repeat steps 2) to 5) until the preset number of iterations is reached or the model converges;

[0035] 7) Use the test set to evaluate the performance of the final model.

[0036] Furthermore, the remaining sub-windows in the time window are specifically all sub-windows after the current sub-window.

[0037] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned operator traffic scheduling optimization method.

[0038] A third aspect of the present invention provides an operator traffic scheduling optimization system, which includes the above-mentioned computer-readable storage medium.

[0039] Compared with the prior art, the beneficial effects of the operator traffic scheduling optimization method, medium and system provided by the present invention are: 1. Dividing the time dimension into time windows of fixed length provides a basis for subsequent traffic analysis and scheduling. This time window division method enables traffic data to be better tracked and predicted, providing support for optimizing scheduling decisions.

[0040] 2. Arrange the processing capacity threshold of the operator-side nodes and the actual traffic rate on the user side in descending order to ensure that the traffic is preferentially allocated to nodes with stronger processing capabilities to maximize resource utilization.

[0041] 3. When the user-side traffic exceeds the maximum threshold of the operator-side node, dynamically select available peak-cutting nodes for processing to improve the overall traffic processing capacity. This dynamic peak-cutting strategy can effectively deal with sudden traffic peaks and ensure service quality.

[0042] 4. The time window is further subdivided into sub-windows, and the traffic forecast is updated in real time to provide support for the next step of traffic scheduling optimization. This fine-grained traffic tracking and prediction makes scheduling decisions more accurate.

[0043] 5. Use the long short-term memory (LSTM) neural network model to predict future user-side traffic trends, adjust resource allocation strategies in advance, and improve overall traffic scheduling efficiency. The LSTM model can effectively capture the long-term dependencies of time series data, thereby making more accurate traffic predictions.

[0044] Compared with the prior art, the traffic scheduling optimization method of the present invention can better meet the user's bandwidth requirements and improve the resource utilization rate on the operator side. Specifically, the method can significantly reduce the congestion caused by traffic peaks and optimize the user experience; dynamically adjust the number of peak-cutting nodes to improve the overall traffic processing capacity; use accurate traffic prediction to make resource allocation in advance and improve scheduling efficiency. In general, the method can effectively solve the problem of mismatch between bandwidth supply on the operator side and bandwidth demand on the user side in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A flow chart of the method provided by the present invention;

[0046] Figure 2 Dynamic adjustment diagram for operator bandwidth node threshold;

[0047] Figure 3 It is a graph of user bandwidth usage;

[0048] Figure 4 It is the flow change diagram within the time window;

[0049] Figure 5 This is a comparison chart of user total traffic prediction and actual value. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0051] like Figure 1 FIG. 1 is a flow chart of a method for optimizing operator traffic scheduling provided by the first aspect of the present invention. The method comprises the following steps:

[0052] S10, dividing the time dimension of the bandwidth node on the operator side into time windows of fixed length, each time window is set to 5 minutes;

[0053] S20, arranging the thresholds of each bandwidth node on the operator side in descending order as threshold sorting, and arranging the bandwidth flow rate on the user side in descending order;

[0054] S30, allocating the user-side traffic to the maximum threshold node on the operator side in descending order, and updating the threshold ranking of the operator-side node after processing each user-side node;

[0055] S40: When the traffic of the user-side node is greater than the current maximum threshold of the operator-side, a peak-shaving node is selected to process the user-side node according to the remaining number of peak-shaving nodes of each operator-side bandwidth node;

[0056] S50, for each operator, further divide each time window into 15 sub-windows of 20 seconds, dynamically calculate the flow value before the start of each sub-window in each time window, and update the flow average value of the remaining sub-windows in the time window in real time;

[0057] S60, recording the actual average value of the peak shaving node when the peak shaving node quota is used, and comparing it with the threshold value of the current operator-side node. If the actual average value of the peak shaving node is lower than the threshold value of the current node, adding a new peak shaving node;

[0058] S70, using a pre-trained user-side traffic prediction model to predict the user-side traffic change trend in a future time window based on historical user-side bandwidth usage data to obtain predicted user traffic;

[0059] S80. Adjust the operator-side bandwidth node threshold sorting and the user-side bandwidth traffic rate sorting in S20 in advance according to the predicted user traffic to prepare for the traffic allocation in the next time window, and then repeat steps S30 to S70.

[0060] The specific implementation methods of the above steps are described in detail below:

[0061] The specific implementation method of step S10 is: divide the time dimension of the bandwidth node on the operator side into time windows of fixed length, and each time window is set to 5 minutes. The purpose of this is to facilitate the subsequent management and analysis of traffic data in different time windows. Dividing the time dimension into time windows of fixed length can better track and predict traffic changes at different times, and provide basic data support for subsequent traffic scheduling and optimization.

[0062] The specific implementation method of step S20 is: first, the thresholds of each bandwidth node on the operator side are arranged in descending order to obtain a threshold ranking. The thresholds of each bandwidth node on the operator side specifically refer to the maximum traffic value that each bandwidth node on the operator side can handle in the current time window. At the same time, the bandwidth traffic rate on the user side is also arranged in descending order, and the bandwidth traffic rate on the user side specifically refers to the average bandwidth usage of each user in the current time window. By arranging in this descending order, it can be ensured that the traffic on the user side is subsequently allocated to the operator side node with stronger processing capability, thereby maximizing the resource utilization on the operator side.

[0063] The specific implementation method of step S30 is: according to the descending order obtained in step S20, the user-side traffic is distributed to the maximum threshold node on the operator side in sequence. After each user-side node is processed, the threshold sorting of the operator-side node needs to be updated to ensure that the traffic is always distributed according to the latest threshold information. The purpose of this is to distribute the user-side traffic to the operator-side node with stronger processing capability as much as possible, thereby improving the overall traffic processing efficiency.

[0064] The specific implementation method of step S40 is: when the traffic of a certain user-side node is greater than the current operator-side maximum threshold, a peak-cutting node is selected to process the user-side node according to the remaining number of peak-cutting nodes of each operator-side bandwidth node. The peak-cutting node specifically refers to the top 5% nodes with the highest bandwidth value that are not included in the billing under the 95 billing mode. The number of nodes currently used for peak cutting of each operator-side bandwidth node is the remaining number of its peak-cutting nodes. Selecting a peak-cutting node to process the user-side traffic that exceeds the current maximum threshold can make the actual traffic of the operator-side node not exceed its threshold, thereby improving the overall traffic processing capacity.

[0065] The specific implementation method of step S50 is: for each operator, each time window is further divided into 15 20-second sub-windows. By dynamically calculating the flow value before the start of each sub-window in each time window, the flow average value of the remaining sub-windows in the time window is updated in real time. The dynamic calculation of the flow value before the start of each sub-window in each time window is specifically calculated based on the actual flow data of the sub-window that has passed and the predicted flow of the remaining sub-windows. The remaining sub-windows in the time window specifically refer to all sub-windows after the current sub-window. The purpose of this is to more finely track and predict the flow changes in each time window, and provide more accurate data support for subsequent flow scheduling.

[0066] The specific implementation of step S60 is: record the actual average value of the peak-shaving node when using the peak-shaving node quota, and compare it with the threshold value of the current operator-side node. If the actual average value of the peak-shaving node is lower than the threshold value of the current node, add a new peak-shaving node. The purpose of this is to dynamically adjust the number of peak-shaving nodes to ensure that there are enough node quotas available when peak shaving is required, thereby improving the overall traffic processing capacity.

[0067] The specific implementation method of step S70 is: using a pre-trained user-side traffic prediction model, based on historical user-side bandwidth usage data, predict the user-side traffic change trend in the future time window, thereby obtaining predicted user traffic. The user-side traffic prediction model adopts a long short-term memory network (LSTM) structure, including an input layer, an LSTM layer, a fully connected layer, and an output layer. Among them, the input layer receives historical time series data, the LSTM layer is used to capture the long-term dependency of the time series, the fully connected layer is used for feature mapping, and the output layer generates a prediction result.

[0068] The specific implementation method of step S80 is: according to the predicted user traffic, the operator-side bandwidth node threshold sorting and the user-side bandwidth traffic rate sorting in step S20 are adjusted in advance to prepare for the traffic allocation of the next time window. Then the operations of steps S30 to S70 are repeated. The purpose of doing so is to use the prediction of future traffic change trends to adjust the resource allocation strategy in advance, so as to better cope with the upcoming traffic changes and improve the overall traffic scheduling efficiency.

[0069] In general, the invention achieves the following key functions:

[0070] 1. Divide the time dimension into time windows of fixed length to provide a basis for subsequent traffic analysis and scheduling.

[0071] 2. Arrange the thresholds of the operator-side nodes and the traffic rates on the user-side in descending order to ensure that traffic is allocated preferentially to nodes with stronger processing capabilities.

[0072] 3. When the user-side traffic exceeds the maximum threshold of the operator-side node, dynamically select available peak-shaving nodes for processing to improve the overall traffic processing capacity.

[0073] 4. Further subdivide the time window into sub-windows, update the traffic forecast in real time, and provide support for the next step of traffic scheduling optimization.

[0074] 5. Dynamically adjust the number of peak shaving nodes to ensure that sufficient quota is available when peak shaving is needed.

[0075] 6. Use the LSTM model to predict future user-side traffic change trends, adjust resource allocation strategies in advance, and improve the overall efficiency of traffic scheduling.

[0076] Through the specific implementation of the above steps, the invention can effectively optimize the operator's traffic scheduling, improve the resource utilization rate and overall traffic processing capacity on the operator side, and ultimately achieve better scheduling and management of user-side traffic.

[0077] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned operator traffic scheduling optimization method.

[0078] A third aspect of the present invention provides an operator traffic scheduling optimization system, which includes the above-mentioned computer-readable storage medium.

[0079] Specifically, the principle of the present invention is: through the three major technical supports of time window division, threshold sorting and traffic prediction, refined management and dynamic optimization scheduling of bandwidth resources on the operator side are realized.

[0080] First, the time dimension is divided into time windows of fixed length to provide basic data support for subsequent traffic analysis and scheduling. This time window division method makes it possible to better track and predict traffic changes at different times, providing a basis for optimizing decisions.

[0081] Secondly, the processing capacity thresholds of each node on the operator side and the actual traffic rate on the user side are arranged in descending order. The purpose of this is to ensure that the traffic on the user side is first allocated to the operator side nodes with stronger processing capabilities, thereby maximizing the overall resource utilization. At the same time, when the user side traffic exceeds the processing capacity of the current maximum threshold node, the available peak-cutting nodes are dynamically selected to process the excess. This flexible peak-cutting strategy can effectively cope with sudden traffic peaks and ensure service quality.

[0082] Thirdly, the present invention further subdivides each time window into multiple sub-windows, dynamically calculates the flow value before the start of each sub-window, and updates the flow average value forecast of the remaining sub-windows in the time window in real time. This fine-grained flow tracking and prediction provides more accurate data support for subsequent flow scheduling optimization.

[0083] Finally, the present invention adopts a long short-term memory (LSTM) neural network model to predict the user-side traffic change trend in the future time window based on the historical user-side bandwidth usage data. The LSTM model is good at capturing the long-term dependencies of time series data and can make more accurate traffic predictions. Using this prediction information, the threshold sorting of the operator-side nodes and the user-side traffic rate sorting can be adjusted in advance to prepare for the traffic scheduling of the next time window and improve the overall scheduling efficiency.

[0084] In general, the traffic scheduling optimization method of the present invention makes full use of the three core technologies of time window, threshold sorting and traffic prediction, and realizes the refined management and dynamic optimization scheduling of bandwidth resources on the operator side. The method can effectively solve the problem in the prior art that there is often a mismatch between the bandwidth resource supply on the operator side and the bandwidth demand on the user side, resulting in low bandwidth resource utilization efficiency.

[0085] In order to better understand and implement the present invention, a specific embodiment 1 of the method of the present invention is provided below. The specific implementation method of each step in this embodiment 1 is described as follows:

[0086] Step S10 mentions dividing the time dimension of the bandwidth node on the operator side into time windows of fixed length, and each time window is set to 5 minutes. The purpose of this is to facilitate the subsequent management and analysis of traffic data in different time windows. Dividing the time dimension into time windows of fixed length can better track and predict traffic changes at different times, and provide basic data support for subsequent traffic scheduling and optimization. Suppose the length of the time window is , then minute.

[0087] In step S20, the thresholds of each bandwidth node on the operator side are arranged in descending order to obtain a threshold ranking. The thresholds of each bandwidth node on the operator side specifically refer to the maximum flow value that each bandwidth node on the operator side can handle within the current time window, which is recorded as ,in Indicates At the same time, the bandwidth flow rate on the user side is also sorted in descending order. The bandwidth flow rate on the user side specifically refers to the average bandwidth usage of each user in the current time window, recorded as ,in Indicates This descending order arrangement ensures that the user-side traffic is subsequently allocated to the operator-side node with stronger processing capabilities, thereby maximizing the resource utilization on the operator side.

[0088] Step S30 mentions that the user-side traffic is distributed to the operator-side maximum threshold node in descending order according to the descending order obtained in step S20. After each user-side node is processed, the threshold sorting of the operator-side node needs to be updated to ensure that traffic is always distributed according to the latest threshold information. The purpose of this is to distribute the user-side traffic to the operator-side node with stronger processing capabilities as much as possible, thereby improving the overall traffic processing efficiency. Assume that the current processing is to the first , then Assigned to the current maximum threshold node , and update , and then re-align Sort in descending order.

[0089] Step S40 mentions that when the traffic of a user-side node is greater than the current operator-side maximum threshold, a peak-cutting node is selected to process the user-side node according to the remaining number of peak-cutting nodes of each operator-side bandwidth node. The peak-cutting node specifically refers to the top 5% highest bandwidth value nodes that are not included in the billing under the 95 billing mode, and is recorded as ,in Indicates The number of nodes currently used for peak shaving at each operator-side bandwidth node is the number of remaining peak shaving nodes, recorded as Selecting a peak-cutting node to process user-side traffic that exceeds the current maximum threshold can make the actual traffic of the operator-side node not exceed its threshold, thereby improving the overall traffic processing capacity. Suppose a user-side traffic is , and the current maximum threshold node , then select Available peak shaving nodes To handle Partial flow.

[0090] Step S50 mentions that for each operator, each time window is further divided into 15 sub-windows of 20 seconds. By dynamically calculating the flow value before the start of each sub-window in each time window, the flow average value of the remaining sub-windows in the time window is updated in real time. Divided into sub-windows, the length of each sub-window is Seconds. sub-windows, and set their start time to , the end time is , then the actual flow in the sub-window is ,in Indicates time The instantaneous flow at the moment. The actual flow and remaining The predicted flow of each sub-window is used to dynamically calculate the average flow of the remaining sub-windows in the current time window. The purpose of this is to more finely track and predict the flow changes within each time window and provide more accurate data support for subsequent flow scheduling.

[0091] Step S60 mentions recording the actual average value of the peak-shaving node when the peak-shaving node quota is used, and comparing it with the threshold value of the current operator-side node. If the actual average value of the peak-shaving node is lower than the threshold value of the current node, a new peak-shaving node is added. Peak Shaving Node To handle the problem of exceeding the operator side node The actual average value of the peak shaving node is If the average value is less than , then add a new peak-shaving node The purpose of this is to dynamically adjust the number of peak-shaving nodes to ensure that there are enough node quotas available when peak-shaving is needed, thereby improving the overall traffic processing capacity.

[0092] Step S70 mentions using a pre-trained user-side traffic prediction model to predict the user-side traffic change trend in the future time window based on historical user-side bandwidth usage data, thereby obtaining predicted user traffic. The user-side traffic prediction model adopts a long short-term memory network (LSTM) structure, including an input layer, an LSTM layer, a fully connected layer, and an output layer. Among them, the input layer receives historical time series data, the LSTM layer is used to capture the long-term dependency of the time series, the fully connected layer is used for feature mapping, and the output layer generates prediction results. The basic structure of the LSTM network is as follows:

[0093] The input layer receives time series data , denoted as ,in Represents the length of historical data. The core formula of the LSTM layer is as follows:

[0094] ;

[0095] in, is the sigmoid activation function, is the hyperbolic tangent activation function, is element-wise multiplication. , , They are forget gate, input gate and output gate respectively. is the candidate cell state, is the current cell state, The current hidden state. , , , is the model parameter.

[0096] The fully connected layer is used to map the output features of the LSTM layer to the final prediction output ,Right now ,in and are the parameters of the fully connected layer.

[0097] Through this LSTM network structure, the long-term dependencies of user-side traffic time series data can be effectively captured, and traffic prediction results within future time windows can be generated.

[0098] In step S80, it is mentioned that according to the predicted user traffic, the operator-side bandwidth node threshold sorting and the user-side bandwidth traffic rate sorting in step S20 are adjusted in advance to prepare for the traffic allocation in the next time window. Then the operations of steps S30 to S70 are repeated. Assume that the predicted user traffic in the current time window is ,in is the predicted number of users. Node threshold on the operator side and user-side traffic rate Reorder and prepare for the traffic allocation in the next time window. The purpose of this is to use the prediction of future traffic change trends to adjust the resource allocation strategy in advance, so as to better cope with the upcoming traffic changes and improve the overall traffic scheduling efficiency.

[0099] In general, this embodiment 1 realizes the following key functions through the specific implementation of the above steps:

[0100] 1. Divide the time dimension into time windows of fixed length to provide a basis for subsequent traffic analysis and scheduling.

[0101] 2. Arrange the thresholds of the operator-side nodes and the traffic rates on the user-side in descending order to ensure that traffic is allocated preferentially to nodes with stronger processing capabilities.

[0102] 3. When the user-side traffic exceeds the maximum threshold of the operator-side node, dynamically select available peak-shaving nodes for processing to improve the overall traffic processing capacity.

[0103] 4. Further subdivide the time window into sub-windows, update the traffic forecast in real time, and provide support for the next step of traffic scheduling optimization.

[0104] 5. Dynamically adjust the number of peak shaving nodes to ensure that sufficient quota is available when peak shaving is needed.

[0105] 6. Use the LSTM model to predict future user-side traffic change trends, adjust resource allocation strategies in advance, and improve the overall efficiency of traffic scheduling.

[0106] Through the specific implementation of the above steps, this embodiment 1 can effectively optimize the operator's traffic scheduling, improve the resource utilization and overall traffic processing capacity on the operator side, and ultimately achieve better scheduling and management of user-side traffic.

[0107] In order to further better understand and implement the present invention, the following provides Example 2 of a specific application scenario of the present invention:

[0108] In the service area of ​​a certain operator A enterprise, the user bandwidth usage during peak hours every day usually exceeds the overall processing capacity of the network, resulting in a decline in some user experience. In order to optimize traffic scheduling and improve user experience, the enterprise decided to adopt the traffic scheduling optimization method proposed in the present invention.

[0109] First, the enterprise divides the time dimension of the bandwidth node on the operator side into fixed-length 5-minute time windows. For example, from 18:00 to 19:00 on a certain day, there are 12 5-minute time windows.

[0110] Next, the enterprise sorts the processing capacity thresholds of each bandwidth node in the period in descending order, and also sorts the actual bandwidth usage rate on the user side in descending order. Assuming that in the hour from 18:00 to 19:00, the enterprise has 4 main bandwidth nodes, whose maximum processing capacities are 100Mbps, 90Mbps, 80Mbps and 70Mbps respectively, the sorted alues ​​are Mbps, Mbps, Mbps and Mbps. At the same time, there are 10 users in this period, and their bandwidth usage rates are Mbps, Mbps, Mbps, Mbps, Mbps, Mbps, Mbps, Mbps, Mbps and Mbps.

[0111] Then, the enterprise distributes the user-side traffic to the maximum threshold node on the operator side in descending order. Taking the first time window 18:00-18:05 as an example, first Mbps allocated to Mbps, at this time Mbps; then Mbps allocated to Mbps, at this time Mbps; Mbps allocated to Mbps, at this time Mbps; finally Mbps allocated to Mbps, at this time Mbps.

[0112] In this process, if the traffic on a user side is greater than the current maximum threshold on the operator side, the enterprise will select a peak-shaving node for processing based on the number of remaining peak-shaving nodes on each operator side. The bandwidth usage rate is Mbps, and the current maximum threshold node Mbps, then exceeds Part Mbps needs to be processed by the peak-shaving nodes. The enterprise has a total of 6 available peak-shaving nodes. to ,in to Used, Remaining and Two nodes. Therefore, the enterprise chooses and To handle The excess traffic makes the actual traffic not exceed the threshold of the current node.

[0113] In order to more precisely track and predict traffic changes within each time window, the enterprise further divides the hour from 18:00 to 19:00 into 15 20-second sub-windows. By dynamically calculating the traffic value before the start of each sub-window, the enterprise can update the traffic average forecast of the remaining sub-windows within the time window in real time. Taking the first time window from 18:00 to 18:05 as an example, the actual traffic data of the first four 20-second sub-windows are 90Mbps, 85Mbps, 80Mbps and 75Mbps respectively. Based on these data, the enterprise can predict that the average traffic of the remaining 11 sub-windows is 72.5Mbps.

[0114] At the same time, the enterprise uses the pre-trained LSTM-based user-side traffic prediction model, combined with historical user bandwidth usage data, to predict the user-side traffic change trend in the future time window. Taking the second time window 18:05-18:10 as an example, the model predicts that the total user-side traffic in this time window is 640Mbps.

[0115] With the above real-time tracking of the traffic status in the current time window and the prediction of the total user-side traffic in the future time window, the enterprise can adjust the threshold sorting of the operator-side nodes and the user-side traffic rate sorting in advance to prepare for the traffic allocation in the next time window. For example, for the third time window 18:10-18:15, the enterprise can adjust the threshold sorting of the operator-side nodes and the user-side traffic rate sorting in advance based on the prediction results. The threshold value is adjusted from 100Mbps to 110Mbps. The threshold is adjusted from 90Mbps to 95Mbps to better meet the upcoming user-side traffic demand.

[0116] In order to ensure that there are enough node quotas available when peak shaving is needed, the enterprise will also dynamically monitor the use of peak shaving nodes. and Two peak-cutting nodes are used to handle part of the traffic. The actual average value of these two nodes is 85Mbps, which is lower than the current maximum threshold node. Mbps, so the enterprise decided to add a new peak-shaving node .

[0117] Through the implementation of the above steps, the company successfully optimized traffic scheduling during the peak period of 18:00-19:00. The following charts show the relevant data intuitively:

[0118] Attached Figure 2 This is a diagram of the operator's bandwidth node threshold dynamic adjustment. This diagram shows the operator's four bandwidth nodes ( arrive ) Dynamic adjustment of the processing capacity threshold in the hour from 18:00 to 19:00. Unlike the original description, it can be seen that the threshold does not remain unchanged, but gradually increases over time. This dynamic adjustment reflects one of the core advantages of the method of the present invention: adjusting the node processing capacity in advance according to the predicted traffic demand to better cope with the upcoming peak traffic. For example, The gradual increase from the initial 100Mbps to 150Mbps reflects the system's adaptive ability.

[0119] Attached Figure 3 This is a user bandwidth usage chart. This chart uses a bar chart to show the actual bandwidth usage rate of 10 users in a specific time period. The chart uses gradient tones to enhance the visual effect, and the specific value is marked on the top of each column. This presentation method not only intuitively shows the difference in bandwidth usage between users, but also provides an important basis for subsequent traffic allocation strategies. Through this visualization, operators can better understand the distribution of user needs and optimize resource allocation.

[0120] Attached Figure 4 This is a flow change diagram within a time window. This figure shows the flow changes in 15 20-second sub-windows within a 5-minute time window. Different from the original description, a more dynamic change process is shown here, rather than a simple average prediction. The curve in the figure shows the real-time change of flow, from the initial downward trend to the later upward trend. This fine-grained flow monitoring enables the system to respond quickly to short-term flow fluctuations and provides a basis for real-time adjustments. The shaded area indicates the possible fluctuation range, reflecting the uncertainty of the prediction.

[0121] Attached Figure 5This is a comparison chart of the total user traffic prediction and actual values. This chart compares the total user traffic predicted by the LSTM model and the actual observed traffic value. The closeness of the two curves reflects the accuracy of the prediction model, while the filled shaded area represents the error between the prediction and the actual value. It can be seen that the predicted value is slightly higher than the actual value overall. This conservative prediction strategy helps the system prepare for possible traffic peaks. The continuous upward trend shown in the figure also shows that during this peak period, the system needs to be constantly adjusted to cope with the increased traffic demand.

[0122] Through the above specific embodiments and data analysis, we can see the specific effect of the traffic scheduling optimization method proposed in the present invention in practical applications:

[0123] 1. Reasonably divide the time window to provide a basis for subsequent traffic analysis and scheduling;

[0124] 2. Allocate traffic according to node processing capacity and actual user needs to maximize resource utilization;

[0125] 3. Dynamically select peak-shaving nodes to cope with traffic peaks and ensure service quality;

[0126] 4. Detailed tracking and prediction of internal traffic changes to provide more accurate data support for scheduling decisions;

[0127] 5. Use the LSTM model to predict future traffic trends, adjust resource allocation strategies in advance, and improve overall efficiency.

[0128] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for optimizing operator traffic scheduling, characterized in that: The following steps are involved: S10, dividing the time dimension of the bandwidth node on the operator side into time windows of fixed length, each time window is set to 5 minutes; S20, arranging the thresholds of each bandwidth node on the operator side in descending order as threshold sorting, and arranging the bandwidth flow rate on the user side in descending order; S30, allocating the user-side traffic to the maximum threshold node on the operator side in descending order, and updating the threshold ranking of the operator-side node after processing each user-side node; S40, when the traffic of the user-side node is greater than the current maximum threshold of the operator-side, select a peak-cutting node to process the user-side node according to the remaining number of peak-cutting nodes of each operator-side bandwidth node; S50, for each operator, further divide each time window into 15 sub-windows of 20 seconds, dynamically calculate the flow value before the start of each sub-window in each time window, and update the flow average value of the remaining sub-windows in the time window in real time; S60, recording the actual average value of the peak shaving node when the peak shaving node quota is used, and comparing it with the threshold value of the current operator-side node. If the actual average value of the peak shaving node is lower than the threshold value of the current node, adding a new peak shaving node; S70, using a pre-trained user-side traffic prediction model to predict the user-side traffic change trend in a future time window based on historical user-side bandwidth usage data to obtain predicted user traffic; S80. Adjust the operator-side bandwidth node threshold sorting and the user-side bandwidth traffic rate sorting in S20 in advance according to the predicted user traffic to prepare for traffic allocation in the next time window, and then repeat steps S30 to S70.

2. The method for optimizing operator traffic scheduling according to claim 1, characterized in that: The threshold of each bandwidth node on the operator side is specifically the maximum flow value that each bandwidth node on the operator side can handle within the current time window.

3. The method for optimizing operator traffic scheduling according to claim 2, characterized in that: The user-side bandwidth flow rate is specifically the average bandwidth usage of each user in the current time window.

4. The method for optimizing operator traffic scheduling according to claim 3, characterized in that: The peak shaving nodes are specifically the top 5% nodes with the highest bandwidth values ​​that are not included in the billing under the 95 billing mode.

5. The method for optimizing operator traffic scheduling according to claim 4, characterized in that: The remaining number of peak shaving nodes specifically refers to the number of nodes currently used for peak shaving by each operator-side bandwidth node; the selection of peak shaving nodes to process the user-side nodes specifically refers to allocating user-side traffic that exceeds the current operator-side maximum threshold to the available peak shaving nodes, so that the actual traffic of the operator-side nodes does not exceed its threshold.

6. The method for optimizing operator traffic scheduling according to claim 5, characterized in that: The dynamic calculation of the flow value before the start of each sub-window within each time window is specifically calculated based on the actual flow data of the sub-window that has passed and the predicted flow of the remaining sub-windows.

7. The method for optimizing operator traffic scheduling according to claim 6, characterized in that: The user-side traffic prediction model adopts a long short-term memory network LSTM structure, specifically including an input layer, an LSTM layer, a fully connected layer and an output layer; the input layer receives historical time series data, the LSTM layer is used to capture the long-term dependencies of the time series, the fully connected layer is used for feature mapping, and the output layer generates prediction results.

8. The method for optimizing operator traffic scheduling according to claim 7, characterized in that: The remaining sub-windows in the time window are specifically all sub-windows after the current sub-window.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed in a computer, they are used to execute the operator traffic scheduling optimization method according to any one of claims 1 to 8.

10. An operator traffic scheduling optimization system, characterized in that: A computer-readable storage medium comprising the computer-readable storage medium of claim 9.

Images