A traffic scheduling method and device based on machine learning and optimization decision

Abstract

The application provides a traffic scheduling method and device based on machine learning and optimization decision, and relates to the field of machine learning. The method comprises the following steps: obtaining historical traffic data of a plurality of target Internet data centers, and constructing a traffic feature set; constructing a traffic prediction machine learning model through the traffic feature set, and outputting traffic prediction results of each target Internet data center; constructing a traffic scheduling optimization model based on the traffic prediction results, and optimizing and solving the traffic distribution relationship between each target Internet data center to obtain a target traffic scheduling scheme; dynamically updating the target traffic scheduling scheme through a preset adaptive feedback mechanism, and performing intelligent traffic scheduling operation based on the updated target traffic scheduling scheme. The application solves the problem that when the traffic in the Internet data center fluctuates rapidly in a short time, the existing traffic scheduling system cannot timely adjust the traffic distribution, thereby causing the user access time delay to increase.

Description

Technical Field

[0001] This application relates to the field of machine learning, and in particular to a traffic scheduling method and apparatus based on machine learning and optimization decision-making. Background Technology

[0002] With the development of cloud computing, content delivery networks, and large-scale internet services, more and more internet service providers are deploying multiple geographically dispersed internet data centers to carry out cross-regional business operations, thereby improving business continuity, network availability, and user access quality. In actual operation, user access requests in different regions will continuously and dynamically change, resulting in obvious periodic fluctuations, regional fluctuations, and sudden changes in traffic load between different internet data centers.

[0003] Existing traffic scheduling systems typically allocate traffic based on fixed-weight strategies, round-robin strategies, or simple real-time load monitoring mechanisms. Since these systems primarily rely on the instantaneous network state at any given moment, they struggle to identify traffic patterns across different time periods when traffic in internet data centers fluctuates rapidly. This can lead to some internet data centers exceeding their minimum bandwidth during peak hours, while others fail to utilize their bandwidth effectively, resulting in significantly increased latency for users.

[0004] Therefore, there is an urgent need for a traffic scheduling method and device based on machine learning and optimization decision-making. Summary of the Invention

[0005] This application provides a traffic scheduling method and apparatus based on machine learning and optimization decision-making, which solves the problem that existing traffic scheduling systems cannot adjust traffic allocation in a timely manner when traffic in Internet data centers fluctuates rapidly in a short period of time, resulting in increased fluctuations in user access latency.

[0006] The first aspect of this application provides a traffic scheduling method based on machine learning and optimization decision-making. The method includes: acquiring historical traffic data of multiple target Internet data centers and constructing a traffic feature set based on the historical traffic data; the traffic feature set includes time features, historical lag features, rolling statistical features, and business status features; constructing a traffic prediction machine learning model through the traffic feature set, and outputting traffic prediction results for each of the target Internet data centers within a preset time window based on the traffic prediction machine learning model; constructing traffic scheduling constraints based on the traffic prediction results, and constructing a traffic scheduling optimization model using the traffic scheduling constraints as the model solution boundary; the traffic scheduling constraints include excess traffic constraints, Internet data center capacity constraints, and access latency constraints; optimizing the traffic allocation relationship between the target Internet data centers through the traffic scheduling optimization model to obtain a target traffic scheduling scheme corresponding to each target Internet data center; dynamically updating the target traffic scheduling scheme through a preset adaptive feedback mechanism, and performing intelligent traffic scheduling operations based on the updated target traffic scheduling scheme.

[0007] Optionally, a traffic feature set is constructed based on historical traffic data, specifically including: extracting time attribute information from historical traffic data and discretizing and encoding the time attribute information to obtain time features; the time features include hourly encoding, weekday encoding, and weekend identifier; extracting multiple outbound bandwidth values ​​corresponding to different historical moments based on the outbound bandwidth sequence of each target Internet data center, and performing time-series correlation processing on the outbound bandwidth values ​​to obtain historical lag features; calculating the mean and standard deviation of the outbound bandwidth sequence respectively, and using the mean and standard deviation as rolling statistical features; obtaining business operation identifier information corresponding to each target Internet data center based on historical traffic data, and performing category encoding processing on the business operation identifier information to obtain business status features; the business operation identifier information includes business type identifier, user access area identifier, and Internet data center node identifier; and fusing and encoding the time features, historical lag features, rolling statistical features, and business status features to obtain a traffic feature set.

[0008] Optionally, a traffic prediction machine learning model is constructed using a set of traffic features, and the traffic prediction results for each target Internet data center within a preset time window are output based on the traffic prediction machine learning model. Specifically, this includes: arranging the traffic feature set in a time series based on the historical traffic change trends corresponding to each target Internet data center to obtain traffic fluctuation differences at different time stages; constructing time-series training samples based on these traffic fluctuation differences; iteratively training the preset machine learning model using the time-series training samples, and verifying the prediction performance of the model parameters based on the operational constraints corresponding to each target Internet data center during the training process, thereby obtaining the traffic prediction machine learning model; the operational constraints include bandwidth capacity status, excess bandwidth cost status, and traffic load offset status; inputting the current traffic feature set corresponding to each target Internet data center into the traffic prediction machine learning model, and extracting time-series change correlations through the traffic prediction machine learning model; the time-series change correlations include periodic change correlations, traffic fluctuation correlations, and cross-time window change correlations; outputting the traffic change sequence corresponding to each target Internet data center within the preset time window based on the time-series change correlations; and performing time window mapping processing on the traffic change sequence to obtain the traffic prediction results for each target Internet data center.

[0009] Optionally, traffic scheduling constraints are constructed based on traffic prediction results, and a traffic scheduling optimization model is constructed using these constraints as the solution boundary. Specifically, this includes: constructing a target optimization problem based on traffic prediction results, and constructing a target optimization function based on the target optimization problem; the target optimization problem includes a target set of internet data centers, a set of user access areas, regional traffic scheduling variables, traffic prediction values, guaranteed bandwidth values, excess bandwidth cost values, capacity upper limits, regional traffic demand values, regional access latency values, and a target latency threshold; constructing traffic scheduling constraints based on the target optimization function; and constructing a traffic scheduling optimization model using the target optimization function as the optimization objective and the traffic scheduling constraints as the constraint boundary.

[0010] Optionally, the target traffic scheduling scheme is dynamically updated through a preset adaptive feedback mechanism, and intelligent traffic scheduling operations are performed based on the updated target traffic scheduling scheme. Specifically, this includes: mapping the target traffic scheduling scheme to traffic scheduling control parameters, and updating the traffic routing policy in the traffic scheduling platform according to the traffic scheduling control parameters; performing cross-data center traffic scheduling for user access requests based on the updated traffic routing policy, and determining in real time whether a data distribution drift event has occurred during the scheduling process; if a data distribution drift event is confirmed, retraining the traffic prediction machine learning model; updating the target traffic scheduling scheme through the retrained traffic prediction machine learning model; and performing intelligent traffic scheduling operations based on the updated target traffic scheduling scheme.

[0011] Optionally, the traffic routing strategy in the traffic scheduling platform is updated according to the traffic scheduling control parameters, specifically including: constructing a traffic scheduling platform; the traffic scheduling platform is used to perform cross-data center routing scheduling of user access requests according to the traffic allocation ratio corresponding to different Internet data centers; updating the traffic routing strategy in the traffic scheduling platform according to the traffic scheduling control parameters; the traffic routing strategy is used to characterize the traffic allocation relationship between user access requests and target Internet data centers.

[0012] Optionally, during the scheduling process, it is determined in real time whether a data distribution drift event has occurred. Specifically, this includes: acquiring real-time running data during the scheduling process; detecting the distribution error between the traffic prediction result and the real-time running data, and determining whether the distribution error is greater than a preset error threshold; if it is confirmed that the distribution error is greater than the preset error threshold, then it is determined that a distribution drift event has occurred.

[0013] A second aspect of this application provides a traffic scheduling apparatus based on machine learning and optimization decision-making. The apparatus includes an acquisition module and a processing module, wherein... The acquisition module is used to acquire historical traffic data from multiple target Internet data centers and construct a traffic feature set based on the historical traffic data. The traffic feature set includes time features, historical lag features, rolling statistical features, and business status features.

[0014] The processing module is used to construct a traffic prediction machine learning model based on a set of traffic features, and output the traffic prediction results for each target Internet data center within a preset time window based on the traffic prediction machine learning model; construct traffic scheduling constraints based on the traffic prediction results, and construct a traffic scheduling optimization model using the traffic scheduling constraints as the solution boundary; the traffic scheduling constraints include excess traffic constraints, Internet data center capacity constraints, and access latency constraints; optimize the traffic allocation relationship between each target Internet data center through the traffic scheduling optimization model to obtain the target traffic scheduling scheme corresponding to each target Internet data center; dynamically update the target traffic scheduling scheme through a preset adaptive feedback mechanism, and perform intelligent traffic scheduling operations based on the updated target traffic scheduling scheme.

[0015] A third aspect of this application provides an electronic device including a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform any of the methods described above.

[0016] A fourth aspect of this application provides a non-transitory computer-readable storage medium storing a computer program, the computer program being executed by a processor using any of the methods described above.

[0017] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. Acquire historical traffic data from multiple target Internet data centers and construct a traffic feature set; build a traffic prediction machine learning model using the traffic feature set and output the traffic prediction results for each target Internet data center; construct a traffic scheduling optimization model based on the traffic prediction results and optimize the traffic allocation relationship between each target Internet data center to obtain the target traffic scheduling scheme; dynamically update the target traffic scheduling scheme through a preset adaptive feedback mechanism and perform intelligent traffic scheduling operations based on the updated target traffic scheduling scheme, thereby improving the adaptability of the Internet data center traffic scheduling process to traffic fluctuations and reducing user access latency fluctuations.

[0018] 2. Construct a target optimization problem based on the traffic prediction results, and then construct a target optimization function based on the target optimization problem. The target optimization problem includes the target set of Internet data centers, the set of user access areas, regional traffic scheduling variables, traffic prediction values, guaranteed bandwidth values, excess bandwidth cost values, capacity upper limit values, regional traffic demand values, regional access latency values, and target latency thresholds. Construct traffic scheduling constraints based on the target optimization function. Use the target optimization function as the optimization objective and the traffic scheduling constraints as the constraint boundaries to construct a traffic scheduling optimization model, thereby reducing the cost of cross-data center traffic scheduling while meeting the Internet data center capacity limits and user access latency requirements.

[0019] 3. Map the target traffic scheduling scheme to traffic scheduling control parameters, and update the traffic routing policy in the traffic scheduling platform according to the traffic scheduling control parameters; perform cross-data center traffic scheduling for user access requests based on the updated traffic routing policy, and determine in real time whether a data distribution drift event has occurred during the scheduling process; if a data distribution drift event is confirmed, retrain the traffic prediction machine learning model; update the target traffic scheduling scheme through the retrained traffic prediction machine learning model; perform intelligent traffic scheduling operations based on the updated target traffic scheduling scheme, thereby improving the adaptability and scheduling stability of the cross-data center traffic scheduling process to changes in business traffic. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a traffic scheduling method based on machine learning and optimization decision-making provided in an embodiment of this application. Figure 2 This is a schematic diagram of a traffic scheduling device based on machine learning and optimization decision-making provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0021] Explanation of reference numerals in the attached figures: 21. Acquisition module; 22. Processing module; 301. Processor; 302. Communication bus; 303. User interface; 304. Network interface; 305. Memory. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0023] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0024] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0025] With the development of cloud computing, content delivery networks, and large-scale internet services, more and more internet service providers are deploying multiple geographically dispersed internet data centers (IDCs) to support cross-regional business operations, thereby improving business continuity, network availability, and user access quality. In actual operation, user access requests from different regions continuously and dynamically change, resulting in significant periodic fluctuations, regional fluctuations, and sudden changes in traffic load across different IDCs. For example, during holidays, large-scale business events, or periods of regional network fluctuation, some IDCs may experience rapid traffic growth, while others remain under low load. At the same time, each IDC typically has different guaranteed bandwidth, excess bandwidth fees, and physical capacity limitations; therefore, internet service providers need to continuously and dynamically schedule traffic across multiple IDCs.

[0026] Existing traffic scheduling methods typically allocate traffic based on fixed-weight strategies, round-robin strategies, or simple real-time load monitoring mechanisms. These methods primarily rely on the instantaneous network state at the current moment, lacking the ability to predict future traffic trends. When traffic in internet data centers fluctuates rapidly within a short period, existing technologies struggle to identify traffic patterns across different time periods. This can easily lead to some internet data centers exceeding their guaranteed bandwidth during peak hours, incurring high excess bandwidth costs, while simultaneously underutilizing bandwidth resources in other data centers. Furthermore, because existing technologies underutilize the temporal correlation characteristics, periodic variation characteristics, and service status characteristics of historical traffic data, traffic prediction results fail to accurately reflect the dynamic traffic relationships between different internet data centers, further impacting the accuracy and stability of subsequent traffic scheduling optimization results.

[0027] To address the aforementioned issues, this application proposes an intelligent traffic scheduling method based on machine learning prediction and optimization decision-making. By extracting and fusing features from historical traffic data corresponding to multiple target Internet data centers, a traffic feature set that can characterize the traffic change patterns at different time stages is constructed. Based on the traffic feature set, a traffic prediction machine learning model is built, thereby providing a reliable data foundation for subsequent cross-Internet data center traffic scheduling optimization.

[0028] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0029] Please refer to Figure 1 The diagram illustrates a flow chart of a traffic scheduling method based on machine learning and optimization decision-making provided in an embodiment of this application. The flow chart mainly includes the following steps: S101 to S105.

[0030] Step S101: Obtain historical traffic data from multiple target Internet data centers and construct a traffic feature set based on the historical traffic data.

[0031] Specifically, historical traffic data corresponding to multiple target Internet data centers is acquired, and a traffic feature set is constructed based on this historical traffic data. Specifically, the historical traffic data includes egress bandwidth data, connection status data, regional access data, and business operation status data corresponding to each target Internet data center. Subsequently, time attribute information, historical correlation traffic information, and window statistics information are extracted from the historical traffic data, and combined with business operation identification information to construct corresponding time features, historical lag features, rolling statistical features, and business status features. Finally, fusion encoding processing is performed on the above different types of features to obtain the traffic feature set, enabling the subsequently constructed traffic prediction machine learning model to simultaneously characterize the time variation patterns, traffic fluctuation patterns, and business status change patterns between different Internet data centers. The traffic feature set includes, but is not limited to, time features, historical lag features, rolling statistical features, and business status features.

[0032] In one possible implementation, step S101 further includes: extracting time attribute information based on historical traffic data, and performing discretization encoding on the time attribute information to obtain time features; the time features include hourly encoding, weekday encoding, and weekend identifier; extracting multiple outbound bandwidth values ​​corresponding to different historical moments based on the outbound bandwidth sequence of each target Internet data center, and performing time-series correlation processing on the outbound bandwidth values ​​to obtain historical lag features; calculating the mean and standard deviation of the outbound bandwidth sequence respectively, and using the mean and standard deviation as rolling statistical features; obtaining business operation identifier information corresponding to each target Internet data center based on historical traffic data, and performing category encoding on the business operation identifier information to obtain business status features; the business operation identifier information includes business type identifier, user access area identifier, and Internet data center node identifier; and fusing and encoding the time features, historical lag features, rolling statistical features, and business status features to obtain a traffic feature set.

[0033] Specifically, after acquiring historical traffic data from multiple target internet data centers, the timestamp information in the historical traffic data is first parsed to extract the corresponding time attribute information. This time attribute information characterizes the time phase characteristics corresponding to different historical traffic data, enabling the subsequent traffic prediction machine learning model to identify traffic change patterns in different time phases. In this embodiment, the time attribute information includes hourly codes, weekday codes, and weekend identifiers. Specifically, hourly codes characterize the differences in business access at different times of the day; weekday codes characterize the periodic traffic change patterns corresponding to different workdays; and weekend identifiers characterize the differences in business access between workdays and weekends.

[0034] Subsequently, the time attribute information is discretized and encoded to generate corresponding time features. Specifically, one-hot encoding, ordinal encoding, or discrete interval mapping can be used to encode the hour, day of the week, and weekend identifiers. For example, the twenty-four hours of a day can be mapped to their corresponding discrete time numbers, and different day of the week can be mapped to their corresponding cycle numbers. Through this processing, time attribute information from different time periods can participate in the subsequent feature fusion process in a unified data format, thereby improving the ability of the traffic prediction machine learning model to identify periodic time patterns.

[0035] After constructing the time features, the egress bandwidth values ​​corresponding to multiple historical moments are extracted based on the egress bandwidth sequence for each target Internet data center, and historical lag features are constructed based on these egress bandwidth values. Specifically, the egress bandwidth sequence is used to characterize the historical traffic changes of the target Internet data center over a continuous time period. Since Internet service traffic typically exhibits significant temporal continuity and historical correlation, the traffic status at different historical moments will influence future traffic changes. Therefore, this embodiment extracts egress bandwidth values ​​corresponding to multiple historical moments and performs time-series correlation processing on the egress bandwidth values ​​at different historical moments to establish traffic correlation relationships between different time periods.

[0036] For example, the outbound bandwidth values ​​can be extracted from the previous time point, the corresponding time point of the previous day, and the corresponding time point of the previous week, and corresponding historical lag features can be constructed based on these outbound bandwidth values. Through the above processing, the traffic prediction machine learning model can learn the historical dependencies between different time periods, thereby improving the adaptability of future traffic prediction results to sudden traffic fluctuations and periodic business changes.

[0037] Furthermore, window statistical processing is performed on the outbound bandwidth sequence to construct corresponding rolling statistical features. Specifically, the mean and standard deviation of the corresponding outbound bandwidth sequence are calculated within a preset continuous time interval. The mean characterizes the average traffic level of the target internet data center within the corresponding time interval; the standard deviation characterizes the degree of traffic fluctuation of the target internet data center within the corresponding time interval. By simultaneously introducing the mean and standard deviation, the subsequent traffic prediction machine learning model can not only identify traffic change trends but also identify changes in traffic stability at different time stages.

[0038] Subsequently, based on historical traffic data, the service operation identification information corresponding to each target Internet Data Center is obtained, and the service operation identification information is categorized and coded to obtain service status characteristics, thereby generating corresponding service status features. The service operation identification information includes service type identifiers, user access area identifiers, and Internet Data Center node identifiers. Specifically, the service type identifier is used to characterize the differences in traffic access patterns corresponding to different service types; the user access area identifier is used to characterize the differences in access distribution among users in different regions; and the Internet Data Center node identifier is used to distinguish the differences in network resource attributes and traffic carrying capacity corresponding to different target Internet Data Centers.

[0039] Specifically, category mapping encoding, one-hot encoding, or vector embedding encoding can be used to classify and encode business operation identification information to obtain business status features. This allows different business status information to be converted into structured feature data that can participate in model training. By introducing these business status features, the traffic prediction machine learning model can further learn the traffic correlation patterns between different business types, different user regions, and different Internet data centers.

[0040] Finally, the time features, historical lag features, rolling statistical features, and business status features are fused and encoded to obtain a traffic feature set. Specifically, feature concatenation, vector fusion, or multi-dimensional mapping can be used to uniformly fuse the above-mentioned different types of features, so that the traffic feature set can simultaneously represent the time variation patterns, historical traffic dependencies, traffic fluctuation patterns, and business status change patterns of the target Internet data center, thereby providing a complete input feature foundation for the subsequent construction of traffic prediction machine learning models.

[0041] Step S102: Construct a traffic prediction machine learning model through a set of traffic features, and output the traffic prediction results of each target Internet data center within a preset time window based on the traffic prediction machine learning model.

[0042] Specifically, a corresponding traffic prediction machine learning model is constructed using a set of traffic features. This model is then used to predict the traffic change trends of multiple target internet data centers within a preset future time window, yielding the corresponding traffic prediction results. The traffic prediction machine learning model is used to learn the traffic change patterns at different time stages and the dynamic traffic relationships between different target internet data centers, thus providing a predictive basis for subsequent traffic scheduling optimization.

[0043] In one possible implementation, step S102 further includes: arranging the traffic feature set in a time series based on the historical traffic change trends corresponding to each target Internet data center to obtain traffic fluctuation differences in different time periods; constructing time-series training samples based on the traffic fluctuation differences; iteratively training a preset machine learning model based on the time-series training samples, and verifying the prediction performance of the model parameters of the preset machine learning model according to the operational constraint states corresponding to each target Internet data center during the training process, thereby obtaining a traffic prediction machine learning model; the operational constraint states include bandwidth capacity state, excess bandwidth cost state, and traffic load offset state; inputting the current traffic feature set corresponding to each target Internet data center into the traffic prediction machine learning model, and extracting time-series change correlations through the traffic prediction machine learning model; the time-series change correlations include periodic change correlations, traffic fluctuation correlations, and cross-time window change correlations; outputting the traffic change sequence corresponding to each target Internet data center within a preset time window based on the time-series change correlations; and performing time window mapping processing on the traffic change sequence to obtain the traffic prediction results corresponding to each target Internet data center.

[0044] Specifically, firstly, based on the historical traffic change trends corresponding to each target Internet data center, the traffic feature set is processed by time series arrangement to form traffic time series data that conforms to the chronological order. Since the business traffic in Internet data centers has obvious periodic change characteristics, sudden fluctuation characteristics, and cross-time window correlation characteristics, it is necessary to maintain the data continuity between different time stages through time series arrangement, thereby improving the ability of subsequent traffic prediction machine learning models to learn the traffic change patterns.

[0045] In this embodiment, the set of traffic features arranged in chronological order is represented as follows: in, This represents the set of traffic features arranged in time series. Indicates the first Traffic feature vectors corresponding to each time period; This represents the total number of events over a historical time period. The traffic feature vector includes time features, historical lag features, rolling statistical features, and business status features.

[0046] Subsequently, the differences in traffic fluctuations across different time periods are calculated based on the time-series arranged traffic feature set. Specifically, the traffic changes between consecutive time periods are calculated to characterize the traffic fluctuations of the target internet data center across different time periods. The corresponding traffic fluctuation differences can be expressed as:

[0047] in, Indicates the first Differences in traffic fluctuations corresponding to different time periods; Indicates the first Outbound bandwidth values ​​during each time period; This represents the outbound bandwidth value in the previous time period. Through the above processing, subsequent models can identify traffic spikes and periodic trends in different time periods.

[0048] Based on the differences in traffic flow fluctuations, corresponding time-series training samples are constructed. Specifically, traffic feature vectors from multiple consecutive time periods and their corresponding traffic flow fluctuation differences are combined to form a training window, and corresponding input samples and target output samples are constructed based on the training window. The corresponding time-series training samples can be represented as follows:

[0049] in, Indicates the first Time-series training samples corresponding to each time stage; Indicates the length of the time window; This represents the flow characteristic sequence within the input time window; This represents the target predicted output value, used to characterize the result of traffic changes in future time periods.

[0050] After constructing the time-series training samples, the preset machine learning model is iteratively trained based on the time-series training samples. In this embodiment, the preset machine learning model can be a Long Short-Term Memory (LSTM) network model, a Gated Recurrent Neural Network (GRU) model, or a time-series prediction model based on an attention mechanism. Since the LSTM model can effectively retain historical state information over a long period of time, this embodiment preferably uses the LSTM model to construct the traffic prediction machine learning model.

[0051] The LSTM model consists of an input layer, a temporal feature extraction layer, a hidden state update layer, and a prediction output layer. The input layer receives feature vectors from the traffic feature set; the temporal feature extraction layer extracts temporal relationships between different time periods; the hidden state update layer preserves historical traffic changes; and the prediction output layer outputs traffic predictions for future time periods.

[0052] During model training, the input information at the current time stage is first written into the hidden state based on the input gate. The calculation expression for the input gate is: in, Indicates the first The input gate output results in each time period; This represents the Sigmoid activation function; This represents the weight matrix corresponding to the input features; This represents the weight matrix corresponding to the historical hidden states; This indicates the hidden state in the previous time period; This represents the bias parameter. This expression controls the degree to which traffic characteristics in the current time period affect the hidden state.

[0053] Subsequently, the degree to which historical flow state information is retained is controlled by the forget gate. The calculation expression for the forget gate is: in, Indicates the first The output results of the forget gate in each time period; , as well as These represent the weight matrix and bias parameters corresponding to the forget gate, respectively. This expression is used to control the proportion of historical traffic state information retained in the current time period.

[0054] Furthermore, the cell state in the current time stage is updated based on the input gate and the forget gate: in, Indicates the first The unit state in each time phase; Indicates the state of the unit in the previous time period; Indicates candidate status information; This represents element-wise vector multiplication. This expression is used to fuse historical state information with input feature information from the current time period.

[0055] Then, the hidden state output result is generated through the output gate: in, Indicates the first Output results of hidden states in each time phase; Indicates the output gate control result; This represents the hyperbolic tangent activation function. This expression is used to generate temporal feature representations for the current time period.

[0056] After completing the hidden state update, the traffic forecast values ​​for future time periods are generated through the prediction output layer: in, Indicates the first Predicted flow values ​​for each time period; This represents the output layer weight matrix; This represents the output layer bias parameters.

[0057] During model training, to improve the adaptability of the traffic prediction machine learning model to actual scheduling scenarios of Internet data centers, this embodiment further verifies the prediction performance of the model by combining the operational constraint states corresponding to each target Internet data center. The operational constraint states include bandwidth capacity state, excess bandwidth cost state, and traffic load offset state. Specifically, the model prediction results are optimized by constructing a constraint-weighted loss function:

[0058] in, This represents the model training loss function; Indicates the number of training samples; This represents the predicted flow rate. This represents the actual traffic volume. This indicates a bandwidth capacity constraint. This represents the excess bandwidth cost constraint. This indicates the flow load offset constraint term; This represents the constraint weight parameter, used to adjust the degree of influence of different constraint terms on the model training results. In this embodiment, All are non-negative real numbers, and satisfy: Among these, the bandwidth capacity constraint term is used to limit the prediction results from exceeding the capacity limit of the Internet data center; the excess bandwidth cost constraint term is used to reduce the risk of excess bandwidth costs corresponding to the prediction results; and the traffic load offset constraint term is used to reduce the degree of load imbalance between different Internet data centers. Through the above constraint optimization methods, the trained traffic prediction machine learning model can not only improve the accuracy of traffic prediction, but also be more adaptable to cross-regional traffic scheduling scenarios in Internet data centers.

[0059] After model training is complete, the current traffic feature set corresponding to each target internet data center is input into the trained traffic prediction machine learning model. The model then extracts the corresponding temporal variation correlations. These temporal variation correlations include periodic variation correlations, traffic fluctuation correlations, and cross-time window variation correlations. Subsequently, based on these temporal variation correlations, the traffic change sequence corresponding to each target internet data center within a preset time window is output. Time window mapping processing is then performed on the traffic change sequences to obtain the traffic prediction results for each target internet data center, thus providing a predictive basis for subsequent traffic scheduling optimization.

[0060] Step S103: Construct traffic scheduling constraints based on traffic prediction results, and construct a traffic scheduling optimization model using the traffic scheduling constraints as the solution boundary.

[0061] Specifically, traffic scheduling constraints are constructed based on traffic prediction results, and a traffic scheduling optimization model is built using these constraints as the solution boundary. The traffic scheduling constraints limit the traffic allocation range between different target Internet data centers, ensuring that subsequent traffic scheduling results simultaneously meet bandwidth capacity limitations, excess bandwidth cost control, and user access latency requirements. Subsequently, the traffic scheduling optimization model is used to optimize the traffic allocation relationship between different target Internet data centers, yielding the corresponding target traffic scheduling scheme, thereby achieving intelligent cross-Internet data center traffic scheduling oriented towards future traffic change trends.

[0062] In one possible implementation, step S103 further includes: constructing a target optimization problem based on the traffic prediction results, and constructing a target optimization function based on the target optimization problem; the target optimization problem includes a target set of Internet data centers, a set of user access areas, regional traffic scheduling variables, traffic prediction values, guaranteed bandwidth values, excess bandwidth cost values, capacity upper limits, regional traffic demand values, regional access latency values, and target latency thresholds; constructing traffic scheduling constraints based on the target optimization function; and constructing a traffic scheduling optimization model using the target optimization function as the optimization objective and the traffic scheduling constraints as the constraint boundaries.

[0063] Specifically, a corresponding target optimization problem is constructed based on the traffic prediction results, and a traffic scheduling optimization model is built on the basis of the target optimization problem to achieve dynamic traffic optimization scheduling among multiple target Internet data centers. Since different target Internet data centers typically have different bandwidth capacities, excess bandwidth costs, and network access latency, the traffic scheduling process not only needs to ensure that the Internet data centers have sufficient traffic carrying capacity, but also needs to reduce excess bandwidth costs and guarantee user access quality. Based on this, this embodiment constructs a traffic scheduling optimization model containing various operational constraints to jointly optimize and solve the traffic allocation relationship among multiple target Internet data centers.

[0064] First, a target optimization problem is constructed based on the traffic prediction results. In this embodiment, the target optimization problem includes the target set of Internet data centers, the set of user access areas, regional traffic scheduling variables, traffic prediction values, guaranteed bandwidth values, excess bandwidth cost values, capacity cap values, regional traffic demand values, regional access latency values, and target latency thresholds.

[0065] The target set of Internet data centers is represented as follows: ,in, Indicates the target set of Internet data centers; Indicates the first One target Internet data center.

[0066] The set of user access areas is represented as follows: ,in, Represents the set of areas accessed by the user; Indicates the first Individual user access areas.

[0067] Furthermore, define the regional traffic scheduling variables: ,in, Indicates from the first The user access area was redirected to the first... The target traffic value for each target Internet data center. Regional traffic scheduling variables are used to characterize the traffic distribution relationship between different user access regions and the target Internet data center.

[0068] Then, the traffic prediction value is defined: ,in, This indicates the first step obtained in step S102. The predicted traffic value corresponding to each target Internet data center.

[0069] Furthermore, define the minimum bandwidth value: ,in, Indicates the first The guaranteed bandwidth value corresponding to each target Internet data center.

[0070] Furthermore, define the excess bandwidth cost value: ,in, Indicates the first The unit excess bandwidth cost value corresponding to each target Internet data center.

[0071] Furthermore, define the upper limit of capacity: ,in, Indicates the first The upper limit of traffic capacity corresponding to a target Internet data center.

[0072] Furthermore, define the regional traffic demand value: ,in, Indicates the first The total traffic demand value corresponding to each user accessing a region.

[0073] Furthermore, define the area access latency value: ,in, Indicates the first The user accesses the area to access the first The network access latency value corresponding to each target Internet data center.

[0074] Furthermore, define the target latency threshold: ,in, Indicates the first The maximum allowed access latency threshold for each user accessing a region.

[0075] After defining the objective optimization problem, an objective optimization function is constructed based on the aforementioned objective parameters. In this embodiment, the objective optimization function is used to minimize the excess bandwidth cost corresponding to multiple target Internet data centers. The corresponding objective optimization function can be expressed as: ,in, This represents the overall traffic scheduling cost target value; Indicates the first The unit excess bandwidth cost value corresponding to each target Internet data center; Indicates the first The excess bandwidth traffic value corresponding to a target Internet Data Center. This expression is used to reduce the overall excess bandwidth cost of an Internet Data Center while meeting the operational constraints of the Internet Data Center.

[0076] Since there is a correlation between excess bandwidth traffic and the target scheduled traffic corresponding to the target Internet data center, excess traffic constraints are further constructed as follows: ,in, Indicates allocation to the first Total regional scheduling traffic for each target Internet data center; This represents the predicted flow rate. This represents the minimum guaranteed bandwidth value. This expression is used to limit the excess traffic range in the target Internet data center after the actual scheduled traffic exceeds the minimum guaranteed bandwidth.

[0077] Furthermore, we construct the capacity constraints for Internet data centers: ,in, Indicates the first The capacity limit value corresponding to each target Internet data center. This expression is used to ensure that the total scheduled traffic in the target Internet data center does not exceed the corresponding traffic capacity limit.

[0078] Furthermore, construct regional traffic demand constraints: ,in, Indicates the first The total traffic demand value corresponding to each user access zone. This expression is used to ensure that the traffic demand of each user access zone can be fully allocated.

[0079] Furthermore, construct access latency constraints: ,in, This indicates the actual access latency between the user's access area and the target Internet data center; This represents the maximum allowed access latency threshold for the corresponding user's access area. This expression is used to ensure that user access requests meet access quality requirements during traffic scheduling.

[0080] Furthermore, construct the non-negativity constraint condition for the flow: ,in, This represents the regional traffic scheduling variable. This expression is used to ensure that the scheduled traffic for each region is non-negative.

[0081] After constructing the objective optimization function and traffic scheduling constraints, a traffic scheduling optimization model is built using the objective optimization function as the optimization objective and the traffic scheduling constraints as the constraint boundaries. In this embodiment, the traffic scheduling optimization model can be constructed using a linear programming optimization model, a constrained optimization model, or a mixed-integer programming optimization model. Since the linear programming optimization model has high solution efficiency, this embodiment preferably uses a linear programming optimization model to optimize the traffic allocation relationship between multiple target Internet data centers.

[0082] Subsequently, a linear programming algorithm is used to solve the traffic scheduling optimization model, outputting the target Internet data center traffic allocation results corresponding to different user access areas. Specifically, the simplex algorithm, interior-point method, or constraint search-based optimization algorithm can be used to solve the traffic scheduling optimization model, thereby obtaining a target traffic scheduling scheme that satisfies bandwidth capacity constraints, access latency constraints, and excess bandwidth cost constraints, providing an optimization decision-making basis for subsequent intelligent traffic scheduling in Internet data centers.

[0083] Step S104: The traffic allocation relationship between each target Internet data center is optimized and solved using the traffic scheduling optimization model to obtain the target traffic scheduling scheme corresponding to each target Internet data center.

[0084] Specifically, the objective optimization function and traffic scheduling constraints are input into the traffic scheduling optimization model. The objective optimization function characterizes the excess bandwidth cost optimization objective in the overall Internet data center traffic scheduling process; the traffic scheduling constraints limit the traffic allocation range between different target Internet data centers to ensure that the target traffic scheduling result meets the Internet data center capacity limits and user access latency requirements.

[0085] Subsequently, the traffic scheduling optimization model is initialized. Specifically, an initial traffic allocation matrix is ​​generated based on the traffic prediction results, bandwidth capacity status, and regional traffic demand status for each target Internet data center. The initial traffic allocation matrix can be represented as:

[0086] in, Represents the initial flow allocation matrix; Indicates the initialization phase from the first The user access area is allocated to the first The initial target traffic value for each target internet data center; Indicates the number of target Internet data centers; This indicates the number of regions accessed by the user.

[0087] Furthermore, the comprehensive operating cost corresponding to each target Internet data center is calculated based on the initial traffic allocation matrix. The corresponding comprehensive operating cost can be expressed as:

[0088] in, Indicates the first The overall operational cost corresponding to each target Internet data center; This represents the cost of excess bandwidth. Indicates the value of region access latency; Indicates the cost of traffic load offset; This represents the cost weighting parameter, used to adjust the degree of influence of different operating cost factors on the overall operating cost.

[0089] in, The range of values ​​is This is used to improve the system's sensitivity to excess bandwidth costs; The range of values ​​is This is used to ensure that user access latency meets quality of service requirements; The range of values ​​is This is used to adjust the load balancing level between different target Internet data centers, and satisfies the following: .

[0090] By using the above comprehensive operational cost calculation method, the traffic scheduling optimization model can simultaneously consider the excess bandwidth cost of Internet data centers, user access quality, and traffic load balancing status.

[0091] Subsequently, the traffic allocation relationship between different target Internet data centers is iteratively optimized using a traffic scheduling optimization model. In this embodiment, linear programming, simplex problem solving, or constraint search-based optimization algorithms can be used to solve the traffic scheduling optimization model. Specifically, in each iteration, the change in the objective optimization function corresponding to the traffic scheduling variables in different regions is calculated, and the corresponding regional traffic scheduling variables are dynamically adjusted based on the change in the objective optimization function to gradually reduce the overall traffic scheduling cost.

[0092] The corresponding regional traffic scheduling variable update expression can be represented as: in, Indicates the first The corresponding regional traffic scheduling variables during each iteration; This represents the updated regional traffic scheduling variable; Indicates the iteration step size parameter; This represents the gradient change of the objective optimization function with respect to the regional traffic scheduling variable.

[0093] in, The range of values ​​is .when When the value is small, the stability of the model solution can be improved; when Larger values ​​can improve the model's solution speed.

[0094] Furthermore, after each iteration, it is determined whether the current regional traffic scheduling variable meets the corresponding traffic scheduling constraints. When it is detected that the current regional traffic scheduling variable exceeds the Internet data center capacity limit or exceeds the target access latency threshold, constraint correction processing is performed on the corresponding regional traffic scheduling variable to ensure that the optimization solution always remains within the traffic scheduling constraint boundary.

[0095] After satisfying the preset convergence conditions, the final optimized flow distribution matrix is ​​output: in, Represents the target traffic allocation matrix; This represents the final optimized regional traffic scheduling result. The target traffic allocation matrix is ​​used to characterize the final traffic allocation relationship between different user access areas and the target Internet data center.

[0096] Finally, based on the target traffic allocation matrix, target traffic scheduling schemes are generated for each target Internet data center, thereby providing a scheduling and control basis for intelligent traffic scheduling across Internet data centers in the subsequent traffic scheduling platform.

[0097] Step S105: Dynamically update the target traffic scheduling scheme through a preset adaptive feedback mechanism, and perform intelligent traffic scheduling operation based on the updated target traffic scheduling scheme.

[0098] Specifically, a pre-set adaptive feedback mechanism is used to dynamically adjust the target traffic scheduling scheme, and intelligent traffic scheduling across Internet data centers is executed based on the adjusted target traffic scheduling scheme. The adaptive feedback mechanism is used to detect errors in the current traffic prediction results based on real-time operational data, and dynamically update the traffic prediction machine learning model and the target traffic scheduling scheme based on the error detection results, thereby improving the adaptability of the Internet data center traffic scheduling process to changes in business traffic.

[0099] In one possible implementation, step S105 further includes: mapping the target traffic scheduling scheme to traffic scheduling control parameters, and updating the traffic routing policy in the traffic scheduling platform according to the traffic scheduling control parameters; performing cross-data center traffic scheduling for user access requests based on the updated traffic routing policy, and determining in real time whether a data distribution drift event has occurred during the scheduling process, specifically including: acquiring real-time running data during the scheduling process; detecting the distribution error between the traffic prediction result and the real-time running data, and determining whether the distribution error is greater than a preset error threshold; if it is confirmed that the distribution error is greater than the preset error threshold, determining that a distribution drift event has occurred; if it is confirmed that a data distribution drift event has occurred, retraining the traffic prediction machine learning model; updating the target traffic scheduling scheme through the retrained traffic prediction machine learning model; and performing intelligent traffic scheduling operations based on the updated target traffic scheduling scheme.

[0100] Specifically, the target traffic scheduling scheme is mapped to the corresponding traffic scheduling control parameters, and the traffic routing policy in the traffic scheduling platform is updated based on the traffic scheduling control parameters. The traffic scheduling control parameters characterize the target traffic allocation ratio among different target Internet data centers; the traffic routing policy controls the traffic forwarding relationship of user access requests among multiple target Internet data centers.

[0101] In this embodiment, the traffic scheduling weights corresponding to each target Internet data center are first calculated based on the target traffic allocation matrix in the target traffic scheduling scheme. The corresponding traffic scheduling weights can be expressed as:

[0102] in, Indicates the first Traffic scheduling weights corresponding to each target Internet data center; Indicates the first The target scheduling traffic value corresponding to each target Internet data center; This represents the number of target Internet data centers. This expression is used to convert target traffic allocation results into uniformly normalized traffic scheduling control parameters.

[0103] Subsequently, the traffic routing policy in the traffic scheduling platform is updated based on the traffic scheduling weights. In this embodiment, the traffic scheduling platform can be a global traffic scheduling platform, an intelligent DNS scheduling platform, or a global load balancing platform. Specifically, the traffic scheduling platform dynamically adjusts the proportion of user access traffic corresponding to different target Internet data centers according to the updated traffic scheduling weights, so as to realize the dynamic traffic distribution of user access requests among multiple target Internet data centers.

[0104] After updating the traffic routing policy, the system performs cross-datacenter traffic scheduling for user access requests based on the updated policy, and acquires corresponding real-time operational data during the scheduling process. This real-time operational data includes real-time egress bandwidth data, user access latency data, internet data center load data, and regional access traffic data. By continuously collecting this real-time operational data, the system can monitor the actual traffic operation status in the current internet data center in real time.

[0105] Furthermore, the distribution error between the traffic prediction result obtained in step S102 and the real-time operational data is detected to determine whether the current traffic prediction machine learning model is still applicable to the current business traffic environment. In this embodiment, the traffic error value between the traffic prediction result and the real-time operational data is first calculated:

[0106] in, Indicates the first Flow error values ​​in each time period; This represents the predicted traffic value output by the traffic prediction machine learning model. This represents the actual flow rate in real-time operational data. This expression characterizes the degree of deviation between the current prediction and the actual operational status.

[0107] Subsequently, the flow error values ​​within the continuous time window are statistically processed to construct the corresponding distribution error index: in, Indicates the distribution error index; Indicates the number of samples within the error statistics time window; This represents the traffic error value within the corresponding time period. This expression is used to characterize the overall prediction bias of the current traffic prediction machine learning model in real-world business scenarios.

[0108] Furthermore, the distribution error index is compared with a preset error threshold to determine whether the distribution error exceeds the preset error threshold. The preset error threshold characterizes the maximum allowable prediction error range for the current traffic prediction machine learning model. If the distribution error is confirmed to exceed the preset error threshold, it is determined that a data distribution drift event has occurred in the current internet service traffic environment.

[0109] In this embodiment, the data distribution drift event is used to characterize a significant deviation between the current internet service traffic distribution and the traffic distribution during the historical training phase. For example, when business activities suddenly increase, the regional network environment changes, or user access behavior changes, the original prediction patterns of the traffic prediction machine learning model may no longer be applicable to the current business scenario, resulting in a significant increase in traffic prediction error.

[0110] After confirming the occurrence of a data distribution drift event, the traffic prediction machine learning model is retrained. Specifically, the corresponding time-series training samples are reconstructed based on the current real-time running data, and the traffic prediction machine learning model is retrained using the updated time-series training samples so that the model can relearn the traffic change patterns in the current business traffic environment.

[0111] Furthermore, the traffic prediction machine learning model is retrained to generate traffic prediction results for future time windows, and the corresponding traffic scheduling optimization model is reconstructed based on the updated traffic prediction results, thereby generating an updated target traffic scheduling scheme.

[0112] Finally, based on the updated target traffic scheduling scheme, the corresponding traffic scheduling control parameters are regenerated, and the intelligent traffic scheduling operation across Internet data centers continues to be executed through the traffic scheduling platform. This forms a dynamic closed-loop scheduling mechanism of "traffic prediction - traffic scheduling - real-time monitoring - drift detection - model retraining - scheduling update", which improves the adaptability and stability of the intelligent traffic scheduling process among multiple target Internet data centers to changes in business traffic.

[0113] In one possible implementation, step S105 further includes: constructing a traffic scheduling platform; the traffic scheduling platform is used to perform cross-data center routing scheduling of user access requests based on the traffic allocation ratio corresponding to different Internet data centers; updating the traffic routing policy in the traffic scheduling platform according to the traffic scheduling control parameters; the traffic routing policy is used to characterize the traffic allocation relationship between user access requests and target Internet data centers.

[0114] Specifically, the traffic scheduling platform is constructed as follows: to achieve intelligent cross-data center traffic scheduling among multiple target Internet data centers. Specifically, the traffic scheduling platform performs dynamic routing allocation for user access requests based on the traffic allocation ratios corresponding to different target Internet data centers, thereby achieving balanced traffic scheduling and collaborative resource scheduling among different target Internet data centers.

[0115] In this embodiment, the traffic scheduling platform can be constructed using a global server load balancing platform, an intelligent DNS scheduling platform, or a traffic scheduling platform based on an application layer gateway. The traffic scheduling platform includes a traffic access layer, a routing decision layer, and a policy enforcement layer.

[0116] The traffic access layer receives user access requests from different user access areas and parses the area source information, access service type information, and network status information corresponding to the user access request. The routing decision layer generates corresponding traffic scheduling control parameters based on the target traffic scheduling scheme and dynamically generates corresponding traffic routing policies based on the traffic scheduling control parameters. The policy execution layer dynamically forwards user access requests to the corresponding target Internet data center based on the updated traffic routing policy.

[0117] Subsequently, the traffic routing policy in the traffic scheduling platform is updated based on the traffic scheduling control parameters. The traffic routing policy characterizes the traffic allocation relationship between user access requests and target Internet data centers. Specifically, the traffic routing policy includes the target traffic mapping relationship between the user access area and the target Internet data center, the area access priority relationship, and the traffic scheduling weight relationship corresponding to the target Internet data center.

[0118] In this embodiment, the traffic routing strategy can be expressed as: in, Represents a set of traffic routing policies; Indicates the first Individual user access areas; Indicates the first One target internet data center; This represents the traffic scheduling weight for the corresponding target Internet data center. This expression is used to characterize the dynamic traffic mapping relationship between user access requests in different user access regions and the target Internet data center.

[0119] Furthermore, during the traffic routing policy update process, the system first acquires real-time network status information corresponding to the current user's access area and operational status information corresponding to the target Internet data center. Based on this information, the system dynamically adjusts the current traffic routing policy. Specifically, when the real-time traffic load in the target Internet data center exceeds a preset load threshold, the traffic scheduling weight of the corresponding target Internet data center is reduced; conversely, when the real-time traffic load in the target Internet data center is below the preset load threshold, the traffic scheduling weight of the corresponding target Internet data center is increased.

[0120] The corresponding dynamic weight update expression can be represented as: in, This indicates the updated traffic scheduling weights; This indicates the traffic scheduling weight before the update; Indicates the first Real-time load utilization rate of each target Internet data center; This indicates the load adjustment parameters.

[0121] in, The range of values ​​is .when A larger value indicates that the system is more sensitive to changes in the load of Internet data centers; when... A smaller value indicates that the system has higher stability to traffic fluctuations.

[0122] By dynamically updating the traffic routing strategy as described above, the traffic scheduling platform can dynamically adjust the traffic allocation relationship corresponding to user access requests based on the real-time operating status of different target Internet data centers, thereby improving the flexibility of traffic scheduling and resource utilization efficiency among multiple target Internet data centers.

[0123] Please refer to Figure 2 This illustration shows a schematic diagram of a traffic scheduling device based on machine learning and optimization decision-making according to an embodiment of this application. The device includes an acquisition module 21 and a processing module 22, wherein... The acquisition module 21 is used to acquire historical traffic data from multiple target Internet data centers and construct a traffic feature set based on the historical traffic data; the traffic feature set includes time features, historical lag features, rolling statistical features, and business status features.

[0124] Processing module 22 is used to construct a traffic prediction machine learning model through a set of traffic features, and output the traffic prediction results of each target Internet data center within a preset time window based on the traffic prediction machine learning model; construct traffic scheduling constraints based on the traffic prediction results, and construct a traffic scheduling optimization model using the traffic scheduling constraints as the solution boundary of the model; the traffic scheduling constraints include excess traffic constraints, Internet data center capacity constraints, and access latency constraints; optimize the traffic allocation relationship between each target Internet data center through the traffic scheduling optimization model to obtain the target traffic scheduling scheme corresponding to each target Internet data center; dynamically update the target traffic scheduling scheme through a preset adaptive feedback mechanism, and perform intelligent traffic scheduling operations based on the updated target traffic scheduling scheme.

[0125] In one possible implementation, the acquisition module 21 is used to construct a traffic feature set based on historical traffic data, specifically including: extracting time attribute information based on historical traffic data, and performing discretization encoding on the time attribute information to obtain time features; the time features include hourly encoding, weekday encoding, and weekend identifier; extracting multiple outbound bandwidth values ​​corresponding to different historical moments based on the outbound bandwidth sequence of each target Internet data center, and performing time-series correlation processing on the outbound bandwidth values ​​to obtain historical lag features; calculating the mean and standard deviation of the outbound bandwidth sequence respectively, and using the mean and standard deviation as rolling statistical features; acquiring business operation identifier information corresponding to each target Internet data center based on historical traffic data, and performing category encoding on the business operation identifier information to obtain business status features; the business operation identifier information includes business type identifier, user access area identifier, and Internet data center node identifier; and fusing and encoding the time features, historical lag features, rolling statistical features, and business status features to obtain a traffic feature set.

[0126] In one possible implementation, the processing module 22 is used to construct a traffic prediction machine learning model through a traffic feature set, and output the traffic prediction results of each target Internet data center within a preset time window based on the traffic prediction machine learning model. Specifically, this includes: arranging the traffic feature set in a time series based on the historical traffic change trends corresponding to each target Internet data center to obtain traffic fluctuation differences in different time stages; constructing time-series training samples based on the traffic fluctuation differences; iteratively training the preset machine learning model on the time-series training samples, and verifying the prediction performance of the model parameters of the preset machine learning model according to the operational constraint states corresponding to each target Internet data center during the training process, thereby obtaining the traffic prediction machine learning model; the operational constraint states include bandwidth capacity state, excess bandwidth cost state, and traffic load offset state; inputting the current traffic feature set corresponding to each target Internet data center into the traffic prediction machine learning model, and extracting time-series change correlations through the traffic prediction machine learning model; the time-series change correlations include periodic change correlations, traffic fluctuation correlations, and cross-time window change correlations; outputting the traffic change sequence corresponding to each target Internet data center within the preset time window based on the time-series change correlations; and performing time window mapping processing on the traffic change sequence to obtain the traffic prediction results corresponding to each target Internet data center.

[0127] In one possible implementation, the processing module 22 is used to construct traffic scheduling constraints based on traffic prediction results, and to construct a traffic scheduling optimization model using the traffic scheduling constraints as the model solution boundary. Specifically, this includes: constructing a target optimization problem based on traffic prediction results, and constructing a target optimization function based on the target optimization problem; the target optimization problem includes a target set of Internet data centers, a set of user access areas, regional traffic scheduling variables, traffic prediction values, guaranteed bandwidth values, excess bandwidth cost values, capacity upper limit values, regional traffic demand values, regional access latency values, and target latency thresholds; constructing traffic scheduling constraints based on the target optimization function; and constructing a traffic scheduling optimization model using the target optimization function as the optimization objective and the traffic scheduling constraints as the constraint boundary.

[0128] In one possible implementation, the processing module 22 is used to dynamically update the target traffic scheduling scheme through a preset adaptive feedback mechanism, and perform intelligent traffic scheduling operations based on the updated target traffic scheduling scheme. Specifically, this includes: mapping the target traffic scheduling scheme to traffic scheduling control parameters, and updating the traffic routing policy in the traffic scheduling platform according to the traffic scheduling control parameters; performing cross-data center traffic scheduling for user access requests based on the updated traffic routing policy, and determining in real time whether a data distribution drift event has occurred during the scheduling process; if a data distribution drift event is confirmed to have occurred, retraining the traffic prediction machine learning model; updating the target traffic scheduling scheme through the retrained traffic prediction machine learning model; and performing intelligent traffic scheduling operations based on the updated target traffic scheduling scheme.

[0129] In one possible implementation, the processing module 22 is used to update the traffic routing policy in the traffic scheduling platform according to the traffic scheduling control parameters, specifically including: constructing the traffic scheduling platform; the traffic scheduling platform is used to perform cross-data center routing scheduling of user access requests according to the traffic allocation ratio corresponding to different Internet data centers; updating the traffic routing policy in the traffic scheduling platform according to the traffic scheduling control parameters; the traffic routing policy is used to characterize the traffic allocation relationship between user access requests and target Internet data centers.

[0130] In one possible implementation, the processing module 22 is used to determine in real time whether a data distribution drift event has occurred during the scheduling process, specifically including: acquiring real-time running data during the scheduling process; detecting the distribution error between the traffic prediction result and the real-time running data, and determining whether the distribution error is greater than a preset error threshold; if it is confirmed that the distribution error is greater than the preset error threshold, then determining that a distribution drift event has occurred.

[0131] It should be noted that the above embodiments of the apparatus are only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0132] This application also provides an electronic device. (See reference...) Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: at least one processor 301, at least one communication bus 302, a user interface 303, at least one network interface 304, and a memory 305.

[0133] The communication bus 302 is used to enable communication between these components.

[0134] The user interface 303 may include a display screen and a camera. Optionally, the user interface 303 may also include a standard wired interface and a wireless interface.

[0135] The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0136] The processor 301 may include one or more processing cores. The processor 301 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 305, and by calling data stored in memory 305. Optionally, the processor 301 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 301 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 301 and may be implemented as a separate chip.

[0137] The memory 305 may include random access memory (RAM) or read-only memory. Optionally, the memory 305 may include a non-transitory computer-readable storage medium. The memory 305 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 305 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 305 may also be at least one storage device located remotely from the aforementioned processor 301. (Refer to...) Figure 3 The memory 305, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a traffic scheduling application based on machine learning and optimization decision-making.

[0138] exist Figure 3 In the illustrated electronic device, the user interface 303 is primarily used to provide an input interface for the user and acquire user input data; while the processor 301 can be used to call the traffic scheduling application based on machine learning and optimization decision-making stored in the memory 305. When executed by one or more processors 301, the electronic device performs one or more of the methods described in the above embodiments. It should be noted that, for the foregoing method embodiments, for the sake of simplicity, they are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0139] This application also provides a non-transitory computer-readable storage medium storing instructions. When executed by one or more processors, these instructions cause an electronic device to perform one or more of the methods described in the above embodiments.

[0140] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0141] In the various embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between apparatuses or units may be electrical or other forms.

[0142] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0143] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0144] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0145] The above description is merely an exemplary embodiment disclosed in this application and should not be construed as limiting the scope of this application. Any equivalent changes and modifications made in accordance with the teachings of this application shall still fall within the scope of this application.

[0146] This application is intended to cover any variations, uses, or adaptations disclosed herein that follow the general principles disclosed herein and include common knowledge or customary technical means in the art that are not described in this application.

Claims

1. A traffic scheduling method based on machine learning and optimization decision-making, characterized in that, The method includes: Historical traffic data from multiple target Internet data centers is acquired, and a traffic feature set is constructed based on the historical traffic data; the traffic feature set includes time features, historical lag features, rolling statistical features, and business status features; A traffic prediction machine learning model is constructed using the traffic feature set, and the traffic prediction results for each of the target Internet data centers within a preset time window are output based on the traffic prediction machine learning model. Based on the traffic prediction results, traffic scheduling constraints are constructed, and a traffic scheduling optimization model is constructed using the traffic scheduling constraints as the solution boundary. The traffic scheduling constraints include excess traffic constraints, Internet data center capacity constraints, and access latency constraints. The traffic allocation relationship between each target Internet data center is optimized and solved by the traffic scheduling optimization model to obtain the target traffic scheduling scheme corresponding to each target Internet data center; The target traffic scheduling scheme is dynamically updated by a preset adaptive feedback mechanism, and intelligent traffic scheduling operations are performed based on the updated target traffic scheduling scheme.

2. The method according to claim 1, characterized in that, The construction of the traffic feature set based on the historical traffic data specifically includes: Based on the historical traffic data, time attribute information is extracted and discretized to obtain the time feature; the time feature includes hour code, weekday code, and weekend identifier; Based on the egress bandwidth sequence of each target Internet data center, extract multiple egress bandwidth values ​​corresponding to different historical moments, and perform time-series correlation processing on the egress bandwidth values ​​to obtain the historical lag feature; The mean and standard deviation of the export bandwidth sequence are calculated respectively, and the mean and standard deviation are used as the rolling statistical features; Based on the historical traffic data, obtain the service operation identification information corresponding to each of the target Internet data centers, and perform category encoding processing on the service operation identification information to obtain the service status characteristics; the service operation identification information includes service type identifier, user access area identifier, and Internet data center node identifier; The time feature, the historical lag feature, the rolling statistical feature, and the business status feature are fused and encoded to obtain the traffic feature set.

3. The method according to claim 1, characterized in that, The step of constructing a traffic prediction machine learning model using the traffic feature set, and outputting traffic prediction results for each of the target Internet data centers within a preset time window based on the traffic prediction machine learning model, specifically includes: Based on the historical traffic change trends corresponding to each of the target Internet data centers, the traffic feature set is arranged in a time series to obtain the traffic fluctuation differences in different time periods. Time-series training samples were constructed based on the aforementioned differences in traffic fluctuations; The preset machine learning model is iteratively trained based on the time-series training samples. During the training process, the model parameters of the preset machine learning model are evaluated for prediction performance based on the operational constraint states corresponding to each target Internet data center, thereby obtaining the traffic prediction machine learning model. The operational constraint states include bandwidth capacity state, excess bandwidth cost state, and traffic load offset state. The current traffic feature set corresponding to each of the target Internet data centers is input into the traffic prediction machine learning model, and the time-series change correlation is extracted through the traffic prediction machine learning model; the time-series change correlation includes periodic change correlation, traffic fluctuation correlation, and cross-time window change correlation. Based on the aforementioned temporal change correlation, output the traffic change sequence corresponding to each of the target Internet data centers within the preset time window; The traffic change sequence is processed by time window mapping to obtain the traffic prediction results corresponding to each of the target Internet data centers.

4. The method according to claim 1, characterized in that, The process of constructing traffic scheduling constraints based on the traffic prediction results, and using these constraints as the model solution boundary to construct a traffic scheduling optimization model, specifically includes: Based on the traffic prediction results, a target optimization problem is constructed, and based on the target optimization problem, a target optimization function is constructed; the target optimization problem includes a target set of Internet data centers, a set of user access areas, regional traffic scheduling variables, traffic prediction values, guaranteed bandwidth values, excess bandwidth cost values, capacity upper limits, regional traffic demand values, regional access latency values, and target latency thresholds; Construct traffic scheduling constraints based on the objective optimization function; The traffic scheduling optimization model is constructed using the objective optimization function as the optimization objective and the traffic scheduling constraints as the constraint boundaries.

5. The method according to claim 1, characterized in that, The step of dynamically updating the target traffic scheduling scheme through a preset adaptive feedback mechanism and performing intelligent traffic scheduling operations based on the updated target traffic scheduling scheme specifically includes: The target traffic scheduling scheme is mapped to traffic scheduling control parameters, and the traffic routing strategy in the traffic scheduling platform is updated according to the traffic scheduling control parameters. Based on the updated traffic routing strategy, user access requests are scheduled across data centers, and data distribution drift events are determined in real time during the scheduling process. If the data distribution drift event is confirmed to have occurred, the traffic prediction machine learning model will be retrained. The target traffic scheduling scheme is updated by retraining the traffic prediction machine learning model; Intelligent traffic scheduling operations are performed based on the updated target traffic scheduling scheme.

6. The method according to claim 5, characterized in that, The step of updating the traffic routing policy in the traffic scheduling platform based on the traffic scheduling control parameters specifically includes: The traffic scheduling platform is constructed; the traffic scheduling platform is used to perform cross-data center routing scheduling of user access requests based on the traffic allocation ratio corresponding to different Internet data centers. The traffic routing policy in the traffic scheduling platform is updated according to the traffic scheduling control parameters; the traffic routing policy is used to characterize the traffic allocation relationship between the user access request and the target Internet data center.

7. The method according to claim 5, characterized in that, The real-time determination of whether a data distribution drift event has occurred during the scheduling process specifically includes: Obtain real-time runtime data during the scheduling process; The distribution error between the traffic prediction result and the real-time running data is detected, and it is determined whether the distribution error is greater than a preset error threshold. If the distribution error is confirmed to be greater than the preset error threshold, then the distribution drift event is determined to have occurred.

8. A traffic scheduling device based on machine learning and optimization decision-making, characterized in that, The device includes an acquisition module and a processing module, wherein, The acquisition module is used to acquire historical traffic data from multiple target Internet data centers and construct a traffic feature set based on the historical traffic data; the traffic feature set includes time features, historical lag features, rolling statistical features, and business status features; The processing module is configured to construct a traffic prediction machine learning model using the traffic feature set, and output traffic prediction results for each of the target Internet data centers within a preset time window based on the traffic prediction machine learning model; construct traffic scheduling constraints based on the traffic prediction results, and construct a traffic scheduling optimization model using the traffic scheduling constraints as the model solution boundary; the traffic scheduling constraints include excess traffic constraints, Internet data center capacity constraints, and access latency constraints; optimize the traffic allocation relationship between each of the target Internet data centers using the traffic scheduling optimization model to obtain the target traffic scheduling scheme corresponding to each target Internet data center; dynamically update the target traffic scheduling scheme through a preset adaptive feedback mechanism, and perform intelligent traffic scheduling operations based on the updated target traffic scheduling scheme.

9. An electronic device, characterized in that, It includes a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the traffic scheduling method based on machine learning and optimization decision-making as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1 to 7.

Images