User traffic distribution method and system based on traffic pool

Abstract

The invention discloses a user traffic distribution method and system based on a traffic pool, and the method comprises the steps: mining the features of real-time traffic use data, historical traffic use data and current traffic pool state data from global and local dimensions through an over-pooling layer; and combining with an attention layer: mining a multi-dimensional association relationship among real-time traffic use data, historical traffic use data and current traffic pool state data through an attention mechanism module and bidirectional LSTM fusion, and outputting a distribution feature vector representing a traffic distribution strategy through a full connection layer. A flow distribution strategy represented by distribution characteristics is obtained, the requirements of users are met, flow resources of a system flow pool are considered at the same time, flow distribution is dynamically adjusted according to the requirements of the users and the system load condition, dynamic and efficient distribution of network resources is achieved, the diversified requirements of the users for flow services in the modern network environment are met, and the user experience is improved. The user experience is guaranteed and the system performance is improved.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a method and system for allocating user traffic based on a traffic pool. Background Art

[0002] In the development of modern communication technology, data cards are an important medium for connecting users to the Internet, and their use scope and frequency are growing rapidly. Data cards allow users to access Internet services in environments without fixed network access points, such as on the move or in remote areas. However, as users' demand for mobile data continues to increase, traditional data distribution and management methods have begun to show some limitations:

[0003] Static traffic allocation: Traditional methods usually allocate traffic based on preset traffic packages. This static allocation method cannot adapt to changes in user needs, especially during peak traffic usage periods, when users may encounter insufficient traffic.

[0004] Uneven resource utilization: Since traffic demands vary greatly among different users and at different time periods, static allocation may result in excess traffic for some time periods or some users, while other time periods or other users face traffic shortages.

[0005] Management complexity: With the increase in user base and the diversification of traffic usage patterns, traditional manual management methods have become increasingly complex and difficult to effectively cope with large-scale traffic management needs.

[0006] Lack of personalized services: Traditional traffic distribution methods lack consideration of users’ personalized needs and are unable to provide customized traffic services, which affects the improvement of user experience.

[0007] Therefore, the market urgently needs a new traffic distribution method that can dynamically adapt to user needs, optimize resource allocation, simplify management processes and provide personalized services. Summary of the invention

[0008] The purpose of the present invention is to provide a user traffic allocation method and system based on a traffic pool to solve the above-mentioned problems existing in the prior art.

[0009] An embodiment of the present invention provides a method for allocating user traffic based on a traffic pool, the method comprising:

[0010] Obtain the user's real-time traffic usage data, historical traffic usage data and current traffic pool status data; the real-time traffic usage data represents the user's real-time traffic usage; the historical traffic usage data represents the user's traffic usage in the past time period; the current traffic pool status data represents the current resource status in the traffic pool;

[0011] Through the traffic allocation model, based on the real-time traffic usage data, historical traffic usage data and current traffic pool status data, the allocation characteristics are obtained; the allocation characteristics represent the traffic allocation strategy, and the traffic allocation strategy includes traffic allocation nodes and traffic allocation amounts;

[0012] Among them, the traffic distribution model includes the first convolutional layer, the second convolutional layer, the third convolutional layer, the pooling layer, the attention layer and the fully connected layer; the attention layer includes the attention mechanism module and the bidirectional LSTM; the bidirectional LSTM includes the forward LSTM and the backward LSTM; the input of the first convolutional layer is the real-time traffic usage data, the input of the second convolutional layer is the historical traffic usage data, and the input of the third convolutional layer is the current traffic pool status data; the pooling layer mines the features of the real-time traffic usage data, the historical traffic usage data and the current traffic pool status data from both global and local dimensions, and combines them with the attention layer: the multi-dimensional correlation between the real-time traffic usage data, the historical traffic usage data and the current traffic pool status data is mined through the attention mechanism module and the bidirectional LSTM fusion, and the allocation feature vector representing the traffic distribution strategy is output through the fully connected layer.

[0013] Optionally, a traffic allocation model is used to obtain a distribution feature vector based on real-time traffic usage data, historical traffic usage data, and current traffic pool status data, including:

[0014] The real-time traffic usage data, the historical traffic usage data and the current traffic pool status data are input into the first convolution layer, the second convolution layer and the third convolution layer respectively; the first convolution layer, the second convolution layer and the third convolution layer respectively output the first convolution data, the second convolution data and the third convolution data;

[0015] Extracting a first main feature vector of the first convolution data through the maximum pooling layer in the pooling layer; extracting a second main feature vector of the second convolution data through the maximum pooling layer in the pooling layer; extracting a third main feature vector of the third convolution data through the maximum pooling layer in the pooling layer;

[0016] Extracting a first global feature vector of the first convolution data through a global pooling layer in the pooling layer; extracting a second global feature vector of the second convolution data through a global pooling layer in the pooling layer; extracting a third global feature vector of the third convolution data through a global pooling layer in the pooling layer;

[0017] The global features and local features are fused through the attention mechanism to obtain fused features: the first local feature vector and the first global feature vector are fused through the attention mechanism to obtain a first fused feature vector; the second local feature vector and the second global feature vector are fused through the attention mechanism to obtain a second fused feature vector; the third local feature vector and the third global feature vector are fused through the attention mechanism to obtain a third fused feature vector;

[0018] Through the attention mechanism combined with bidirectional LSTM, the allocation feature vector is obtained based on the first fused feature vector, the second fused feature vector and the third fused feature.

[0019] Optionally, an attention mechanism is combined with a bidirectional LSTM to obtain an allocation feature vector based on the first fused feature vector, the second fused feature vector, and the third fused feature, including:

[0020] Obtain a first time dimension feature vector based on the first fused feature vector through a bidirectional LSTM;

[0021] Obtain a second time dimension feature vector based on the second fused feature vector through a bidirectional LSTM;

[0022] Obtain a third time dimension feature vector based on the third fused feature vector through a bidirectional LSTM;

[0023] Obtaining, through the attention mechanism, a first weight matrix of the feature vector of the first time dimension, a second weight matrix of the feature vector of the second time dimension, and a third weight matrix of the feature vector of the third time dimension;

[0024] The first time dimension feature, the second time dimension feature and the third time dimension feature are integrated based on the first weight matrix, the second weight matrix and the third weight matrix to obtain the attention feature;

[0025] The attention features are extracted through the fully connected layer to obtain the assigned feature vector.

[0026] Optionally, obtaining a first weight matrix of the feature vector of the first time dimension, a second weight matrix of the feature vector of the second time dimension, and a third weight matrix of the feature vector of the third time dimension through an attention mechanism includes:

[0027] Through the attention mechanism, a first weight matrix of the first time dimension feature vector is obtained, including: for the kth first time dimension feature t1 of the first time dimension feature vector k And the second time dimension feature vector t2, calculate the kth first time dimension feature t1 k The first similarity score e1 with the second time dimension feature vector t2 k =f(t1 k , t2), where f(t1 k , t2) represents the change of t1 k Perform point convolution operation with t2;

[0028] The first attention weight is obtained based on the first similarity score, specifically:

[0029]

[0030] Among them, α1 k represents the attention weight of the kth first time dimension feature, e1 s Represents the sth first time dimension feature t1 s The first similarity score with the second time dimension feature vector t2; e1 k Represents the kth first time dimension feature t1 k The first similarity score with the second time dimension feature vector t2; S represents the number of first time dimension features;

[0031] Based on the first attention weight, the first weight matrix is ​​obtained: W1 = [α11, α12, ..., α1 S ]

[0032] Among them, W1 is the first weight matrix; α11 represents the first attention weight of the first first time dimension feature, α12 represents the first attention weight of the second first time dimension feature, α1 S Represents the first attention weight of the Sth first time dimension feature;

[0033] Through the attention mechanism, a second weight matrix of the second time dimension feature vector is obtained, including: for the nth second time dimension feature t2 of the second time dimension feature vector n And the third time dimension feature vector t3, calculate the nth second time dimension feature t2 n The second similarity score e2 with the third time dimension feature vector t3 n =f(t2 n , t3), where f(t2 n , t3) represents the change of t2 n Perform point convolution operation with t3;

[0034] The second attention weight is obtained based on the second similarity score, specifically:

[0035]

[0036] Among them, α2 n represents the second attention weight of the nth second time dimension feature, e2 u Represents the u-th second time dimension feature t2 u The second similarity score with the third time dimension feature vector t3; e2 n Represents the nth second time dimension feature t2 n The second similarity score with the third time dimension feature vector t3; U represents the number of second time dimension features;

[0037] Based on the second attention weight, the second weight matrix is ​​obtained: W2 = [α21, α22, ..., α2 U ]

[0038] W2 represents the second weight matrix, α21, α22, α2 U They respectively represent the second attention weight of the first second time dimension feature, the second attention weight of the second second time dimension feature, and the second attention weight of the Uth second time dimension feature;

[0039] Through the attention mechanism, a third weight matrix of the feature vector of the third time dimension is obtained, including: for the mth third time dimension feature t3 of the feature vector of the third time dimension m And the first time dimension feature vector t1, calculate the mth third time dimension feature t3 m The third similarity score e3 with the first time dimension feature vector t1 m =f(t3 m , t1), where f(t3 m , t1) represents the difference between t3 m Perform point convolution operation with t1;

[0040] The second attention weight is obtained based on the second similarity score, specifically:

[0041]

[0042] Among them, α3 m represents the third attention weight of the nth third time dimension feature, e3 m The mth third time dimension feature t3 m The third similarity score with the first time dimension feature vector t1; e3 y Represents the yth third time dimension feature t3 y The third similarity score with the first time dimension feature vector t1; Y represents the number of third time dimension features;

[0043] The third weight matrix is ​​obtained based on the third attention weight: W3 = [α31, α32, ..., α3 Y ]

[0044] W2 represents the second weight matrix, α3, α32, α3 Y They respectively represent the third attention weight of the first third time dimension feature, the third attention weight of the second third time dimension feature, and the third attention weight of the Yth third time dimension feature.

[0045] Optionally, the first time dimension feature, the second time dimension feature, and the third time dimension feature are fused based on the first weight matrix, the second weight matrix, and the third weight matrix to obtain the attention feature, including:

[0046] Introduce the attention mechanism and obtain the attention features:

[0047] F=σ(W1·avgpool(X1))·X1+σ(W2·avgpool(X2))·X2+σ(W3·avgpool(X3))·X3

[0048] Among them, F represents the attention feature, avgpool() represents the average pooling operation on the data in (), σ() is the activation function, which means that the data in () is activated by the activation function, · represents the dot product operation, X1, X2 and X3 represent the first time dimension feature, the second time dimension feature and the third time dimension feature respectively; W1, W2, W3 represent the first weight matrix, the second weight matrix and the third weight matrix respectively.

[0049] Optionally, the global features and local features are fused through the attention mechanism to obtain fused features, including:

[0050] Specifically, the global features and local features are fused through the attention mechanism to obtain the fused features as follows: for the i-th local feature l i And the global feature g, calculate the i-th local feature l i Similarity score e with global feature g i =f(l i , g), where f(l i , g) indicates l i Perform point convolution operation with g;

[0051] The attention weight is obtained based on the similarity score, specifically:

[0052] Among them, α i represents the attention weight of the i-th local feature, e j represents the jth local feature l j Similarity score with the global feature g; e i represents the i-th local feature l i The similarity score with the global feature g; J represents the number of local features;

[0053] The global features and local features are weighted summed according to the calculated attention weights to obtain the fused features:

[0054]

[0055] Among them, h represents the fusion feature.

[0056] Optionally, the global features and the local features in the above operation of fusing the global features and the local features are replaced with a first local feature vector and a first global feature vector, respectively, to obtain a first fused feature vector;

[0057] Replacing the global features and the local features in the above operation of fusing the global features and the local features with the second local feature vector and the second global feature vector, respectively, to obtain a second fused feature vector;

[0058] The global features and the local features in the above operation of fusing the global features and the local features are replaced with the third local feature vector and the third global feature vector respectively to obtain a third fused feature vector.

[0059] Optionally, the method further includes:

[0060] Establishment of a traffic pool: determine a central controller and deploy multiple distributed nodes in the network; the central controller is used to centrally manage and schedule resources in the traffic pool; the distributed nodes are responsible for collecting traffic data and executing the instructions of the central controller; establish a data storage system, which is used to store the configuration information, traffic data and performance indicators of the traffic pool.

[0061] Optionally, the method further includes: obtaining a traffic distribution strategy corresponding to the distribution feature from a database.

[0062] The embodiment of the present invention also provides a user traffic allocation system based on a traffic pool, the system comprising:

[0063] The acquisition module is used to obtain the user's real-time traffic usage data, historical traffic usage data and current traffic pool status data; the real-time traffic usage data represents the user's real-time traffic usage; the historical traffic usage data represents the user's traffic usage in the past time period; the current traffic pool status data represents the current resource status in the traffic pool;

[0064] The prediction module is used to obtain the allocation characteristics through the traffic allocation model based on the real-time traffic usage data, the historical traffic usage data and the current traffic pool status data; the allocation characteristics represent the traffic allocation strategy, and the traffic allocation strategy includes the traffic allocation node and the traffic allocation amount;

[0065] Among them, the traffic distribution model includes the first convolutional layer, the second convolutional layer, the third convolutional layer, the pooling layer, the attention layer and the fully connected layer; the attention layer includes the attention mechanism module and the bidirectional LSTM; the bidirectional LSTM includes the forward LSTM and the backward LSTM; the input of the first convolutional layer is the real-time traffic usage data, the input of the second convolutional layer is the historical traffic usage data, and the input of the third convolutional layer is the current traffic pool status data; the pooling layer mines the features of the real-time traffic usage data, the historical traffic usage data and the current traffic pool status data from both global and local dimensions, and combines them with the attention layer: the multi-dimensional correlation between the real-time traffic usage data, the historical traffic usage data and the current traffic pool status data is mined through the attention mechanism module and the bidirectional LSTM fusion, and the allocation feature vector representing the traffic distribution strategy is output through the fully connected layer.

[0066] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:

[0067] The embodiment of the present invention provides a method and system for distributing user traffic based on a traffic pool. The pooling layer mines the features of real-time traffic usage data, historical traffic usage data and current traffic pool status data from global and local dimensions, and combines them with the attention layer: the multi-dimensional correlation between real-time traffic usage data, historical traffic usage data and current traffic pool status data is mined through the attention mechanism module and the bidirectional LSTM fusion, and the distribution feature vector representing the traffic distribution strategy is output through the fully connected layer. That is, the user's traffic usage characteristics are mined from time, space, horizontal and vertical dimensions, and combined with the influence of the time, space, horizontal and vertical multi-dimensional characteristics of the traffic pool on the user's traffic usage characteristics, the traffic distribution strategy represented by the distribution characteristics is obtained, which fits the user's needs while taking into account the traffic resources of the system traffic pool, dynamically meets the user's traffic needs and efficiently utilizes the traffic resources, and can dynamically adjust the traffic distribution strategy according to the real-time needs and resource conditions, with high flexibility and high resource utilization; through intelligent scheduling, the utilization of server and network resources is improved. At the same time, the response speed and service quality of user requests are improved. That is, the technical solution provided by the embodiment of the present invention at least achieves the following beneficial effects:

[0068] Improve system performance and stability; optimize user experience and reduce latency; adaptively adjust to suit different scenarios; and flexibly respond to traffic fluctuations during peak hours.

[0069] In summary, the embodiments of the present invention can monitor and analyze user traffic data in real time, dynamically adjust traffic distribution according to user needs and system load conditions, and realize dynamic and efficient allocation of network resources to meet the diverse needs of users for traffic services in a modern network environment, thereby ensuring user experience and improving system performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 It is a flow chart of a method for allocating user traffic based on a traffic pool provided in an embodiment of the present invention.

[0071] Figure 2 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention.

[0072] Markings in the figure: 500 - bus; 501 - receiver; 502 - processor; 503 - transmitter; 504 - memory; 505 - bus interface. DETAILED DESCRIPTION

[0073] The present invention will be described in detail below in conjunction with the accompanying drawings.

[0074] Example

[0075] The embodiment of the present invention provides a method for allocating user traffic based on a traffic pool. Figure 1 As shown, the user traffic allocation method based on the traffic pool includes:

[0076] S101: Obtain the user's real-time traffic usage data, historical traffic usage data and current traffic pool status data.

[0077] Real-time traffic usage data characterizes the user's real-time traffic usage; historical traffic usage data characterizes the user's traffic usage in the past time period; and the current traffic pool status data characterizes the current resource situation in the traffic pool. In an embodiment of the present invention, the real-time traffic usage data includes real-time traffic usage, real-time time, real-time location, and real-time device type; the historical traffic usage data includes multiple historical traffic usages, multiple historical times, multiple historical locations, and multiple historical device types; the current traffic pool status data includes available bandwidth, network load, delay, packet loss rate, jitter, number of connections, CPU usage, memory usage, and network device status. Available bandwidth: the amount of unused bandwidth in the current network, usually expressed in Mbps or Gbps. Network load: the traffic load situation in the current network, including overall traffic, peak traffic, and average traffic, etc. Delay: the delay of the current network, that is, the time required for data to be transmitted from one point to another, usually in milliseconds (ms). Packet loss rate: the packet loss rate in the current network, expressed as a percentage. Jitter: the jitter in the current network, that is, the change in the packet transmission time, usually in milliseconds (ms). Number of connections: The number of active connections in the current network, including user connections and server connections. CPU usage: The CPU usage of network devices (such as routers, switches) and servers, expressed as a percentage. Memory usage: The memory usage of network devices and servers, expressed as a percentage. Network device status: The operating status and health of network devices, including temperature, failure rate, etc.

[0078] S102: Obtaining allocation characteristics through a traffic allocation model based on real-time traffic usage data, historical traffic usage data and current traffic pool status data, wherein the allocation characteristics represent a traffic allocation strategy. Optionally, the traffic allocation strategy includes a traffic allocation node and a traffic allocation amount.

[0079] In an embodiment of the present invention, a traffic distribution model includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a pooling layer, an attention layer and a fully connected layer; the attention layer includes an attention mechanism module and a bidirectional LSTM; the bidirectional LSTM includes a forward LSTM and a backward LSTM; the input of the first convolutional layer is real-time traffic usage data, the input of the second convolutional layer is historical traffic usage data, and the input of the third convolutional layer is current traffic pool status data; the features of real-time traffic usage data, historical traffic usage data and current traffic pool status data are mined from both global and local dimensions through the pooling layer, and combined with the attention layer: the multi-dimensional correlation relationship between real-time traffic usage data, historical traffic usage data and current traffic pool status data is mined through the attention mechanism module and the bidirectional LSTM fusion, and the allocation feature vector representing the traffic distribution strategy is output through the fully connected layer.

[0080] By adopting the above scheme, the user's traffic usage characteristics are mined from time, space, horizontal and vertical dimensions, and the influence of the time, space, horizontal and vertical characteristics of the traffic pool on the user's traffic usage characteristics is combined to obtain a traffic allocation strategy characterized by the allocation characteristics, which meets the needs of users while taking into account the traffic resources of the system traffic pool, dynamically meets the user's traffic needs and efficiently utilizes traffic resources. It can dynamically adjust the traffic allocation strategy according to real-time needs and resource conditions, with high flexibility and high resource utilization; through intelligent scheduling, the utilization rate of servers and network resources is improved. At the same time, the response speed and service quality of user requests are improved. That is, the technical solution provided by the embodiment of the present invention at least achieves the following beneficial effects:

[0081] Improve system performance and stability; optimize user experience and reduce latency; adaptively adjust to suit different scenarios; and flexibly respond to traffic fluctuations during peak hours.

[0082] Optionally, a traffic allocation model is used to obtain a distribution feature vector based on real-time traffic usage data, historical traffic usage data, and current traffic pool status data, including:

[0083] The real-time traffic usage data, historical traffic usage data and current traffic pool status data are input into the first convolution layer, the second convolution layer and the third convolution layer respectively. The first convolution layer, the second convolution layer and the third convolution layer output the first convolution data, the second convolution data and the third convolution data respectively.

[0084] The first main eigenvector of the first convolution data is extracted through the maximum pooling layer in the pooling layer, the second main eigenvector of the second convolution data is extracted through the maximum pooling layer in the pooling layer, and the third main eigenvector of the third convolution data is extracted through the maximum pooling layer in the pooling layer.

[0085] The first global feature vector of the first convolution data is extracted through the global pooling layer in the pooling layer, the second global feature vector of the second convolution data is extracted through the global pooling layer in the pooling layer, and the third global feature vector of the third convolution data is extracted through the global pooling layer in the pooling layer.

[0086] The global features and local features are fused through the attention mechanism to obtain fused features, including: fusing the first local feature vector and the first global feature vector through the attention mechanism to obtain a first fused feature vector, fusing the second local feature vector and the second global feature vector through the attention mechanism to obtain a second fused feature vector, and fusing the third local feature vector and the third global feature vector through the attention mechanism to obtain a third fused feature vector.

[0087] Among them, the global features and local features are fused through the attention mechanism to obtain fused features, including:

[0088] Specifically, the global features and local features are fused through the attention mechanism to obtain the fused features as follows: for the i-th local feature l i And the global feature g, calculate the i-th local feature l i Similarity score e with global feature g i =f(l i , g), where f(l i , g) indicates l i Perform point convolution operation with g;

[0089] The attention weight is obtained based on the similarity score, specifically:

[0090] Among them, α i represents the attention weight of the i-th local feature, e j represents the jth local feature l j Similarity score with the global feature g; e i represents the i-th local feature l i The similarity score with the global feature g; J represents the number of local features;

[0091] The global features and local features are weighted summed according to the calculated attention weights to obtain the fused features:

[0092]

[0093] Among them, h represents the fusion feature.

[0094] Specifically, the global features and the local features in the above operation of fusing the global features and the local features are replaced with the first local feature vector and the first global feature vector, respectively, to obtain a first fused feature vector;

[0095] Replacing the global features and the local features in the above operation of fusing the global features and the local features with the second local feature vector and the second global feature vector, respectively, to obtain a second fused feature vector;

[0096] The global features and the local features in the above operation of fusing the global features and the local features are replaced with a third local feature vector and a third global feature vector respectively to obtain a third fused feature vector.

[0097] Through the attention mechanism combined with bidirectional LSTM, the allocation feature vector is obtained based on the first fused feature vector, the second fused feature vector and the third fused feature.

[0098] Among them, the attention mechanism is combined with the bidirectional LSTM to obtain the allocation feature vector based on the first fused feature vector, the second fused feature vector and the third fused feature, including:

[0099] The first time dimension feature vector is obtained based on the first fused feature vector by the bidirectional LSTM, the second time dimension feature vector is obtained based on the second fused feature vector by the bidirectional LSTM, and the third time dimension feature vector is obtained based on the third fused feature vector by the bidirectional LSTM. Specifically, the first fused feature is forward predicted by the forward LSTM of the bidirectional LSTM to obtain the first forward feature, the first fused feature is backward predicted by the backward LSTM of the bidirectional LSTM to obtain the first backward feature, the first forward feature and the first backward feature are fused to obtain the first time dimension feature vector, and the specific fusion method can be weighted summation, or the first forward feature and the first backward feature are concatenated to obtain the first time dimension feature vector. The second fused feature is forward predicted by the forward LSTM of the bidirectional LSTM to obtain the second forward feature, the second fused feature is backward predicted by the backward LSTM of the bidirectional LSTM to obtain the second backward feature, the second forward feature and the second backward feature are fused to obtain the second time dimension feature vector, and the specific fusion method can be weighted summation, or the second forward feature and the second backward feature are concatenated to obtain the second time dimension feature vector. The third fused feature is forward predicted by the forward LSTM of the bidirectional LSTM to obtain the third forward feature, the third fused feature is backward predicted by the backward LSTM of the bidirectional LSTM to obtain the third backward feature, the third forward feature and the third backward feature are fused to obtain a third time dimension feature vector. The specific fusion method can be weighted summation or concatenation of the third forward feature and the third backward feature to obtain the third time dimension feature vector.

[0100] Through the attention mechanism, a first weight matrix of the feature vector of the first time dimension, a second weight matrix of the feature vector of the second time dimension, and a third weight matrix of the feature vector of the third time dimension are obtained. Specifically, through the attention mechanism, a first weight matrix of the feature vector of the first time dimension, a second weight matrix of the feature vector of the second time dimension, and a third weight matrix of the feature vector of the third time dimension are obtained, including:

[0101] Through the attention mechanism, a first weight matrix of the first time dimension feature vector is obtained, including: for the kth first time dimension feature t1 of the first time dimension feature vector k And the second time dimension feature vector t2, calculate the kth first time dimension feature t1 k The first similarity score e1 with the second time dimension feature vector t2 k =f(t1 k , t2), where f(t1 k , t2) represents the change of t1 k Perform point convolution operation with t2;

[0102] The first attention weight is obtained based on the first similarity score, specifically:

[0103]

[0104] Among them, α1 k represents the attention weight of the kth first time dimension feature, e1 s Represents the sth first time dimension feature t1 s The first similarity score with the second time dimension feature vector t2; e1 k Represents the kth first time dimension feature t1 k The first similarity score with the second time dimension feature vector t2; S represents the number of first time dimension features;

[0105] Based on the first attention weight, the first weight matrix is ​​obtained: W1 = [α11, α12, ..., α1 S ]

[0106] Among them, W1 is the first weight matrix; α11 represents the first attention weight of the first first time dimension feature, α12 represents the first attention weight of the second first time dimension feature, α1 S Represents the first attention weight of the Sth first time dimension feature;

[0107] Through the attention mechanism, a second weight matrix of the second time dimension feature vector is obtained, including: for the nth second time dimension feature t2 of the second time dimension feature vector n And the third time dimension feature vector t3, calculate the nth second time dimension feature t2 n The second similarity score e2 with the third time dimension feature vector t3 n =f(t2 n , t3), where f(t2 n , t3) represents the change of t2 n Perform point convolution operation with t3;

[0108] The second attention weight is obtained based on the second similarity score, specifically:

[0109]

[0110] Among them, α2 n represents the second attention weight of the nth second time dimension feature, e2 u Represents the u-th second time dimension feature t2 u The second similarity score with the third time dimension feature vector t3; e2 n Represents the nth second time dimension feature t2 nThe second similarity score with the third time dimension feature vector t3; U represents the number of second time dimension features;

[0111] Based on the second attention weight, the second weight matrix is ​​obtained: W2 = [α21, α22, ..., α2 U ]

[0112] W2 represents the second weight matrix, α21, α22, α2 U They respectively represent the second attention weight of the first second time dimension feature, the second attention weight of the second second time dimension feature, and the second attention weight of the Uth second time dimension feature;

[0113] Through the attention mechanism, a third weight matrix of the feature vector of the third time dimension is obtained, including: for the mth third time dimension feature t3 of the feature vector of the third time dimension m And the first time dimension feature vector t1, calculate the mth third time dimension feature t3 m The third similarity score e3 with the first time dimension feature vector t1 m =f(t3 m , t1), where f(t3 m , t1) represents the difference between t3 m Perform point convolution operation with t1;

[0114] The second attention weight is obtained based on the second similarity score, specifically:

[0115]

[0116] Among them, α3 m represents the third attention weight of the nth third time dimension feature, e3 m The mth third time dimension feature t3 m The third similarity score with the first time dimension feature vector t1; e3 y Represents the yth third time dimension feature t3 y The third similarity score with the first time dimension feature vector t1; Y represents the number of third time dimension features;

[0117] The third weight matrix is ​​obtained based on the third attention weight: W3 = [α31, α32, ..., α3 Y ]

[0118] W2 represents the second weight matrix, α3, α32, α3 Y They respectively represent the third attention weight of the first third time dimension feature, the third attention weight of the second third time dimension feature, and the third attention weight of the Yth third time dimension feature.

[0119] Based on the first weight matrix, the second weight matrix and the third weight matrix, the first time dimension feature, the second time dimension feature and the third time dimension feature are fused to obtain the attention feature. Specifically, the attention mechanism is introduced to obtain the attention feature:

[0120] F=σ(W1·avgpool(X1))·X1+σ(W2·avgpool(X2))·X2+σ(W3·avgpool(X3))·X3

[0121] Among them, F represents the attention feature, avgpool() represents the average pooling operation on the data in (), σ() is the activation function, which means that the data in () is activated by the activation function, · represents the dot product operation, X1, X2 and X3 represent the first time dimension feature, the second time dimension feature and the third time dimension feature respectively; W1, W2, W3 represent the first weight matrix, the second weight matrix and the third weight matrix respectively.

[0122] The attention features are extracted through the fully connected layer to obtain the assigned feature vector.

[0123] Optionally, the method for allocating user traffic based on the traffic pool also includes:

[0124] Establishment of a traffic pool: determine a central controller and deploy multiple distributed nodes in the network; the central controller is used to centrally manage and schedule resources in the traffic pool; the distributed nodes are responsible for collecting traffic data and executing the instructions of the central controller; establish a data storage system, which is used to store the configuration information, traffic data and performance indicators of the traffic pool.

[0125] As an optional implementation, the method for distributing user traffic based on traffic pool further includes: obtaining a traffic distribution strategy corresponding to the distribution feature from a database. In an embodiment of the present invention, a traffic distribution strategy corresponding to the distribution feature is pre-stored in the database.

[0126] In the embodiment of the present invention, the traffic distribution model can be used directly without training. The traffic distribution model can also be trained by the following method:

[0127] Obtain training set: The training set includes real-time test data of users, historical traffic training data, and traffic pool status training data;

[0128] The real-time test data, the historical traffic training data and the traffic pool status training data are respectively input into the first convolution layer, the second convolution layer and the third convolution layer to obtain the first training feature, the second training feature and the third training feature respectively, and then the first training feature, the second training feature and the third training feature are respectively locally and globally pooled through the pooling layer to obtain the first pooling feature, the second pooling feature and the third pooling feature respectively;

[0129] The first pooling feature, the second pooling feature and the third pooling feature are fused through the attention layer to obtain the fused attention training feature; the fused attention training feature is extracted through the fully connected layer to obtain the output feature;

[0130] Obtain the loss function based on the output features:

[0131]

[0132] Among them, Loss represents the loss function of the model, B represents the dimension of the output feature, that is, the number of features in the output feature, x1 b Represents the bth feature in the output feature, x0 b represents the bth feature of the first training feature, is the weight matrix A b The transposed matrix of b represents the second pooling feature, H b represents the third pooling feature, δ and β are weight parameters, and the values ​​of δ and β are in the range of [0.01, 0.1, 1, 10]. The method for obtaining the weight matrix is ​​the same as the method for obtaining the third weight matrix mentioned above. For details, please refer to the description of the method for obtaining the third weight matrix mentioned above, which will not be repeated here.

[0133] In the embodiment of the present invention, the output feature and the first training feature are further dimensionally aligned through an embedding layer. Specifically, the output feature and the first training feature are convolved to obtain output features and the first training feature of the same dimension, and then the loss function is calculated.

[0134] If the loss function converges, it is determined that the training of the traffic distribution model is completed.

[0135] By adopting the above scheme, the pooling layer can mine the features of real-time traffic usage data, historical traffic usage data and current traffic pool status data from both global and local dimensions, and combine them with the attention layer: the multi-dimensional correlation between real-time traffic data, historical traffic data and traffic pool status data is mined through the attention mechanism module and the bidirectional LSTM fusion, and the distribution feature vector representing the traffic distribution strategy is output through the fully connected layer. The reliability and security of traffic distribution are improved.

[0136] Based on the above-mentioned user traffic allocation method based on traffic pool, an embodiment of the present invention provides a user traffic allocation system based on traffic pool, which is used to execute the above-mentioned user traffic allocation method based on traffic pool. The user traffic allocation system based on traffic pool includes:

[0137] The acquisition module is used to obtain the user's real-time traffic usage data, historical traffic usage data and current traffic pool status data; the real-time traffic usage data represents the user's real-time traffic usage; the historical traffic usage data represents the user's traffic usage in the past time period; the current traffic pool status data represents the current resource situation in the traffic pool; the real-time traffic usage data includes real-time traffic usage, real-time time, real-time location, and real-time device type; the historical traffic usage data includes multiple historical traffic usages, multiple historical times, multiple historical locations, and multiple historical device types; the current traffic pool status data includes available bandwidth, network load, delay, packet loss rate, jitter, number of connections, CPU usage, memory usage, and network device status.

[0138] The prediction module is used to obtain allocation characteristics through a traffic allocation model based on real-time traffic usage data, historical traffic usage data and current traffic pool status data; the allocation characteristics represent the traffic allocation strategy, and the traffic allocation strategy includes traffic allocation nodes and traffic allocation amounts.

[0139] Optionally, the system further includes:

[0140] The traffic pool establishment module is used to establish the traffic pool: determine the central controller and deploy multiple distributed nodes in the network; the central controller is used to centrally manage and schedule resources in the traffic pool; the distributed nodes are responsible for collecting traffic data and executing the instructions of the central controller; establish a data storage system, which is used to store the configuration information, traffic data and performance indicators of the traffic pool.

[0141] Optionally, the system further includes: a traffic distribution module, configured to obtain a traffic distribution strategy corresponding to the distribution feature from a database.

[0142] Based on the above-mentioned user traffic allocation method based on traffic pool, an embodiment of the present invention provides a user traffic allocation system based on traffic pool, which is used to execute the above-mentioned user traffic allocation method based on traffic pool. The user traffic allocation system based on traffic pool includes:

[0143] The specific implementation of the functions of each module of the above system is as described in the user traffic allocation method based on the traffic pool, which will not be repeated here.

[0144] The embodiment of the present invention also provides an electronic device for integrating the above-mentioned user traffic allocation system based on traffic pool, such as Figure 2 As shown, the electronic device includes a memory 504, a processor 502, and a computer program stored in the memory 504 and executable on the processor 502. When the processor 502 executes the program, the steps of any of the above-mentioned methods for allocating user traffic based on the traffic pool are implemented.

[0145] Among them, Figure 2 In the embodiment of the present invention, a bus architecture (represented by bus 500) is shown, which may include any number of interconnected buses and bridges, and bus 500 links various circuits including one or more processors represented by processor 502 and memory represented by memory 504. Bus 500 may also link various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. Bus interface 505 provides an interface between bus 500 and receiver 501 and transmitter 503. Receiver 501 and transmitter 503 may be the same element, namely a transceiver, providing a unit for communicating with various other devices over a transmission medium. Processor 502 is responsible for managing bus 500 and general processing, while memory 504 may be used to store data used by processor 502 when performing operations.

[0146] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-mentioned methods for allocating user traffic based on a traffic pool.

[0147] The algorithm and display provided herein are not inherently related to any particular computer, virtual system or other device. Various general purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing such systems. In addition, the present invention is not directed to any specific programming language either. It should be understood that various programming languages ​​can be utilized to realize the content of the present invention described herein, and the description of the above specific languages ​​is for disclosing the best mode of the present invention.

[0148] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0149] Similarly, it should be understood that in order to streamline the present disclosure and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of the present invention, various features of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed method should not be interpreted as reflecting the intention that the claimed invention requires more features than those explicitly recited in each claim. More specifically, as reflected in the claims below, inventive aspects lie in less than all the features of the individual embodiments disclosed above. Therefore, the claims that follow the specific embodiment are hereby expressly incorporated into the specific embodiment, with each claim itself serving as a separate embodiment of the present invention.

[0150] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and in addition they may be divided into a plurality of submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this manner may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.

[0151] In addition, those skilled in the art will appreciate that, although some embodiments herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present invention and form different embodiments. For example, in the claims below, any one of the claimed embodiments may be used in any combination.

[0152] The various component embodiments of the present invention may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or digital signal processor (DSP) may be used in practice to implement some or all of the functions of some or all of the components in the apparatus according to an embodiment of the present invention. The present invention may also be implemented as a device or device program (e.g., a computer program and a computer program product) for executing part or all of the methods described herein. Such a program implementing the present invention may be stored on a computer-readable medium, or may have the form of one or more signals. Such a signal may be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

Claims

1. A method for allocating user traffic based on a traffic pool, characterized in that: The method comprises: Obtain the user's real-time traffic usage data, historical traffic usage data and current traffic pool status data; the real-time traffic usage data represents the user's real-time traffic usage; the historical traffic usage data represents the user's traffic usage in the past time period; the current traffic pool status data represents the current resource status in the traffic pool; Through the traffic allocation model, based on the real-time traffic usage data, historical traffic usage data and current traffic pool status data, the allocation characteristics are obtained; the allocation characteristics represent the traffic allocation strategy, and the traffic allocation strategy includes traffic allocation nodes and traffic allocation amounts; Among them, the traffic distribution model includes the first convolutional layer, the second convolutional layer, the third convolutional layer, the pooling layer, the attention layer and the fully connected layer; the attention layer includes the attention mechanism module and the bidirectional LSTM; the bidirectional LSTM includes the forward LSTM and the backward LSTM; the input of the first convolutional layer is the real-time traffic usage data, the input of the second convolutional layer is the historical traffic usage data, and the input of the third convolutional layer is the current traffic pool status data; the pooling layer mines the features of the real-time traffic usage data, the historical traffic usage data and the current traffic pool status data from both global and local dimensions, and combines them with the attention layer: the multi-dimensional correlation between the real-time traffic usage data, the historical traffic usage data and the current traffic pool status data is mined through the attention mechanism module and the bidirectional LSTM fusion, and the allocation feature vector representing the traffic distribution strategy is output through the fully connected layer.

2. The method for allocating user traffic based on traffic pool according to claim 1, characterized in that: Through the traffic allocation model, based on the real-time traffic usage data, historical traffic usage data and current traffic pool status data, the allocation feature vector is obtained, including: The real-time traffic usage data, the historical traffic usage data and the current traffic pool status data are input into the first convolution layer, the second convolution layer and the third convolution layer respectively; the first convolution layer, the second convolution layer and the third convolution layer respectively output the first convolution data, the second convolution data and the third convolution data; Extracting a first main feature vector of the first convolution data through the maximum pooling layer in the pooling layer; extracting a second main feature vector of the second convolution data through the maximum pooling layer in the pooling layer; extracting a third main feature vector of the third convolution data through the maximum pooling layer in the pooling layer; Extracting a first global feature vector of the first convolution data through a global pooling layer in the pooling layer; extracting a second global feature vector of the second convolution data through a global pooling layer in the pooling layer; extracting a third global feature vector of the third convolution data through a global pooling layer in the pooling layer; The global features and local features are fused through the attention mechanism to obtain fused features: the first local feature vector and the first global feature vector are fused through the attention mechanism to obtain a first fused feature vector; the second local feature vector and the second global feature vector are fused through the attention mechanism to obtain a second fused feature vector; the third local feature vector and the third global feature vector are fused through the attention mechanism to obtain a third fused feature vector; Through the attention mechanism combined with bidirectional LSTM, the allocation feature vector is obtained based on the first fused feature vector, the second fused feature vector and the third fused feature.

3. The method for distributing user traffic based on traffic pool according to claim 2, characterized in that: By combining the attention mechanism with the bidirectional LSTM, the allocation feature vector is obtained based on the first fused feature vector, the second fused feature vector and the third fused feature, including: Obtain a first time dimension feature vector based on the first fused feature vector through a bidirectional LSTM; Obtain a second time dimension feature vector based on the second fused feature vector through a bidirectional LSTM; Obtain a third time dimension feature vector based on the third fused feature vector through a bidirectional LSTM; Obtaining, through the attention mechanism, a first weight matrix of the feature vector of the first time dimension, a second weight matrix of the feature vector of the second time dimension, and a third weight matrix of the feature vector of the third time dimension; The first time dimension feature, the second time dimension feature and the third time dimension feature are integrated based on the first weight matrix, the second weight matrix and the third weight matrix to obtain the attention feature; The attention features are extracted through the fully connected layer to obtain the assigned feature vector.

4. The method for distributing user traffic based on traffic pool according to claim 3, characterized in that: Through the attention mechanism, a first weight matrix of the feature vector of the first time dimension, a second weight matrix of the feature vector of the second time dimension, and a third weight matrix of the feature vector of the third time dimension are obtained, including: Through the attention mechanism, a first weight matrix of the first time dimension feature vector is obtained, including: for the kth first time dimension feature t1 of the first time dimension feature vector k And the second time dimension feature vector t2, calculate the kth first time dimension feature t1 k The first similarity score e1 with the second time dimension feature vector t2 k =f(t1 k , t2), where f(t1 k , t2) represents the change of t1 k Perform point convolution operation with t2; The first attention weight is obtained based on the first similarity score, specifically: Among them, α1 k represents the attention weight of the kth first time dimension feature, e1 s Represents the sth first time dimension feature t1 s The first similarity score with the second time dimension feature vector t2; e1 k Represents the kth first time dimension feature t1 k The first similarity score with the second time dimension feature vector t2; S represents the number of first time dimension features; Based on the first attention weight, the first weight matrix is ​​obtained: W1 = [α11, α12, ..., α1 S ] Among them, W1 is the first weight matrix; α11 represents the first attention weight of the first first time dimension feature, α12 represents the first attention weight of the second first time dimension feature, α1 S Represents the first attention weight of the Sth first time dimension feature; Through the attention mechanism, a second weight matrix of the second time dimension feature vector is obtained, including: for the nth second time dimension feature t2 of the second time dimension feature vector n And the third time dimension feature vector t3, calculate the nth second time dimension feature t2 n The second similarity score e2 with the third time dimension feature vector t3 n =f(t2 n , t3), where f(t2 n , t3) represents the change of t2 n Perform point convolution operation with t3; The second attention weight is obtained based on the second similarity score, specifically: Among them, α2 n represents the second attention weight of the nth second time dimension feature, e2 u Represents the u-th second time dimension feature t2 u The second similarity score with the third time dimension feature vector t3; e2 n Represents the nth second time dimension feature t2 n The second similarity score with the third time dimension feature vector t3; U represents the number of second time dimension features; based on the second attention weight, the second weight matrix is ​​obtained: W2 = [α21, α22, ..., α2 U ] W2 represents the second weight matrix, α21, α22, α2 U They respectively represent the second attention weight of the first second time dimension feature, the second attention weight of the second second time dimension feature, and the second attention weight of the Uth second time dimension feature; Through the attention mechanism, a third weight matrix of the feature vector of the third time dimension is obtained, including: for the mth third time dimension feature t3 of the feature vector of the third time dimension m And the first time dimension feature vector t1, calculate the mth third time dimension feature t3 m The third similarity score e3 with the first time dimension feature vector t1 m =f(t3 m , t1), where f(t3 m , t1) represents the difference between t3 m Perform point convolution operation with t1; The second attention weight is obtained based on the second similarity score, specifically: Among them, α3 m represents the third attention weight of the nth third time dimension feature, e3 m The mth third time dimension feature t3 m The third similarity score with the first time dimension feature vector t1; e3 y Represents the yth third time dimension feature t3 y The third similarity score with the first time dimension feature vector t1; Y represents the number of third time dimension features; The third weight matrix is ​​obtained based on the third attention weight: W3 = [α31, α32, ..., α3 Y ] W2 represents the second weight matrix, α3, α32, α3 Y They respectively represent the third attention weight of the first third time dimension feature, the third attention weight of the second third time dimension feature, and the third attention weight of the Yth third time dimension feature.

5. The method for distributing user traffic based on traffic pool according to claim 2, characterized in that: The first time dimension feature, the second time dimension feature and the third time dimension feature are fused based on the first weight matrix, the second weight matrix and the third weight matrix to obtain the attention feature, including: Introduce the attention mechanism and obtain the attention features: F=σ(W1·avgpool(X1))·X1+σ(W2·avgpool(X2))·X2+σ(W3·avgpool(X3))·X3 Among them, F represents the attention feature, avgpool() represents the average pooling operation on the data in (), σ() is the activation function, which means that the data in () is activated by the activation function, · represents the dot product operation, X1, X2 and X3 represent the first time dimension feature, the second time dimension feature and the third time dimension feature respectively; W1, W2, W3 represent the first weight matrix, the second weight matrix and the third weight matrix respectively.

6. The method for distributing user traffic based on traffic pool according to claim 2, characterized in that: Through the attention mechanism, global features and local features are fused to obtain fused features, including: Specifically, the global features and local features are fused through the attention mechanism to obtain the fused features as follows: for the i-th local feature l i And the global feature g, calculate the i-th local feature l i Similarity score e with global feature g i =f(l i , g), where f(l i , g) indicates l i Perform point convolution operation with g; The attention weight is obtained based on the similarity score, specifically: Among them, α i represents the attention weight of the i-th local feature, e j represents the jth local feature l j Similarity score with the global feature g; e i represents the i-th local feature l i The similarity score with the global feature g; J represents the number of local features; the global features and local features are weighted and summed according to the calculated attention weights to obtain the fused features: Among them, h represents the fusion feature.

7. The method for distributing user traffic based on traffic pool according to claim 6, characterized in that: Replacing the global features and the local features in the operation of fusing the global features and the local features in claim 6 with the first local feature vector and the first global feature vector, respectively, to obtain a first fused feature vector; Replacing the global features and the local features in the operation of fusing the global features and the local features in claim 6 with the second local feature vector and the second global feature vector, respectively, to obtain a second fused feature vector; The global features and local features in the operation of fusing the global features and the local features in claim 6 are replaced with a third local feature vector and a third global feature vector respectively to obtain a third fused feature vector.

8. The method for distributing user traffic based on traffic pool according to claim 1, characterized in that: The method further comprises: Establishment of a traffic pool: determine a central controller and deploy multiple distributed nodes in the network; the central controller is used to centrally manage and schedule resources in the traffic pool; the distributed nodes are responsible for collecting traffic data and executing the instructions of the central controller; establish a data storage system, which is used to store the configuration information, traffic data and performance indicators of the traffic pool.

9. The method for distributing user traffic based on traffic pool according to claim 1, characterized in that: The method further includes: obtaining a flow distribution strategy corresponding to the distribution feature from a database.

10. A user traffic distribution system based on traffic pool, characterized in that: The system comprises: The acquisition module is used to obtain the user's real-time traffic usage data, historical traffic usage data and current traffic pool status data; the real-time traffic usage data represents the user's real-time traffic usage; the historical traffic usage data represents the user's traffic usage in the past time period; the current traffic pool status data represents the current resource status in the traffic pool; The prediction module is used to obtain the allocation characteristics through the traffic allocation model based on the real-time traffic usage data, the historical traffic usage data and the current traffic pool status data; the allocation characteristics represent the traffic allocation strategy, and the traffic allocation strategy includes the traffic allocation node and the traffic allocation amount; Among them, the traffic distribution model includes the first convolutional layer, the second convolutional layer, the third convolutional layer, the pooling layer, the attention layer and the fully connected layer; the attention layer includes the attention mechanism module and the bidirectional LSTM; the bidirectional LSTM includes the forward LSTM and the backward LSTM; the input of the first convolutional layer is the real-time traffic usage data, the input of the second convolutional layer is the historical traffic usage data, and the input of the third convolutional layer is the current traffic pool status data; the pooling layer mines the features of the real-time traffic usage data, the historical traffic usage data and the current traffic pool status data from both global and local dimensions, and combines them with the attention layer: the multi-dimensional correlation between the real-time traffic usage data, the historical traffic usage data and the current traffic pool status data is mined through the attention mechanism module and the bidirectional LSTM fusion, and the allocation feature vector representing the traffic distribution strategy is output through the fully connected layer.

Images