Training method and device of traffic prediction model, and storage medium

Abstract

The invention relates to a traffic prediction model training method and device, and a storage medium. The method comprises the following steps: acquiring topological relation data corresponding to a plurality of sample nodes in a target network, and acquiring traffic historical data of each sample node; for each sample node, determining traffic time data of the sample node according to the traffic historical data of the sample node; and training a preset traffic prediction model by using the topological relation data and the traffic historical data and the traffic time data of the plurality of sample nodes so as to execute a traffic prediction task by using the trained traffic prediction model. According to the invention, the flow prediction model carries out modeling based on the topological relation, can sense the periodic change of flow processing, the trend change of flow processing and the periodic change of sampling time of each sample node, and can sense the topological relevance and the flow circulation state between the sample nodes. The reference data of the traffic prediction model is more comprehensive, and the accuracy of the traffic prediction model is improved.

Description

Technical Field

[0001] The present application relates to the technical field of traffic prediction, and in particular to a training method, device and storage medium for a traffic prediction model. Background Art

[0002] With the improvement of communication technology and the rapid development of network technology, users are using more and more network services. In order to cope with the increase in network load, load balancing is imperative. Furthermore, in different scenarios, the uncertainty of user behavior will lead to the emergence of super busy areas and super idle areas in the network. In super busy areas, the congestion rate is high and the service quality is poor. In super idle areas, the resource utilization rate is low and there is waste. Load balancing can achieve a balance between service quality and resource utilization, so that the load sharing in each area is balanced.

[0003] The traditional load balancing method is to balance the load according to the busyness of the current network traffic, but it only considers the current traffic situation and does not consider the development and change of traffic, which makes the traditional load balancing method lack of foresight. In order to deal with this problem, a traffic prediction method is currently designed. The core of this method is to use historical traffic to predict future traffic and perform load balancing tasks based on the traffic prediction results, so that the load balancing results are predictable.

[0004] However, the current traffic prediction method is based on a single network element device and the accuracy of the traffic prediction results is low. Research has found that this is because the current traffic prediction process only considers the traffic data of the network element device itself, and does not consider the factors that cause the fluctuation of the network element device traffic. Therefore, the data used in the traffic prediction process is too one-sided, which makes the traffic prediction results limited and leads to low accuracy of the traffic prediction results. Summary of the invention

[0005] The present application provides a training method, device and storage medium for a traffic prediction model to solve the problem of low accuracy of traffic prediction results using current traffic prediction methods.

[0006] In view of the above technical problems, the technical solution of this application is solved by the following embodiments:

[0007] An embodiment of the present application provides a method for training a traffic prediction model, comprising: in a target network, obtaining topological relationship data corresponding to multiple sample nodes and obtaining traffic history data of each of the sample nodes; for each of the sample nodes, determining the traffic time data of the sample node based on the traffic history data of the sample node; using the topological relationship data and the traffic history data and traffic time data of the multiple sample nodes to train a preset traffic prediction model, so as to perform traffic prediction tasks using the trained traffic prediction model.

[0008] Wherein, the type of each of the sample nodes is a network element port or a network tunnel; before obtaining the traffic history data of each of the sample nodes, it also includes: for each of the sample nodes, respectively performing the following steps: when the type of the sample node is a network element port, collecting the traffic statistics data reported by the network element to which the sample node belongs; when the type of the sample node is a network tunnel, collecting the traffic statistics data reported by the starting network element of the sample node; from the collected traffic statistics data, extracting the traffic processing volume data of the sample node at different sampling times as the traffic history data of the sample node; wherein the traffic history data of the sample node includes multiple data entries; each of the data entries corresponds to the traffic processing volume data of the sample node at a sampling time.

[0009] Among them, determining the traffic time data of the sample node based on the traffic history data of the sample node includes: for each data entry in the traffic history data of the sample node, according to a plurality of pre-set time granularities, performing time granularity conversion processing on the sampling time corresponding to the data entry, determining the sampling time data corresponding to the data entry at each of the time granularities, and forming the traffic time data of the sample node.

[0010] Wherein, the traffic history data of the sample node includes the traffic history data within a first preset time period and the traffic history data within a second preset time period; wherein, in terms of time, the first preset time period precedes the second preset time period; wherein the use of the topological relationship data and the traffic history data and traffic time data of multiple sample nodes to train a preset traffic prediction model includes: using the topological relationship data, the traffic history data of each of the multiple sample nodes within the first preset time period and the traffic time data of each of the sample nodes to train the traffic prediction model; wherein the traffic history data of each of the multiple sample nodes within the second preset time period is used as training label data.

[0011] Wherein, the traffic prediction model is used to perform the following steps: determine the topological relationship network of the plurality of sample nodes according to the topological relationship data corresponding to the plurality of sample nodes; aggregate the edge feature data corresponding to each sample node according to the topological relationship network and the node feature data corresponding to each sample node; wherein, for each sample node, the traffic history data of the sample node in the first preset time period and the traffic time data of the sample node are used as the node feature data of the sample node in the topological relationship network; for each sample node, generate the global feature data corresponding to the sample node according to the node feature data corresponding to the sample node and the edge feature data corresponding to the sample node; wherein the dimension of the global feature data corresponding to the sample node is the same as the dimension of the node feature data corresponding to the sample node; predict the traffic prediction data of each sample node in the second preset time period according to the global feature data corresponding to each sample node, the traffic history data of each sample node in the first preset time period and the traffic time data of each sample node in the second preset time period; wherein the traffic time data of each sample node includes the traffic time data in the first preset time period and the traffic time data in the second preset time period.

[0012] Among them, the edge feature data corresponding to each sample node is aggregated according to the topological relationship network and the node feature data corresponding to each sample node: including: for each sample node, according to the type of the sample node, determining whether the sample node is a sending node or a receiving node; when the sample node is a sending node, determining the receiving node connected to the sending node in the topological relationship network, aggregating the node feature data corresponding to the receiving node connected to the sending node, and obtaining the edge feature data corresponding to the sending node; when the sample node is a receiving node, determining the sending node connected to the receiving node in the topological relationship network, aggregating the node feature data corresponding to the sending node connected to the receiving node, and obtaining the edge feature data corresponding to the receiving node.

[0013] Among them, generating the global feature data corresponding to the sample node according to the node feature data corresponding to the sample node and the edge feature data corresponding to the sample node includes: splicing the node feature data corresponding to the sample node and the edge feature data corresponding to the sample node to obtain the global feature data corresponding to the sample node; using a deep learning component to adjust the dimension of the global feature data so that the dimension of the global feature data is the same as the dimension of the node feature data corresponding to the sample node.

[0014] Wherein, the use of the topological relationship data and the traffic history data and traffic time data of multiple sample nodes to train a preset traffic prediction model includes: dividing the multiple sample nodes into multiple batches according to the topological relationship data corresponding to the multiple sample nodes; wherein, in the topological relationship network corresponding to the topological relationship data, the subnetwork formed by the sample nodes in the same batch is a fully connected network; and applying the traffic history data and traffic time data of the sample nodes in each batch to a round of training of the traffic prediction model.

[0015] An embodiment of the present application also provides a training device for a traffic prediction model, comprising: at least one communication interface; at least one bus connected to the at least one communication interface; at least one processor connected to the at least one bus; and at least one memory connected to the at least one bus, wherein the processor is configured to: execute a training program for the traffic prediction model stored in the memory to implement a training method for a traffic prediction model described in any one of the above items.

[0016] An embodiment of the present application also provides a computer-readable storage medium, which stores computer-executable instructions, and the computer-executable instructions are executed to implement the training method of the traffic prediction model described in any one of the above items.

[0017] The above technical solution provided by the embodiment of the present application has the following advantages compared with the prior art: the method provided by the embodiment of the present application can obtain the topological relationship data corresponding to multiple sample nodes in the target network and obtain the traffic history data of each sample node; for each sample node, determine the traffic time data of the sample node according to the traffic history data of the sample node; use the topological relationship data and the traffic history data and traffic time data of multiple sample nodes to train the preset traffic prediction model, so as to perform the traffic prediction task using the trained traffic prediction model. In the embodiment of the present application, when making traffic predictions, the traffic prediction model refers to the topological relationship data in addition to the traffic history data and traffic time data. The traffic prediction model is modeled based on the topological relationship between each sample node, which can not only perceive the periodic changes in traffic processing, the trend changes in traffic processing and the periodic changes in sampling time of each sample node, but also perceive the topological correlation and traffic flow status between each sample node, so that the reference data of the traffic prediction model is more comprehensive and the accuracy of the traffic prediction model is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0019] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0020] One or more embodiments are exemplarily described by pictures in the corresponding drawings, and these exemplified descriptions do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings represent similar elements, and unless otherwise stated, the figures in the drawings do not constitute proportional limitations.

[0021] Figure 1 is a flow chart of a method for training a traffic prediction model according to an embodiment of the present application;

[0022] Figure 2 is a schematic diagram of a target network according to an embodiment of the present application;

[0023] Figure 3 A multi-dimensional traffic entity relationship diagram according to an embodiment of the present application;

[0024] Figure 4 is a schematic diagram of a topological relationship network according to an embodiment of the present application;

[0025] Figure 5 A structural diagram of a training device for a traffic prediction model according to an embodiment of the present application;

[0026] Figure 6 is a sample schematic diagram according to an embodiment of the present application;

[0027] Figure 7 A flow chart of execution steps of a traffic prediction model according to an embodiment of the present application;

[0028] Figure 8 is a schematic diagram of the operation of a primary update module according to an embodiment of the present application;

[0029] Fig. 9 A flow chart of the steps of a traffic prediction task according to an embodiment of the present application;

[0030] Fig.10 It is a structural diagram of a training device for a traffic prediction model according to an embodiment of the present application. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0032] The disclosure below provides many different embodiments or examples to implement the different structures of the present application. In order to simplify the disclosure of the present application, the parts and settings of specific examples are described below. Of course, they are only examples, and the purpose is not to limit the present application. In addition, the application can repeat reference numbers and / or letters in different examples. This repetition is for the purpose of simplification and clarity, and does not itself indicate the relationship between the various embodiments and / or settings discussed.

[0033] The embodiment of the present application provides a method for training a traffic prediction model, wherein the traffic prediction model is used to predict traffic data of a target network within a period of time in the future.

[0034] In practical applications, the traffic data of the target network in the future can be used directly as reference data, and the reference data can also be used to perform predictive load balancing and network expansion on the target network. Furthermore, the prediction results of the traffic prediction model can reflect the development and changes of the traffic in the target network. In data analysis, the prediction results have a high reference value. In load balancing and network expansion tasks, the prediction results are forward-looking, which can make the execution of load balancing and network expansion more reasonable and accurate.

[0035] However, both the reference value of the prediction results and the effects of load balancing and network expansion are positively correlated with the prediction accuracy of the traffic prediction model. Based on this, the present application intends to provide a traffic prediction model with higher prediction accuracy through model training.

[0036] Figure 1 The present invention is a flowchart of a method for training a traffic prediction model according to an embodiment of the present application.

[0037] Step S110: in the target network, obtaining topological relationship data corresponding to a plurality of sample nodes and obtaining flow history data of each of the sample nodes.

[0038] The target network refers to the network for which traffic prediction is required.

[0039] The sample node refers to a network node in the target network, which can be a network element device, a network element port, a network tunnel and / or a data object.

[0040] Topological relationship data is used to describe each sample node in the target network and the relationship between each sample node.

[0041] Traffic history data refers to the traffic volume data of a sample node within a historical period of time. Furthermore, the traffic history data of a sample node includes multiple data entries; each data entry corresponds to the traffic volume data of the sample node at a sampling time. The traffic volume can be network traffic, pedestrian traffic, vehicle traffic or stock fluctuation volume.

[0042] In an embodiment of the present application, topological relationship data corresponding to multiple sample nodes and flow history data of each sample node can be obtained from a network management room device or a designated network device. The network management room device and the designated network device are used to store the topological relationship data of the target network and the flow history data of each node in the target network.

[0043] Step S120: for each of the sample nodes, determine the flow time data of the sample node according to the flow history data of the sample node.

[0044] Traffic time data is used to record the sampling time data of the sample node within a period of historical time. Furthermore, the traffic history data of the sample node within a period of historical time includes multiple data entries, and each data entry corresponds to a sampling time. The traffic time data is the statistics of the sampling time corresponding to each data entry. Then, it can be understood that the traffic time data is the sampling time data of each traffic processing volume data (each data entry) of the sample node within the period of historical time.

[0045] Step S130, using the topological relationship data and the traffic history data and traffic time data of the plurality of sample nodes, a preset traffic prediction model is trained so as to perform a traffic prediction task using the trained traffic prediction model.

[0046] The traffic prediction model is used to predict the traffic processing volume data of multiple sample nodes in the future. Furthermore, the topological relationship data and the traffic history data and traffic time data of the multiple sample nodes are input into the traffic prediction model, and the traffic prediction model can output the traffic prediction data of the multiple sample nodes in the future. The traffic prediction data of each sample node includes multiple prediction items, and each prediction item corresponds to the traffic processing volume data of a prediction time.

[0047] The specific model training process will be described in detail later, so it will not be repeated here.

[0048] In an embodiment of the present application, topological relationship data corresponding to multiple sample nodes are obtained in the target network and the traffic history data of each of the sample nodes is obtained; for each of the sample nodes, the traffic time data of the sample node is determined based on the traffic history data of the sample node; and a preset traffic prediction model is trained using the topological relationship data and the traffic history data and traffic time data of the multiple sample nodes, so that the trained traffic prediction model can be used to perform traffic prediction tasks.

[0049] In the embodiment of the present application, the traffic prediction model can not only perform traffic prediction for a single node in the target network, but also perform traffic prediction for multiple nodes in the target network together. The traffic prediction model has a wide range of application and is global in nature.

[0050] In an embodiment of the present application, the topological relationship data can describe the topological association relationship between each sample node in the target network, the traffic history data can reflect the periodic changes in traffic processing and the trend changes in traffic processing of the sample node within a period of historical time, and the traffic time data can reflect the periodic changes in the sampling time of the sample node within the period of historical time. In this way, when the traffic prediction model makes traffic predictions, in addition to referring to the traffic history data and traffic time data, it also refers to the topological relationship data. The topological relationship data can be used to analyze the topological association and traffic flow status between each sample node, which is a key factor affecting the traffic fluctuations of each sample node. When training the traffic prediction model in an embodiment of the present application, the traffic prediction model is modeled based on the topological relationship between each sample node, which can not only perceive the periodic changes in traffic processing, the trend changes in traffic processing and the periodic changes in sampling time of each sample node, but also perceive the topological association and traffic flow status between each sample node, so that the reference data of the traffic prediction model is more comprehensive, and the accuracy and generalization of the traffic prediction model are enhanced.

[0051] In order to make the embodiments of the present application clearer, each step of the embodiments of the present application will be described in more detail below.

[0052] The following describes the steps of obtaining topological relationship data corresponding to multiple sample nodes.

[0053] The target network includes multiple nodes. There is a connection relationship or no connection relationship between the multiple nodes. For example, in a communication network, there is a connection relationship between network elements (nodes). Another example is that in a stock network, there is no connection relationship between stocks (nodes).

[0054] In the target network, all or part of the nodes are selected as sample nodes. The target network can be a heterogeneous network, and nodes of different types or the same type are selected as sample nodes.

[0055] The types of sample nodes include, but are not limited to: network element devices, network element ports, network tunnels, and data objects. The network element devices may be switches, terminals, etc. required to form an IPV4 or IPV6 network. Network tunnels are used to transmit traffic between network element devices. Data objects are, for example, stocks.

[0056] In the target network, nodes can be selected as needed. For example, nodes of the network element port and network tunnel type are selected as sample nodes. Another example is: nodes of the network element device type are selected as sample nodes. Another example is: nodes of the data object type are selected as sample nodes.

[0057] The following describes the process of generating topology relationship data with reference to the case where the sample node type is a network element port or a network tunnel:

[0058] like Figure 2 As shown, it is a schematic diagram of a target network according to an embodiment of the present application. The target network includes 8 network element devices, each of which is provided with a network element port. After configuration, 3 tunnels (black lines) are established on the 8 network element devices, and each tunnel is connected to multiple ports in turn. In the embodiment of the present application, tunnels and ports are selected as sample nodes, and then a multi-dimensional flow entity relationship diagram can be abstracted based on the relationship between tunnels and ports.

[0059] like Figure 3 As shown, it is a multi-dimensional flow entity relationship diagram according to an embodiment of the present application. In the multi-dimensional flow entity relationship diagram, only the relationship between the tunnels and ports in the target network is reflected, and the network element devices in the target network are ignored. Among them, tunnel 1 passes through ports 0, 1, 2, 3, 4 and 10 in sequence, tunnel 2 passes through ports 0, 5, 6, 7, 4 and 11 in sequence, and tunnel 3 passes through ports 8, 3, 4 and 9 in sequence.

[0060] like Figure 4 As shown, it is a schematic diagram of a topological relationship network according to an embodiment of the present application. According to the type of sample nodes, the sample nodes are clustered into different groups. In an embodiment of the present application, the tunnels and ports in the target network are used as sample nodes and separated on both sides according to the type, and according to the multidimensional flow entity relationship diagram, a unidirectional edge from the port to the tunnel is abstracted (the edge points from the port to its associated tunnel), thereby forming a topological relationship network of the tunnel and the port. Because the edges in the topological relationship network are only used to describe the transmission of traffic, the edges in the topological relationship network are unidirectional, and the direction of the edges can be set according to the type of sample nodes, for example: the ports are all sending nodes and the tunnels are all receiving nodes, or the ports are all receiving nodes and the tunnels are all sending nodes, and the direction of the edges can be clarified according to the sending and receiving relationship of the nodes.

[0061] After obtaining the topological relationship network, topological relationship data may be generated according to the topological relationship network. In the case where the type of the sample node is a network element port or a network tunnel, the topological relationship data includes but is not limited to: port asset data, tunnel asset data and asset relationship data.

[0062] The port asset data includes, but is not limited to: data of network element devices in the target network, and data of network element ports included in each network element device. The data of network element devices at least includes the unique code of the network device. The data of network element ports at least includes the unique code of the network element ports.

[0063] The tunnel asset data includes, but is not limited to, data of the network tunnel included in the target network. The data of the network tunnel at least includes a unique code of the network tunnel.

[0064] Asset relationship data, including but not limited to: network element port data of each tunnel and each hop in the target network.

[0065] The topology relationship data may be generated at regular intervals to cope with changes in network element devices, network element ports and tunnels in the target network. Further, after the topology relationship data is generated, the topology relationship data may be stored and retrieved when needed.

[0066] In the embodiment of the present application, by selecting the sample node type, the topological relationship network can be focused on the tunnel and port level, while the nodes of the traditional network topology map are focused on the network element devices. In this way, the node granularity of the topological relationship network is smaller than the node granularity in the network topology map. Using topological relationship data with smaller node granularity to train the traffic prediction model will make the traffic prediction results more targeted and more accurate.

[0067] Furthermore, in the target network, a single network element device may include multiple network element ports, which may belong to different network tunnels. A single network element device needs to carry the traffic of multiple network tunnels. Of course, the same network element port can also carry the traffic of multiple network tunnels. Moreover, the traffic passes through the tunnel from the port of network element A (starting point) to the port of network element Z (terminal point). In addition to the ports of network element A and network element Z, the tunnel also passes through multiple intermediate ports that are not perceived by the business. In this way, the path of the network tunnel and the traffic size of the network tunnel will affect the traffic of the network element device. In scenarios with high requirements for load balancing, if load balancing is only performed on network element devices, the traffic pressure of network element ports or network tunnels cannot be effectively alleviated. Therefore, when the traffic prediction model makes traffic predictions for network tunnels and network element ports, it can more accurately locate the specific location with high traffic pressure.

[0068] The following describes the steps for obtaining the flow history data of each sample node. In the embodiment of the present application, the topological relationship data corresponding to the multiple sample nodes and the flow history data of each sample node are obtained. These two steps can be performed simultaneously or sequentially. The order of the two steps is not limited.

[0069] Before obtaining the traffic history data of each sample node, the following steps are performed for each sample node: when the type of the sample node is a network element device, directly collect the traffic statistics data reported by the sample node; when the type of the sample node is a network element port, collect the traffic statistics data reported by the network element to which the sample node belongs; when the type of the sample node is a network tunnel, collect the traffic statistics data reported by the starting network element of the sample node; when the sample node is a data object, collect the traffic statistics data reported by the data device corresponding to the data object; from the collected traffic statistics data, extract the traffic processing volume data of the sample node at different sampling times as the traffic history data of the sample node.

[0070] The traffic statistics data of the sample node include one or more statistical entries; each statistical entry corresponds to the network traffic data of the sample node in a statistical period. The duration of the statistical period corresponding to different statistical entries is the same. The duration of the statistical period can be set according to demand. For example: the duration of the statistical period is 1s, that is, the network traffic data is counted every 1s. The network traffic data includes multiple fields related to traffic processing volume. The multiple fields are used to store traffic processing volume data. The traffic processing volume data includes but is not limited to: the average traffic sending value (avgSendSpeed) and the average traffic receiving value (avgRecvSpeed). Extract the traffic processing volume data in the multiple fields in a network traffic data. If a field is missing data or the data is abnormal, the preset quantity value can be used to replace the data in the field.

[0071] The traffic history data of the sample node includes multiple data entries; each data entry corresponds to the traffic processing volume data of the sample node at a sampling time. Furthermore, for a sample node, each data entry in the traffic history data of the sample node corresponds to a sampling period. The duration of the sampling periods corresponding to different data entries is equal. The duration of the statistical period can be set according to demand. The sampling time corresponding to each data entry is the start time, middle time, end time of the sampling period, or the time determined according to preset rules.

[0072] Furthermore, the duration of the sampling period is greater than or equal to the duration of the statistical period. When the duration of the sampling period is equal to the duration of the statistical period, the traffic statistical data and the traffic historical data may be equivalent data. When the duration of the sampling period is greater than the duration of the statistical period, it is necessary to extract the traffic processing volume data from multiple statistical entries, and perform aggregation calculation on the traffic processing volume data in the multiple statistical entries, so as to aggregate the traffic processing volume data of multiple statistical periods into the traffic processing volume data of one sampling period. For example: aggregate 15 (statistical entries) of 1 minute of traffic processing volume data into 1 (data entry) of 15 minutes of traffic processing volume data.

[0073] In the embodiment of the present application, in the flow history data of multiple sample nodes, the sampling time of the data entries at corresponding positions is the same. For example, the sampling time of the first item is 0:00, and the sampling time of the second item is 0:01.

[0074] Furthermore, the statistical time periods of multiple sample nodes can be set to be the same, and the two end values ​​of the sampling time period can be determined based on the end values ​​of the statistical time period. Since the two end values ​​of the statistical time period corresponding to each statistical item in the traffic statistical data are known, the minimum end value of the statistical time period included in the sampling time period can be used as the start time of the sampling time period; the maximum end value of the statistical time period included in the sampling time period can be used as the end time of the sampling time period. In this way, the collection time of the data items can be kept consistent.

[0075] Furthermore, the sampling time can be set as the time for storing data entries. By controlling the storage time of the data entries of each sample node, the sampling time of the data entries at the corresponding positions of multiple sample nodes can be guaranteed to be the same. For example, starting from the same time point, the data entries of each sample are stored every 1 minute.

[0076] In the embodiment of the present application, the execution subject of the training method of the traffic prediction model is a training device of the traffic prediction model. The training device can be set in a network management room, or can be specifically set according to needs.

[0077] Figure 5 It is a structural diagram of a training device for a traffic prediction model according to an embodiment of the present application.

[0078] The training device of the traffic prediction model includes, but is not limited to: a southbound interface, a data sorting module, a data storage module and a traffic prediction model training module.

[0079] Traffic statistics of each sample node can be collected on the southbound interface. The southbound interface is a module that interacts with network element devices, including the OpenFlow protocol parsing interface.

[0080] The data sorting module is used to sort the collected traffic statistics data, extract data from the traffic statistics data, and form traffic history data.

[0081] The data storage module is used to store the historical traffic data of each sample node.

[0082] The data storage module can also be used to store the flow time data of each sample node, pre-store the topological relationship data corresponding to multiple sample nodes, or pre-store the topological relationship data corresponding to all nodes in the target network. When training the flow prediction model, the topological relationship data corresponding to multiple sample nodes can be extracted from the topological relationship data corresponding to all nodes.

[0083] After acquiring the flow history data of each sample node, the flow time data of the sample node can be determined.

[0084] The following further describes how to determine the flow time data of the sample node based on the flow history data of the sample node.

[0085] For each of the multiple sample nodes, the following steps may be performed:

[0086] For each data entry in the flow history data of the sample node, according to the preset multiple time granularities, the time granularity conversion process is performed on the sampling time corresponding to the data entry, and the sampling time data corresponding to the data entry under each of the time granularities is determined to form the flow time data of the sample node. In this way, the number of entries in the flow time data is the same as the number of data entries in the flow history data (the same number of rows).

[0087] The time granularity can be determined according to the requirements. The time granularity includes but is not limited to: sampling minutes (0-59 minutes), sampling hours (0-23 hours), sampling weeks (0-6) and holidays (0 / 1). Among them, the sampling weeks are 0-6, representing Monday to Sunday respectively. Holidays are 0 or 1, which are used to indicate whether the sampling time is a holiday.

[0088] For example: the sampling time is 9:50 on September 25, 2023, then the sampling points are 50, the sampling time is 9, the sampling week is 0, and the holiday is 0.

[0089] In order to facilitate setting of the sample set, when acquiring the flow history data of the sample node, the flow history data of two time parts can be acquired. In this way, the flow history data of the sample node includes: the flow history data within the first preset time period and the flow history data within the second preset time period. When setting the sample set, the flow history data of the sample node within the second preset time period is used as the label data, thereby avoiding the error caused by manual labeling.

[0090] Further, in terms of time, the first preset time period precedes the second preset time period. The first preset time period and the second preset time period may form a continuous time. For example, the first preset time period is the day before yesterday, and the second preset time period is yesterday. The length of the first preset time period is equal to the length of the second preset time period, and the number of data entries of the flow history data within the first preset time period is the same as the number of data entries of the flow history data within the second preset time period.

[0091] Furthermore, when acquiring the traffic history data, the traffic history data of the two time parts may be acquired separately, or the traffic history data may be acquired as a whole and then split into the traffic history data of the two time parts.

[0092] Specifically, the flow history data of the sample node within the first preset time period and the flow history data within the second preset time period are obtained; the flow time data of the sample node within the first preset time period is generated according to the flow history data of the sample node within the first preset time period; the flow time data of the sample node within the second preset time period is generated according to the flow history data of the sample node within the second preset time period; the flow history data of the sample node within the first preset time period, the flow time data of the sample node within the first preset time period and the flow time data within the second preset time period are used as a sample for training a flow prediction model; the flow history data of the sample node within the second preset time period is used as label data of the sample, so as to determine the loss of the flow prediction model.

[0093] like Figure 6 As shown, it is a sample schematic diagram according to an embodiment of the present application. In this example, the traffic statistics data is equivalent to the traffic history data. The traffic history data of the sample node is collected t times a day, and a total of x+y days are collected. x days are taken as the first preset time, and y days are taken as the second preset time period. Then, the traffic history data of the sample node within x days includes t×x rows of data entries; the traffic time data of the sample node within x is also t×x rows. The traffic history data of the sample node within y days includes t×y rows of data entries (not shown in the figure); the traffic time data of the sample node within y days is also t×y rows. The traffic history data of the sample node within x days (t×x rows) and the traffic data time of the sample node within x+y days (t×x rows + t×y rows) are taken as a sample, and the traffic history data of the sample node within y days (t×y rows) are taken as the label data of the sample.

[0094] After the samples in the sample set are prepared, the traffic prediction model can be trained.

[0095] The following further describes the training process of the traffic prediction model.

[0096] In an embodiment of the present application, the traffic prediction model may be a pre-set basic model, a pre-trained model, or a migration model migrated from other scenarios or fields. Among them, the basic model needs to be trained to obtain traffic prediction capabilities. The pre-trained model has preliminary traffic prediction capabilities, but the accuracy is low, and training is required to improve the accuracy of traffic prediction. The migration model is a prediction model used in other scenarios or fields. The prediction model may be a model that has similarities with the traffic prediction model. Training the migration model in the traffic prediction scenario can enable the migration model to have traffic prediction capabilities.

[0097] Specifically, the training process of the traffic prediction model can be divided into a model training phase and a model inference phase. The model training phase is used to adjust the parameters in the traffic prediction model. The model inference phase is used to evaluate the prediction performance of the traffic prediction model. The model training phase and the model inference phase can use the same sample set or different sample sets.

[0098] For example, when the same sample set is used, the sample set includes: topological relationship data corresponding to multiple sample nodes, flow history data and flow time data of each sample node.

[0099] For example, when using different sample sets, the sample set in the model training phase includes: the topological relationship data corresponding to some sample nodes in the target network, the traffic history data and traffic time data of these sample nodes. The sample set in the model reasoning phase includes: the topological relationship data corresponding to another part of the sample nodes in the target network, the traffic history data and traffic time data of another part of the sample nodes.

[0100] The following is a description of the model training phase:

[0101] The traffic prediction model is trained using topological relationship data, traffic history data of each sample node in a first preset time period, and traffic time data of each sample node; wherein the traffic history data of each sample node in a second preset time period is used as training label data. The topological relationship data may be the topological relationship data corresponding to the first preset time period.

[0102] When the amount of data in the sample set is large, the multiple sample nodes can be divided into multiple batches according to the topological relationship data corresponding to the multiple sample nodes; wherein, in the topological relationship network corresponding to the topological relationship data, the subnetwork formed by the sample nodes in the same batch is a fully connected network. A fully connected network is one in which both the network tunnel and the network element port are abstracted as nodes in the topological relationship graph. Starting from any node in the topological relationship graph, all other nodes in the topological relationship graph can be reached. This batch division method will not destroy the topological relationship between nodes, and can enable the traffic prediction model to better perceive the topological relationship between nodes. Furthermore, the BFS (Breadth First Search) algorithm, DFS (Depth First Search) or Dijkstra (Dijkstra algorithm) can be used to divide multiple sample nodes into multiple fully connected networks.

[0103] The flow history data and flow time data of the sample nodes in each batch are applied to a round of training of the flow prediction model, and a batch of data is input in each round of training. Further, a preset round of training is performed for the flow prediction model. In each round of training, the topological relationship data corresponding to a batch of sample nodes and the flow history data of each sample node in the current batch within the first preset time period and the flow time data of each sample node are sequentially obtained; the topological relationship data, the flow history data of each sample node in the current batch within the first preset time period and the flow time data of each sample node are input into the flow prediction model; the flow prediction data of multiple sample nodes output by the flow prediction model within the second preset time period are obtained; the loss value of the flow history data and flow prediction data of each sample node within the second preset time period is determined by using a preset loss function; when the loss value is greater than a preset loss threshold, the parameters in the flow prediction model are adjusted, and the next round of training is entered; when the loss value is less than or equal to the loss threshold, the next round of training is directly entered.

[0104] In the embodiment of the present application, the loss function may be a MAE (Mean Absolute Error) loss function.

[0105] In an embodiment of the present application, the traffic prediction model may be a deep learning model, which may be a neural network with an encoder-process-decode structure.

[0106] Furthermore, the neural network includes an encode-process network and a decode network. The encoding network includes at least one one-time update module. The output of each one-time update module is connected to the decoding network. The one-time update module can sense the flow of sample nodes in the topological relationship network based on the topological relationship data.

[0107] The neural network includes multiple network layers, so as to realize automatic feature extraction and feature learning by using the neural network with multiple network layers. The number of network layers in the neural network can be determined according to the demand, and after training, the multiple layers have the ability of traffic prediction.

[0108] The neural network can be CNN (convolutional neural networks), RNN (Recurrent Neural Network) or transformer (deep neural network based on attention mechanism).

[0109] An update module can be a GCN (Convolutional Neural Network) layer or a GAT (Graph Attention Network) layer.

[0110] The following describes the execution steps of the traffic prediction model in a round of training. In the embodiment of the present application, the traffic prediction model can be trained using SGD (Stochastic Gradient Descent) to correct the parameters in the traffic prediction model.

[0111] Figure 7 The figure is a flow chart of the execution steps of the traffic prediction model according to one embodiment of the present application. Figure 8 Schematic diagram of the operation of an update module according to an embodiment of the present application.

[0112] Step S710: A primary updating module determines a topological relationship network of a plurality of sample nodes according to the topological relationship data corresponding to the plurality of sample nodes.

[0113] The multiple sample nodes are sample nodes in a batch. The topological relationship data are topological relationship data of the multiple sample nodes in the batch.

[0114] The flow history data of the sample node within the first preset time period and the flow time data of the sample node may be spliced ​​into multi-dimensional data, and the multi-dimensional data is used as the node characteristic data of the sample node.

[0115] like Figure 8As shown, each circle represents a sample node. In each slice, the upper 3 circles represent network tunnels, and the lower 8 circles represent network element ports. Each slice corresponds to a sampling time, so the data of circles at the same position in multiple slices constitute the node feature data of a sample node. Figure 8 Only three slices are schematically shown in FIG. In each slice, the association relationship of each sample node (circle) is determined according to the topological relationship data of multiple sample nodes, thereby forming a topological relationship network.

[0116] Step S720, an update module aggregates edge feature data corresponding to each sample node based on the topological relationship network and the node feature data corresponding to each sample node; wherein, for each sample node, the flow history data of the sample node within the first preset time period and the flow time data of the sample node are used as the node feature data of the sample node in the topological relationship network.

[0117] The following steps can be used to aggregate edge feature data:

[0118] Step S1: for each sample node, determine whether the sample node is a sending node or a receiving node according to the type of the sample node.

[0119] In the embodiment of the present application, the network tunnel is a receiving node and the network element port is a sending node.

[0120] When generating the topological relationship data, the direction of the edge in the topological relationship network has been set according to the type of the sample node. Therefore, the sample node can also be determined as a sending node or a receiving node according to the topological relationship data.

[0121] Step S2, when the sample node is a sending node, in the topological relationship network, determine the receiving node connected to the sending node, aggregate the node feature data corresponding to the receiving node connected to the sending node, and obtain the edge feature data corresponding to the sending node.

[0122] In a topological relationship network, each receiving node broadcasts its own node feature data to the associated edges, and the sending node aggregates the node feature data on the edges associated with it, so that the sending node can perceive the data of the associated receiving nodes in the topological relationship network.

[0123] Aggregation can be maximum, average, or sum.

[0124] Step S3, when the sample node is a receiving node, in the topological relationship network, determine the sending node connected to the receiving node, aggregate the node feature data corresponding to the sending node connected to the receiving node, and obtain the edge feature data corresponding to the receiving node.

[0125] In a topological relationship network, each sending node broadcasts its own node feature data to the associated edges, and the receiving node aggregates the node feature data on the edges associated with it, so that the receiving node can perceive the data of the associated sending nodes in the topological relationship network.

[0126] Step S730, the primary update module generates global feature data corresponding to each sample node according to the node feature data corresponding to the sample node and the edge feature data corresponding to the sample node; wherein the dimension of the global feature data corresponding to the sample node is the same as the dimension of the node feature data corresponding to the sample node.

[0127] For each sample node, the node feature data corresponding to the sample node and the edge feature data corresponding to the sample node are spliced ​​to obtain the global feature data corresponding to the sample node; the dimension of the global feature data is adjusted using the deep learning component so that the dimension of the global feature data is the same as the dimension of the node feature data corresponding to the sample node.

[0128] The deep learning component can be used to update the state of global feature data. For example, adjust the dimension of the global feature vector. The deep learning component can be LSTM (Long Short-Term Memory Network).

[0129] Step S740, the decoding network predicts the traffic prediction data of each sample node in the second preset time period based on the global feature data corresponding to each sample node, the traffic history data of each sample node in the first preset time period, and the traffic time data of each sample node in the second preset time period; wherein the traffic time data of each sample node includes the traffic time data in the first preset time period and the traffic time data in the second preset time period.

[0130] The decoding network learns the global feature data corresponding to each sample node, the traffic history data of each sample node in the first preset time period, and the traffic time data of each sample node in the second preset time period. The shape of the features is adjusted through convolution, flattening and fully connected layers, and finally the traffic prediction data of each sample node in the second preset time period can be output.

[0131] After the decoding network outputs the traffic prediction data of each sample node in the second prediction time period, based on the traffic prediction data of each sample node in the second prediction time period and the traffic history data of each sample node in the second prediction time period, a preset loss function is used to calculate the loss value between the traffic prediction data and the traffic history data.

[0132] When the loss value is greater than the preset loss threshold, the parameters in the traffic prediction model are adjusted, and then the next round of training is entered; otherwise, the next round of training is directly entered; until the preset round of training ends.

[0133] After the preset round-robin training is completed, the model training phase ends and the model inference phase can begin.

[0134] In the model inference stage, the sample set is stored in the data storage module. The sample set includes at least: topological relationship data corresponding to multiple sample nodes, traffic history data of each sample node in the first preset time period, and traffic history data in the second preset time period. Among them, the traffic history data of each sample node in the second preset time period is still used as a label. Of course, the sample set can also include: traffic history data of each sample node in the second preset time period, and traffic time data of each sample node in the first preset time period and in the second preset time period.

[0135] The following describes the steps of the model reasoning phase of the traffic prediction model.

[0136] From the sample set, obtain the topological relationship data corresponding to the multiple sample nodes, the flow history data of each sample node in the first preset time period, and the flow history data of each sample node in the second preset time period; for each sample node, determine the flow time data of the sample node in the first preset time period according to the flow history data of the sample node in the first preset time period; determine the flow time data of each sample node in the second preset time period according to the flow history data of the sample node in the second preset time period; input the topological relationship data corresponding to the multiple sample nodes, the flow history data of each sample node in the first preset time period, and the flow time data of each sample node in the first preset time period and the second preset time period into the flow prediction model, and obtain the flow prediction data of each sample node in the second preset time period output by the flow prediction model; determine the accuracy, precision and recall of the flow prediction model according to the flow history data of each sample node in the second preset time period and the flow prediction data of each sample node in the second preset time period; when the accuracy, precision and recall of the flow prediction model all meet the preset standard conditions, determine that the flow prediction model training is completed. Among them, the compliance conditions include: the accuracy is greater than the preset accuracy threshold, the precision is greater than the preset precision threshold, and the recall rate is greater than the preset recall rate threshold.

[0137] Furthermore, when the sample set includes the flow time data of each sample node within the first preset time period and the second preset time period, it can be directly acquired.

[0138] Furthermore, if the data volume of the sample set is large, the multiple sample nodes triggered in the target network may be divided into multiple batches, and the data in the sample set may be input into the traffic prediction model in batches.

[0139] After the model training phase and the model inference phase, the traffic prediction model training is completed, and the file corresponding to the traffic prediction model is stored in the data storage module. The traffic prediction model can be retrained at regular intervals. For example, a timer is set, and when the time set by the timer arrives, the training of the traffic prediction model is triggered.

[0140] After the traffic prediction model is trained, it is not only applicable to traffic prediction of a single node, but can be applied to the entire target network and is global.

[0141] After the traffic prediction model is trained, the trained traffic prediction model can be used to perform traffic prediction tasks.

[0142] Fig. 9 The figure is a flow chart of the steps of a traffic prediction task according to an embodiment of the present application.

[0143] Step S910: display target network data in a preset display interface.

[0144] The target network data includes but is not limited to: a plurality of target nodes included in the target network.

[0145] Step S920: Detect multiple triggered target nodes in the target network according to the target network data.

[0146] The triggered multiple target nodes can form a fully connected network.

[0147] Step S930, obtaining topological relationship data corresponding to the multiple target nodes and traffic history data of each sample node within a third preset time period.

[0148] The end time of the third preset time period is the current time. The start time of the third preset time period is determined according to the duration to be predicted.

[0149] The topological relationship data corresponding to the multiple target nodes and the flow history data of each target node within a third preset time period are obtained from the data storage module.

[0150] Step S940: for each target node, determine the flow time data of the target node at the third preset time according to the flow history data of the target node at the third preset time.

[0151] The steps for determining the flow time data based on the flow history data can refer to the above description and will not be repeated here.

[0152] Step S950: Determine the flow time data of each target node within a fourth preset time period based on the flow time data of the target node within the third preset time period.

[0153] The time difference between the time start point of the fourth preset time period and the time end point of the third preset time period is a preset time value, and the time length of the fourth preset time period is the same as the time length of the third preset time period.

[0154] The preset time value is the time difference between the end time of the first preset time period and the start time of the second preset time period.

[0155] The sampling time of the target node at different time granularities (sampling time every minute, every hour, every day, every week) has a periodic pattern. Based on the traffic history data of the target node within the third preset time period, the traffic time data of the target node within the third preset time period is determined, and based on the traffic time data of the target node within the third preset time period, the traffic time data of the target node within the fourth preset time period is determined.

[0156] For example: in the traffic time data within the third preset time period, the sampling time at the minute granularity is every 15 minutes, with a cycle of 0-59 minutes; the sampling time at the hourly granularity is every 1 hour, with a cycle of 0-23 hours; based on the regularity of the sampling time at each time granularity, the sampling time at each granularity in each subsequent entry can be derived.

[0157] Step S960, input the topological relationship data corresponding to the multiple target nodes, the traffic history data of each target node within the third preset time period, and the traffic time data of each target node within the third preset time period and the fourth preset time period into the traffic prediction model, and obtain the traffic prediction data of each target node within the fourth preset time period output by the traffic prediction model.

[0158] In an embodiment of the present application, the traffic prediction model refers to topological relationship data, traffic history data and traffic time data, and can perceive both the association relationship between nodes and the traffic flow status between nodes, and on this basis predict the traffic in the future with high accuracy.

[0159] The embodiment of the present application can be applied to the network operation and maintenance product of the bearer network to predict the flow trend in the future. The flow is not limited to network flow, but can also be human flow, vehicle flow, stock fluctuation, etc. Therefore, the flow prediction model of the embodiment of the present application can be applied to network flow prediction, human flow prediction, vehicle flow prediction, stock prediction and other scenarios.

[0160] The present application also provides a training device for a traffic prediction model, such as Fig.10 , which is a structural diagram of a training device for a traffic prediction model according to an embodiment of the present application.

[0161] The training device of the traffic prediction model includes: a processor 1010, a communication interface 1020, a memory 1030 and a communication bus 1040. The processor 1010, the communication interface 1020 and the memory 1030 communicate with each other through the communication bus 1040.

[0162] The memory 1030 is used to store computer programs.

[0163] In one embodiment of the present application, the processor 1010, when executing the program stored in the memory 1030, implements the training method of the traffic prediction model provided by any of the aforementioned method embodiments, including: in the target network, obtaining topological relationship data corresponding to multiple sample nodes and obtaining the traffic history data of each of the sample nodes; for each of the sample nodes, determining the traffic time data of the sample node based on the traffic history data of the sample node; using the topological relationship data and the traffic history data and traffic time data of the multiple sample nodes to train a preset traffic prediction model, so as to use the trained traffic prediction model to perform traffic prediction tasks.

[0164] Wherein, the type of each of the sample nodes is a network element port or a network tunnel; before obtaining the traffic history data of each of the sample nodes, it also includes: for each of the sample nodes, respectively performing the following steps: when the type of the sample node is a network element port, collecting the traffic statistics data reported by the network element to which the sample node belongs; when the type of the sample node is a network tunnel, collecting the traffic statistics data reported by the starting network element of the sample node; from the collected traffic statistics data, extracting the traffic processing volume data of the sample node at different sampling times as the traffic history data of the sample node; wherein the traffic history data of the sample node includes multiple data entries; each of the data entries corresponds to the traffic processing volume data of the sample node at a sampling time.

[0165] Among them, determining the traffic time data of the sample node based on the traffic history data of the sample node includes: for each data entry in the traffic history data of the sample node, according to a plurality of pre-set time granularities, performing time granularity conversion processing on the sampling time corresponding to the data entry, determining the sampling time data corresponding to the data entry at each of the time granularities, and forming the traffic time data of the sample node.

[0166] Wherein, the traffic history data of the sample node includes the traffic history data within a first preset time period and the traffic history data within a second preset time period; wherein, in terms of time, the first preset time period precedes the second preset time period; wherein the use of the topological relationship data and the traffic history data and traffic time data of multiple sample nodes to train a preset traffic prediction model includes: using the topological relationship data, the traffic history data of each of the multiple sample nodes within the first preset time period and the traffic time data of each of the sample nodes to train the traffic prediction model; wherein the traffic history data of each of the multiple sample nodes within the second preset time period is used as training label data.

[0167] Wherein, the traffic prediction model is used to perform the following steps: determine the topological relationship network of the plurality of sample nodes according to the topological relationship data corresponding to the plurality of sample nodes; aggregate the edge feature data corresponding to each sample node according to the topological relationship network and the node feature data corresponding to each sample node; wherein, for each sample node, the traffic history data of the sample node in the first preset time period and the traffic time data of the sample node are used as the node feature data of the sample node in the topological relationship network; for each sample node, generate the global feature data corresponding to the sample node according to the node feature data corresponding to the sample node and the edge feature data corresponding to the sample node; wherein the dimension of the global feature data corresponding to the sample node is the same as the dimension of the node feature data corresponding to the sample node; predict the traffic prediction data of each sample node in the second preset time period according to the global feature data corresponding to each sample node, the traffic history data of each sample node in the first preset time period and the traffic time data of each sample node in the second preset time period; wherein the traffic time data of each sample node includes the traffic time data in the first preset time period and the traffic time data in the second preset time period.

[0168] Among them, the edge feature data corresponding to each sample node is aggregated according to the topological relationship network and the node feature data corresponding to each sample node: including: for each sample node, according to the type of the sample node, determining whether the sample node is a sending node or a receiving node; when the sample node is a sending node, determining the receiving node connected to the sending node in the topological relationship network, aggregating the node feature data corresponding to the receiving node connected to the sending node, and obtaining the edge feature data corresponding to the sending node; when the sample node is a receiving node, determining the sending node connected to the receiving node in the topological relationship network, aggregating the node feature data corresponding to the sending node connected to the receiving node, and obtaining the edge feature data corresponding to the receiving node.

[0169] Among them, generating the global feature data corresponding to the sample node according to the node feature data corresponding to the sample node and the edge feature data corresponding to the sample node includes: splicing the node feature data corresponding to the sample node and the edge feature data corresponding to the sample node to obtain the global feature data corresponding to the sample node; using a deep learning component to adjust the dimension of the global feature data so that the dimension of the global feature data is the same as the dimension of the node feature data corresponding to the sample node.

[0170] Wherein, the use of the topological relationship data and the traffic history data and traffic time data of multiple sample nodes to train a preset traffic prediction model includes: dividing the multiple sample nodes into multiple batches according to the topological relationship data corresponding to the multiple sample nodes; wherein, in the topological relationship network corresponding to the topological relationship data, the subnetwork formed by the sample nodes in the same batch is a fully connected network; and applying the traffic history data and traffic time data of the sample nodes in each batch to a round of training of the traffic prediction model.

[0171] The embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method for training a traffic prediction model provided in any of the aforementioned method embodiments are implemented. Since the training method for the traffic prediction model has been described in detail above, for any details not fully described in the description of this embodiment, please refer to the relevant descriptions in the aforementioned embodiments, and no further description will be given here.

[0172] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0173] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the relevant technology can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0174] It should be understood that the terms used herein are only for the purpose of describing specific example embodiments and are not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms "one", "an" and "said" as used herein may also be meant to include plural forms. The terms "include", "comprise", "contain", and "have" are inclusive, and therefore specify the existence of stated features, steps, operations, elements and / or parts, but do not exclude the existence or addition of one or more other features, steps, operations, elements, parts, and / or combinations thereof. The method steps, processes, and operations described herein are not interpreted as necessarily requiring them to be performed in the specific order described or illustrated, unless the execution order is clearly indicated. It should also be understood that additional or alternative steps may be used.

[0175] The above description is only a specific implementation of the present application, so that those skilled in the art can understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest range consistent with the principles and novel features applied for herein.

Claims

1. A method for training a traffic prediction model, characterized in that: include: In the target network, topological relationship data corresponding to a plurality of sample nodes are obtained and flow history data of each of the sample nodes is obtained; For each of the sample nodes, determine the flow time data of the sample node according to the flow history data of the sample node; The preset traffic prediction model is trained using the topological relationship data and the traffic history data and traffic time data of the plurality of sample nodes, so that the traffic prediction task can be performed using the trained traffic prediction model.

2. The method according to claim 1, characterized in that The type of each of the sample nodes is a network element port or a network tunnel; Before acquiring the traffic history data of each sample node, the method further includes: For each of the sample nodes, perform the following steps respectively: When the type of the sample node is a network element port, collecting traffic statistics data reported by the network element to which the sample node belongs; When the type of the sample node is a network tunnel, collecting traffic statistics data reported by a starting network element of the sample node; From the collected traffic statistics data, the traffic processing volume data of the sample node at different sampling times are extracted as the traffic history data of the sample node; wherein the traffic history data of the sample node includes multiple data entries; each of the data entries corresponds to the traffic processing volume data of the sample node at a sampling time.

3. The method according to claim 2, characterized in that The determining the flow time data of the sample node according to the flow history data of the sample node includes: For each data entry in the traffic history data of the sample node, time granularity conversion processing is performed on the sampling time corresponding to the data entry according to multiple pre-set time granularities, and the sampling time data corresponding to the data entry at each time granularity is determined to form the traffic time data of the sample node.

4. The method according to claim 1, characterized in that: The traffic history data of the sample node includes traffic history data within a first preset time period and traffic history data within a second preset time period; wherein, in terms of time, the first preset time period is prior to the second preset time period; The method of using the topological relationship data and the traffic history data and traffic time data of the plurality of sample nodes to train a preset traffic prediction model includes: The traffic prediction model is trained using the topological relationship data, the traffic history data of each of the multiple sample nodes within the first preset time period, and the traffic time data of each of the sample nodes; wherein the traffic history data of each of the multiple sample nodes within the second preset time period is used as training label data.

5. The method according to claim 4, characterized in that The traffic prediction model is used to perform the following steps: Determine a topological relationship network of the plurality of sample nodes according to the topological relationship data corresponding to the plurality of sample nodes; Aggregate edge feature data corresponding to each sample node according to the topological relationship network and the node feature data corresponding to each sample node; wherein, for each sample node, the flow history data of the sample node within the first preset time period and the flow time data of the sample node are used as the node feature data of the sample node in the topological relationship network; For each of the sample nodes, generating global feature data corresponding to the sample node according to the node feature data corresponding to the sample node and the edge feature data corresponding to the sample node; wherein the dimension of the global feature data corresponding to the sample node is the same as the dimension of the node feature data corresponding to the sample node; Based on the global feature data corresponding to each of the sample nodes, the traffic history data of each of the sample nodes in the first preset time period and the traffic time data of each of the sample nodes in the second preset time period, the traffic prediction data of each of the sample nodes in the second preset time period is predicted; wherein the traffic time data of each of the sample nodes includes the traffic time data in the first preset time period and the traffic time data in the second preset time period.

6. The method according to claim 5, characterized in that The step of aggregating edge feature data corresponding to each sample node according to the topological relationship network and the node feature data corresponding to each sample node comprises: For each of the sample nodes, determining, according to the type of the sample node, whether the sample node is a sending node or a receiving node; When the sample node is a sending node, in the topological relationship network, a receiving node connected to the sending node is determined, and node feature data corresponding to the receiving node connected to the sending node is aggregated to obtain edge feature data corresponding to the sending node; When the sample node is a receiving node, in the topological relationship network, a sending node connected to the receiving node is determined, and node feature data corresponding to the sending nodes connected to the receiving node are aggregated to obtain edge feature data corresponding to the receiving node.

7. The method according to claim 5, characterized in that The generating the global feature data corresponding to the sample node according to the node feature data corresponding to the sample node and the edge feature data corresponding to the sample node comprises: Concatenate the node feature data corresponding to the sample node and the edge feature data corresponding to the sample node to obtain the global feature data corresponding to the sample node; The dimension of the global feature data is adjusted using a deep learning component so that the dimension of the global feature data is the same as the dimension of the node feature data corresponding to the sample node.

8. The method according to claim 1, characterized in that The method of using the topological relationship data and the traffic history data and traffic time data of the plurality of sample nodes to train a preset traffic prediction model includes: According to the topological relationship data corresponding to the plurality of sample nodes, the plurality of sample nodes are divided into a plurality of batches; wherein, In the topological relationship network corresponding to the topological relationship data, the subnetwork formed by the sample nodes in the same batch is a fully connected network; The traffic history data and traffic time data of the sample nodes in each batch are applied to a round of training of the traffic prediction model.

9. A training device for a traffic prediction model, characterized in that: include: at least one communication interface; at least one bus connected to the at least one communication interface; at least one processor coupled to the at least one bus; At least one memory connected to the at least one bus, wherein the processor is configured to: execute a training program of a traffic prediction model stored in the memory to implement the training method of the traffic prediction model described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are executed to implement the training method of the traffic prediction model described in any one of claims 1-8.

Images