Time-sensitive data stream cross-domain aggregation method, system, device and storage medium

Abstract

The embodiment of the present invention discloses a method, system, device and storage medium for cross-domain aggregation of time-sensitive data streams, including: obtaining data packet information and network status information in a cross-domain network, and constructing an input feature vector based on the data packet information and network status information; inputting the input feature vector into an aggregation neural network model, and generating a cross-domain aggregation configuration table based on the aggregation result output by the aggregation neural network model; wherein the cross-domain aggregation configuration table contains at least one time slot offset information that allows the aggregation of time-sensitive data streams during cross-domain transmission; and sending the cross-domain aggregation configuration table to the communication nodes of each network data plane in the cross-domain network to complete the cross-domain aggregation transmission of each allowed aggregation time-sensitive data stream in the cross-domain network. This ensures that the time-sensitive data streams are aggregated from local to wide area in accordance with the cross-domain aggregation configuration table, while completing the balance between the time efficiency and load capacity of the cross-domain transmission of the time-sensitive data streams.

Description

Technical Field

[0001] The present invention relates to the field of network communication technology, and in particular to a method, system, device and storage medium for cross-domain aggregation of time-sensitive data streams. Background Art

[0002] The use of Time-Sensitive Networks (TSN) in localized industrial systems to transmit time-sensitive (TS) data streams (such as automation control data streams) has become a key development direction for unmanned factories. Similar to local TSN, Deterministic Internet Protocol (DIP) networks have also been developed for wide-area transmission of TS data streams. Integrated TSN and DIP networks, which utilize TSN as the access network and DIP as the aggregation and core network, are becoming a key development trend in cross-domain, distributed, unmanned factory data transmission infrastructure.

[0003] However, integrated TSN and DIP networks face the problem of converging multiple TSN domains into a single DIP network domain. Improper convergence can significantly reduce network utility. The time slot mapping from multiple TSN domains to the DIP domain relies on a cross-domain aggregation mechanism. Existing cross-domain aggregation mechanisms from TSN domains to DIP network domains are generally manually designed. This design typically requires constructing multiple TSN-to-DIP time slot mapping functions, while also considering the trade-off between utility and load capacity. This makes the design difficult, and manual design may result in insufficient exploration of the design space and the abandonment of the optimal design, making it difficult to achieve optimal settings for utility and load. Summary of the Invention

[0004] The present invention provides a method, system, device and storage medium for cross-domain aggregation of time-sensitive data streams, which reduces human participation in determining the cross-domain aggregation mechanism of time-sensitive data streams, optimizes the balance between utility and load in the aggregation mechanism, and improves the efficiency of cross-domain aggregation of time-sensitive data streams.

[0005] In a first aspect, an embodiment of the present invention provides a method for cross-domain aggregation of time-sensitive data streams, including:

[0006] Obtaining data packet information and network status information in the cross-domain network, and constructing an input feature vector based on the data packet information and network status information;

[0007] Inputting the input feature vector into the aggregation neural network model, and generating a cross-domain aggregation configuration table according to the aggregation result output by the aggregation neural network model; wherein the cross-domain aggregation configuration table includes at least one time slot offset information allowing the aggregated time-sensitive data stream to be transmitted across domains;

[0008] Send the cross-domain aggregation configuration table to the communication nodes of each network data plane in the cross-domain network to complete the cross-domain aggregation transmission of each time-sensitive data flow allowed to be aggregated in the cross-domain network;

[0009] Among them, the cross-domain network includes at least one time-sensitive network TSN network data plane and a deterministic internet protocol DIP network data plane; the aggregated neural network model is based on the input feature vector samples extracted by the online sampling mechanism and trained through a competitive mechanism.

[0010] In a second aspect, an embodiment of the present invention further provides a time-sensitive data stream cross-domain aggregation system, comprising: a software-defined networking controller, at least one time-sensitive network TSN network data plane and a deterministic internet protocol DIP network data plane;

[0011] The software-defined networking controller includes at least a user controller, an aggregator, and a network configurator;

[0012] The user controller is used to obtain data packet information and network status information in the cross-domain network, construct an input feature vector based on the data packet information and the network status information, and send the input feature vector to the aggregator;

[0013] The aggregator is an aggregation neural network model trained based on input feature vector samples extracted based on an online sampling mechanism and with output data stream index labels. It is used to generate aggregation results based on the input feature vectors and send the aggregation results to the network configurator.

[0014] The network configurator is used to generate a cross-domain aggregation configuration table based on the aggregation results, and send the cross-domain aggregation configuration table to the intermediate nodes of each TSN network data plane and the aggregation node of the DIP network data plane;

[0015] Each TSN network data plane and DIP network data plane is used to complete the cross-domain transmission of the aggregated time-sensitive data stream corresponding to the cross-domain aggregation configuration table based on the cross-domain aggregation configuration table.

[0016] In a third aspect, an embodiment of the present invention further provides a device for cross-domain aggregation of time-sensitive data streams, including:

[0017] at least one processor; and

[0018] a memory communicatively connected to at least one processor; wherein,

[0019] The memory stores a computer program that can be executed by at least one processor. The computer program is executed by the at least one processor so that the at least one processor can execute the time-sensitive data stream cross-domain aggregation method provided by an embodiment of the present invention.

[0020] In a fourth aspect, an embodiment of the present invention further provides a storage medium comprising computer-executable instructions, which, when executed by a computer processor, are used to execute the time-sensitive data stream cross-domain aggregation method provided by an embodiment of the present invention.

[0021] An embodiment of the present invention provides a method, system, device and storage medium for cross-domain aggregation of time-sensitive data streams, which obtains data packet information and network status information in a cross-domain network, and constructs an input feature vector based on the data packet information and network status information; inputs the input feature vector into an aggregation neural network model, and generates a cross-domain aggregation configuration table based on the aggregation result output by the aggregation neural network model; wherein the cross-domain aggregation configuration table contains at least one time slot offset information allowing the aggregation of time-sensitive data streams during cross-domain transmission; the cross-domain aggregation configuration table is sent down to the communication nodes of each network data plane in the cross-domain network to complete the cross-domain aggregation transmission of each allowed aggregation time-sensitive data stream in the cross-domain network; wherein the cross-domain network includes at least one time-sensitive network TSN network data plane and a deterministic Internet Protocol DIP network data plane; the aggregation neural network model is based on input feature vector samples extracted by an online sampling mechanism and trained through a competitive mechanism. By adopting the above technical solution, for the cross-domain aggregation transmission scenario of TS data streams in multiple TSN network data planes to the DIP network data plane, the input feature vector samples extracted by the online sampling mechanism are used to train an aggregation neural network model for determining the TS data stream aggregation mechanism from multiple TSN domains to the DIP domain that meets both the optimal quality and load constraints. This makes it unnecessary to manually design a cross-domain aggregation mechanism when performing cross-domain aggregation transmission of TS data streams. Instead, the aggregation neural network model constructs the input feature vector based on the obtained data packet information and network status information, and outputs the aggregation result that meets both the highest transmission quality and the load constraints. Based on the aggregation results output by the aggregation neural network model, the allowed aggregated TS data streams in each TSN network data plane that can be aggregated and sent to the DIP network data plane are determined, and a cross-domain aggregation configuration table containing the time slot offset information of the allowed aggregated TS data streams is generated. The cross-domain aggregation configuration table is distributed to each network data plane so that the allowed aggregated TS data streams can be aggregated and transmitted across domains. This ensures that while the TS data streams are aggregated from local to wide area in a "many-to-one" manner according to the cross-domain aggregation configuration table, the time sensitivity of the data streams is retained, and the balance between the time efficiency of the cross-domain transmission of the TS data streams and the cross-domain network load capacity is achieved, thereby improving the efficiency of the cross-domain aggregation of the TS data streams.

[0022] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 A flowchart of a method for cross-domain aggregation of time-sensitive data streams provided in Example 1 of the present invention;

[0025] Figure 2 A flowchart of a method for cross-domain aggregation of time-sensitive data streams provided in Example 2 of the present invention;

[0026] Figure 3 A flowchart illustrating a method for determining at least one time-sensitive data flow to be aggregated based on occurrence probabilities, data packet length values ​​corresponding to the occurrence probabilities in the input feature vector, and output restriction conditions, provided in the second embodiment of the present invention;

[0027] Figure 4 A schematic diagram of the structure of a cross-domain aggregation system for time-sensitive data streams provided in Example 3 of the present invention;

[0028] Figure 5 A schematic diagram of the structure of a cross-domain aggregation device for time-sensitive data streams provided in Example 4 of the present invention. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] Example 1

[0032] Figure 1 A flowchart of a method for cross-domain aggregation of time-sensitive data streams provided in the first embodiment of the present invention is applicable to the situation where time-sensitive data streams are aggregated and transmitted across domains in multiple TSN network domains and a DIP network domain. The method can be executed by a software-defined network (SDN) controller in a time-sensitive data stream cross-domain aggregation system. The time-sensitive data stream cross-domain aggregation system can be implemented by software and / or hardware, and the time-sensitive data stream cross-domain aggregation system can be configured in a time-sensitive data stream cross-domain aggregation device. Optionally, the time-sensitive data stream cross-domain aggregation device can be a notebook, desktop computer, smart tablet, etc., which is not limited by the embodiment of the present invention.

[0033] like Figure 1 As shown, an embodiment of the present invention provides a method for cross-domain aggregation of time-sensitive data streams, which specifically includes the following steps:

[0034] S101: Obtain data packet information and network status information in a cross-domain network, and construct an input feature vector based on the data packet information and network status information.

[0035] The cross-domain network includes at least one time-sensitive network TSN network data plane and a deterministic internet protocol DIP network data plane.

[0036] In this embodiment, the cross-domain network can be specifically understood as a network that includes both a local area network and a wide area network. In the embodiment of the present invention, it can be understood as a network composed of at least one local area network type TSN network data plane and a wide area network type DIP network data plane. The data packet information can be specifically understood as the parameter information of the data packet composed of the TS data stream that needs to be transmitted across the domain based on each TSN network data plane in the cross-domain network. The network status information can be specifically understood as the parameter information determined based on the current demand and the load capacity limit of the DIP network data plane in the cross-domain network, which is used to characterize the ability of the DIP network to receive TS data streams. The input feature vector can be specifically understood as a vector containing the TS data stream information that needs to be transmitted in the cross-domain network, as well as the overall status information of the cross-domain network.

[0037] Specifically, through manual input or automatic reading, the cross-domain network extracts packet information of the TS data streams required for cross-domain aggregation transmission from each TSN network data plane. Furthermore, the cross-domain network extracts network status information related to the load capacity of the DIP network data plane. The packet information and network status information are then converted and constructed to obtain an input feature vector suitable for the input requirements of the aggregated neural network model.

[0038] S102: Input the input feature vector into the aggregation neural network model, and generate a cross-domain aggregation configuration table according to the aggregation result output by the aggregation neural network model.

[0039] The cross-domain aggregation configuration table includes at least one time slot offset information that allows the aggregated time-sensitive data stream to be transmitted across domains.

[0040] In this embodiment, the aggregation neural network model can be specifically understood as a neural network model that generates an aggregation mechanism that satisfies the optimal utility of cross-domain network load and TS data stream transmission based on an input feature vector input therein, which includes cross-domain network load information and TS data stream information to be transmitted. The aggregation result can be specifically understood as the output result of the aggregation neural network model. Optionally, the aggregation result can be the index of the TS data stream that is allowed to be transmitted across domains among the TS data streams that need to be transmitted across domains in the input aggregation neural network model. It can be understood that there can be multiple TS data streams that need to be transmitted across domains, and the load capacity of the DIP network data plane is limited. In order to achieve the optimal utility of TS data stream transmission under limited load capacity, it is necessary to select from each TS data stream, and the TS data stream selected based on the aggregation result is the time-sensitive data stream that allows aggregation. The cross-domain aggregation configuration table can be specifically understood as a configuration table generated based on the timing offset information of each aggregated TS data stream allowed to be transmitted across domains in a cross-domain network, and used to be sent down to each network data plane to implement TS data stream scheduling. It can also be understood as information that is sent down to each network data plane containing a cross-domain aggregation mode for each network data plane to complete the configuration.

[0041] Specifically, the input feature vector is input into the aggregation neural network model to obtain an aggregation result containing the index of TS data streams that can be allowed to be transmitted across domains. The TSN network data plane in the cross-domain network that can perform cross-domain aggregation transmission of TS data streams can be determined based on the aggregation result, and the TS data streams that need to be transmitted by the TSN network data plane can be determined as TS data streams that are allowed to be aggregated. According to actual needs, an appropriate time slot offset determination method is selected to determine the time slot offset information of each allowed aggregated TS data stream when performing cross-domain transmission, and based on each time slot offset information, a cross-domain aggregation configuration table is constructed for sending to each network data plane to realize TS data stream scheduling.

[0042] In an embodiment of the present invention, by inputting an input feature vector containing cross-domain network information into the aggregation neural network model, an aggregation result that meets the cross-domain network load demand and utility demand is automatically obtained, and then the allowed aggregation TS data stream that can be cross-domain aggregated and the time slot offset information corresponding to the allowed aggregation TS data stream can be determined based on the aggregation result, thereby reducing the degree of human participation in the aggregation mechanism determination process, so that the corresponding cross-domain aggregation configuration table generated based on each time slot offset information is more in line with the performance requirements of cross-domain aggregation transmission of TS data streams.

[0043] Optionally, the aggregate neural network model is trained based on input feature vector samples extracted by an online sampling mechanism and a competitive mechanism.

[0044] In this embodiment, the online sampling mechanism can be specifically understood as a training sample generation mechanism that does not rely on data samples stored in a database, but rather performs sampling in accordance with certain random rules in the network.

[0045] Optionally, before training the aggregated neural network model, a preset number of input feature training samples may be generated based on a random rule or a deterministic rule provided by the user to form an input feature training sample set, and two initial neural network models with the same parameters may be randomly initialized. The input feature training samples are respectively input into the two initial neural network models. A loss function is constructed based on the logarithmic probability and utility output by the two models that meet the load constraint. The two neural network models are trained through a competitive mechanism based on the loss function until a pre-set convergence condition is met to obtain an aggregated neural network model. The convergence condition may be that the average utility reaches the expected value or reaches the maximum number of iterations, etc., which is not limited in this embodiment of the present invention.

[0046] In this embodiment of the present invention, an online sampling mechanism is used to construct input feature vector samples, reducing the amount of data required for the database and the number of samples that need to be stored in the database. Furthermore, the online sampling mechanism allows for the extraction of new input feature vector samples during each iterative training, reducing the probability of sample duplication and improving the accuracy of the aggregate neural network model training.

[0047] S103: Send the cross-domain aggregation configuration table to the communication nodes of each network data plane in the cross-domain network, so as to complete the cross-domain aggregation transmission of each allowed aggregation time-sensitive data flow in the cross-domain network.

[0048] In this embodiment, the communication node can be specifically understood as a node used by each network data plane in the cross-domain network to send or receive transmission data streams to the outside, such as a TSN switch in the TSN network data plane, a DIP router in the DIP network data plane, etc.

[0049] Specifically, the SDN controller will send the determined cross-domain aggregation configuration table to the communication nodes of each network data plane in the cross-domain network, so that each network data plane can clearly know the allowed aggregated TS data streams that can be transmitted through cross-domain aggregation, as well as the time slot offset information of each allowed aggregated TS data stream, so that each allowed aggregated TS data stream can complete cross-domain aggregation transmission to the DIP network data plane according to the corresponding time slot offset information.

[0050] The technical solution of this embodiment obtains data packet information and network status information in a cross-domain network, and constructs an input feature vector based on the data packet information and network status information; inputs the input feature vector into an aggregation neural network model, and generates a cross-domain aggregation configuration table based on the aggregation result output by the aggregation neural network model; wherein the cross-domain aggregation configuration table includes at least one time slot offset information that allows aggregation of time-sensitive data streams during cross-domain transmission; the cross-domain aggregation configuration table is sent down to the communication nodes of each network data plane in the cross-domain network to complete the cross-domain aggregation transmission of each allowed aggregation time-sensitive data stream in the cross-domain network; wherein the cross-domain network includes at least one time-sensitive network TSN network data plane and a deterministic Internet Protocol DIP network data plane; the aggregation neural network model is trained based on the input feature vector samples extracted by the online sampling mechanism through a competitive mechanism. By adopting the above technical solution, for the cross-domain aggregation transmission scenario of TS data streams in multiple TSN network data planes to the DIP network data plane, the input feature vector samples extracted by the online sampling mechanism are used to train a neural network model for determining the TS data stream aggregation mechanism from multiple TSN domains to the DIP domain that simultaneously meets the optimal quality and load constraints. This makes it unnecessary to manually design a cross-domain aggregation mechanism when performing cross-domain aggregation transmission of TS data streams. Instead, the aggregation neural network model constructs the input feature vector based on the obtained data packet information and network status information, and outputs the aggregation result that simultaneously meets the highest transmission quality and meets the load constraints. Based on the aggregation results output by the aggregation neural network model, the allowed aggregated TS data streams in each TSN network data plane that can be aggregated and sent to the DIP network data plane are determined, and a cross-domain aggregation configuration table containing the time slot offset information of the allowed aggregated TS data streams is generated. The cross-domain aggregation configuration table is completed for each network data plane so that the allowed aggregated TS data streams can be aggregated and transmitted across domains. This ensures that while the TS data streams are aggregated from local to wide area in a "many-to-one" manner according to the cross-domain aggregation configuration table, the time sensitivity of the data streams is retained, and the balance between the time efficiency of the cross-domain transmission of the TS data stream and the cross-domain network load capacity is achieved, thereby improving the efficiency of the cross-domain aggregation of the TS data streams.

[0051] Example 2

[0052] Figure 2A flowchart of a method for cross-domain aggregation of time-sensitive data streams provided in Example 2 of the present invention is provided. The technical solution of the embodiment of the present invention is further optimized based on the above-mentioned optional technical solutions. After determining the network status by obtaining the queue length value and the maximum queue length value of the DIP network data plane, and determining the data packet information in the cross-domain network by obtaining the maximum utility value and the maximum packet length value of the cross-domain network, as well as the data packet feature information corresponding to each TSN network data plane, the obtained data packet information and network status information are first verified for data validity, and an input feature vector for inputting into the aggregation neural network model is generated only when the data is valid. The output multi-head attention layer of the aggregation neural network model contains output restriction conditions determined according to the network status information, which is used to assist in determining the allowed aggregation TS data streams, and then generate a cross-domain aggregation configuration table based on the aggregation results output by the aggregation neural network model. The cross-domain aggregation configuration table is sent down to the intermediate nodes of each TSN network data plane and the aggregation node of the DIP network data plane that is independent of other DIP routers, so that the allowed aggregation TS data streams can realize cross-domain aggregation transmission in the cross-domain network. After completing the above transmission, the data packet feature information corresponding to each allowed aggregation TS data stream can be displayed to the outside world, so that the user can clearly understand which TSN network data plane the TS data stream that finally completes the cross-domain aggregation transmission belongs to, as well as the utility of the transmission data stream. By setting output restriction conditions in the aggregation neural network model, the allowed aggregation TS data stream corresponding to the output aggregation result can meet the load requirements of the cross-domain network during cross-domain transmission. At the same time, before the aggregation neural network model outputs the aggregation result, each TS data stream is sorted according to the probability of occurrence to ensure that the output utility of the allowed aggregation TS data stream corresponding to the aggregation result is optimal, so that the TS data stream will not lose its time-sensitive characteristics after cross-domain transmission, thereby improving the efficiency of cross-domain aggregation of TS data streams and making the cross-domain aggregation transmission results intuitively displayed to the outside world.

[0053] like Figure 2 As shown, a method for cross-domain aggregation of time-sensitive data streams provided in the second embodiment of the present invention specifically includes the following steps:

[0054] S201: Obtain a queue length value and a maximum queue length value of a DIP network data plane in an inter-domain network, and determine the queue length value and the maximum queue length value as network status information.

[0055] In this embodiment, the queue length value can be specifically understood as the load capacity value that the DIP network data plane can provide in the current scenario, which is set according to actual needs. The maximum queue length value can be specifically understood as the maximum load capacity value that the cross-domain network can provide.

[0056] Specifically, when performing cross-domain aggregated transmission of TS data streams, the SDN controller can receive the load capacity currently available for cross-domain transmission of TS data streams, set according to actual needs, and use it as the queue length value. At the same time, the SDN controller can read the overall network capabilities of the cross-domain network, determine the maximum queue length value based on the maximum load capacity that the cross-domain network can provide, and determine the queue length value and the maximum queue length value as network status information. It should be clarified that the maximum queue length value is the maximum load capacity that the cross-domain network can provide, that is, the queue length value must be less than or equal to the maximum queue length value.

[0057] S202: Obtain the maximum utility value and maximum packet length value of the cross-domain network, as well as the data packet feature information corresponding to each TSN network data plane.

[0058] The data packet characteristic information includes the data packet length value and data packet utility value of the data packet corresponding to the time-sensitive data flow to be transmitted in the TSN network data plane.

[0059] In this embodiment, the TS data stream to be transmitted can be specifically understood as the TS data stream that the TSN network data plane needs to perform cross-domain transmission in the current scenario. The maximum utility value can be specifically understood as the maximum utility value that may appear in a single TS data stream that can be transmitted across domains in the current scenario. The maximum packet length value can be specifically understood as the maximum packet length of a data packet corresponding to a single TS data stream that can be transmitted across domains in the current scenario. It is understood that the maximum utility value and maximum packet length value can be directly obtained from external input according to actual needs, or can be determined based on the currently acquired TS data streams to be transmitted of each TSN network data plane. In other words, the maximum length value of the data packet in each TS data stream to be transmitted is determined as the maximum data packet length value, and the maximum utility value in each TS data stream to be transmitted is determined as the maximum utility value. Different utility values ​​represent different meanings, such as a utility value of 1 represents throughput, a utility value of 2 represents sending status and reward value, etc. Those skilled in the art can clearly understand the specific meanings represented by different utility values, and the embodiments of the present invention do not impose any detailed restrictions on this. The data packet characteristic information can be specifically understood as the parameter information generated when the TS data stream to be transmitted performs data transmission.

[0060] Specifically, when executing cross-domain aggregation transmission of TS data streams, the SDN controller can receive the maximum utility value and maximum packet length value that the cross-domain network can support, which are set according to actual needs, or can obtain the packet length value and packet utility value of the data packet corresponding to the TS data stream to be transmitted on each TSN network data plane, and then determine the maximum utility value and maximum packet length of the cross-domain network in the current scenario based on each packet length value and packet utility value, and determine a group of packet length values ​​and packet utility values ​​of the same data packet as the packet characteristic information of the TSN network data plane corresponding to the data packet.

[0061] It is understandable that there is no obvious order relationship between S201 and S202. In the embodiment of the present invention, only the order of S201-S202 is taken as an example, and the specific execution order is not limited in the embodiment of the present invention.

[0062] S203: Determine the characteristic information, maximum utility value, and maximum packet length value of each data packet as data packet information.

[0063] S204: Determine the data validity of each data packet characteristic information according to the maximum utility value, the maximum packet length value, and the maximum queue length value.

[0064] Specifically, in order to ensure that the TS data stream aggregation transmission process meets the load requirements of the cross-domain network, and each data stream to be transmitted should not exceed the maximum range allowed for transmission across the domain network, the data validity of each data packet characteristic information can be determined according to the maximum utility value, maximum packet length value and maximum queue length value to ensure that each TS data stream to be transmitted is data that can be aggregated and transmitted across domains.

[0065] For example, for a TS data stream to be transmitted, the corresponding data packet length value is determined as w i , the utility value of the data packet is determined as u i , the maximum utility value is determined as u max , the maximum packet length value is determined to be w max , the maximum queue length value is determined to be c max , then when 0<u i ≤u max 、0<w i ≤w max and w i ≤c max , it can be considered that the data of the data packet characteristic information corresponding to the TS data stream to be transmitted is valid.

[0066] S205: Determine the data validity of the queue length value according to the maximum queue length value.

[0067] Specifically, the queue length should be set within the range allowed by the current cross-domain network load. Therefore, to ensure that the data subsequently input into the aggregated neural network model is valid, it is necessary to first determine whether the queue length set for the current scenario is valid. Since the maximum queue length corresponds to the maximum load supported by the cross-domain network, the queue length set for the current scenario is considered valid when the queue length is less than or equal to the maximum queue length.

[0068] In an embodiment of the present invention, data validity is verified for each parameter used for cross-domain aggregation transmission based on the obtained maximum packet length value, maximum utility value, and maximum queue length value, and an input feature vector for inputting the aggregation neural network model is generated only when the data is valid, thereby ensuring that the aggregation result output by the aggregation neural network model is an aggregation mechanism available for the current scenario, thereby improving the execution stability of cross-domain aggregation transmission of TS data streams.

[0069] It is understandable that there is no obvious order relationship between S204 and S205. In the embodiment of the present invention, only the order of S204-S205 is taken as an example, and the specific execution order is not limited in the embodiment of the present invention.

[0070] S206. If the characteristic information of each data packet and the queue length value are valid, construct a single data flow characteristic vector corresponding to each time-sensitive data flow to be transmitted based on the characteristic information of each data packet, the queue length value, the maximum utility value, the maximum packet length value and the maximum queue length value.

[0071] Specifically, when the characteristic information of each data packet and the queue length value are valid, it can be considered that the cross-domain aggregation transmission of the TS data stream can be completed based on the obtained data packet information and network status information in the current scenario. At this time, for the TS data stream to be transmitted corresponding to a TSN network data plane, a corresponding single data stream feature vector can be constructed.

[0072] For example, assuming that the cross-domain network includes M TSN network data planes, that is, includes M TS data streams to be transmitted, then for the i-th TS data stream to be transmitted, the corresponding data packet length value is determined as w i , the utility value of the data packet is determined as u i , the queue length value is determined as c, and the maximum utility value is determined as u max , the maximum packet length value is determined to be w max , the maximum queue length value is determined to be c max , the constructed single data stream feature vector x i It can be expressed as:

[0073]

[0074] Where i=0,1,…,M-1.

[0075] S207: Determine the sequence matrix formed by the feature vectors of each single data stream as the input feature vector.

[0076] Specifically, after the construction of each single data stream feature vector is completed, the single data stream feature vectors are integrated into a sequence matrix, and the sequence matrix is ​​determined as the input feature vector. iThe sequence matrix is ​​composed of M×3 dimensions.

[0077] S208: Input the input feature vector into the aggregate neural network model for dimension mapping, feature extraction and integration, and determine the occurrence probability of the time-sensitive data flow to be transmitted corresponding to each TSN network data plane in the input feature vector.

[0078] Among them, the output multi-head attention layer of the aggregated neural network model contains output constraints determined according to network state information.

[0079] In this embodiment, the output restriction condition can be specifically understood as a restriction condition set according to the currently available cross-domain network load capacity, which is used to determine the number and size of TS data streams allowed to be transmitted across domains. It can be understood that in the embodiment of the present invention, the queue length value is the load capacity value that the DIP network data plane can provide in the current scenario, which can also be understood as the cross-domain network load capacity that the cross-domain network can provide in the current scenario. When each TSN network data plane sends data packets corresponding to the TS data stream across domains, the DIP network data plane only allows data packets whose sum of data packet length values ​​is less than the queue length value to be transmitted across domains, so the output restriction condition can be set according to the queue length value.

[0080] Optionally, the aggregated neural network model may include a linear embedding layer, a placeholder layer, an encoding layer, and a decoding layer. The encoding layer structure may sequentially include multiple sub-encoding layers with the same structure, each sub-encoding layer including a multi-head attention layer, a residual layer, a layer normalization, a linear rectification layer, a linear layer, a residual layer, and a layer normalization, etc.; the decoding layer structure is similar to the encoding layer structure, including multiple sub-decoding layers with the same structure and an output multi-head attention layer, the sub-encoding layers sequentially including a multi-head attention layer, a residual layer, a layer normalization, a linear rectification layer, a linear layer, a residual layer, and a layer normalization, etc., wherein the output multi-head attention layer includes output constraints determined according to network state information.

[0081] Continuing with the above example, when the input feature vector is a sequence matrix of dimension M×3, the input feature vector is input into the linear embedding layer in the aggregate neural network model, and the dimension M×3 is mapped to M×n1, where n1 is the output dimension of the linear embedding layer; the output of the linear embedding layer is input into the placeholder layer, which is cascaded with a random placeholder to obtain a placeholder layer output of dimension (M+1)×n1; the placeholder layer output is input into the encoding layer for feature extraction and integration, and the corresponding encoding result matrix h of dimension (M+1)×n1 is obtained. enc o der ; and input the encoding result into the decoding layer for decoding, through h encoder, the last decoding index is embedded and the position is embedded to obtain the probability of occurrence of the TS data stream to be transmitted in each TSN network data plane corresponding to the input feature vector, or it can be understood as the probability that the TS data stream to be transmitted in each TSN network data plane is worthy of cross-domain aggregation transmission.

[0082] Furthermore, the encoding layer and decoding layer in the aggregated neural network model in the embodiment of the present invention only use layer normalization to process data instead of batch normalization, which reduces the amount of data processing, improves the training speed of the aggregated neural network model, and the data calculation speed during use, thereby reducing the amount of data calculation.

[0083] S209: Determine at least one time-sensitive data flow that is allowed to be aggregated according to each occurrence probability, the data packet length value corresponding to each occurrence probability in the input feature vector, and the output restriction condition.

[0084] Specifically, since the output restriction condition is determined based on the network status information in the current scenario and satisfies the restriction condition of the cross-domain network load capacity that can be provided in the current scenario, it can also be understood as a restriction condition for limiting the total length of the data packets corresponding to the cross-domain aggregated transmission TS data stream. Therefore, according to each occurrence probability, the TS data stream to be transmitted with better utility and higher priority in the cross-domain aggregated transmission can be first determined, and the data packet length value corresponding to each TS data stream to be transmitted can be determined. The length values ​​of each data packet are summed up in turn and compared with the output restriction condition to obtain a solution with the largest sum of each occurrence probability based on satisfying the output restriction condition, and the TS data streams to be transmitted contained in the solution are determined as allowed aggregated TS data streams.

[0085] Optional, Figure 3 The second embodiment of the present invention provides a flow chart for determining at least one time-sensitive data flow that is allowed to be aggregated based on each occurrence probability, the packet length value corresponding to each occurrence probability in the input feature vector, and the output restriction condition, as shown in FIG. Figure 3 As shown, the specific steps include:

[0086] S2091. Reorder the data packet length values ​​corresponding to each occurrence probability in the input feature vector in descending order of occurrence probability.

[0087] Specifically, since the occurrence probability represents the utility contribution of the corresponding TS data stream to be transmitted to the overall cross-domain transmission when cross-domain aggregation transmission is performed, the TS data streams to be transmitted can be sorted in descending order according to the occurrence probability, that is, the data packet length values ​​corresponding to each TS data stream to be transmitted in the input feature vector are reordered.

[0088] S2092: Add the length values ​​of the reordered data packets in sequence. When the sum is greater than the queue length value for the first time, determine the occurrence probability corresponding to the currently added data packet length value as the critical occurrence probability.

[0089] Specifically, after completing the reordering of the length values ​​of each data packet, the length values ​​of each data packet are added in order of occurrence probability from high to low. After each addition, the obtained sum is compared with the queue length value. If the sum is less than the queue length value, it can be considered that the TS data streams to be transmitted corresponding to the sum can be aggregated and transmitted across domains on the basis of meeting the cross-domain network load. When the sum is greater than the queue length value for the first time, the last data packet length value corresponding to the sum can be considered to be a critical value, that is, the TS data streams to be transmitted before this addition can be aggregated and transmitted across domains on the basis of meeting the cross-domain network load, but after adding the TS data stream to be transmitted added this time, the load requirements of the cross-domain network cannot be met. Therefore, the occurrence probability corresponding to the current added data packet length value can be determined as the critical occurrence probability.

[0090] S2093: Determine the time-sensitive data flow to be transmitted whose occurrence probability is greater than the critical occurrence probability as the time-sensitive data flow allowed to be aggregated.

[0091] Specifically, since the addition of the length values ​​of each data packet is performed in order from high to low according to the probability of occurrence, before reaching the critical probability of occurrence, the probability of occurrence corresponding to the length values ​​of the data packets that have been added should be greater than the critical probability of occurrence, and the TS data stream to be transmitted corresponding to the length values ​​of the data packets that have been added can be considered to be the TS data stream transmission combination with the best utility based on meeting the cross-domain network load. At this time, the TS data stream to be transmitted with a probability of occurrence greater than the critical probability of occurrence can be determined as a time-sensitive data stream that allows aggregation.

[0092] S210: Determine a set of data flow indexes corresponding to each time-sensitive data flow that is allowed to be aggregated as an aggregation result.

[0093] In this embodiment, the index can be specifically understood as a data structure used to characterize the logical pointing relationship so as to efficiently obtain data. The data stream index in the embodiment of the present invention can be specifically understood as a data structure used to characterize the pointing relationship of the TS data stream to be transmitted in the TSN network data plane.

[0094] Specifically, since each allowed aggregated TS data stream is a part of the TS data stream to be transmitted given by different TSN network data planes, and there is a corresponding relationship between each TS data stream to be transmitted and the TSN network data plane that sends it, the corresponding TS data stream to be transmitted can be represented by the number of the TSN network data plane, that is, the number of the TSN network data plane corresponding to the allowed aggregated TS data stream can be determined as the data stream index of the allowed aggregated TS data stream, and then the set of data stream indexes corresponding to all allowed aggregated TS data streams can be determined as the output result of the aggregation neural network model, which is used to point to each TSN network data plane that can perform cross-domain aggregation transmission of TS data streams.

[0095] S211. Generate a cross-domain aggregation configuration table based on the aggregation results output by the aggregation neural network model.

[0096] Specifically, according to the index information of each data stream contained in the aggregation result, the allowed aggregated TS data streams that can be aggregated and transmitted across domains in each TSN network data plane are determined, and then according to the packet length value and other information corresponding to each allowed aggregated TS data stream, the time slot offset information required for cross-domain transmission to the DIP network data plane is determined. According to the corresponding relationship between each time slot offset information and each allowed aggregated TS data stream and the TSN network data plane, a cross-domain aggregation configuration table is constructed.

[0097] S212: Send the cross-domain aggregation configuration table to the intermediate nodes of each TSN network data plane.

[0098] In this embodiment, the intermediate node can be specifically understood as a TSN switch in the TSN network data plane for sending and receiving data to the outside. It can be understood that the TSN switch exists independently of other TSN switches in the TSN network data plane, and the TS data stream transmission between the TSN switches is based on the Cyclic Queuing and Forwarding (CQF) protocol.

[0099] Specifically, after generating the cross-domain aggregation configuration table, the SDK controller sends the cross-domain aggregation configuration table to the intermediate nodes of each TSN network data plane based on the cross-domain network architecture, so that the intermediate node completes the cross-domain aggregation configuration based on the cross-domain aggregation configuration table, and then performs the corresponding transmission operation in the subsequent cross-domain aggregation transmission of the TS data stream.

[0100] S213: Send the cross-domain aggregation configuration table to the aggregation node of the DIP network data plane.

[0101] In this embodiment, the aggregation node can be specifically understood as a DIP router in the DIP network data plane for sending and receiving data externally. It can be understood that the aggregation node exists independently of other DIP routers in the DIP network data plane, and the TS data stream transmission within the DIP network data plane is implemented based on the DIP protocol.

[0102] Specifically, after generating the cross-domain aggregation configuration table, the SDK controller sends the cross-domain aggregation configuration table to the aggregation node of the DIP network data plane based on the cross-domain network architecture, completes the configuration of time slot offsets of different TS data streams in the aggregation node, and then performs corresponding transmission operations in the subsequent cross-domain aggregation transmission of the TS data stream.

[0103] In an embodiment of the present invention, by setting an intermediate node in the TSN network data plane and a convergence node in the DIP network data plane, the cross-domain aggregation configuration only needs to occur between the above-mentioned nodes without affecting the transmission protocol for TS data streams in each TSN network data plane and DIP network data plane, thereby reducing the impact of the cross-domain aggregation configuration on the overall data transmission of the cross-domain network.

[0104] It is understandable that there is no obvious order relationship between S212 and S213. In the embodiment of the present invention, only the order of S212-S213 is taken as an example, and the specific execution order is not limited in the embodiment of the present invention.

[0105] Exemplarily, the embodiment of the present invention further provides a training method for an aggregate neural network model, which can generate n size The training sample set D1 of input feature vector samples is generated at the same time. size The test sample set D2 of input feature vector samples, the samples in D1 and D2 are different, and each input feature vector sample contains the maximum utility value u max , the maximum packet length value w max , the maximum queue length value c max , the queue length value is determined to be c, and there are M data packet length values ​​w i and the packet utility value u i , where i = 0, 1,…, M-1.

[0106] After the training sample set and the test sample set are determined, the specific training steps are as follows: (a) Randomly initialize two initial aggregation neural network models with the same parameters, denoted as TF1 and TF2. (b) Calculate the sample s in the training sample set D1 through TF1. j , output the logarithmic probability and P of the TS data stream index that meets the output constraint conditions jand utility U j (TF1), where s j ∈D1, there is a corresponding relationship between the logarithmic probability and the data stream index; similarly, the sample s in the training sample set D1 is calculated by TF2 j , output the utility U that meets the output constraint conditions j (TF2); The loss function is used to iterate the connection weights in TF1 through the stochastic gradient algorithm, and then re-sample online to generate a new training sample set D1. The above iteration for TF1 is repeated until the preset number of iterations is reached.

[0107] (c) After completing a complete iterative training for TF1, n test The set of test sample sets D2 is determined as Calculate the average utility U(TF1) and U(TF2) of TF1 and TF2 in C respectively, where the average utility is calculated as follows: If U(TF1)>U(TF2), the connection weights of TF2 are updated to be the same as those of TF1. If U(TF1) is greater than the preset expected value or the training for TF1 has reached the maximum number of iterations, training is stopped and TF1 is determined as the aggregate neural network model; otherwise, the process returns to re-execute operations (b) and (c) until training is stopped.

[0108] Furthermore, after completing the cross-domain transmission of each allowed aggregation TS data stream in the cross-domain network, it also includes: displaying data packet feature information corresponding to each allowed aggregation TS data stream.

[0109] Specifically, after completing the transmission of each allowed aggregated TS data stream, the data packet feature information corresponding to each allowed aggregated TS data stream can be sent to the data module for external display through the SDN controller, displaying the TSN network data platform currently performing cross-domain aggregated data transmission and the utility brought by the overall transmission, so that users can clearly understand which TSN network data plane the TS data stream that finally completes the cross-domain aggregated transmission belongs to, as well as the utility of the transmission data stream, thereby enhancing intuitiveness and interactivity.

[0110] The technical solution of this embodiment determines the network status by obtaining the queue length and maximum queue length values ​​of the DIP network data plane, and determines the data packet information in the cross-domain network by obtaining the maximum utility value and maximum packet length value of the cross-domain network, as well as the data packet feature information corresponding to each TSN network data plane. The obtained data packet information and network status information are first verified for data validity, and input feature vectors for input into the aggregation neural network model are generated only when the data is valid. This ensures that the aggregation result output by the aggregation neural network model is an aggregation mechanism that is applicable to the current scenario, thereby improving the execution stability of cross-domain aggregation transmission of TS data streams. When constructing the aggregation neural network model, output constraints are set based on network status information so that the sum of the outputs of the aggregation neural network model meets the load requirements of the cross-domain network, assisting in determining the TS data streams that can be aggregated. A cross-domain aggregation configuration table is then generated based on the aggregation results output by the aggregation neural network model, and the cross-domain aggregation configuration table is distributed to the intermediate nodes of each TSN network data plane and the aggregation node of the DIP network data plane that exists independently of other DIP routers. This allows the TS data streams that can be aggregated to achieve cross-domain aggregate transmission in the cross-domain network. After the above transmission is completed, the data packet feature information corresponding to each TS data stream that can be aggregated can be displayed externally, so that users can clearly understand which TSN network data plane the TS data stream that finally completes the cross-domain aggregate transmission belongs to, as well as the utility of the transmitted data stream. Before the aggregation neural network model outputs the aggregation result, each TS data stream is sorted based on the probability of occurrence to ensure that the output utility of the TS data stream that can be aggregated corresponding to the aggregation result is optimal. This ensures that the TS data stream does not lose its time-sensitive characteristics after cross-domain transmission, improves the efficiency of TS data stream cross-domain aggregation, and allows the cross-domain aggregate transmission results to be intuitively displayed externally.

[0111] Example 3

[0112] Figure 4 A schematic diagram of the structure of a time-sensitive data stream cross-domain aggregation system provided in Example 3 of the present invention is shown as follows: Figure 4 As shown, the time-sensitive data flow cross-domain aggregation system 3 includes an SDN controller 31, at least one TSN network data plane 32 and a DIP network data plane 33.

[0113] The SDN controller includes at least a user controller 311, an aggregator 312, and a network configurator 313. Each TSN network data plane 32 includes multiple TSN switches 321 and a server 322, with one TSN switch 321 serving as an intermediate node 3211 for external data transmission and reception. The DIP network data plane 33 includes multiple DIP routers 331, with one DIP router 331 serving as an aggregation node 3311 for external data transmission and reception. In this embodiment of the present invention, N TSN network data planes 32 are used as an example, with each TSN network data plane 32 including two TSN switches 321 and the DIP network data plane 33 including four DIP routers 331.

[0114] The user controller 311 is used to obtain data packet information and network status information in the cross-domain network, and construct an input feature vector based on the data packet information and network status information, and send the input feature vector to the aggregator 312; wherein the cross-domain network is composed of at least one TSN network data plane 32 and one DIP network data plane 33.

[0115] The aggregator 312 is an aggregation neural network model obtained by training input feature vector samples with output data stream index labels extracted based on an online sampling mechanism, and is used to generate aggregation results based on the input feature vectors and send the aggregation results to the network configurator 313.

[0116] The network configurator 313 is used to generate a cross-domain aggregation configuration table according to the aggregation result, and send the cross-domain aggregation configuration table to the intermediate node 3211 of each TSN network data plane 32 and the aggregation node 3311 of the DIP network data plane 33 respectively.

[0117] Each TSN network data plane 32 and DIP network data plane 33 is used to complete the cross-domain transmission of the aggregated time-sensitive data stream corresponding to the cross-domain aggregation configuration table based on the cross-domain aggregation configuration table.

[0118] It can be understood that the entities in the time-sensitive data flow cross-domain aggregation system 3 all support their traditional functions, for example: the server 322 provides data flow time slot offset, the same-domain TSN switch 321 supports time synchronization, the TSN switch 321 supports cyclic queuing forwarding, the DIP router 331 supports traditional time slot mapping function, the SDN controller 31 provides the function of collecting network information and distributing scheduling signaling, routers and switches support wired connection capabilities, nodes (TSN switches and DIP routers) have computing resources and the ability to execute calculation results, and TSN switches and DIP routers have deterministic time slot mapping functions (see IEEE standard [IEEE Std 802.1Qch-2017] and IEEE draft standard [draft-qiang-detnet-large-scale-detnet-05]).

[0119] For example, it is assumed that when the SDN controller extracts samples and trains the aggregated neural network model according to the online sampling mechanism, the dimension of the model linear embedding layer is n1=128, and the aggregated neural network model obtained after training is recorded as TF. Assume that the time-sensitive data flow cross-domain aggregation system includes five TSN network data planes, where the intermediate nodes of each TSN network data plane are recorded as v1, v2, v3, v4 and v5 respectively. During training, v1 to v5 randomly generate a total of 10 TS data streams participating in the aggregation, and the TS data stream utilities are randomly and uniformly sampled from {1, 2, 3, ..., 9}, and the time-sensitive data flow packet length is set to w i =1, the queue length of the aggregation node v6 in the DIP network data plane is set to c=3, and the maximum packet length value obtained based on the above setting is u max =9, the maximum utility value is W max =1, the maximum queue length is c max =3. Based on the above settings, TF training can be completed. TF training can be implemented through a competitive mode or through any other neural network model training method, which is not limited in this embodiment of the present invention.

[0120] The following two scenarios test the cross-domain aggregation capability of TF for TS data streams after training:

[0121] 1) Aggregation capability test when network status information is the same as the training sample set

[0122] During the test, 20 test samples were generated. In each sample, v1, v2, v3, v4, and v5 generated 3, 3, 3, 1, and 0 TS data streams, respectively (a total of 10). The utilities of the 10 TS data streams were randomly and uniformly sampled from {1, 2, 3, ..., 9}. The TS data stream packet length was set to 1 unit, and the queue length of the DIP sink node v6 was c = 3. The utilities of the 20 test samples were:

[0123] u 1 =[7, 9, 1, 2, 9, 7, 4, 3, 6, 4] (the superscript 1 indicates the first test sample, where the data means the utilities of TS data streams 0-9 are u0=7, ..., u9=4 respectively);

[0124] u 2 =[1,8,7,3,2,9,6,7,5,7];

[0125] u 3 =[9,5,5,7,2,9,2,1,8,3];

[0126] u 4 =[5,1,8,3,8,6,6,5,7,3];

[0127] u 5 =[9,5,9,9,6,7,9,8,4,8];

[0128] u 6 =[5,1,1,5,3,4,4,5,9,2];

[0129] u 7 =[2,2,8,2,3,9,2,2,7,9];

[0130] u 8 =[3,7,5,2,5,8,1,9,9,8];

[0131] u 9 =[4,3,4,7,5,3,8,4,4,9];

[0132] u 10 =[2,4,5,5,6,7,9,8,8,6];

[0133] u 11 =[7,4,3,4,3,4,8,1,9,6];

[0134] u 12 =[2,6,4,5,6,6,4,6,6,5];

[0135] u 13=[6,6,2,6,9,4,2,2,7,6];

[0136] u 14 =[5,7,2,3,3,9,8,5,3,2];

[0137] u 15 =[4,4,3,6,4,2,9,9,5,8];

[0138] u 16 =[4,8,7,9,2,6,6,8,2,9];

[0139] u 17 =[9,1,4,8,2,2,3,2,6,1];

[0140] u 18 =[4,4,6,8,7,7,2,6,9,1];

[0141] u 19 =[6,9,2,6,4,6,2,7,2,8];

[0142] u 20 =[6,9,8,1,9,6,4,9,5,7].

[0143] Based on the above 20 test samples, the aggregated results of TF calculation after training can be expressed as: 1 =[1, 4, 0] (superscript 1 indicates the first test sample, [1, 4, 0] indicates that the aggregated TS data stream indexes are 1, 4, and 0 respectively), f 2 =[5, 1, 2], f 3 =[0,5,8],f 4 =[2, 4, 8], f 5 =[0, 2, 3], f 6 =[8,0,3],f 7 =[5,9,2],f 8 =[7,8,5],f 9 =[9,6,3],f 10 =[6,7,8],f 11 =[8,6,0],f 12 =[1, 4, 5], f 13 =[4,8,0],f 14 =[5, 6, 1], f 15 =[6, 7, 9], f 16 =[3,9,1],f 17 =[0,3,8],f 18 =[8,3,4],f 19=[1,9,7],f 20 =[1, 4, 7]. It can be verified that the total utility of the output data stream is the optimal aggregate utility of the corresponding test sample, and the output result satisfies the load constraint of no more than 3 (c=3) data streams.

[0144] 2) Aggregation capability test when network status information and training sample set are different

[0145] During the test, 20 test samples were generated. In each sample, v1, v2, v3, v4, and v5 generated 2, 2, 2, 2, and 2 TS data streams, respectively (a total of 10). The packet utilities and packet lengths in these samples were the same as those in test case 1, but the queue length at DIP sink node v6 was c = 2. It is understandable that the queue length indicates a network state, and the network state in this test is completely different from that in the training scenario.

[0146] Based on the above 20 test samples, the aggregated results of TF calculation after training can be expressed as: 1 =[1, 4] (superscript 1 indicates the first test sample, [1, 4] indicates that the aggregated data stream indexes are 1 and 4 respectively), f 2 =[5,1],f 3 =[0,5],f 4 =[2,4],f 5 =[0,2],f 6 =[8,0],f 7 =[5,9],f 8 =[7,8],f 9 =[9,6],f 10 =[6,7],f 11 =[8,6],f 12 =[1,4],f 13 =[4,8],f 14 =[5,6],f 15 =[6,7],f 16 =[3,9],f 17 =[0,3],f 18 =[8,3],f 19 =[1,9],f 20 = [1, 4]. It can be verified that the total utility of the output data stream is the optimal aggregated utility for the corresponding test sample, and the output result satisfies the load constraint of no more than two data streams. Since the network state during testing is completely different from that during training, TS data stream aggregation with optimal utility and satisfying the load constraint can still be achieved. Therefore, it can be considered that the trained aggregator has the ability to generalize and identify unknown network conditions.

[0147] The time-sensitive data stream cross-domain aggregation system provided by the embodiment of the present invention can execute the time-sensitive data stream cross-domain aggregation method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0148] Example 4

[0149] Figure 5 A schematic structural diagram of a time-sensitive data stream cross-domain aggregation device provided for embodiment four of the present invention. The time-sensitive data stream cross-domain aggregation device 40 may be an electronic device intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are provided as examples only and are not intended to limit the implementation of the present invention described and / or required herein.

[0150] like Figure 5 As shown, the time-sensitive data stream cross-domain aggregation device 40 includes at least one processor 41, and a memory connected to the at least one processor 41, such as a read-only memory (ROM) 42, a random access memory (RAM) 43, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 41 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 42 or the computer program loaded from the storage unit 48 to the random access memory (RAM) 43. Various programs and data required for the operation of the time-sensitive data stream cross-domain aggregation device 40 can also be stored in the RAM 43. The processor 41, ROM 42 and RAM 43 are connected to each other via a bus 44. An input / output (I / O) interface 45 is also connected to the bus 44.

[0151] Multiple components in the time-sensitive data flow cross-domain aggregation device 40 are connected to an I / O interface 45, including: an input unit 46, such as a keyboard, a mouse, etc.; an output unit 47, such as various types of displays, speakers, etc.; a storage unit 48, such as a magnetic disk, an optical disk, etc.; and a communication unit 49, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 49 allows the time-sensitive data flow cross-domain aggregation device 40 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0152] Processor 41 can be various general-purpose and / or specialized processing components with processing and computing capabilities. Some examples of processor 41 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. Processor 41 performs the various methods and processes described above, such as the method for cross-domain aggregation of time-sensitive data streams.

[0153] In some embodiments, the time-sensitive data stream cross-domain aggregation method can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 48. In some embodiments, part or all of the computer program can be loaded and / or installed on the time-sensitive data stream cross-domain aggregation device 40 via the ROM 42 and / or the communication unit 49. When the computer program is loaded into the RAM 43 and executed by the processor 41, one or more steps of the time-sensitive data stream cross-domain aggregation method described above can be performed. Alternatively, in other embodiments, the processor 41 can be configured to execute the time-sensitive data stream cross-domain aggregation method by any other appropriate means (for example, by means of firmware).

[0154] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0155] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the computer program is executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program may be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0156] In the context of the present invention, computer-readable storage media can be tangible media that can contain or store a computer program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Computer-readable storage media can include but are not limited to electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, computer-readable storage media can be machine-readable signal media. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0157] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0158] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0159] A computing system may include clients and servers. The clients and servers are typically remote from each other and typically interact via a communication network. This client-server relationship arises through computer programs running on the respective computers, creating a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host. This server is a hosting product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosting and VPS services.

[0160] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This is not limited herein.

[0161] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A method for cross-domain aggregation of time-sensitive data streams, characterized in that: include: Acquire data packet information and network status information in the cross-domain network, and construct an input feature vector based on the data packet information and the network status information; Inputting the input feature vector into an aggregation neural network model, and generating a cross-domain aggregation configuration table according to an aggregation result output by the aggregation neural network model; wherein the cross-domain aggregation configuration table includes at least one time slot offset information allowing the aggregated time-sensitive data stream to be transmitted across domains; Sending the cross-domain aggregation configuration table to the communication nodes of each network data plane in the cross-domain network, so as to complete the cross-domain aggregation transmission of each of the time-sensitive data flows allowed to be aggregated in the cross-domain network; The cross-domain network includes at least one time-sensitive network (TSN) network data plane and a deterministic internet protocol (DIP) network data plane; the aggregated neural network model is trained based on input feature vector samples extracted by an online sampling mechanism and a competitive mechanism; The data packet information is parameter information of a data packet composed of a time-sensitive TS data stream that needs to be transmitted across domains in the data plane of each TSN network in the cross-domain network; The network status information is parameter information that is determined based on current demand and the load capacity limit of the DIP network data plane in the cross-domain network and is used to characterize the ability of the DIP network to receive TS data streams; The output multi-head attention layer of the aggregated neural network model includes output constraints determined according to the network state information; The step of inputting the input feature vector into the aggregate neural network model includes: Inputting the input feature vector into the aggregate neural network model for dimension mapping, feature extraction and integration, and determining the occurrence probability of the time-sensitive data flow to be transmitted corresponding to each of the TSN network data planes in the input feature vector; Determine at least one time-sensitive data flow that is allowed to be aggregated according to each of the occurrence probabilities, a data packet length value corresponding to each of the occurrence probabilities in the input feature vector, and the output restriction condition; A set of data stream indexes corresponding to the time-sensitive data streams allowed to be aggregated is determined as an aggregation result.

2. The method according to claim 1, characterized in that The obtaining of data packet information and network status information in the cross-domain network includes: Obtaining a queue length value and a maximum queue length value of the DIP network data plane in the inter-domain network, and determining the queue length value and the maximum queue length value as network status information; Obtaining the maximum utility value and maximum packet length value of the cross-domain network, and data packet feature information corresponding to each of the TSN network data planes; Determining each of the data packet characteristic information, the maximum utility value, and the maximum packet length value as data packet information; The data packet characteristic information includes a data packet length value and a data packet utility value of a data packet corresponding to a time-sensitive data stream to be transmitted in the TSN network data plane.

3. The method according to claim 2, characterized in that The constructing an input feature vector according to the data packet information and the network status information includes: Determining data validity of each of the data packet characteristic information according to the maximum utility value, the maximum packet length value, and the maximum queue length value; Determining data validity of the queue length value based on the maximum queue length value; If each of the data packet characteristic information and the queue length value is valid, constructing a single data flow characteristic vector corresponding to each of the time-sensitive data flows to be transmitted according to each of the data packet characteristic information, the queue length value, the maximum utility value, the maximum packet length value, and the maximum queue length value; A sequence matrix composed of the single data stream feature vectors is determined as an input feature vector.

4. The method according to claim 1, wherein Determining at least one time-sensitive data flow allowed to be aggregated according to each of the occurrence probabilities, a data packet length value corresponding to each of the occurrence probabilities in the input feature vector, and the output restriction condition, comprising: Reordering the data packet length values ​​corresponding to each occurrence probability in the input feature vector in descending order of occurrence probability; Adding the length values ​​of the reordered data packets in sequence, and when the sum of the sums is greater than the queue length value of the DIP network data plane for the first time, determining the occurrence probability corresponding to the summed data packet length value as the critical occurrence probability; The to-be-transmitted time-sensitive data flow with an occurrence probability greater than the critical occurrence probability is determined as a time-sensitive data flow allowed to be aggregated.

5. The method according to claim 1, wherein The sending of the cross-domain aggregation configuration table to the communication nodes of each network data plane in the cross-domain network includes: Send the cross-domain aggregation configuration table to the intermediate nodes of each TSN network data plane; The cross-domain aggregation configuration table is sent to the aggregation node of the DIP network data plane.

6. The method according to claim 1, characterized in that After completing the cross-domain aggregation transmission of each of the time-sensitive data flows that are allowed to be aggregated in the cross-domain network, the method further includes: Display the data packet feature information corresponding to each of the time-sensitive data flows that are allowed to be aggregated.

7. A time-sensitive data stream cross-domain aggregation system, characterized in that: include: A software-defined networking controller, at least one time-sensitive networking (TSN) network data plane and one deterministic internet protocol (DIP) network data plane; The software-defined networking controller includes at least a user controller, an aggregator and a network configurator; The user controller is used to obtain data packet information and network status information in the cross-domain network, construct an input feature vector according to the data packet information and the network status information, and send the input feature vector to the aggregator; The aggregator is an aggregation neural network model obtained by training input feature vector samples with output data stream index labels extracted based on an online sampling mechanism, and is used to generate an aggregation result based on the input feature vector and send the aggregation result to the network configurator; The network configurator is used to generate a cross-domain aggregation configuration table according to the aggregation result, and send the cross-domain aggregation configuration table to the intermediate nodes of each TSN network data plane and the aggregation node of the DIP network data plane respectively; Each of the TSN network data plane and the DIP network data plane is used to complete the cross-domain transmission of the aggregated time-sensitive data stream corresponding to the cross-domain aggregation configuration table based on the cross-domain aggregation configuration table; The data packet information is parameter information of a data packet composed of a time-sensitive TS data stream that needs to be transmitted across domains in the data plane of each TSN network in the cross-domain network; The network status information is parameter information that is determined based on current demand and the load capacity limit of the DIP network data plane in the cross-domain network and is used to characterize the ability of the DIP network to receive TS data streams; The output multi-head attention layer of the aggregated neural network model includes output constraints determined according to the network state information; The step of sending the input feature vector to the aggregator includes: Inputting the input feature vector into the aggregate neural network model for dimension mapping, feature extraction and integration, and determining the occurrence probability of the time-sensitive data flow to be transmitted corresponding to each of the TSN network data planes in the input feature vector; Determine at least one time-sensitive data flow that is allowed to be aggregated according to each of the occurrence probabilities, a data packet length value corresponding to each of the occurrence probabilities in the input feature vector, and the output restriction condition; A set of data stream indexes corresponding to the time-sensitive data streams allowed to be aggregated is determined as an aggregation result.

8. A time-sensitive data stream cross-domain aggregation device, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the time-sensitive data stream cross-domain aggregation method according to any one of claims 1 to 6.

9. A storage medium containing computer-executable instructions, characterized in that: When executed by a computer processor, the computer executable instructions are used to perform the time-sensitive data stream cross-domain aggregation method as described in any one of claims 1 to 6.

Images