Time sensitive network traffic scheduling method and device

Abstract

The application provides a time-sensitive network traffic scheduling method and device, which comprises: updating the basic action probability mask of each to-be-scheduled time-triggered flow with the current scheduled time-triggered flow set and the remaining to-be-scheduled time-triggered flow set, obtaining the current action probability mask, and selecting the next to-be-scheduled time-triggered flow to join the current scheduling sequence based on the mask, generating an initial scheduling sequence and evaluating it to obtain a time slot scheduling result, and generating a conflict heat map in the link and time window dimensions based on the result; identifying high-conflict time-triggered flows and associated flows based on the conflict heat map, generating neighborhood candidate solutions, and searching and optimizing each neighborhood candidate solution to output a target scheduling sequence; and generating a gate control list according to the target scheduling sequence and the time slot scheduling result of the target scheduling sequence and issuing it to the time-sensitive network. The application can improve the solving efficiency, scheduling quality and convergence stability of large-scale time-triggered flow scheduling.

Description

Technical Field

[0001] This application relates to the field of network communication technology, and in particular to a time-sensitive network traffic scheduling method and device. Background Technology

[0002] Time-Sensitive Networking (TSN) is a set of deterministic communication technologies built on top of standard Ethernet. It provides low-latency, low-jitter, and highly reliable data transmission capabilities for scenarios such as industrial control, automotive communication, avionics, and smart manufacturing through mechanisms such as time synchronization, traffic shaping, gating scheduling, and resource management. In TSN networks, the Time-Aware Shaper (TAS) defined in IEEE 802.1Qbv controls the opening or closing of different queues within a specified time window through a Gate Control List (GCL), ensuring that time-triggered flows (TT flows) can be transmitted deterministically according to the predetermined time window. To generate a GCL that satisfies the constraints of collision-free, low-latency, and deadline-based transmission, it is necessary to determine the transmission time window of each TT flow on each output port it passes through. Since different TT flows may share links, share output ports, or compete for transmission resources in adjacent time windows, the scheduling order of TT flows directly affects the subsequent time slot allocation results and the final GCL quality.

[0003] Existing TSN scheduling methods can be broadly categorized into exact solution methods and heuristic optimization methods. Exact solution methods are typically based on integer linear programming, constrained programming, or satisfiability modular theory, achieving good or even optimal results for small-scale problems. However, as the number of TT flows and the network size increase, the solution time rises rapidly, making it difficult to adapt to the rapid reconfiguration requirements in engineering scenarios. Heuristic optimization methods, such as genetic algorithms, particle swarm optimization, simulated annealing, and tabu search, offer better computational efficiency, but their search performance usually depends on the quality of the initial solution and the neighborhood generation method. In summary, existing technologies lack guidance from multi-dimensional scheduling-sensitive features in the initial flow sorting, which can easily lead to unreasonable sorting of critical flows, high-conflict flows, or low-relaxation flows; the Q-Learning action selection process lacks prior constraints on the scheduling features of TT flows, which can easily result in invalid or low-quality explorations; the tabu search neighborhood generation process lacks time slot scheduling conflict feedback, resulting in a scattered search direction and insufficient effectiveness of candidate solutions; and the Q-Learning initial sorting stage and the tabu search optimization stage lack collaborative information, making it difficult to form a closed-loop optimization oriented towards actual conflict locations. Summary of the Invention

[0004] In view of this, embodiments of this application provide a time-sensitive network traffic scheduling method and device to eliminate or improve one or more defects existing in the prior art.

[0005] One aspect of this application provides a time-sensitive network traffic scheduling method, the method comprising the following steps: The basic action probability mask of each scheduled time trigger flow is updated using the current set of scheduled time trigger flows and the set of remaining scheduled time trigger flows to obtain the corresponding current action probability mask. Based on the current action probability mask, the next scheduled time trigger flow is selected to be added to the current scheduling sequence to generate the initial scheduling sequence. Based on the initial scheduling sequence, a time slot scheduling evaluation is performed to obtain the corresponding time slot scheduling result, and a conflict heatmap of the link and time window dimensions is generated based on the time slot scheduling result. Based on the conflict heatmap of the link and time window dimensions, high-conflict time-triggered flows and associated flows are identified, and neighborhood candidate solutions are generated based on the high-conflict time-triggered flows and associated flows. The preset induced tabu search algorithm is used to search and optimize each neighborhood candidate solution, and the corresponding target scheduling sequence is output. Based on the target scheduling sequence and the corresponding time slot scheduling result, a gating control list is generated and sent to the time-sensitive network.

[0006] In some embodiments of this application, the generation process of the basic action probability mask includes: A directed network topology graph is constructed based on the network topology information obtained in the time-sensitive network; and the transmission path of each time-triggered stream in the obtained time-triggered stream set is determined according to the attribute information of each time-triggered stream to be scheduled and the directed network topology graph. Based on the directed network topology and the transmission paths of each scheduled time-triggered stream, the scheduling-sensitive features of each scheduled time-triggered stream are extracted. The scheduling sensitivity features of each scheduled time-triggered flow are normalized, and a scheduling weight vector is generated based on the normalized scheduling sensitivity features. The basic action probability mask is generated based on the scheduling weight vector.

[0007] In some embodiments of this application, the scheduling-sensitive feature includes: Path overlap is used to characterize the degree of resource contention between the current time-triggered stream and other time-triggered streams on a shared link or shared output port; Temporal slack is used to characterize the time slack that the current time-triggered flow has to satisfy the deadline constraint; Link resource occupancy is used to characterize the intensity of link transmission resource occupancy by the current time-triggered stream; Potential conflict level is used to characterize the likelihood that the current time-triggered stream will cause a transmission window conflict during subsequent time slot allocation; The scheduling weight vector is positively correlated with the normalized path overlap, link resource occupancy, and potential conflict, and negatively correlated with the normalized temporal relaxation.

[0008] In some embodiments of this application, the step of updating the basic action probability mask of each pre-set time-triggered time-triggered stream with the currently scheduled set of time-triggered streams and the remaining set of time-triggered streams to obtain the corresponding current action probability mask, and selecting the next time-triggered stream to be added to the scheduling sequence based on the current action probability mask to generate an initial scheduling sequence includes: Set the time trigger streams in the currently scheduled time trigger stream set to an unselectable state, and correct them according to the action probability of each time trigger stream in the remaining time trigger stream set to obtain the corresponding current action probability mask. Based on the current action probability mask, select the next time trigger stream to be scheduled as a newly added time trigger stream and add it to the scheduling sequence; Based on the current partial scheduling sequence, the preset time slot estimator is invoked to estimate the end-to-end delay increment, link conflict penalty, and deadline violation risk caused by the newly added time-triggered flow. The instant reward is calculated and obtained by combining the scheduling weight vector of the newly added time-triggered flow or the preset position benefit. The Q-value table is updated using the instant reward until all time-triggered streams are added to the scheduling sequence to obtain a trained Q-value table, and the initial scheduling sequence is obtained based on the trained Q-value table and the current action probability mask.

[0009] In some embodiments of this application, selecting the next time-triggered stream to be scheduled as a newly added time-triggered stream to the scheduling sequence based on the current action probability mask includes: In exploration mode, actions are sampled according to the probability distribution formed by the current action probability mask, and the next time trigger flow to be scheduled is selected. In the utilization mode, the Q value is fused with the current action probability mask, and the scheduled time trigger flow with the highest comprehensive score is selected as the next scheduled time trigger flow.

[0010] In some embodiments of this application, the conflict heatmap of the link and time window dimensions includes a first dimension and a second dimension; wherein, the first dimension is a link or output port identifier, the second dimension is a discrete time window identifier, and the heatmap element is a conflict intensity value; the conflict intensity value is determined based on at least one of the following: the number of time-triggered streams competing to be sent by the same link or output port in the same or adjacent time windows, the length of the overlapping time windows, the degree of deadline violation, and the waiting delay increment.

[0011] In some embodiments of this application, the identification of high-conflict time-triggered flows and associated flows based on the conflict heatmap of the link and time window dimensions includes: The dynamic conflict contribution value of each scheduled time-triggered flow is calculated and determined based on the conflict heatmap of the link and time window dimensions. If the dynamic conflict contribution value is greater than a preset threshold or is located in the top preset proportion of all time-triggered flows, then the time-triggered flow to be scheduled corresponding to the dynamic conflict contribution value is marked as a high-conflict time-triggered flow; and the time-triggered flows that share a link, share an output port, have adjacent transmission time windows, or are adjacent in sorting position with the high-conflict time-triggered flow are marked as associated flows. The dynamic conflict contribution value is expressed by the following formula: in, HC i Indicates the dynamic conflict contribution value; i Indicates the index of the time-triggered stream; e Identifier for the link or output port; E i This represents the set of links or output ports traversed by the i-th time-triggered stream; b k This represents the boundary or identifier of the k-th discrete-time window; TW i,e This indicates the transmission time window of the time-triggered stream on the link or output port; H[e,k] This indicates the conflict intensity value of the corresponding link or output port in the conflict heatmap within the time window.

[0012] In some embodiments of this application, generating neighborhood candidate solutions based on the high-collision time-triggered flow and the associated flow includes: The current scheduling sequence is transformed according to a preset transformation method, and the transformed scheduling sequence is used as a neighborhood candidate solution; wherein, the transformation method includes: Move the high-conflict time-triggered stream forward in the current scheduling sequence; The high-conflict-time trigger stream is shifted to the later part of the current scheduling sequence; Swap the positions of the high-conflict-time triggering stream and the associated stream in the current scheduling sequence; Swap the positions of the two time-triggered streams that share a critical link in the current scheduling sequence.

[0013] In some embodiments of this application, the step of using a preset induced tabu search algorithm to search and optimize each of the neighborhood candidate solutions and outputting the corresponding target scheduling sequence includes: For each of the aforementioned neighborhood candidate solutions, the time slot scheduling is re-evaluated, and the objective function value and constraint violation degree of the candidate scheduling sequence are calculated. The system uses a pre-defined taboo table to record recently executed move operations to determine whether the move operation corresponding to each neighborhood candidate solution is in a taboo state. If the move operation corresponding to the neighborhood candidate solution is taboo but the neighborhood candidate solution is better than the historical best solution, then the amnesty criterion is triggered and the neighborhood candidate solution is accepted; if there is no neighborhood candidate solution that meets the above conditions, then the neighborhood candidate solution with the best comprehensive evaluation is selected from the non-taboo neighborhood candidate solutions as the current scheduling sequence; and the taboo table and the conflict heatmap of the link and time window dimensions are updated. When the preset stopping condition is met, the search stops and the historical best scheduling sequence is output as the target scheduling sequence.

[0014] Another aspect of this application provides an electronic device including a processor and a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the time-sensitive network traffic scheduling method when executing the computer program.

[0015] A third aspect of this application provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the time-sensitive network traffic scheduling method described above.

[0016] A fourth aspect of this application provides a computer program product comprising a computer program that, when executed by a processor, implements the time-sensitive network traffic scheduling method described above.

[0017] The time-sensitive network traffic scheduling method of this application includes the following steps: updating the basic action probability mask of each scheduled time-triggered flow with the currently scheduled set of time-triggered flows and the remaining set of time-triggered flows to obtain the corresponding current action probability mask, and selecting the next scheduled time-triggered flow to be added to the current scheduling sequence based on the current action probability mask to generate an initial scheduling sequence; performing time slot scheduling evaluation based on the initial scheduling sequence to obtain the corresponding time slot scheduling result, and generating a conflict heatmap of the link and time window dimensions based on the time slot scheduling result; identifying high-conflict time-triggered flows and associated flows based on the conflict heatmap of the link and time window dimensions, generating neighborhood candidate solutions based on the high-conflict time-triggered flows and associated flows, and using a preset induced tabu search algorithm to search and optimize each neighborhood candidate solution to output the corresponding target scheduling sequence; generating a gating control list based on the target scheduling sequence and the time slot scheduling result corresponding to the target scheduling sequence and sending it to the time-sensitive network. This approach addresses the problems of unstable initial sequence quality, lack of scheduling feature constraints in reinforcement learning action selection, and lack of conflict feedback guidance in tabu search neighborhood generation in existing flow-ordering-based TSN scheduling methods. Without altering the TT flow ordering scheduling background and retaining the Q-LTS hybrid optimization framework (Q-LTS represents a hybrid scheduling framework consisting of Q-Learning initial flow ordering and tabu search local optimization), it introduces scheduling-sensitive features of TT flows, such as path overlap, temporal relaxation, link resource occupancy, and potential conflict, into the initial ordering stage. Furthermore, it feeds back the link-time window conflict distribution generated during time slot scheduling to the tabu search stage, thereby improving the solution efficiency, scheduling quality, and convergence stability of large-scale TT flow scheduling.

[0018] Additional advantages, objectives, and features of this application will be set forth in part in the description which follows, and will in part become apparent to those skilled in the art upon review of the following description, or may be learned by practice of the application. The objectives and other advantages of this application can be realized and obtained by means of the structures specifically pointed out in the specification and drawings.

[0019] Those skilled in the art will understand that the purposes and advantages that can be achieved with this application are not limited to those specifically described above, and that the above and other purposes that this application can achieve will be more clearly understood from the following detailed description. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, do not constitute a limitation thereof. The components in the drawings are not drawn to scale but are merely for illustrating the principles of this application. For ease of illustration and description of certain parts of this application, corresponding portions in the drawings may be enlarged, i.e., may appear larger relative to other components in an exemplary device actually manufactured according to this application. In the drawings: Figure 1 This is a schematic diagram of the first process of a time-sensitive network traffic scheduling method in one embodiment of this application.

[0021] Figure 2 This is a schematic diagram of a second process of a time-sensitive network traffic scheduling method in one embodiment of this application.

[0022] Figure 3 This is a schematic diagram of the third process of the time-sensitive network traffic scheduling method in one embodiment of this application.

[0023] Figure 4 This is a schematic diagram of the overall scheduling process of a Q-LTS scheduling method for time-sensitive network traffic scheduling in a specific example of this application.

[0024] Figure 5 This is a schematic diagram of the initial flow sorting process of Q-Learning driven by the action probability mask in a specific example of the time-sensitive network traffic scheduling method in this application.

[0025] Figure 6 This is a schematic diagram of the tabu search optimization process induced by the link-time window conflict heatmap in the time-sensitive network traffic scheduling method of this application, as shown in a specific example.

[0026] Figure 7 This is a schematic diagram of a time-sensitive network traffic scheduling system as a specific example in this application. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and their descriptions are used to explain this application, but are not intended to limit it.

[0028] It should also be noted that, in order to avoid obscuring this application with unnecessary details, only the structures and / or processing steps closely related to the solution according to this application are shown in the accompanying drawings, while other details that are not closely related to this application are omitted.

[0029] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0030] It should also be noted that, unless otherwise specified, the term "connection" in this article can refer not only to a direct connection, but also to an indirect connection involving an intermediary.

[0031] In the following description, embodiments of the present application will be illustrated with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts, or the same or similar steps.

[0032] It should be noted that existing TSN scheduling methods based on flow ranking typically generate a TT flow ranking first, followed by slot allocation and conflict resolution. However, random ranking, fixed priority ranking, or single-index ranking cannot simultaneously consider factors such as path overlap, timing urgency, link resource consumption, and potential conflicts. Ordinary Q-Learning action exploration lacks prior constraints on scheduling features, and ordinary tabu search neighborhood generation lacks slot conflict feedback, resulting in scattered search directions and insufficient effectiveness of candidate solutions. Therefore, a scheduling method is needed that can incorporate the scheduling-sensitive features of TT flows and slot conflict feedback into the ranking and local optimization processes. Based on this, the inventors of this application first conceived of transforming the scheduling-sensitive features of TT flows themselves into reinforcement learning action selection constraints while retaining the flow ranking scheduling framework and the Q-LTS hybrid optimization approach, and feeding back the slot scheduling conflict distribution to the tabu search neighborhood generation process. The time-sensitive network traffic scheduling method provided in this application is deployed in a centralized network configuration unit (CNC). The CNC receives TT flow service requests submitted by the centralized user configuration unit (CUC), obtains TSN network topology information, calculates the target scheduling sequence, and generates an IEEE 802.1Qbv gated control list (GCL) which is then sent to the TSN switching equipment. Compared with existing random sorting, fixed priority sorting, single rule sorting, and conventional Q-Learning and tabu search cascaded scheduling methods, this application uses scheduling-sensitive features to generate an action mask to guide the initial Q-learning sorting, improving the quality of the initial sequence; it utilizes the link-time window conflict heatmap generated by time slot scheduling to induce tabu search directional optimization, reducing invalid neighborhood searches. The overall approach forms a closed-loop optimization, reducing the risk of port conflicts and deadline violations, and improving scheduling convergence speed and deterministic transmission capability.

[0033] The following examples will provide a detailed description.

[0034] This application provides a time-sensitive network traffic scheduling method. See also... Figure 1 The method includes the following steps: Step 100: Update the basic action probability mask of each scheduled time trigger stream with the current set of scheduled time trigger streams and the set of remaining scheduled time trigger streams to obtain the corresponding current action probability mask, and select the next scheduled time trigger stream to be added to the current scheduling sequence based on the current action probability mask to generate the initial scheduling sequence. In step 100, a basic action probability mask is constructed based on the scheduling weight vector. This mask represents the prior selection tendency of each TT flow as a candidate ranking action, enabling Q-Learning to prioritize TT flows with higher scheduling sensitivity during the action selection phase. A Markov decision process model is established, using the currently scheduled TT flow set and the remaining unscheduled TT flow set as the state. The next TT flow is selected from the remaining unscheduled TT flows and added to the scheduling sequence as an action.

[0035] Step 200: Perform time slot scheduling evaluation based on the initial scheduling sequence to obtain the corresponding time slot scheduling results, and generate a conflict heatmap of the link and time window dimensions based on the time slot scheduling results; In step 200, a time slot scheduling evaluation is performed based on the initial scheduling sequence. Specifically, according to the order of TT streams in the initial scheduling sequence, a transmission time window is allocated to each output port on its transmission path for each TT stream, and port mutual exclusion constraints, link bandwidth constraints, flow conservation constraints, and deadline constraints are checked. During the evaluation process, the conflict situation of different links or output ports in different time windows is recorded, generating a conflict heatmap in the link-time window dimension. The conflict heatmap in the link-time window dimension can be represented as a two-dimensional matrix, a sparse table, or a key-value set, where one dimension is the link or output port identifier, the other dimension is the discrete time window identifier, and the matrix elements are conflict intensity values. The conflict intensity value can be comprehensively determined based on the number of TT streams competing for transmission in the same or adjacent time windows of the same link or output port, the overlap length of the transmission time windows, the degree of deadline violation, and the waiting delay increment.

[0036] Step 300: Identify high-conflict time-triggered flows and associated flows based on the conflict heatmap of the link and time window dimensions, generate neighborhood candidate solutions based on the high-conflict time-triggered flows and associated flows, and use a preset induced tabu search algorithm to search and optimize each neighborhood candidate solution, and output the corresponding target scheduling sequence. In step 300, the candidate solution for the neighborhood can be a new complete TT flow scheduling sequence formed by performing forward, backward, or swap operations on the high-conflict TT flow and its associated flows based on the current scheduling sequence. The target scheduling sequence can be the historically optimal complete TT flow scheduling sequence that is optimal in terms of objective function value and constraint satisfaction after iterative optimization through induced tabu search.

[0037] Step 400: Based on the target scheduling sequence and the time slot scheduling result corresponding to the target scheduling sequence, generate a gating control list and send it to the time-sensitive network.

[0038] In step 400, based on the target scheduling sequence and the corresponding time slot scheduling results, the start and end times of transmission for each TT stream on each output port are calculated. For the same output port, the transmission time windows of all TT streams are mapped to IEEE 802.1Qbv gating control entries in chronological order to generate a gating control list. The generated gating control list is then encapsulated into configuration data recognizable by the switching equipment and distributed to the TSN switching equipment via NETCONF, the YANG model configuration interface, or other southbound configuration protocols.

[0039] As described above, the time-sensitive network traffic scheduling method provided in this application evaluates the scheduling sensitivity features of TT flows through path overlap, temporal slack, link resource occupancy, and potential conflict. This ensures that the initial flow ranking no longer relies solely on random order or a single priority, but comprehensively reflects the impact of different TT flows on link resource contention, time constraints, and potential conflicts, thereby improving the rationality and stability of the initial scheduling sequence. Furthermore, a basic action probability mask is generated based on the scheduling weight vector, and the current action probability mask is dynamically updated during Q-Learning state transitions. This allows the reinforcement learning action selection process to filter already scheduled TT flows and prioritize candidate TT flows with higher scheduling sensitivity, thereby reducing invalid action selection and low-quality exploration, and improving the efficiency and effectiveness of Q-Learning in generating the initial sequence. Additionally, a link-time window conflict heatmap is generated during the time slot scheduling evaluation process. This transforms the link conflicts, port contention, time window overlap, and deadline violation risks generated during actual scheduling into feedback information that can be utilized by tabu search, preventing subsequent local searches from blindly revolving around random neighborhoods and instead prioritizing optimization of high-conflict areas. This application also identifies high-conflict TT flows and their associated flows based on conflict heatmaps, and generates directional neighborhood candidate solutions around these flows. This improves the correlation between the candidate solutions and the actual conflict locations, reduces the number of tabu search searches in low-value neighborhoods, and thus improves the convergence speed and search quality in the local optimization stage. The Q-Learning stage utilizes static scheduling-sensitive features to improve the quality of the initial sequence, while the tabu search stage uses dynamic conflict heatmaps to correct the scheduling sequence. These two stages form a synergistic relationship, which helps reduce the risks of port conflicts, link conflicts, and deadline violations. This application does not change the TT flow set itself, nor does it merge or split the TT flows. Instead, it enhances feature perception and conflict feedback capabilities while maintaining the flow ordering scheduling framework and the Q-LTS algorithm structure. Therefore, it is easy to deploy in a centralized network configuration unit (CNC) and is compatible with existing TSN configuration mechanisms such as the IEEE 802.1Qbv gated control list generation process, the NETCONF interface, and the YANG model configuration interface. The target scheduling sequence generated in this application can be further used to calculate the start and end times of transmission for each TT stream on different output ports, and mapped to a gated control list (GCL). Compared to scheduling methods with low initial sequence quality or no clear direction in neighborhood search, this application helps reduce GCL gating switching conflicts, improves the deterministic transmission capability of time-triggered streams, and enhances network resource utilization efficiency. In summary, this application, while retaining the stream ordering scheduling framework and the Q-LTS hybrid optimization approach, introduces multi-dimensional scheduling-sensitive features of TT streams to generate action probability masks, and utilizes time slot scheduling conflict heatmaps to induce tabu search-oriented optimization, thereby improving the solution efficiency, scheduling quality, and convergence stability of large-scale TSN scheduling.

[0040] To further reduce invalid action selection and low-quality exploration, and improve the efficiency and effectiveness of Q-Learning in generating initial sequences, a time-sensitive network traffic scheduling method is provided in this application embodiment, see [link to relevant documentation]. Figure 2 The generation process of the basic action probability mask includes: Step 010: Construct a directed network topology graph based on the network topology information obtained in the time-sensitive network; and determine the transmission path of each scheduled time-triggered stream in the obtained time-triggered stream set according to the attribute information of each scheduled time-triggered stream and the directed network topology graph. In step 010, information such as switches, terminal nodes, port connection relationships, link bandwidth, and propagation delay in the TSN network is obtained through the southbound interface to construct a directed network topology graph. The attribute information can include the source node, destination node, frame length, period, deadline, priority, and service identifier of the event-triggered flow to be scheduled. The transmission path can be determined using the shortest path algorithm, the K-shortest path algorithm, or a pre-configured path.

[0041] Step 020: Based on the directed network topology graph and the transmission paths of each scheduled time-triggered flow, extract the scheduling-sensitive features of each scheduled time-triggered flow; Step 030: Normalize the scheduling sensitivity features of each scheduled time-triggered stream, and generate a scheduling weight vector based on the normalized scheduling sensitivity features. In step 030, the scheduling weight vector can be W={w1,w2,...,w...} n}, where w i The scheduling weight of the i-th TT stream can be determined as follows: w i =a1·Norm(OR i )+a2·(1-Norm(SL i ))+a3·Norm(RU i )+a4·Norm(PC i ), where OR i For path overlap, SL i For temporal relaxation, RU i For link resource occupancy, PC i The potential conflict level is represented by a1 to a4, which are weighting coefficients.

[0042] Step 040: Generate the basic action probability mask based on the scheduling weight vector.

[0043] In step 040, the basic action probability mask can be generated from the scheduling weight vector. Specifically, the scheduling weight of each TT flow is first calculated based on path overlap, temporal slack, link resource occupancy, and potential conflict. Then, it is converted into a basic action probability mask through a normalization function or a softmax function (normalization exponential function), which is used to characterize the prior selection tendency of each TT flow as a candidate ranking action.

[0044] To further improve the rationality and stability of the initial scheduling sequence, a time-sensitive network traffic scheduling method is provided in this application embodiment, see [link to relevant documentation]. Figure 3 Step 100 includes: Step 110: Set the time trigger streams in the currently scheduled time trigger stream set to an unselectable state, and correct them according to the action probability of each time trigger stream in the remaining time trigger stream set to obtain the corresponding current action probability mask. In step 110, before each action selection, the basic action probability mask is updated statefully based on the currently scheduled time trigger stream set and the remaining unscheduled time trigger stream set to obtain the current action probability mask. The current action probability mask sets the scheduled time trigger streams to an unselectable state and corrects the action probability or action score of the remaining unscheduled time trigger streams.

[0045] Step 120: Based on the current action probability mask, select the next time trigger stream to be scheduled as a newly added time trigger stream and add it to the scheduling sequence; Step 130: Based on the current partial scheduling sequence, call the preset time slot estimator to estimate the end-to-end delay increment, link conflict penalty, and deadline violation risk caused by the newly added time-triggered flow, and calculate and obtain the instant reward by combining the scheduling weight vector of the newly added time-triggered flow or the preset position benefit. In step 130, the time slot estimator can be a lightweight scheduling simulation and constraint detection module deployed in a CNC. This module, based on the input TT flow scheduling sequence, sequentially allocates transmission time windows for each TT flow on the links or output ports it passes through, and detects port mutual exclusion constraints, link bandwidth constraints, flow conservation constraints, and deadline constraints. Simultaneously, it outputs end-to-end delay, conflict penalty, waiting delay increment, and a link-time window conflict heatmap.

[0046] Step 140: Update the Q-value table using the instant reward until all time-triggered streams are added to the scheduling sequence to obtain a trained Q-value table, and obtain the initial scheduling sequence based on the trained Q-value table and the current action probability mask.

[0047] In step 140, after multiple rounds of training, a complete initial scheduling sequence is generated based on the Q-value table obtained from the training and the current action probability mask.

[0048] In one or more embodiments of this application, before each action selection, the scheduled streams are set to be unselectable based on the set of scheduled TT streams and the set of remaining unscheduled TT streams, and the action probability or action score of the remaining TT streams is corrected; then, the next TT stream is selected to be added to the scheduling sequence based on the current action probability mask, until a complete initial scheduling sequence is formed.

[0049] To further improve the rationality and stability of the initial scheduling sequence, in a time-sensitive network traffic scheduling method provided in this application embodiment, step 120 includes: In exploration mode, actions are sampled according to the probability distribution formed by the current action probability mask, and the next time trigger flow to be scheduled is selected. In the utilization mode, the Q value is fused with the current action probability mask, and the scheduled time trigger flow with the highest comprehensive score is selected as the next scheduled time trigger flow.

[0050] In one or more embodiments of this application, the comprehensive score can be a fusion score of the Q-value and the current action probability mask. For candidate TT stream f i Its overall score can be expressed as: Among them, Q(s) t ,f i ) represents state s t Choose f i State-action value, m t (i) represents f in the current action probability mask. i The mask value, λ m Here, δ represents the masking fusion coefficient, and δ is a positive number to avoid the undefined logarithm.

[0051] To further improve the rationality and stability of the initial scheduling sequence, in a time-sensitive network traffic scheduling method provided in this application embodiment, the conflict heatmap of the link and time window dimensions includes a first dimension and a second dimension; wherein, the first dimension is the link or output port identifier, the second dimension is the discrete time window identifier, and the heatmap element is the conflict intensity value; the conflict intensity value is determined based on at least one of the following: the number of time-triggered flows competing to be sent by the same link or output port in the same or adjacent time windows, the length of the overlapping time windows, the degree of deadline violation, and the waiting delay increment.

[0052] In one or more embodiments of this application, the conflict heatmap of the link and time window dimensions can be represented as a two-dimensional structure, wherein the first dimension is the link or output port identifier, the second dimension is the discrete time window identifier, and the heatmap elements are the conflict intensity values ​​of the corresponding link or port within the corresponding time window. The conflict intensity can be determined based on factors such as the number of competing TT streams, the overlap length of the transmission time window, the degree of deadline violation, and the waiting delay increment.

[0053] To further improve the correlation between neighborhood candidate solutions and actual conflict locations, reduce the number of tabu searches in low-value neighborhoods, and thus improve the convergence speed and search quality in the local optimization stage, in a time-sensitive network traffic scheduling method provided in this application embodiment, the identification of high-conflict time-triggered flows and associated flows based on the conflict heatmap of the link and time window dimensions includes: Step 310: Calculate and determine the dynamic conflict contribution value of each scheduled time-triggered flow based on the conflict heatmap of the link and time window dimensions; In step 310, the dynamic conflict contribution value is expressed by the following formula: in, HC i Indicates the dynamic conflict contribution value; i Indicates the index of the time-triggered stream; e Identifier for the link or output port; E i This represents the set of links or output ports traversed by the i-th time-triggered stream; b k This represents the boundary or identifier of the k-th discrete-time window; TW i,e This indicates the transmission time window of the time-triggered stream on the link or output port; H[e,k] This indicates the conflict intensity value of the corresponding link or output port in the conflict heatmap within the time window. This indicates that the k-th discrete time window overlaps with the sending time window of this stream; only overlapping time windows are included in the summation; non-overlapping time windows are not counted.

[0054] Step 320: If the dynamic conflict contribution value is greater than a preset threshold or is in the top preset proportion of all time-triggered flows, then the time-triggered flow to be scheduled corresponding to the dynamic conflict contribution value is marked as a high-conflict time-triggered flow; the time-triggered flows that share a link, share an output port, have adjacent transmission time windows, or are adjacent in sorting position with the high-conflict time-triggered flow are marked as associated flows; In step 320, the preset threshold can be a fixed conflict contribution threshold, or it can be adaptively determined based on the dynamic conflict contribution values ​​of all TT flows, such as using the mean plus standard deviation multiple, quantile threshold, or historical experience threshold. The top preset proportion among all time-triggered flows can be understood as selecting the top ρ proportion of TT flows after sorting all TT flows from high to low according to their dynamic conflict contribution values, where ρ can be set according to the network size and search budget, for example, 5% to 30%.

[0055] To further improve the correlation between neighborhood candidate solutions and actual conflict locations, reduce the number of tabu search searches in low-value neighborhoods, and thus improve the convergence speed and search quality in the local optimization stage, a time-sensitive network traffic scheduling method provided in this application embodiment includes generating neighborhood candidate solutions based on the high-conflict time-triggered flow and the associated flow, comprising: Step 330: Transform the current scheduling sequence based on a preset transformation method, and use the transformed scheduling sequence as a neighborhood candidate solution; wherein, the transformation method includes: Move the high-conflict time-triggered stream forward in the current scheduling sequence; The high-conflict-time trigger stream is shifted to the later part of the current scheduling sequence; Swap the positions of the high-conflict-time triggering stream and the associated stream in the current scheduling sequence; Swap the positions of the two time-triggered streams that share a critical link in the current scheduling sequence.

[0056] In one or more embodiments of this application, when generating neighborhood candidate solutions using tabu search, directional neighborhood candidate solutions are preferentially generated around high-collision TT flows and their associated flows. Directional neighborhood operations include moving high-collision TT flows forward, moving high-collision TT flows backward, swapping the positions of high-collision TT flows and associated flows, and swapping the positions of two TT flows sharing a critical link. Neighborhood operations involving high-collision flows or associated flows can be given higher sampling priority to improve the targeting of local searches. If necessary, a small number of ordinary swap operations can also be retained to maintain search diversity.

[0057] To further improve the correlation between neighborhood candidate solutions and actual conflict locations, reduce the number of tabu search searches in low-value neighborhoods, and thus improve the convergence speed and search quality in the local optimization stage, a time-sensitive network traffic scheduling method provided in this application embodiment includes the following step: using a preset induced tabu search algorithm to search and optimize each of the neighborhood candidate solutions and outputting the corresponding target scheduling sequence. Step 340: Re-evaluate the time slot scheduling for each of the neighborhood candidate solutions, and calculate the objective function value and constraint violation degree of the candidate scheduling sequence; In step 340, for each candidate solution, the time slot scheduling evaluation is re-executed to calculate the objective function value and constraint violation degree. The objective function value may include indicators such as the total end-to-end delay of all network TT flows, the maximum end-to-end delay, the deadline violation penalty, the link conflict penalty, and the number of GCL entries. The objective function can be a comprehensive evaluation function, expressed as follows: in, This represents the total end-to-end latency of the entire network's TT streams. Indicates the maximum end-to-end delay. This indicates the penalty for violating the deadline. Indicates a penalty for link or output port conflicts. This indicates the penalty for the number of GCL entries or the number of gating switches. to These are the weighting coefficients of the objective function.

[0058] Step 350: Use a preset taboo table for recording recently executed move operations to determine whether the move operation corresponding to each neighborhood candidate solution is in a taboo state; Step 360: If the move operation corresponding to the neighborhood candidate solution is in a taboo state but the neighborhood candidate solution is better than the historical best solution, then the amnesty criterion is triggered and the neighborhood candidate solution is accepted; if there is no neighborhood candidate solution that meets the above conditions, then the neighborhood candidate solution with the best comprehensive evaluation is selected from the non-taboo neighborhood candidate solutions as the current scheduling sequence; and the taboo table and the conflict heatmap of the link and time window dimensions are updated. In step 360, a taboo table is maintained to record recently executed move operations, preventing the search from oscillating repeatedly around the same scheduling sequence. When a candidate solution is taboo but better than the historical best solution, the amnesty criterion is triggered and the candidate solution is accepted; otherwise, the candidate solution with the best comprehensive evaluation is selected from the non-taboo candidate solutions as the current scheduling sequence.

[0059] Step 370: When the preset stopping condition is met, stop the search and output the historical best scheduling sequence as the target scheduling sequence.

[0060] In step 370, when the maximum number of iterations is reached, the objective function converges, no improvement is achieved for several consecutive rounds, or the preset scheduling quality threshold is met, the tabu search stops, and the historical best scheduling sequence is output as the target scheduling sequence.

[0061] In one or more embodiments of this application, the target scheduling sequence can be generated iteratively by induced tabu search. Using the initial scheduling sequence as the initial solution, directional neighborhood candidate solutions are generated based on the conflict heatmap. The objective function value and constraint violation degree of each candidate solution are calculated, and the current solution and historical best solution are updated in conjunction with the tabu list and the amnesty criterion. After the termination condition is met, the historical best scheduling sequence is output as the target scheduling sequence.

[0062] In a specific example of the time-sensitive network traffic scheduling method of this application, see [link to example]. Figure 4 The method includes the following steps: Step 1: The CNC acquires the TSN network topology information. This topology information includes terminal nodes, TSN switches, switch port connections, link bandwidth, link propagation delay, and switching node processing delay. Based on this information, the CNC constructs a directed network topology model, as shown in the following equation: in, Represents a set of nodes. This represents the link set. The CNC receives the TT stream service request submitted by the CUC and constructs the TT stream set, as shown in the following formula: For any TT stream It records the source node, destination node, frame length, period, deadline, priority, and service identifier, and determines its transmission path based on the network topology. The transmission path can be the shortest path, K-shortest path, or a pre-configured path.

[0063] Step two, the CNC performs a scheduling sensitivity feature evaluation on each TT flow. The scheduling sensitivity features include path overlap. Temporal relaxation Link resource occupancy and potential conflict level .

[0064] For TT stream Let the set of links it traverses be denoted as . The output port set is The normalization operator is , To prevent positive numbers with a denominator of zero.

[0065] Path overlap is used to characterize TT flow. The degree of resource contention with other TT streams on a shared link or shared output port can be determined according to the following relationship: Temporal relaxation is used to characterize TT flow. The time margin required to meet the deadline constraint. First, calculate the minimum end-to-end transmission time under contention-free conditions, as shown in the following formula: The temporal relaxation is then calculated, as shown in the following formula: Link resource occupancy is used to characterize TT flow. The intensity of link transmission resource occupancy can be determined according to the following relationship: Potential conflict degree is used to characterize TT flow The likelihood of transmission window conflicts occurring during subsequent time slot allocation can be determined according to the following relationship: in, Indicates TT stream Frame length, Indicates period, Indicates the deadline. Indicates link bandwidth, Indicates the link propagation delay. Indicates the number of exchange nodes traversed. Indicates node processing latency. Indicates TT stream In the link The set of feasible sending time windows.

[0066] Step 3: The CNC generates a scheduling weight vector based on scheduling sensitivity characteristics. TT flows with high path overlap, high link resource occupancy, high potential conflict, or low timing slack are assigned higher scheduling weights. The scheduling weights can be determined according to the following relationship: in, These are the feature weight coefficients, and Subsequently, the CNC generates a basic action probability mask based on the scheduling weight vector. The basic action probability mask represents the prior selection tendency of each TT flow as a candidate ranking action. In one implementation, the basic action probability can be determined by the softmax function, as shown in the following equation: in, This is the mask temperature coefficient.

[0067] Step four: The CNC executes the initial sorting of Q-Learning driven by the motion probability mask. See also... Figure 5 The CNC will set the currently scheduled TT streams. and the remaining set of TT streams to be scheduled Together as a reinforcement learning state, it will... The next TT stream is selected as the action. Before each action selection, the CNC updates the probability mask of the current action based on the current scheduling state, as shown in the following formula: in, Indicates the current state of TT flow The action mask value. If If so, the corresponding TT stream will not participate in the current action selection. In exploration mode, the agent determines the action based on... The resulting probability distribution is used to sample actions; in exploitation mode, the agent fuses the Q-value with the current action probability mask and selects the TT stream with the highest overall score. The overall score can be determined according to the following relationship: in, Representing state Select TT stream State-action value This represents the mask fusion coefficient.

[0068] Step 5: The CNC calculates the immediate reward based on the impact of the newly added TT stream on the current partial scheduling sequence. The immediate reward can comprehensively consider the end-to-end latency increment, link or port conflict penalty, deadline violation penalty, and location gain, as expressed in the following formula: in, This represents the end-to-end latency increment caused by the addition of the TT stream. Indicates a port or link conflict penalty. This indicates the penalty for violating the deadline constraint. Indicates location-based revenue. This is the reward weighting coefficient. Location-based rewards can be determined according to the following relationship: in, Indicates TT stream Relative position within the current part of the sequence. This represents the total number of TT streams. The CNC updates the Q value based on the immediate reward, repeats the training process, and continues until the training round limit is reached, the objective function converges, or the preset stopping condition is met, then outputs the initial scheduling sequence.

[0069] Step 6: The CNC performs time slot scheduling evaluation based on the initial scheduling sequence. The CNC allocates a transmission time window to each output port on the transmission path of each TT stream according to the order of the TT streams in the initial scheduling sequence, and checks the port mutual exclusion constraints, link bandwidth constraints, flow conservation constraints, and deadline constraints.

[0070] Step 7: The CNC generates a link-time window conflict heatmap based on the time slot scheduling evaluation results. The scheduling timeout is divided into... Discrete time window Conflict heatmap elements Indicates a link or output port. In the time window The intensity of conflict within the area. Conflict intensity can be determined according to the following relationship: in, Indicates on the link or output port time window The number of TT streams sent during internal competition. This indicates the total length of the overlapping transmission time windows. This indicates the penalty for violating the deadline. This represents the increase in delay caused by conflict waiting. This is the conflict intensity weighting coefficient.

[0071] Step 8: The CNC calculates the dynamic conflict contribution value for each TT flow based on the conflict heatmap. For TT flows... Its dynamic conflict contribution value can be determined according to the following relationship: in, Indicates TT stream At the link or output port The transmission time window is used. If the dynamic collision contribution value of a TT flow exceeds a preset threshold, or is within the top preset proportion of the dynamic collision contribution value of all TT flows, then the TT flow is marked as a high-collision TT flow. TT flows that share a link with a high-collision TT flow, share an output port, have adjacent transmission time windows, or are adjacent in sorting position are marked as associated flows.

[0072] Step nine: The CNC performs tabu search optimization induced by the conflict heatmap. See also Figure 6 The CNC generates directional neighborhood candidate solutions around the high-collision TT flow and its associated flows. Directional neighborhood operations include moving forward, moving backward, swapping the positions of the high-collision TT flow and its associated flows, and swapping the positions of two TT flows sharing a critical link. The sampling probability of neighborhood operations can be determined according to the following relationship: If neighborhood operation The set of TT streams involved is The sampling probability of the directional neighborhood operation can be determined according to the following relationship: in, The induction coefficient, This represents the sampling probability of a normal exchange operation. Represents a set of highly conflicting TT streams. Represents a set of associated streams.

[0073] Step 10: For each candidate solution, the CNC re-executes the time-slot scheduling evaluation and calculates the objective function value and constraint violation degree. Candidate Scheduling Sequence The comprehensive evaluation function can be determined according to the following relationship: in, This represents the total end-to-end latency of the entire network's TT streams. Indicates the maximum end-to-end delay. This indicates the penalty for violating the deadline. Indicates a penalty for link or output port conflicts. This indicates the penalty for the number of GCL entries or the number of gating switches. to These are the weighting coefficients of the objective function.

[0074] Step 11: The CNC selects candidate solutions based on the taboo list and the amnesty criterion. If the move operation corresponding to a candidate solution is taboo, but the candidate solution is better than the historical best solution, the amnesty criterion is triggered and the candidate solution is accepted; otherwise, the candidate solution with the best overall evaluation is selected from the non-taboo candidate solutions as the current scheduling sequence. After each iteration, the CNC updates the taboo list and conflict heatmap.

[0075] Step 12: When the maximum number of iterations is reached, the objective function converges, no improvement is achieved for several consecutive rounds, or the preset scheduling quality threshold is met, the tabu search stops, and the CNC outputs the historical best scheduling sequence as the target scheduling sequence.

[0076] Step thirteen: The CNC generates the GCL based on the target scheduling sequence. Specifically, the CNC calculates the start and end times of transmission for each TT stream on each output port, and maps the transmission time windows on the same output port to IEEE 802.1Qbv gating control entries in chronological order. Finally, the CNC distributes the GCL to the TSN switching equipment via NETCONF, the YANG model configuration interface, or other southbound configuration protocols.

[0077] In another specific example of this application, see [link to example]. Figure 7This application also provides a time-sensitive network traffic scheduling system. This system can be deployed in a centralized network configuration unit (CNC) 100 and interacts with a centralized user configuration unit (CUC) 200 and the TSN network infrastructure. The system includes a network topology awareness module 101, a TT flow feature evaluation module 1021, a weight and mask generation module 1022, a Q-Learning initial sorting module 1023, a conflict heatmap feedback module 104, an induced tabu search optimization module 1024, and a GCL generation and distribution module 103.

[0078] The network topology awareness module 101 is used to acquire network status information such as TSN switch 302, terminal node 301, port connection relationship, link bandwidth and link propagation delay, construct TSN network topology model, and receive TT flow service requirements submitted by CUC.

[0079] The TT flow feature evaluation module 1021 is used to calculate the path overlap, timing slack, link resource occupancy and potential conflict of each TT flow based on the network topology, TT flow attributes and transmission path, and output the scheduling sensitivity features of each TT flow.

[0080] The weight and mask generation module 1022 is used to normalize and weight the TT stream scheduling sensitive features, generate a scheduling weight vector, and generate a basic action probability mask based on the scheduling weight vector.

[0081] The Q-Learning initial sorting module 1023 is used to update the current action probability mask based on the current set of scheduled TT streams and the set of remaining unscheduled TT streams, and to perform reinforcement learning training or inference based on the current action probability mask and the Q value to generate an initial scheduling sequence.

[0082] The conflict heatmap feedback module 104 is used to perform time slot scheduling evaluation based on the initial scheduling sequence or the current scheduling sequence during the tabu search process, to count the conflict intensity of different links or output ports in different time windows, and to generate a link-time window conflict heatmap.

[0083] The induced tabu search optimization module 1024 is used to identify high-conflict TT flows and associated flows based on the conflict heatmap, generate directional neighborhood candidate solutions around the high-conflict TT flows and their associated flows, and perform local search optimization by combining the tabu table and the amnesty criterion, and output the target scheduling sequence.

[0084] The GCL generation and distribution module 103 is used to generate an IEEE 802.1Qbv gating control list based on the target scheduling sequence and the corresponding time slot scheduling result, and then distribute the gating control list to the TSN switching equipment.

[0085] In a third specific example of this application, a time-sensitive network controller is also provided, comprising a communication interface, a memory, and a processor. The communication interface receives TT flow service requests sent by the CUC, collects TSN network topology information and link status information, and sends GCL configuration data to the TSN switching equipment. The memory stores network topology data, TT flow attribute data, transmission path data, scheduling-sensitive feature data, scheduling weight vectors, basic action probability masks, current action probability masks, Q-value tables, conflict heatmaps, taboo tables, scheduling sequences, and a scheduler. The processor invokes the scheduler stored in the memory to execute the aforementioned time-sensitive network traffic scheduling method.

[0086] This application also provides an electronic device, which may include a processor, a memory, a receiver, and a transmitter. The processor is used to execute the time-sensitive network traffic scheduling method mentioned in the above embodiments. The processor and the memory can be connected via a bus or other means, taking a bus connection as an example. The receiver can be connected to the processor and the memory via wired or wireless means.

[0087] The processor can be a central processing unit (CPU). The processor can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.

[0088] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the time-sensitive network traffic scheduling method described in the embodiments of this application. The processor executes various functional applications and data processing by running the non-transitory software programs, instructions, and modules stored in the memory, thereby implementing the time-sensitive network traffic scheduling method described in the above method embodiments.

[0089] The memory may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor, etc. Furthermore, the memory may include high-speed random access memory and non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0090] The one or more modules are stored in the memory, and when executed by the processor, the time-sensitive network traffic scheduling method described in the embodiment is executed.

[0091] In some embodiments of this application, the user equipment may include a processor, a memory, and a transceiver unit. The transceiver unit may include a receiver and a transmitter. The processor, memory, receiver, and transmitter may be connected via a bus system. The memory is used to store computer instructions, and the processor is used to execute the computer instructions stored in the memory to control the transceiver unit to send and receive signals.

[0092] As one implementation method, the functions of the receiver and transmitter in this application can be implemented by transceiver circuits or dedicated transceiver chips, and the processor can be implemented by dedicated processing chips, processing circuits or general-purpose chips.

[0093] As another implementation approach, the server provided in this application embodiment can be implemented using a general-purpose computer. That is, the program code implementing the processor, receiver, and transmitter functions is stored in memory, and the general-purpose processor implements the processor, receiver, and transmitter functions by executing the code in memory.

[0094] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the aforementioned time-sensitive network traffic scheduling method. The computer-readable storage medium can be a tangible storage medium, such as random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, floppy disks, hard disks, removable storage disks, CD-ROMs, or any other form of storage medium known in the art.

[0095] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the aforementioned time-sensitive network traffic scheduling method.

[0096] Those skilled in the art will understand that the exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application. When implemented in hardware, it can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. The programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave.

[0097] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0098] In this application, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or in place of features of other embodiments.

[0099] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to the embodiments of this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A time-sensitive network traffic scheduling method, characterized in that, The method includes: The basic action probability mask of each scheduled time trigger flow is updated using the current set of scheduled time trigger flows and the set of remaining scheduled time trigger flows to obtain the corresponding current action probability mask. Based on the current action probability mask, the next scheduled time trigger flow is selected to be added to the current scheduling sequence to generate the initial scheduling sequence. Based on the initial scheduling sequence, a time slot scheduling evaluation is performed to obtain the corresponding time slot scheduling result, and a conflict heatmap of the link and time window dimensions is generated based on the time slot scheduling result. Based on the conflict heatmap of the link and time window dimensions, high-conflict time-triggered flows and associated flows are identified, and neighborhood candidate solutions are generated based on the high-conflict time-triggered flows and associated flows. The preset induced tabu search algorithm is used to search and optimize each neighborhood candidate solution, and the corresponding target scheduling sequence is output. Based on the target scheduling sequence and the corresponding time slot scheduling result, a gating control list is generated and sent to the time-sensitive network.

2. The method according to claim 1, characterized in that, The generation process of the basic action probability mask includes: A directed network topology graph is constructed based on the network topology information obtained in the time-sensitive network; and the transmission path of each time-triggered stream in the obtained time-triggered stream set is determined according to the attribute information of each time-triggered stream to be scheduled and the directed network topology graph. Based on the directed network topology and the transmission paths of each scheduled time-triggered stream, the scheduling-sensitive features of each scheduled time-triggered stream are extracted. The scheduling sensitivity features of each scheduled time-triggered flow are normalized, and a scheduling weight vector is generated based on the normalized scheduling sensitivity features. The basic action probability mask is generated based on the scheduling weight vector.

3. The method according to claim 2, characterized in that, The scheduling-sensitive features include: Path overlap is used to characterize the degree of resource contention between the current time-triggered stream and other time-triggered streams on a shared link or shared output port; Temporal slack is used to characterize the time slack that the current time-triggered flow has to satisfy the deadline constraint; Link resource occupancy is used to characterize the intensity of link transmission resource occupancy by the current time-triggered stream; Potential conflict level is used to characterize the likelihood that the current time-triggered stream will cause a transmission window conflict during subsequent time slot allocation; The scheduling weight vector is positively correlated with the normalized path overlap, link resource occupancy, and potential conflict, and negatively correlated with the normalized temporal relaxation.

4. The method according to claim 1, characterized in that, The process involves updating the basic action probability mask of each scheduled time trigger stream using the currently scheduled set of time trigger streams and the remaining set of time trigger streams to obtain the corresponding current action probability mask, and then selecting the next scheduled time trigger stream to be added to the scheduling sequence based on the current action probability mask to generate an initial scheduling sequence, including: Set the time trigger streams in the currently scheduled time trigger stream set to an unselectable state, and correct them according to the action probability of each time trigger stream in the remaining time trigger stream set to obtain the corresponding current action probability mask. Based on the current action probability mask, select the next time trigger stream to be scheduled as a newly added time trigger stream and add it to the scheduling sequence; Based on the current partial scheduling sequence, the preset time slot estimator is invoked to estimate the end-to-end delay increment, link conflict penalty, and deadline violation risk caused by the newly added time-triggered flow. The instant reward is calculated and obtained by combining the scheduling weight vector of the newly added time-triggered flow or the preset position benefit. The Q-value table is updated using the instant reward until all time-triggered streams are added to the scheduling sequence to obtain a trained Q-value table, and the initial scheduling sequence is obtained based on the trained Q-value table and the current action probability mask.

5. The method according to claim 4, characterized in that, The step of selecting the next time trigger stream to be scheduled as a newly added time trigger stream to the scheduling sequence based on the current action probability mask includes: In exploration mode, actions are sampled according to the probability distribution formed by the current action probability mask, and the next time trigger flow to be scheduled is selected. In the utilization mode, the Q value is fused with the current action probability mask, and the scheduled time trigger flow with the highest comprehensive score is selected as the next scheduled time trigger flow.

6. The method according to claim 1, characterized in that, The conflict heatmap of the link and time window dimensions includes a first dimension and a second dimension; wherein, the first dimension is the link or output port identifier, the second dimension is the discrete time window identifier, and the heatmap element is the conflict intensity value; the conflict intensity value is determined based on at least one of the following: the number of time-triggered streams competing to be sent by the same link or output port in the same or adjacent time windows, the length of the overlapping time windows, the degree of deadline violation, and the waiting delay increment.

7. The method according to claim 1, characterized in that, The identification of high-conflict time-triggered flows and associated flows based on the conflict heatmap of the link and time window dimensions includes: The dynamic conflict contribution value of each scheduled time-triggered flow is calculated and determined based on the conflict heatmap of the link and time window dimensions. If the dynamic conflict contribution value is greater than a preset threshold or is located in the top preset proportion of all time-triggered flows, then the time-triggered flow to be scheduled corresponding to the dynamic conflict contribution value is marked as a high-conflict time-triggered flow; and the time-triggered flows that share a link, share an output port, have adjacent transmission time windows, or are adjacent in sorting position with the high-conflict time-triggered flow are marked as associated flows. The dynamic conflict contribution value is expressed by the following formula: in, HC i Indicates the dynamic conflict contribution value; i Indicates the index of the time-triggered stream; e Identifier for the link or output port; E i This represents the set of links or output ports traversed by the i-th time-triggered stream; b k This represents the boundary or identifier of the k-th discrete-time window; TW i,e This indicates the transmission time window of the time-triggered stream on the link or output port; H[e,k] This indicates the conflict intensity value of the corresponding link or output port in the conflict heatmap within the time window.

8. The method according to claim 1, characterized in that, The generation of neighborhood candidate solutions based on the high-collision time-triggered flow and the associated flow includes: The current scheduling sequence is transformed according to a preset transformation method, and the transformed scheduling sequence is used as a neighborhood candidate solution; wherein, the transformation method includes: Move the high-conflict time-triggered stream forward in the current scheduling sequence; The high-conflict-time trigger stream is shifted to the later part of the current scheduling sequence; Swap the positions of the high-conflict-time triggering stream and the associated stream in the current scheduling sequence; Swap the positions of the two time-triggered streams that share a critical link in the current scheduling sequence.

9. The method according to claim 1, characterized in that, The step of using a preset induced tabu search algorithm to search and optimize each of the neighborhood candidate solutions and outputting the corresponding target scheduling sequence includes: For each of the aforementioned neighborhood candidate solutions, the time slot scheduling is re-evaluated, and the objective function value and constraint violation degree of the candidate scheduling sequence are calculated. The system uses a pre-defined taboo table to record recently executed move operations to determine whether the move operation corresponding to each neighborhood candidate solution is in a taboo state. If the move operation corresponding to the neighborhood candidate solution is taboo but the neighborhood candidate solution is better than the historical best solution, then the amnesty criterion is triggered and the neighborhood candidate solution is accepted; if there is no neighborhood candidate solution that meets the conditions, then the neighborhood candidate solution with the best comprehensive evaluation is selected from the non-taboo neighborhood candidate solutions as the current scheduling sequence; and the taboo table and the conflict heatmap of the link and time window dimensions are updated. When the preset stopping condition is met, the search stops and the historical best scheduling sequence is output as the target scheduling sequence.

10. An electronic device, characterized in that, It includes a processor and a memory; when the processor executes the running program stored in the memory, it implements the time-sensitive network traffic scheduling method as described in any one of claims 1 to 9.

Images