Self-adaptive redundant path optimization method for dynamic network service reliability

Abstract

The invention relates to the technical field of communication networks, in particular to a dynamic network service reliability-oriented self-adaptive redundant path optimization method, which comprises the following steps of: acquiring a flow of a TSN network, and constructing a topological graph of the TSN network; generating a candidate redundant path set of each stream in the topological graph; processing the topological graph of the TSN network by adopting a graph neural network to obtain structural features of the topological graph; according to the structural characteristics of the topological graph and the candidate redundant path set of each flow, distributing an optimal redundant path set for each flow by adopting a dual deep Q network; scheduling each flow according to the optimal redundant path set to obtain a scheduling result of each flow; according to the method, the reliability of the path is calculated according to the intersection degree of partial disjoint links in the path, hidden single-point fault hidden dangers can be identified, and the effectiveness of path selection is improved; according to the method, the graph neural network and deep reinforcement learning methods are introduced, and the structural dependency relationship and the link attribute characteristics between the nodes are effectively extracted.

Description

Technical Field

[0001] The present invention relates to the technical field of communication networks, and in particular to an adaptive redundant path optimization method oriented to dynamic network service reliability. Background Art

[0002] With the rapid development of fields such as the Industrial Internet and intelligent transportation, TSN has been widely adopted as a key technology for ensuring real-time and reliable data transmission. However, current TSN (Time-Sensitive Networking) networks still face many challenges when facing complex and changing network environments. On the one hand, link failures and traffic congestion are common in the network, which seriously impacts the reliability and real-time performance of TSN networks. On the other hand, as the demands of various businesses on TSN networks become increasingly diverse, data streams of different priorities need to be properly allocated paths and resources within the network to ensure the service quality of critical businesses. However, existing path selection and resource allocation strategies struggle to meet this requirement.

[0003] Existing TSN network redundant path selection and scheduling solutions have certain limitations. Some solutions focus on static rule-based path planning and are unable to make effective adjustments in a timely manner when network topology or traffic changes dynamically, resulting in reduced network performance. Some solutions that use traditional machine learning methods have difficulty accurately capturing the complex characteristics of network topology when processing large-scale, high-dimensional network data, resulting in poor path selection accuracy and optimization effects. In addition, most existing solutions lack refined scheduling of data streams of different priorities, cannot fully guarantee the transmission quality of high-priority streams, and are difficult to adapt to scenarios such as industrial control and autonomous driving that have extremely high requirements for reliability and real-time performance. Therefore, a method that can adapt to dynamic network changes and accurately optimize redundant path selection and scheduling is needed to improve the overall performance and reliability of TSN networks. Summary of the Invention

[0004] To solve the above-mentioned problems in the prior art, the present invention adopts an adaptive redundant path optimization method for dynamic network service reliability, comprising:

[0005] S1. Obtain the flow of the TSN network and build the topology graph G of the TSN network;

[0006] S2. Generate a set of candidate redundant paths for each flow in the topology graph G;

[0007] S3. Use a graph neural network to process the topology graph of the TSN network to obtain the structural characteristics of the topology graph G of the TSN network;

[0008] S4. Based on the structural characteristics of the topology graph G and the candidate redundant path set of each flow, deep reinforcement learning is used to assign the optimal redundant path set to each flow;

[0009] S5. Schedule each flow according to the optimal redundant path set to obtain a scheduling result for each flow.

[0010] Beneficial effects:

[0011] 1. In a TSN network, two paths may partially intersect at the core switch. Traditional methods cannot distinguish the reliability difference between such situations and completely independent paths. The present invention calculates the degree of intersection of some non-intersecting links in the path, and calculates the reliability of the path based on the degree of intersection of some non-intersecting links in the path. It can identify hidden single-point failure risks more carefully, and then screen them according to the reliability of the path, overcoming the traditional path selection scheme that only considers simple topologies that are completely non-intersecting, thereby improving the effectiveness of path selection; 2. The present invention introduces graph neural networks and deep reinforcement learning methods to effectively extract structural dependencies and link attribute characteristics between nodes, explore the optimal scheduling strategy for different flows on redundant paths, and has the advantages of high reliability, high resource utilization, and strong real-time performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A flowchart of a method for adaptive redundant path optimization for dynamic network service reliability provided by an embodiment of the present invention;

[0013] Figure 2 A schematic diagram of determining a redundant path set using the Yen's KSP algorithm provided in an embodiment of the present invention;

[0014] Figure 3 Schematic diagram of some disjoint redundant paths and FRER mechanism provided by an embodiment of the present invention;

[0015] Figure 4 A schematic diagram of the GNN graph neural network convolution provided by an embodiment of the present invention;

[0016] Figure 5 This is a diagram of the DDQN deep reinforcement learning framework provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments 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 are within the scope of protection of the present invention.

[0018] like Figure 1 As shown, the embodiment of the present invention adopts an adaptive redundant path optimization method for dynamic network service reliability, including:

[0019] S1. Obtain the flow of the TSN network and construct the topology graph G = (V, E) of the TSN network. The topology graph G of the TSN network includes a node set V and an edge set E. The node set includes terminals and switches of the TSN network, and the edge set includes links of the TSN network.

[0020] The attribute of a node is the failure probability of the node, and the attributes of an edge include: the failure probability of the link, link load, link bandwidth, and link delay, which are usually obtained through statistical data or equipment manuals. Each flow in a TSN network includes a source node, a destination node, bandwidth conditions, and delay conditions.

[0021] S2. Generate a set of candidate redundant paths for each flow in the topology graph G;

[0022] The candidate redundant path set for each flow generated in the topology graph G includes:

[0023] S21, using Yen's KSP algorithm to generate the initial candidate redundant path set for each flow l in the topology graph G Where l is the index of the flow, indicating the priority of the flow. The initial candidate redundant path set includes multiple paths, and each path includes multiple links.

[0024] like Figure 2 Specifically, Yen's KSP algorithm is used to determine the path with the shortest number of hops between the source node and the destination node of flow l, and it is stored as the first path in the initial candidate redundant path set. For the subsequent generation of redundant paths, some edges or nodes on the path are disabled, and a new shortest path is found in the modified network structure as a candidate path. Similarly, the initial candidate redundant path set of flow l is obtained. Among them, Yen's KSP algorithm is Yen's K shortest path algorithm.

[0025] S22. Calculate the initial candidate redundant path set for each flow l The cost function value Cost(P l,k );

[0026] like Figure 3As shown in Figure 1, in communication networks, the FRER (Frame Replication and Elimination for Reliability) mechanism is a technology that improves data transmission reliability through "frame replication and elimination". Its core idea is to provide redundant transmission paths for key data frames and eliminate duplicate frames at the receiving end, thereby avoiding data loss caused by a single path failure.

[0027] In a TSN network, two paths may partially intersect at the core switch. Traditional methods cannot distinguish the reliability difference between such a situation and completely independent paths. However, the present invention calculates the degree of intersection of some non-intersecting links in the path, and calculates the reliability of the path based on the degree of intersection of some non-intersecting links in the path, which can more finely identify hidden single point failure hazards. The cost function value Cost(P l,k ) is calculated as:

[0028] Cost(p l,k )=w1Length(P l,k )+w2R total (P l,k )

[0029]

[0030] Among them, Length(P l,k ) represents the initial candidate redundant path set of flow l The number of hops on the kth path in R total (P l,k ) represents the initial candidate redundant path set of flow l The overall reliability of the kth path in , θ l,k Represents the initial candidate redundant path set of flow l The degree of intersection between the kth path in the flow l and the rest of the paths. If the paths do not overlap at all, that is, they are independent, then θ i =1,P l,k is the initial candidate redundant path set of flow l The kth path in P l,g is the initial candidate redundant path set of flow l The g-th path in ∪ g≠k P l,g Represents the initial candidate redundant path set of flow l Eliminate path P l,k The link set of all paths other than |p l,k ∩(∪ g≠k p l,g )| represents the path p l,k and the link set ∪ g≠k p l,g Number of overlapping links, |p l,k | represents the initial candidate redundant path set for flow l The number of links in the kth path, R(p l,k,i ) is the initial candidate redundant path set of flow l The reliability of the i-th link of the k-th path, reliability R(p l,k,i )=1-Er(p l,k,i ), Er(p l,k,i ) is the failure probability of the link.

[0031] In one embodiment, the failure probability of a link Er(p l,k,i )=e αt , gradually increases with time t, and α is generally a small constant.

[0032] S23, the initial candidate redundant path set for each flow l according to the cost function value Filter each path in the to obtain a set of candidate redundant paths B for each flow l l .

[0033] Specifically, a threshold N is set, and N candidate redundant paths with the largest cost function values ​​in the initial candidate redundant path set of flow l are screened.

[0034] S3. Use a graph neural network to process the topology graph G of the TSN network to obtain the structural characteristics of the topology graph G of the TSN network;

[0035] like Figure 4 As shown in the figure, the use of graph neural network to process the topology graph of TSN network includes: obtaining a pre-trained graph neural network, which includes multiple layers of graph convolution layers; constructing a node feature matrix according to the attributes of the node, and constructing an adjacency matrix according to the attributes of the link; graph convolution aggregates the node's own features with the related information of the adjacent nodes according to the node feature matrix and the adjacency matrix, and then obtains the embedded representation of the node after graph convolution through activation function, such as Relu function. After aggregation of multiple layers of graph convolution, the node can obtain high-order neighbor features, thereby fully capturing the structural dependencies between nodes and the state propagation information between links in the network topology.

[0036] In one embodiment, constructing a node feature matrix based on node attributes includes: encoding the failure rate of each node to obtain a feature of each node, and combining the features of all nodes to obtain a node feature matrix.

[0037] In one embodiment, constructing the adjacency matrix according to the attributes of the links includes: calculating the weights of the links according to the link failure rate, link load, link bandwidth, and link delay, and constructing the adjacency matrix according to the link weights.

[0038] S4. Based on the structural characteristics of the topology graph G and the candidate redundant path set of each flow, deep reinforcement learning is used to assign the optimal redundant path set to each flow;

[0039] Assigning the optimal redundant path set to each flow includes:

[0040] S41. Construct state space, action space and reward function;

[0041] State space S = {s t},s t =(V,E,F); where s t represents the state at time t, (V, E, F) is the structural feature of the topological graph G, and F is the node feature set of the topological graph G output by the graph convolutional network;

[0042] Action Space Among them, a t represents the action at time t, a l is the action of flow l, representing the optimal redundant path set of flow l, B l is the set of candidate redundant paths for flow l, and L is the number of flows;

[0043] Reward Function w4.R finish );in, is the resource feasibility reward, which represents the optimal redundant path set a selected by flow l l The ratio of bandwidth occupied by all paths in the total bandwidth, W l represents the bandwidth occupied by all paths in the optimal redundant path set selected for flow l, W l,all R represents the total bandwidth of all paths in the optimal redundant path set selected by flow l; total (l) is the path reliability reward, which represents the sum of the total reliability of all paths in the optimal redundant path set selected by flow l; R delay (l) is the path delay reward, which is expressed as the sum of the delays of flow l from the starting node to the destination node on all paths in the optimal redundant path set. The delay includes: propagation delay, queuing delay and processing delay. R finish The reward for completing the full flow is usually set to a large constant, usually set to 100-500, or even higher, adjusted according to the needs of the scenario to let the agent know that it has completed all the selected tasks.

[0044] S42, constructing a dual deep Q network, training the dual deep Q network based on the state space, the action space, and the reward function, and obtaining a trained dual deep Q network;

[0045] like Figure 5 As shown, the training process of the dual deep Q network includes:

[0046] S421. Initialize the experience replay pool D, collect experience samples by interacting with the TSN network based on the state space, action space, and reward function, and store the experience samples in the experience replay pool D.

[0047] Collecting experience samples includes: taking the structural features of the current topology G as the current state s t , the current state s t Input the main network and get the optimal action a t , that is, the optimal redundant path set for each flow, performing the optimal action a t , get reward r t and the next state s t+1 ; The experience sample (s t ,a t ,s t+1 ,r t ) is stored in the experience replay pool D.

[0048] S422. Sampling experience samples from the experience replay pool D, and training a dual-depth network based on the sampled experience samples to obtain a trained dual-depth Q network; the dual-depth Q network includes: a main network and a target network;

[0049] The training process of the dual deep Q network based on the sampled empirical samples includes:

[0050] The state s in the experience sample t Input the main network and get the optimal action a t+1 ; Calculate the Q value Q(s) of the main network t ,a t ; θ); where θ is the parameter of the main network;

[0051] The next state s in the experience sample t+1 and the optimal action a t+1 Input the target network to estimate the target Q value y t =r t -γQ(s t+1 ,a t+1 ;θ′); where γ is the discount factor and θ′ is the parameter of the target network;

[0052] Calculating the loss function Where E is the expectation;

[0053] Update the parameters of the dual deep Q network through the gradient descent method according to the loss function;

[0054] The gradient descent formula is:

[0055]

[0056] The update formula is:

[0057]

[0058] in, is the gradient and a is the learning rate.

[0059] Through continuous iterative training, the loss function gradually reaches a convergence state, thereby representing the strategy to achieve an optimal strategy, thereby achieving continuous optimization of the path selection strategy.

[0060] S43. Input the structural features of the topology graph G and the candidate redundant path set of each flow into the trained dual-depth Q network to obtain the optimal redundant path set of each flow.

[0061] S5. Schedule each flow according to the optimal redundant path set to obtain a scheduling result for each flow.

[0062] Scheduling each flow based on the optimal redundant path set includes:

[0063] S51, scheduling each flow l in turn according to the optimal redundant path set based on the flow priority to obtain a scheduling set for each flow l;

[0064] Scheduling each flow l in turn includes:

[0065] S511, obtain the time slot resource table table0, schedule the flow with the highest priority according to the optimal redundant path set B1′ and the time slot resource table table0, and obtain the scheduling set SR1 of the flow with the highest priority and the time slot resource table table1 after scheduling;

[0066] The time slot resource table table0 is obtained by globally scanning the network time slot resources. It pre-collects basic information such as the available bandwidth, occupancy status, and reserved resources of each link in the TSN network in different time slots.

[0067] S512, schedule the second priority flow according to the optimal redundant path set B2' and the time slot resource table table1, and obtain the scheduling set SR2 of the second priority flow and the scheduled time slot resource table table2;

[0068] S513, based on the optimal redundant path set B l ' and time slot resource table l-1Schedule flow l and obtain the scheduling set SR of flow l l And the time slot resource table after scheduling l ;

[0069] According to the optimal redundant path set B l ' and time slot resource table l-1 Scheduling flow l includes:

[0070] Step 1: Obtain the optimal redundant path set B′ l The first path P in l,1 ′, determine the first path P l,1 'In the time slot resource table l-1 Is there a time slot set ts1 that meets the bandwidth and delay conditions of flow l? If so, allocate flow l to the time slot set ts1 and update the time slot resource table table l-1 , get the scheduling result SR of flow l l，1 and time slot resource table l-1，1 ; Among them, the scheduling result SR l,1 ={P l,1 ′,ts1}, including the scheduled path P l,1 ′ and the allocated time slot set ts1;

[0071] Step 2: Obtain the optimal redundant path set B′ l The next path P′ in l,m , determine the path P′ l,m In the time slot resource table l-1,m-1 Is there a time slot set ts that satisfies the bandwidth and delay conditions of stream l? m , if it exists, then the path P′ of flow l l,m Assigned to time slot set ts m And update the time slot resource table l-1,m-1 , get the scheduling result SR of flow l l,m and time slot resource table l-1,m ; Among them, the scheduling result SR l,m ={P′ l,m ,ts m}, including the scheduled path P′ l,m and the allocated time slot set ts m , m is the optimal redundant path set B′ l The index of the path in ;

[0072] Determine path P′ l,m In the time slot resource table l-1,m-1 Is there a time slot set ts that satisfies the bandwidth and delay conditions of stream l? mThe process includes: extracting the bandwidth condition and delay condition of stream l, traversing each time slot, performing bandwidth check and delay check on each time slot according to the bandwidth condition and delay condition of stream l, and determining whether there is a time slot set ts that passes both bandwidth check and delay check. m , if it exists, then we get the path P′ l,m Allocated time slot set ts m ; Among them, the time slot set ts m ={ts m,1 ,ts m,2 ,...,ts m,N}, ts m,i Represented as path P′ l,m The time slot allocated to the i-th link.

[0073] The delay check includes: checking whether the accumulated delay of the link corresponding to each time slot of the time slot set meets the delay condition of the flow, that is, whether it is less than the maximum end-to-end delay of the flow.

[0074] Bandwidth verification includes: checking whether the links corresponding to each time slot in the time slot set meet the bandwidth conditions, that is, whether the available bandwidth of the links is greater than the bandwidth required by the flow.

[0075] Update the time slot resource table l-1,m-1 Including: According to the flow scheduling result SR l,m In the table l-1,m-1 The time slot position of the corresponding link in the time slot is deducted, the allocated bandwidth is marked, and the occupied status is marked, thereby completing the update of the time slot resource table, realizing dynamic synchronization of resource status, and providing accurate real-time resource data for subsequent flow scheduling.

[0076] Step 3: Repeat step 2 until all redundant path set B is traversed. l ′, and obtain all the scheduling results of flow l and the time slot resource table after scheduling. l , combine all scheduling results of flow l into the scheduling set of flow l.

[0077] S514, repeat step S513 until the scheduling set of the lowest priority flow and the scheduled time slot resource table table are obtained L .

[0078] S52. Perform a global conflict check on the scheduling sets of all flows. If there are conflicting flows, combine the conflicting flows to obtain a set ST. Remove the flow with the highest priority from the set ST and add the flows whose scheduling set is an empty set to obtain a set ST′. Otherwise, stop scheduling. Conflicting flows are flows that occupy the same link in the same time slot.

[0079] S53, rescheduling each flow in the set ST′ in turn according to the optimal redundant path set based on the priority of the flow, to obtain a new scheduling set for each flow in the set ST′;

[0080] The scheduling process of this step is consistent with that of step S51, except that the time slot resource table is the remaining time slot resource table obtained after removing the time slots allocated to the remaining flows.

[0081] S54, perform global conflict detection on the new scheduling set of each flow in the set ST'. If there are conflicting flows, combine the conflicting flows to obtain the set In the collection Eliminate the flow with the highest priority and add the flows whose scheduling result is an empty set in set ST′ to obtain set ST″; otherwise, stop scheduling;

[0082] S55, deleting some links in the optimal redundant path set of each flow in the set ST", to obtain the optimal remaining redundant path set of each flow in the set ST";

[0083] Deleting some links of the optimal redundant path set of each flow in the set ST″ includes: obtaining the set The path P of the flow with the highest priority max , delete the optimal redundant path set of each flow in the set ST″ and the path P max For intersecting links, if deleting a link only disconnects part of the path and there are still other redundant paths in the set, then delete the link; if deleting a link directly disconnects all paths and there are no other redundant paths in the set, then skip the link.

[0084] S56. Reschedule each flow in the set ST″ in turn according to the optimal remaining redundant path set based on the priority of the flow to obtain a new scheduling set for each flow in the set ST″;

[0085] The scheduling process of this step is consistent with step S51, except that the time slot resource table is the remaining time slot resource table obtained after removing the time slots allocated to the remaining flows, and the optimal redundant path set is replaced by the optimal remaining redundant path set.

[0086] S57. Evaluate the new scheduling set for each flow in the set ST″, combine the flows whose evaluation results do not meet the requirements, and obtain the set ST″′; if the set ST″′ is an empty set, stop scheduling; otherwise, disable some nodes and edges in the topology graph G, and return to step S2 to regenerate the candidate redundant path set for each flow in the set ST″′.

[0087] Specifically, the overall reliability and bandwidth utilization of all paths in the new scheduling set of each flow in the set ST″′ are evaluated; a reliability threshold and a bandwidth utilization threshold are set, and flows whose overall reliability and bandwidth utilization are both less than the corresponding thresholds are combined to obtain the set ST″′.

[0088] The reliability threshold is set to 0.8-0.9, and the bandwidth utilization threshold is generally set to 0.6-0.7.

[0089] The bandwidth utilization of all paths scheduled for each flow is the ratio of the bandwidth used by all paths scheduled for each flow to the total bandwidth of the TSN network.

[0090] Disabling some nodes and links in the topology graph G includes: obtaining all paths of all current flow scheduling, calculating the nodes and links with the most scheduling times according to all paths of all flow scheduling, and disabling the nodes and links with the most scheduling times in the topology graph G.

[0091] The above embodiments further illustrate the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above embodiments are only preferred implementation plans of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An adaptive redundant path optimization method for dynamic network service reliability, characterized in that: include: S1. Obtain the flow of the TSN network and build the topology graph G of the TSN network; S2. Generate a set of candidate redundant paths for each flow in the topology graph G; S3. Use a graph neural network to process the topology graph G of the TSN network to obtain the structural characteristics of the topology graph G of the TSN network; S4. Based on the structural characteristics of the topology graph G and the candidate redundant path set of each flow, deep reinforcement learning is used to assign the optimal redundant path set to each flow; S5. Schedule each flow according to the optimal redundant path set to obtain a scheduling result for each flow.

2. The adaptive redundant path optimization method for dynamic network service reliability according to claim 1, characterized in that: The candidate redundant path set for each flow generated in the topology graph G includes: S21, using Yen's KSP algorithm to generate the initial candidate redundant path set for each flow l in the topology graph G Where l is the index of the flow, indicating the priority of the flow. The initial candidate redundant path set includes multiple paths, and each path includes multiple links. S22. Calculate the initial candidate redundant path set for each flow l The cost function value of each path in; S23, the initial candidate redundant path set for each flow l according to the cost function value Filter each path in the to obtain a set of candidate redundant paths B for each flow l l .

3. The adaptive redundant path optimization method for dynamic network service reliability according to claim 2, characterized in that: Calculate the cost function value Cost(P l,k )include: Cost(P l,k )=w1Length(P l,k )+w2R total (P l,k ) Among them, Length(P l,k ) represents the initial candidate redundant path set of flow l The number of hops on the kth path in R total (P l,k ) represents the initial candidate redundant path set of flow l The overall reliability of the kth path in , w1 and w2 are weights.

4. The adaptive redundant path optimization method for dynamic network service reliability according to claim 3, characterized in that: The initial candidate redundant path set of flow l The overall reliability R of the kth path in total (P l,k )for: Among them, θ l,k Represents the initial candidate redundant path set of flow l The degree of intersection of the kth path in the flow l with the rest of the paths, P l,k is the initial candidate redundant path set of flow l The kth path in p l,k,i is the initial candidate redundant path set of flow l The i-th link in the k-th path in P l,g is the initial candidate redundant path set of flow l The g-th path in ∪ g≠k P l,g Represents the initial candidate redundant path set of flow L Eliminate path P l,k The set of links of all paths other than |P l,k | represents the initial candidate redundant path set for flow l The number of links in the kth path, R(p l,k,i ) is the initial candidate redundant path set of flow l The reliability of the i-th link in the k-th path in .

5. The adaptive redundant path optimization method for dynamic network service reliability according to claim 1, characterized in that: Assigning the optimal redundant path set to each flow includes: S41. Construct state space, action space and reward function; S42, constructing a dual deep Q network, training the dual deep Q network based on the state space, the action space, and the reward function, and obtaining a trained dual deep Q network; S43. Input the structural features of the topology graph G and the candidate redundant path set of each flow into the trained dual-depth Q network to obtain the optimal redundant path set of each flow.

6. The adaptive redundant path optimization method for dynamic network service reliability according to claim 5, characterized in that: State space S = {s t },s t =(V,E,F), action space A = {a t },a t =(a1,a2,...,a L ), Among them, s t represents the state at time t, (V, E, F) is the structural feature of the topological graph G, F is the node feature set of the topological graph G, V represents the node set of the topological graph G, E represents the edge set of the topological graph G, a t represents the action at time t, a l is the action of flow l, representing the optimal redundant path set of flow l, B l is the set of candidate redundant paths for flow l, L is the number of flows, and l is the index of the flow, indicating the priority of the flow.

7. The adaptive redundant path optimization method for dynamic network service reliability according to claim 6, characterized in that: Reward Function Among them, R feasibility (l) is the resource feasibility reward of flow l, which represents the optimal redundant path set a selected by flow l. l The ratio of bandwidth occupied by all paths in the total bandwidth, R total (l) is the path reliability reward of flow l, which represents the optimal redundant path set a selected by flow l. l The sum of the total reliabilities of all paths in R delay (l) is the path delay reward of flow l, which means that flow l is in the optimal redundant path set a l The sum of the delays on all paths in R finish Assign rewards for full flow completion.

8. The adaptive redundant path optimization method for dynamic network service reliability according to claim 1, characterized in that: Scheduling each flow based on the optimal redundant path set includes: S51. Schedule each flow l in turn according to the optimal redundant path set based on the flow priority to obtain a scheduling set for each flow l; where l is the index of the flow, indicating the priority of the flow; S52. Perform a global conflict check on the scheduling sets of all flows. If there are conflicting flows, combine the conflicting flows to obtain a set ST. Remove the flow with the highest priority from the set ST and add the flows whose scheduling set is an empty set to obtain a set ST′. Otherwise, stop scheduling. S53, rescheduling each flow in the set ST′ in turn according to the optimal redundant path set based on the priority of the flow, to obtain a new scheduling set for each flow in the set ST′; S54, perform global conflict detection on the new scheduling set of each flow in the set ST'. If there are conflicting flows, combine the conflicting flows to obtain the set In the collection Eliminate the flow with the highest priority and add the flows whose new scheduling set is an empty set in set ST′ to obtain set ST″; otherwise, stop scheduling; S55, deleting some links in the optimal redundant path set of each flow in the set ST", to obtain the optimal remaining redundant path set of each flow in the set ST"; S56. Reschedule each flow in the set ST″ in turn according to the optimal remaining redundant path set based on the priority of the flow to obtain a new scheduling set for each flow in the set ST″; S57. Evaluate the new scheduling set for each flow in the set ST″, combine the flows whose evaluation results do not meet the requirements, and obtain the set ST″′; if the set ST″′ is an empty set, stop scheduling; otherwise, disable some nodes and edges in the topology graph G, and return to step S2 to regenerate the candidate redundant path set for each flow in the set ST″′.

9. The adaptive redundant path optimization method for dynamic network service reliability according to claim 8, characterized in that: Scheduling each flow l in turn includes: S511, obtain the time slot resource table table0, schedule the flow with the highest priority according to the optimal redundant path set B1′ and the time slot resource table table0, and obtain the scheduling set SR1 of the flow with the highest priority and the time slot resource table table1 after scheduling; S512, schedule the second priority flow according to the optimal redundant path set B2' and the time slot resource table table1, and obtain the scheduling set SR2 of the second priority flow and the scheduled time slot resource table table2; S513, based on the optimal redundant path set B l ' and time slot resource table l-1 Schedule flow l and obtain the scheduling set SR of flow l l And the time slot resource table after scheduling l ; S514: Repeat step S513 until the scheduling set of the flow with the lowest priority is obtained.

10. The adaptive redundant path optimization method for dynamic network service reliability according to claim 9, characterized in that: Scheduling flow l includes: Step 1: Obtain the optimal redundant path set B′ l The first path P in l,1 ′, determine the first path P l,1 'In the time slot resource table l-1 Is there a time slot set ts1 that meets the bandwidth and delay conditions of stream l? If so, update the time slot resource table table l-1 , get the scheduling result SR of flow l l，1 and time slot resource table l-1，1 ; Step 2: Obtain the optimal redundant path set B′ l The next path P′ in l,m , determine the path P′ l,m In the time slot resource table l-1,m-1 Is there a time slot set ts that satisfies the bandwidth and delay conditions of stream l? m If it exists, update the time slot resource table l-1,m-1 , get the scheduling result SR of flow l l,m and time slot resource table l-1,m ; Among them, the scheduling result SR l,m ={P′ l,m ,ts m }, m is the optimal redundant path set B′ l The index of the path in ; Step 3: Repeat step 2 until all redundant path sets B′ are traversed. l All paths in the table are used to obtain all scheduling results of flow l and the time slot resource table after scheduling. l , combine all scheduling results of flow l into the scheduling set of flow l.

Images