Differential service flow deterministic transmission scheduling method and system for satellite-ground integrated network

Abstract

The application provides a kind of star-ground fusion network differentiated service flow deterministic transmission scheduling method and system, belongs to communication network technical field, according to the total demand of service flow, based on the elastic transmission algorithm of cache utilization of genetic, calculate the complementary star-ground cross-domain peak cache resource reserved, and carry out the adaptive optimization of cache resource in domain, obtain the domain node resource allocation result;According to node resource allocation result, based on the transient routing and variable queue algorithm of deep reinforcement learning, the two-dimensional scheduling of routing-queue is carried out to service flow, and deterministic transmission scheduling is carried out to meet the differentiated cache capacity demand and time demand of service flow.The application makes up the defects of traditional STIN network architecture, such as lack of differentiated deterministic service concept and loose adaptation between protocol stack levels, significantly improves the deterministic communication guarantee capability of network in different scenarios, and provides strong support for efficient operation and reliable transmission of STIN.

Description

Technical Field

[0001] This invention relates to the field of communication network technology, and specifically to a deterministic transmission scheduling method and system for differentiated service flows in a space-ground converged network. Background Technology

[0002] Driven by massive terminal access, 5G networks have developed rapidly and been deployed on a large scale. Against this backdrop, the need for geographically extensive coverage in communication networks has become increasingly urgent. At the same time, the focus of network functions has shifted from simply "reliable transmission" to providing "differentiated and customizable services." New network applications have placed more stringent requirements on the quality of service standards for network latency and jitter, prompting communication networks to evolve from the traditional best-effort service model to a deterministic service model.

[0003] However, several challenges remain. First, existing terrestrial mobile communication systems are limited by their coverage area, making it difficult to achieve seamless global communication. While Satellite-Terrestrial Integrated Networks (STINs) can provide universal, consistent, and scalable services by integrating satellite and terrestrial networks, effectively alleviating coverage blind spots to some extent, STINs themselves lack the capability to support customized and differentiated services. Second, the traditional internet can only reduce end-to-end average latency to the tens of milliseconds and cannot guarantee strict latency limits. While existing deterministic technologies have made some progress in specific scenarios, a mature transmission scheduling scheme suitable for the vast STIN system has not yet been developed. Third, the loose adaptation between protocol stack layers and the lack of global control and local optimization methods result in a lack of coordination between upper-layer network services and lower-layer network resources, and between large-scale transmission and small-scale scheduling, making it difficult to guarantee the stability and continuity of transmission scheduling.

[0004] Given the aforementioned shortcomings, there is an urgent need to explore and construct a more complete space-ground converged network architecture to meet the differentiated needs of massive services for deterministic services. Furthermore, due to the interdisciplinary complexity of communication networks, a top-down system design based on application requirements is essential to ensure the widespread applicability of the results. Summary of the Invention

[0005] The purpose of this invention is to provide a deterministic transmission scheduling method and system for differentiated service flows in a space-ground converged network, so as to solve at least one of the technical problems existing in the background art, such as the inability of traditional STIN network architecture to achieve deterministic communication guarantee.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a deterministic transmission scheduling method for differentiated service flows in a space-ground converged network, comprising: calculating and reserving complementary cross-domain peak buffer resources for space-ground based on a genetic buffer utilization elastic transmission algorithm according to the total demand of the service flows, and performing adaptive optimization of buffer resources within the domain to obtain the node resource allocation results for the domain; based on the node resource allocation results, performing two-dimensional routing-queue scheduling for the service flows based on a deep reinforcement learning transient routing and variable queue algorithm to perform deterministic transmission scheduling, so as to meet the differentiated buffer capacity and time requirements of the service flows; the deep reinforcement learning-based transient routing and variable queue algorithm includes: on the basis of a deep Q network, introducing the idea of ​​Double DQN to form a dual network structure in which action selection and target Q value calculation are separated. The two networks have the same structure but different parameters. Action selection is completed by the online network, while the target Q value calculation is completed by the target network.

[0008] Secondly, this invention provides a deterministic transmission scheduling system for differentiated service flows in a space-ground converged network, comprising: a resource adaptation module, used to calculate and reserve complementary cross-domain peak cache resources based on a genetic buffer utilization elastic transmission algorithm according to the total demand of the service flow, and to perform adaptive optimization of cache resources within the domain to obtain the domain node resource allocation result; and a differentiated scheduling module, used to perform two-dimensional routing-queue scheduling of the service flow based on the node resource allocation result and a transient routing and variable queue algorithm based on deep reinforcement learning, to perform deterministic transmission scheduling to meet the differentiated buffer capacity and time requirements of the service flow; wherein, the transient routing and variable queue algorithm based on deep reinforcement learning includes: on the basis of a deep Q network, introducing the idea of ​​Double DQN to form a dual network structure in which action selection and target Q value calculation are separated. The two networks have the same structure but different parameters. Action selection is completed by the online network, while the target Q value calculation is completed by the target network.

[0009] Thirdly, the present invention provides a non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the deterministic transmission scheduling method for differentiated service flows in a space-ground converged network as described in the first aspect.

[0010] Fourthly, the present invention provides a computer device including a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the deterministic transmission scheduling method for differentiated service flows in a space-ground converged network as described in the first aspect.

[0011] Fifthly, the present invention provides an electronic device, comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the deterministic transmission scheduling method for differentiated service flows in the space-ground converged network as described in the first aspect.

[0012] The beneficial effects of this invention are as follows: Starting from actual business needs, it satisfies the differentiated buffer capacity and time requirements of transmission scheduling through two modules: resource adaptation and differentiated scheduling, thereby achieving end-to-end low-latency and high-reliability transmission of differentiated service flows. This transmission scheduling method differentially and in parallel supports applications and services with different deterministic requirements. While maintaining low network configuration overhead, it improves the effectiveness and efficiency of service flow transmission scheduling, enhances network operability and resource utilization, and compensates for the scalability shortcomings of strict time scheduling mechanisms.

[0013] The advantages of additional aspects of the invention will be set forth more clearly in the following description or will be learned by practice of the invention. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of the deterministic satellite-ground fusion network architecture described in an embodiment of the present invention.

[0016] Figure 2 This is a flowchart illustrating the deterministic technique described in an embodiment of the present invention.

[0017] Figure 3 This is a flowchart of the resource adaptation module described in an embodiment of the present invention.

[0018] Figure 4 This is a flowchart of the differentiated scheduling module described in an embodiment of the present invention. Detailed Implementation

[0019] To facilitate understanding of the present invention, the present invention will be further explained and described below with reference to the accompanying drawings and specific embodiments. However, the specific embodiments do not constitute a limitation on the embodiments of the present invention.

[0020] Those skilled in the art should understand that the accompanying drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily essential for implementing the present invention.

[0021] This invention provides a Deterministic Satellite-terrestrial Network (DetSTIN) architecture and a service flow scheduling method. The DetSTIN architecture includes a deterministic technology framework and a Satellite-terrestrial Network (STIN) entity. The deterministic technology framework provides layered deterministic services, enabling smooth interconnection and integration of heterogeneous networks. The STIN entity network consists of fixed-mobile terminals, terrestrial networks, and satellite networks, achieving end-to-end low-latency and high-reliability transmission of massive differentiated services under the support of the deterministic technology framework. The service flow scheduling method includes: 1) Resource adaptation: dynamically allocating node cache resources through inter-domain peak cache resource reservation and intra-domain adaptive cache optimization to minimize resource allocation overhead while maintaining network performance; 2) Differentiated scheduling: optimizing transmission scheduling efficiency while ensuring service flow transmission effectiveness by dynamically deciding on transmission paths and queue selection. The DetSTIN architecture and transmission scheduling method provided by this invention make up for the shortcomings of traditional STIN network architecture, such as the lack of differentiated deterministic service concepts and loose adaptation between protocol stack layers. It significantly improves the deterministic communication guarantee capability of the network in different scenarios and provides strong support for the efficient operation and reliable transmission of STIN.

[0022] Example 1

[0023] In this embodiment 1, a deterministic transmission scheduling system for differentiated service flows in a space-ground converged network is first provided, including: a resource adaptation module (i.e., a Centralized Intelligent Control Center, CICC), used to calculate and reserve complementary cross-domain peak cache resources based on a genetic buffer utilization elastic transmission algorithm according to the total demand of the service flow, and to perform adaptive optimization of the cache resources within the domain to obtain the domain node resource allocation results; wherein, the genetic buffer utilization elastic transmission algorithm considers the cache resource configuration overhead. Throughput τ and resource utilization η, i.e.:

[0024]

[0025] stτ≥τ req ,η≥η req

[0026] Where Λ, Γ, and Π are the weight coefficients of each indicator, and Λ>Γ>Π, Λ+Γ+Π=1; This represents the overhead of normalized cache resource configuration.

[0027] The differentiated scheduling module (i.e., the Domain Controller, DC) is used to perform two-dimensional routing-queue scheduling of service flows based on node resource allocation results and a transient routing and variable queue algorithm based on deep reinforcement learning. This deterministic transmission scheduling is used to meet the differentiated buffer capacity and time requirements of service flows. The transient routing and variable queue algorithm based on deep reinforcement learning includes: on the basis of deep Q-network, the idea of ​​Double DQN is introduced to form a dual network structure in which action selection and target Q-value calculation are separated. The two networks have the same structure but different parameters. Action selection is completed by the online network, while the target Q-value calculation is completed by the target network.

[0028] In this embodiment 1, the above-described system is used to implement a differentiated service flow transmission scheduling method based on a space-ground converged network. This method includes: using a resource adaptation module to calculate, based on the total demand of the service flow and a genetically-based elastic transmission algorithm for cache utilization, reserving complementary cross-domain peak cache resources for space-ground communication; and performing adaptive optimization of cache resources within the domain to obtain the domain-specific node resource allocation results. The genetically-based elastic transmission algorithm for cache utilization considers the overhead of cache resource configuration. Throughput τ and resource utilization η, i.e.:

[0029]

[0030] stτ≥τ req ,η≥η req

[0031] Where Λ, Γ, and Π are the weight coefficients of each indicator, and Λ>Γ>Π, Λ+Γ+Π=1; This represents the normalized cache resource configuration overhead. Based on node resource allocation results, a differentiated scheduling module uses a transient routing and variable queue algorithm based on deep reinforcement learning to perform two-dimensional routing-queue scheduling on service flows, enabling deterministic transmission scheduling to meet the differentiated cache capacity and time requirements of service flows. The transient routing and variable queue algorithm based on deep reinforcement learning includes: introducing the concept of Double DQN on top of a deep Q-network, forming a dual-network structure where action selection and target Q-value calculation are separated. These two networks have the same structure but different parameters; action selection is performed by the online network, while the target Q-value calculation is performed by the target network.

[0032] The genetically based caching algorithm utilizes an elastic transfer algorithm to eliminate node degree constraints during initialization by randomly generating the initial population within a reasonable range. During crossover, a reflexive operation is introduced for individuals that do not undergo crossover, further enriching population diversity. During mutation, the mutation magnitude is adaptively adjusted according to the node's degree value, ensuring that nodes with higher degree values ​​are allocated to higher resource allocations, as nodes with higher degree values ​​typically require more resources due to their central role in the network. Conversely, nodes with lower degree values ​​require less resources and are allocated to lower resource allocations.

[0033] In the transient routing and variable queue algorithm of deep reinforcement learning, two state matrices are constructed based on the environmental input. The first state matrix describes the transmission requirements of the business flow, including the source address, destination address, flow size and transmission latency requirements. The second state matrix records the queue buffer capacity of each node across the domain. Based on these two state matrices, a decision is made to be executed in two stages.

[0034] The first stage involves global routing decisions, which select a suitable end-to-end transmission path based on the transmission latency requirements of the service flow. This path selection is carried out in two steps: first, a suitable transmission medium is selected, and then a specific route is determined based on the selected medium to ensure that the path meets the latency and reliability requirements of the service flow.

[0035] The second phase involves local queue scheduling decisions; selecting a suitable sending queue at the source node to determine the sending time of the service flow; for AVB and BE flows, selecting a maximum waiting delay to achieve time-isolated queuing of differentiated service flows; if the local scheduling decision cannot meet the latency requirements or the node's buffer capacity is insufficient, a compensation strategy is triggered to attempt rescheduling; within an acceptable latency range, service flows that fail to be scheduled according to the initial decision due to upstream gating failure or arrive early are rescheduled to a relatively suboptimal queue and suffer additional latency penalties.

[0036] After completing the scheduling decision, the system receives the corresponding reward based on the decision; the current state, action, and reward information are stored in the experience replay pool, and then the system transitions to the next state to continue the decision-making process.

[0037] Example 2

[0038] In this second embodiment, utilizing the concepts and characteristics of intelligent collaborative networks, a deterministic space-ground fusion network (DetSTIN) architecture is provided to support deterministic communication, such as... Figure 1 As shown, the architecture consists of two parts: a deterministic technology framework and a Space-Ground Integrated Network (STIN) entity. The deterministic technology framework provides layered deterministic services for the STIN entity network, enabling smooth interconnection and integration of heterogeneous networks.

[0039] In this embodiment, the deterministic technical framework is responsible for making transmission scheduling decisions and supports the STIN entity through deterministic transmission scheduling, including:

[0040] The Centralized Intelligent Control Center (CICC) is responsible for overall management and control, determining peak resource allocation, and making global routing decisions.

[0041] CICC is responsible for global control, achieving centralized management of the entire network through a deterministic technical framework. It oversees the global time slot-period mapping, ensuring that each time slot is allocated and used according to the predetermined period, thereby providing time guarantees for deterministic transmission;

[0042] CICC determines peak resource allocation by rationally allocating peak resources based on the overall network resource status and service requirements, ensuring that the network can still operate stably during peak service traffic periods and meet the requirements of deterministic transmission.

[0043] CICC makes global routing decisions, taking into account factors such as network topology, link status and service priority, and selects the optimal transmission path to achieve deterministic latency performance in end-to-end (E2E).

[0044] Domain Controllers (DCs) work in conjunction with CICCs to be responsible for the control and management within the domain;

[0045] The DC receives global routing decisions and resource allocation instructions from the CICC and refines them to specific links and nodes within its domain, ensuring the rational use of resources and the orderly transmission of services within the domain.

[0046] The DC is responsible for dynamically optimizing peak resource allocation, monitoring the resource usage of links and nodes within the domain in real time, and flexibly adjusting resource allocation strategies according to changes in service traffic to cope with sudden service peaks and ensure transmission performance and stability within the domain.

[0047] DC uses segmented delay quotas allocated by CICC to precisely allocate delay for links within each domain and across domains, ensuring that each link can transmit data according to predetermined delay requirements, thereby achieving deterministic streaming within the domain.

[0048] In this embodiment, the STIN (Space-Ground Convergence Network) entity is responsible for the end-to-end transmission of differentiated service flows generated by fixed-mobile terminals, including:

[0049] Fixed-mobile terminals include a large number of devices equipped with dual-mode communication chips. Through the automatic switching function of the dual-mode communication chips, seamless switching of service flows between terrestrial networks and satellite networks is achieved, thereby ensuring the efficient and accurate forwarding of service flows.

[0050] The terrestrial network comprises a terrestrial communication network, gateway stations, and a Central Intelligent Control Center (CICC). The gateway stations are responsible for establishing satellite-to-ground laser communication links with the satellite network, facilitating efficient data exchange between the ground and satellite systems. Furthermore, the CICC, as DetSTIN's global control and management center, supports deterministic transmission scheduling and provides high-quality services between the satellite and the ground.

[0051] The satellite network comprises geostationary orbit (GEO) satellites, medium Earth orbit (MEO) satellites, and low Earth orbit (LEO) satellites. GEO satellites are equipped with advanced onboard monitoring and control capabilities, acting as logical sub-controllers. They receive peak resource allocation and global routing decisions from the ground-based CICC (Central Integrated Control Center), and based on these decisions, formulate control commands and distribute them to the MEO and LEO satellites. The MEO and LEO satellites perform deterministic forwarding of service flows via highly reliable inter-satellite laser links, enabling seamless and efficient streaming across satellites.

[0052] In this embodiment, based on the above architecture, a service flow transmission scheduling method under the deterministic satellite-ground converged network (DetSTIN) architecture is provided. The method includes two main steps: resource adaptation and differentiated scheduling.

[0053] like Figure 3 As shown, the resource adaptation process includes the following steps:

[0054] Step S101: Inter-domain peak buffer resource reservation. Specifically, based on the overall service flow buffer capacity requirements, peak buffer resources are predicted and reserved based on historical data characteristics to ensure the stability and sufficiency of inter-domain transmission. This includes allocating sufficient buffer resources to end-domain nodes to meet the overall service flow buffer requirements, and resource complementarity between the ground forwarding domain and the satellite forwarding domain to collaboratively support transmission.

[0055] Step S102: Intra-domain cache resource configuration initialization. Specifically, a genetic-based buffer utilization algorithm for elastic transmission (BUDGET) is used. A population is used to represent the set of all intra-domain cache resource configuration schemes, with each individual in the population representing one of these schemes. Individuals are represented by chromosomes, and the genes (encoding) on ​​the chromosomes represent the amount of cache resources allocated to a particular node.

[0056] Using chromosome C i To represent an individual in a population, Where LT is the length of the chromosome, and chromosome encoding uses natural number encoding to associate point positions with network node indices, and genes on the chromosome. 1≤K≤LT. It should be noted that when iterating through domain-wide cache resource configuration decisions based on BUDGET, This represents the cache resource configuration decision for the k-th node within the domain in this scheme. For With (0.1C) max C max The value is characterized by random numbers within a certain range. Specifically, the gateway node is fixed at C. max .in, This represents the overall business flow caching resource requirements. The total amount of caching resources configured within the domain shall not exceed the peak caching resources reserved in step S101. Different caching resource configuration schemes within the domain correspond to different initial individuals, and the initial individuals form the initial population.

[0057] Individuals form a population in Let C be the individual in the population after the j-th iteration. i , where p is the number of individuals in the population. For an individual have:

[0058]

[0059] When iterating through in-domain cache resource configuration decisions based on BUDGET, this population corresponds to the potential solutions for in-domain cache resource configuration decisions. The algorithm begins by initializing the population, i.e.

[0060] Step S103: Optimize intra-domain cache resource configuration. Specifically, a genetically based cache utilization elastic transfer algorithm (BUDGET) is used to select the best individual in the current population, and crossover / reflexive mutation operations are performed on the selected individuals to obtain potential solutions. Specifically, the following three processes are executed to iteratively approach the optimal solution.

[0061] Step S1031: Select the optimal individual based on fitness. Specifically, the fitness function is used as the indicator for selecting the optimal individual. Its corresponding value is calculated to measure each individual in the population, excluding individuals with low fitness. The formula for calculating the fitness function is:

[0062]

[0063] stτ≥τ req ,η≥η req

[0064] Where Λ, Γ, and Π are the weight coefficients of each indicator, and Λ>Γ>∏, Λ+Γ+∏=1. This represents the normalized cache resource allocation overhead. Fitness directly reflects network performance; under the premise of maintaining a certain throughput and resource utilization, lower resource allocation overhead corresponds to higher fitness.

[0065] Individuals with high fitness have a greater chance of passing on their genes to the next generation. Therefore, the selection process employs a probability-based roulette wheel algorithm. First, an elite strategy is used, directly replicating the top 5% of fittest individuals into the offspring population.

[0066] Elites = Top σ (N1(SortDesc(fitness)))

[0067] Then parental selection is performed. Survival rate is used to evaluate the solution quality of the genetic iteration process, i.e. Clearly, higher survivability implies higher fitness, resulting in a better individual. Individuals with higher survivability have more opportunities for reproduction in the next generation, and vice versa. A roulette wheel selection method can be used to determine which individuals meet these criteria. Individuals r∈[0,1) are considered as parental individuals.

[0068] Step S1032: Parental chromosome crossover / reflexivity. Specifically, once the parents are selected, a non-gateway node is randomly chosen as the operation point for probability-based operations. When probability p... c When the probability is 1-p, perform a single-point crossover; otherwise, when the probability is 1-p... c In the case of crossover, the genes of offspring 1 are inherited from parent 1 before the operation point and from parent 2 after the operation point, while the genes of offspring 2 are inherited from parent 2 in the opposite way. In the case of reflexivity, the genes of each offspring are mirrored at the operation point, producing two self-reflecting offspring.

[0069] Step S1033: Offspring chromosome gene mutation. Specifically, for the resulting offspring population, including retained elite individuals and individuals generated through parental chromosome crossing over or reflexivity, gene mutations are performed with an adaptation probability p. m Adaptive mutation is performed based on the population's individual populations and average fitness. The mutation rate is dynamically adjusted. Individuals with lower fitness are assigned a higher probability of significant mutation, aiming to discover new potential solutions. The specific form of the mutation operator is:

[0070]

[0071] in, This represents the difference between the maximum and minimum mutation probabilities. Meanwhile, the magnitude of the mutation is negatively correlated with the node degree. The larger the node degree, the less resources are reduced or the more resources are added with each change. By taking into account this adaptive mutation based on node degree, the designed algorithm can converge quickly.

[0072] Step S104: Optimal Solution Search and Iteration. Repeat step S103 to perform elite retention / parental selection, chromosome crossover / reflexivity, and chromosome gene mutation until the iteration terminates. This will yield the optimal solution for the allocation of cached resources within the domain, i.e., the optimal resource adaptation decision.

[0073] Thanks to its unique adaptive mutation operator technique, the resource adaptation algorithm, which incorporates elite preservation and innovative reflexive operations, outperforms the classic genetic algorithm in terms of population diversity, convergence speed, and convergence accuracy.

[0074] like Figure 4 As shown, the process of differentiated scheduling includes the following steps:

[0075] Step S201: Network Status and Differentiated Service Flow Transmission Demand Awareness. Specifically, the domain's DC senses the transmission requirements of service flows, including key information such as source address, destination address, flow size, and transmission latency requirements, and reports this data to CICC. Simultaneously, the domain's cache resource configuration is also reported to CICC for overall coordination and optimization management. The D3QN agent deployed at CICC constructs two state matrices based on the environmental input. The first state matrix describes the service flow transmission requirements, which can be represented as:

[0076]

[0077] The second state matrix characterizes the queue cache capacity of each node across domains. It can be represented as:

[0078]

[0079] in, and Representing fields respectively Middle node v i and v j The remaining cache capacity of the s-th queue, D represents Δ Node v in the domain k The remaining buffer capacity of the s-th queue, Δ∈{β,γ}.

[0080] Step S202: Global Route Selection Decision. Specifically, the agent first selects a suitable transmission medium (i.e., terrestrial or satellite network) based on the transmission latency requirements of the service flow. Based on this, a flexible latency ceiling is allocated to each domain, and a suitable segment route is selected from all feasible segment routes for the service flow. These segment routes form a global route to ensure that the path meets the latency and reliability requirements of the service flow. Furthermore, the end-to-end latency ceiling of the service flow is reasonably allocated to each domain. Within a domain, segment route selection can be flexibly performed according to the latency requirements of the service flow, filtering suitable routes from all feasible segment routes. These segment routes are then concatenated and integrated to form a global route. For example, for time-sensitive service flows, segment routes with relatively fewer hops are prioritized to reduce latency; while for best-effort service flows, segment routes with more hops can be selected. Through this differentiated path selection strategy, latency optimization of the service flow can be achieved on a large scale.

[0081] Step S203: Local queue scheduling decision. Specifically, queue scheduling distinguishes between end-domain scheduling and forwarding-domain scheduling.

[0082] The endpoint uses a Cycling Queuing and Forwarding (CQF) scheduling mechanism. Through a cyclic gating list, the two queues are scheduled in a cyclic manner. At the beginning of a cycle, the receive gate of queue Q0 opens first, preparing for receiving service flows. Simultaneously, queue Q0, which is in the sending state, transmits all service flows in its queue to the downstream node within that cycle. After a single cycle, the working state of the two queues switches according to the gating list. This mechanism implicitly incorporates link latency into the upstream and downstream cycles. The upstream node's queue Q... c The service flow (c∈{0,1}) issued in period Tξ will be in the Q of the downstream node. (c+1)％2 The queue is received, and in period T ξ+1 Move to the next hop. The agent selects an appropriate sending queue based on the latency requirements of the service flow and the network status to determine the sending time of the service flow, that is, to enter the current queue Q0 to send in the next cycle, or to enter the next queue Q1 to wait for one cycle before sending.

[0083] The forwarding domain employs a wide-area deterministic IP (DIP) scheduling mechanism. The opening and closing of the transmit and receive gates are controlled by a periodically rotating gating list to switch the working states of m queues. In one cycle, only one queue is in the transmit state, represented by 0, while the remaining M-1 queues are in the receive state, represented by 1. Then queue Q... m During the transmission period T ξ The working status is:

[0084]

[0085] In regular nodes, a DIP mechanism with standard mapping between upstream and downstream node queues is used, where the upstream node queue Q... c The service flow originating from (c∈{0,1,…,M-1}) experiences a propagation delay. After reaching the downstream node, the queue's working state has switched. Therefore, the service flow enters the queue of the downstream node. wait The data is then sent to the next node. At the aggregation node, given that indiscriminate mapping of mixed service flows to the same queue in downstream nodes could potentially lead to latency inversion, a priority-based service time window configuration is constructed based on the standard DIP mechanism. For TT flows with low latency requirements, due to their strict latency requirements, they are always exempt from secondary scheduling; while for AVB and BE flows, the agent selects a maximum waiting latency to achieve time-isolated queuing of differentiated service flows. That is, high-bandwidth audio and video service flows are decided by the agent to be delayed, and at most delayed by one queue; best-effort flows are decided by the agent to be delayed, and at most delayed by two queues. By adding an upper limit constraint to the queuing offset, it is possible to effectively avoid low-priority service flows from being trapped in an infinite delay, and prevent them from "starving" due to long-term occupation of queue resources by high-priority service flows.

[0086] Step S204: Compensation Strategy Rescheduling. Specifically, if the local scheduling decision cannot meet the latency requirements or the node's buffer capacity is insufficient, the DC will trigger a compensation strategy to attempt rescheduling. Within acceptable latency ranges, service flows that failed to be scheduled according to the initial decision due to upstream gating failure or arrived early can be rescheduled to a relatively suboptimal queue and incur additional latency penalties. This ensures that even with scheduling deviations, the DC maintains a degree of flexibility to handle unexpected situations while minimizing transmission interruptions. Notably, the rescheduling process still considers the service flow priority and the current load status of available queues, ensuring that high-priority service flows are minimally affected. This strategy takes effect automatically, ensuring that service flows are transmitted within the global latency requirements without further intervention from the agent.

[0087] In summary, the Deterministic Deterministic STIN Network (DetSTIN) architecture and service flow scheduling method provided in this embodiment solves the problems of traditional network architectures lacking differentiated deterministic service concepts and having loose adaptation between protocol stack layers by adopting a deterministic technical framework to support layered deterministic services for STIN network entities. Simultaneously, through resource adaptation and differentiated scheduling, it minimizes cache configuration overhead while ensuring network performance, and achieves end-to-end low-latency and high-reliability transmission of massive differentiated service flows, achieving both effectiveness and efficiency.

[0088] Example 3

[0089] In this embodiment 3, a deterministic satellite-terrestrial network (DetSTIN) architecture is first provided, including a deterministic technical framework and a satellite-terrestrial network (STIN) entity. The deterministic technical framework provides layered deterministic services, is responsible for making transmission scheduling decisions, and supports the STIN entity through deterministic transmission scheduling, enabling smooth interconnection and integration of heterogeneous networks. The STIN entity interfaces with network services and network nodes, and is responsible for the end-to-end transmission of differentiated service flows generated by fixed-mobile terminals. Supported by the deterministic technical framework, it achieves low-latency, high-reliability transmission, thereby ensuring the determinism of network service communication processes.

[0090] Specifically, the deterministic technical framework includes:

[0091] The Centralized Intelligent Control Center (CICC) is responsible for global management and control, determining peak resource allocation, and making global routing decisions. CICC manages the entire network centrally through a deterministic technology framework. It oversees the global time slot-period mapping, ensuring that each time slot is allocated and used according to a predetermined period, thus providing time guarantees for deterministic transmission. CICC determines peak resource allocation, rationally allocating peak resources based on the network's global resource status and service demands, ensuring stable network operation during peak traffic periods and meeting the requirements of deterministic transmission. CICC makes global routing decisions, comprehensively considering factors such as network topology, link status, and service priorities to select the optimal transmission path to achieve end-to-end deterministic latency performance. Domain Controllers (DCs) work in conjunction with CICC, responsible for control and management within their domains. DCs receive global routing decisions and resource allocation instructions from CICC and refine them to specific links and nodes within their domain, ensuring the rational utilization of resources and the orderly transmission of services within the domain. The Data Center (DC) is responsible for dynamically optimizing peak resource allocation, monitoring the resource usage of links and nodes within its domain in real time, and flexibly adjusting resource allocation strategies based on changes in service traffic to cope with sudden service peaks and ensure transmission performance and stability within the domain. Based on the segmented delay quotas allocated by CICC, the DC performs precise delay allocation for each link within the domain and for cross-domain links, ensuring that each link can transmit data according to predetermined delay requirements, thereby achieving deterministic streaming transmission within the domain.

[0092] Specifically, the STIN entity includes: Fixed-Mobile Terminals (PCTs) comprising a large number of devices equipped with dual-mode communication chips. Through the automatic switching function of these chips, seamless switching of service flows between the terrestrial and satellite networks is achieved, ensuring efficient and accurate forwarding of service flows. The terrestrial network includes a terrestrial communication network, gateway stations, and CICCs. The gateway stations are responsible for establishing satellite-to-ground laser communication links with the satellite network, facilitating efficient data exchange between the terrestrial and satellite systems. Furthermore, the CICC, as the global control and management center for DetSTIN, supports deterministic transmission scheduling and provides high-quality services between satellites and the ground. The satellite network includes Geosynchronous Earth Orbit (GEO) satellites, Medium Earth Orbit (MEO) satellites, and Low Earth Orbit (LEO) satellites. GEO satellites are equipped with advanced onboard monitoring and control capabilities. As logical sub-controllers, they receive peak resource allocation and global routing decisions from the ground-based CICC and formulate control commands based on these decisions, distributing them to MEO and LEO satellites. MEO and LEO satellites perform deterministic forwarding of service flows through highly reliable inter-satellite laser links, enabling seamless and efficient streaming across satellites.

[0093] For a more concrete description, in this embodiment, DetSTIN is modeled as an undirected graph G(V,ε), where the sets of nodes and links are respectively defined by V = {v0, v1, ..., v...} n} and ε={e∣e kj =(v k ,v j ),e jl =(v j ,v l ),…}express.

[0094] In this embodiment, a service flow transmission scheduling method is implemented based on the DetSTIN architecture described above, which includes two main steps: resource adaptation and differentiated scheduling.

[0095] The resource adaptation module ensures that the buffer capacity of the service flow transmission nodes meets the requirements. This module deploys a buffer utilization-based algorithm for elastic transmission (BUDGET), which reserves complementary cross-domain peak buffer resources based on the total demand of the service flow and performs adaptive optimization of buffer resources within the domain. At the same time, it passes the node resource allocation results to the differential scheduling module. The differential scheduling module performs two-dimensional intelligent scheduling of the service flow using routing and queuing. This module deploys a transient routing and variable queue algorithm (TRAVEL) based on deep reinforcement learning. Based on the node resource allocation results generated by the resource adaptation module, it performs deterministic transmission scheduling to meet the differentiated buffer capacity and time requirements of the service flow.

[0096] Specifically, the resource adaptation module includes:

[0097] CICC-driven inter-domain peak buffer resource reservation, from the perspective of ensuring transmission, analyzes the characteristics of cross-domain service flows and historical data to reasonably predict and reserve peak buffer resources, thereby ensuring the stability and sufficiency of inter-domain transmission. In this module, CICC allocates sufficient buffer resources to end-domain nodes to fully meet the buffering requirements of service flows, ensuring the reliability and integrity of data during end-to-end transmission. Resources between the terrestrial forwarding domain and the satellite forwarding domain are allocated by CICC in a complementary manner to collaboratively support transmission.

[0098] Domain-controlled adaptive caching optimization, from a resource-saving perspective, dynamically adjusts cache resource allocation based on real-time network conditions and changes in domain traffic, thereby achieving efficient resource utilization and rapid adaptation. Domain node caching is complementaryly allocated by DCs deploying the BUDGET algorithm, jointly completing multipath transmission of flows.

[0099] Specifically, the differentiated scheduling module includes:

[0100] The global routing plan developed by CICC, from the perspective of large-scale latency optimization, first decides on the forwarding medium based on the size and latency requirements of differentiated service flows, i.e., whether each service flow forwards via satellite network or terrestrial network. Simultaneously, the DC returns feasible paths within the domain to CICC, which then performs segment routing selection and allocates segment latency quotas to domain and inter-domain links based on the routing selection results. Global determinism is achieved through segmented determinism. Appropriate routing planning can reduce unnecessary node hops, thereby reducing overall latency and maximizing the fulfillment of end-to-end transmission requirements for overall service flows.

[0101] Differentiated service flows: The set of service flows generated by massive fixed-mobile terminals is as follows:

[0102]

[0103] Among them, f i Indicates size is The i-th business flow in the flow set, and These are the source node and the destination node, respectively. Furthermore, and These represent the maximum size of the service flow and the maximum latency, respectively. This refers to the unit benefit of end-to-end latency optimization. Based on the size and latency requirements of the service flow, massive service flows are divided into latency-sensitive Time-Triggered (TT) flows, relatively bandwidth-intensive Audio / Video Bridgging (AVB) flows, and Best-Effort (BE) flows that offer no guarantees regarding latency, reliability, or other performance characteristics. For service flow f... i Its end-to-end actual delay It must be less than the latency limit of this service flow. Typically, the scheduling benefits resulting from unit transmission delay optimization are expressed as ξ. TT >ξ AVB >ξ BE Therefore, the time priority of the three streams in transmission scheduling is usually TT > AVB > BE;

[0104] Global routing planning: When mixed service flows are injected into DetSTIN, CICC selects a forwarding medium for each service flow. This indicates that, for differentiated scheduling, TT streams with strict latency requirements typically choose shorter paths, while BE streams with more lenient latency requirements can choose longer paths. Specifically, all service streams generated by airborne monitoring missions and BE streams originating from the ground can be relayed via satellite, denoted as... Other ground flows are forwarded through ground nodes, denoted as... This provides flexibility in balancing network load and optimizing resource utilization. For the selection of path p f Business flow f i Its end-to-end actual delay Due to link latency and node latency Composition, that is

[0105]

[0106] Where e is a directed arc segment connecting adjacent nodes, and l is the global link formed by e.

[0107] Intra-domain segment routing selection: Service flow f i The complete transmission path from the source end domain to the destination end domain via a ground or satellite relay domain can be: Among them, end domain D α The path segment (v0, v1, ..., v) λ )and It is usually fixed, while the ground forwarding domain D β or satellite relay domain D γ Segment routing (v) λ+1 ,v λ+2 ,…,v μ This is flexible and varied. Here, λ∈(0,|p f |) and μ∈(0,|p f |) represents the node index. Therefore, the link delay can be calculated from... To measure, among which e kj Length, This indicates the speed of propagation; different forwarding media have different propagation speeds, i.e.

[0108]

[0109] The local queue selection mechanism, defined by the Data Center (DC), starts from a small-scale latency optimization perspective. Based on the size and latency requirements of differentiated service flows, it selectively injects service flows into designated queues at each node within the domain for intra-domain forwarding cycle decisions, with higher-priority queues corresponding to earlier forwarding cycles. This fine-grained queue selection can effectively utilize reserved buffer resources, flexibly adapt to mixed service flows, and improve the latency determinism and predictability of end-to-end transmission.

[0110] Local queue selection: This method guides service flows to be forwarded periodically at each node through a deterministic transport scheduling mechanism. Essentially, it employs a deterministic timeslot queue management approach, enabling highly flexible per-flow scheduling and effectively utilizing node timeslot buffers. Service flow f i At node v j The delay at the point is determined by the transmission delay. Queue delay Waiting delay It consists of three parts, namely

[0111]

[0112] Among them, transmission delay in Indicates from node v j The emitted arc segment e jlBandwidth. Queue latency. Where f j This indicates that the position in the queue is f. i Previous business processes required waiting for the preceding business process to be sent before they could proceed. i The transmission delay and queuing delay can approach the nanosecond level, considering the high bandwidth of laser and fiber optic communications. Furthermore, the waiting delay... It is related to the intrinsic gating period of the queue.

[0113] Deterministic transmission scheduling mechanism: Used for queuing and forwarding service flows among forwarding devices in a network. By implementing a deterministic gating strategy, it ensures that service flows are sent within a predetermined period, thereby achieving precise time control. Specifically, M gating queues are set up at the forwarding device port. Each queue has two working states: sending and receiving, controlled by sending and receiving gates respectively. The switching of queue working states is achieved by controlling the opening and closing of the sending and receiving gates through a periodically rotating gating list. The time is divided into equal-length transmission periods T. In one period, only one queue is in the sending state, represented by 0, and the remaining M-1 queues are in the receiving state, represented by 1. Then queue Q m During the transmission period T ζ The working status is:

[0114]

[0115] Given this queue configuration, set the period duration accordingly. To maintain send-receive alignment between the two mechanisms, i.e.

[0116]

[0117] This balance ensures that service flows can be seamlessly coordinated and forwarded within domains with different cycle lengths, thereby guaranteeing the determinism of service flow transmission.

[0118] End-domain deterministic transmission scheduling: In the end-domain, a Cycling Queuing and Forwarding (CQF) mechanism with precise time synchronization is employed. Two queues (M=2) periodically switch between sending and receiving states to alternately transmit service flows, avoiding queue congestion and uncertain latency. Using a cyclic gating list, the two queues are scheduled cyclically. At the beginning of a cycle, the receive gate of queue Q0 opens first, preparing for receiving service flows. Simultaneously, queue Q0, in the sending state, transmits all service flows in the queue to the downstream node within that cycle. After a single cycle, the working state of the two queues switches according to the gating list. This mechanism implicitly incorporates link latency into the upstream and downstream cycles. The upstream node's queue Q... c In period T ζThe sent service flow will be in the Q of the downstream node. (c+1)％2 The queue is received, and in period T ζ+1 Send to the next hop. Due to business flow f i The routes in the endpoint domain are (v0, v1, ..., v λ )and The transmission times are respectively

[0119]

[0120] Forwarding Domain Deterministic Transport Scheduling: In wide area networks, a Deterministic IP (DIP) mechanism combining Differentiated Service (DiffServ) principles is adopted. Multiple queues (M≥3) are used to achieve highly flexible flow-by-flow relative queue scheduling for the massive differentiated service flows aggregated in the forwarding domain. The periodic buffering of nodes is effectively utilized, and these flows are scheduled according to priority. This ensures that among flows mapped to the same period, TT flows take precedence over AVB flows, and AVB flows take precedence over BE flows, thereby achieving fine-grained flow differentiation scheduling and effectively guaranteeing the transmission performance and service quality requirements of different flow types. This mechanism introduces four key enhancements:

[0121] First, it transforms traditional time-slot-based absolute queue scheduling into period-based relative queue scheduling. This eliminates the dependency on high-precision clock synchronization between nodes and allows upstream or downstream nodes to operate with misaligned period timings.

[0122] Second, based on the period offset per hop, different types of service flows are assigned to relative queues with specific maximum waiting delays to achieve temporal isolation. The maximum waiting delay is measured by ψ∈{1,2,…,M-1}.

[0123] Third, the wide area nodes are divided into regular nodes and aggregation nodes.

[0124] In regular nodes, a DIP mechanism with standard mapping between upstream and downstream node queues is used, where the upstream node queue Q... c The sent business flow, when the business flow undergoes propagation delay After reaching the downstream node, the queue's working state has switched. or Therefore, the service flow enters the queue of the downstream node. (or ),wait (or Afterwards, it is sent to the next node. Since the link delay between adjacent nodes may not be an integer multiple of the period, we consider the queuing delay as part of the total queuing waiting delay. Therefore, we know that the service flow f...i At node v j The waiting time is:

[0125]

[0126] At the aggregation node, combining the DiffServ concept, upstream and downstream mappings with differentiated maximum wait times are used for different priority service flows before queuing. The maximum wait time setting isolates different types of service flows in terms of time. Specifically, TT flows enter the relative queue without offset, with a maximum wait time of one DIP cycle, ensuring the stability and low latency of TT flow transmission. For AVB flows, they can enter the relative queue with a maximum wait time of one or two DIP cycles. While ensuring a certain level of transmission efficiency, it provides relatively flexible scheduling space, allowing adaptive scheduling based on real-time network conditions to balance resource consumption and transmission timeliness. For BE flows, they are arranged into the relevant queue with a maximum wait time of two or three DIP cycles, allowing BE flows to be scheduled for transmission within a relatively relaxed time limit. This differentiated scheduling ensures that high-priority flows experience minimal latency while maintaining the flexibility of low-priority flows (i.e., AVB and BE), ensuring that BE flows have a fair chance for forwarding and preventing BE flows from being starved. Thus, service flow f i At node v j The waiting time is

[0127]

[0128] in, (or ) represents the business flow f i The relative queue to which it is mapped.

[0129] Fourth, a compensation policy is used to handle special cases. Within acceptable latency ranges, service flows that fail to be scheduled according to the initial decision or arrive early due to upstream gating failure can be rescheduled to a relatively suboptimal queue and incur additional latency penalties. This ensures that even with scheduling deviations, the data center maintains a degree of flexibility to handle unexpected situations while minimizing transmission interruptions. It is worth noting that the rescheduling process still considers the service flow priority and the current load status of available queues, ensuring that high-priority service flows are minimally affected.

[0130] In this embodiment, the genetically based cache-utilizing elastic transfer algorithm (BUDGET) includes:

[0131] An improved genetic algorithm based on fitness, where fitness comprehensively considers the overhead of cache resource allocation. Throughput τ and resource utilization η, i.e.:

[0132]

[0133] stv≥τ req ,η≥η req

[0134] Where Λ, Γ, and ∏ are the weight coefficients of each indicator, and Λ>Γ>∏ and Λ+Γ+∏=1. This represents the normalized cache resource configuration overhead. Inspired by the principles of biological evolution, genetic algorithms simulate the concept of "survival of the fittest." In the process of population evolution, through operations such as selection, crossover, and mutation, only individuals that adapt to the environment are selected and retained, thus making them suitable for global optimization decisions.

[0135] Unlike conventional genetic algorithms, BUDGET eliminates node degree constraints during initialization by randomly generating the initial population within a reasonable range. During crossover, BUDGET introduces a reflexive operation for individuals that do not undergo crossover, further enriching population diversity. In the mutation phase, the algorithm adaptively adjusts the mutation magnitude based on node degree values, ensuring that nodes with higher degree values ​​are allocated higher resources, as these nodes typically require more resources due to their central role in the network. Conversely, nodes with lower degree values ​​require fewer resources and are allocated lower resources. Furthermore, the adaptive mutation operator is fitness-related, assigning higher mutation probabilities to low-fitness individuals with poor performance to encourage exploration. Combined with an elite preservation strategy, this increases population diversity, improves the targeting of exploration, and reduces the required number of generations.

[0136] Population initialization: The domain-wide cache resource configuration scheme is abstracted as an individual. The cache configuration of each node within the domain is characterized by genes on the individual's chromosome. Chromosome encoding uses natural numbers to associate point locations with network node indices. For each point on the chromosome, in (0.1C... max C max Generate a random number within the range to represent the amount of cached resources on the node. The gateway node is fixed at C. max .in, This represents the overall business flow caching resource requirements. Different caching resource configuration schemes within different domains correspond to different initial individuals, and these initial individuals form the initial population.

[0137] Selection: The selection process employs a probabilistic roulette wheel algorithm, where individuals with higher fitness have a greater chance of passing on their genes to the next generation. Fitness directly reflects network performance; under the premise of maintaining a certain throughput and resource utilization, lower resource allocation overhead corresponds to higher fitness. An elitist strategy is used in the selection process, directly replicating the top 5% of fittest individuals into the offspring population, followed by parental selection. This strategy aims to ensure that the genetic information of the optimal individuals is preserved, thereby enhancing the algorithm's stability and convergence speed.

[0138] Reflexive / Crossover: Once a parent node is chosen, a non-gateway node is randomly selected as the operation point. When probability p c When the probability is 1-p, perform a single-point crossover; otherwise, when the probability is 1-p... c At this point, a single-point reflexive operation is performed. In the crossover case, the genes of offspring 1 are inherited from parent 1 before the operation point and from parent 2 after the operation point, while the genes of offspring 2 are inherited from parent 2 in the opposite way. For reflexive operations, the genes of each offspring are mirrored at the operation point, producing two self-reflecting offspring.

[0139] Mutation: For the resulting offspring population, including retained elite individuals and individuals produced through parental crossover or reflexivity, gene mutations occur with an adaptation probability p. m Adaptive mutation is performed based on the population's individual populations and average fitness. The mutation rate is dynamically adjusted. Individuals with lower fitness are assigned a higher probability of significant mutation, aiming to discover new potential solutions. The specific form of the mutation operator is as follows:

[0140]

[0141] in, This represents the difference between the maximum and minimum mutation probabilities. Meanwhile, the magnitude of the mutation is negatively correlated with the node degree. The larger the node degree, the less resources are reduced or the more resources are added with each change. By taking into account this adaptive mutation based on node degree, the designed algorithm can converge quickly.

[0142] This resource-adaptive algorithm achieves a good balance between exploration and utilization. The adaptive mutation strategy plays a key role in exploring new solutions, while the elite preservation strategy emphasizes utilizing existing optimal solutions.

[0143] The transient routing and variable queue (TRAVEL) algorithm based on deep reinforcement learning includes:

[0144] The Dueling Double Deep Q Network (D3QN), based on deep reinforcement learning algorithms, introduces the concept of Double DQN into the Deep Q Network (DQN). This creates a dual-network structure where action selection and target Q-value calculation are separate. These two networks have the same structure but different parameters. Action selection is performed by the online network, while the target Q-value calculation is performed by the target network. This allows for more accurate Q-value estimation and avoids overestimation.

[0145] The D3QN agent constructs two state matrices based on environmental input, providing a foundation for subsequent scheduling decisions. The first state matrix describes the transmission requirements of the service flow, including source address, destination address, flow size, and transmission latency requirements. The second state matrix records the queue buffer capacity of each node across domains, and can be represented as:

[0146]

[0147] in, and Representing fields respectively Middle node v i and v j The remaining cache capacity of the s-th queue, D represents Δ Node v in the domain k The remaining buffer capacity of the s-th queue, Δ∈{β,γ}. The agent formulates a decision a, executed in two phases, based on these two state matrices. t .

[0148]

[0149] Wherein, δ(v) j ) is node v j The number of queues is [number].

[0150]

[0151] The first phase involves global routing decisions, where the agent selects a suitable end-to-end transmission path based on the latency requirements of the service flow. This path selection is performed in two steps: first, a suitable transmission medium (i.e., terrestrial or satellite network) is selected; then, a specific route is determined based on the selected medium to ensure that the path meets the latency and reliability requirements of the service flow.

[0152] The second phase involves local queue scheduling decisions. The agent selects a suitable sending queue at the source node to determine the transmission time of the service flow. For AVB and BE flows, the agent selects a maximum waiting delay to achieve time-isolated queuing of differentiated service flows. If the local scheduling decision cannot meet the latency requirements or the node's buffer capacity is insufficient, the DC will trigger a compensation strategy to attempt rescheduling. Within acceptable latency ranges, service flows that failed to be scheduled according to the initial decision due to upstream gating failure or arrived early can be rescheduled to a relatively suboptimal queue and incur additional latency penalties. This ensures that even with scheduling deviations, the DC maintains a degree of flexibility to handle unexpected situations while minimizing transmission interruptions. Notably, the rescheduling process still considers the service flow priority and the current load status of available queues, ensuring that high-priority service flows are minimally affected. This strategy takes effect automatically, ensuring that service flows are transmitted within the global latency requirements without further intervention from the agent.

[0153] After making a scheduling decision, the agent receives a corresponding reward based on its decision.

[0154]

[0155] This reward is directly related to the latency optimization of the traffic flow; compared to the BE flow, the TT flow receives a higher reward for latency optimization per unit. A uniform negative reward is given in the event of transmission failure. Therefore, the agent prioritizes the optimization of TT and AVB flows, placing latency optimization of the BE flow in a lower priority. A well-designed reward function helps accelerate the agent's learning process, enabling it to converge efficiently to the optimal scheduling strategy.

[0156] Finally, the current state, actions, and reward information are stored in the experience replay pool, and then the process transitions to the next state to continue the decision-making process. Through iterative optimization, the agent gradually improves its scheduling performance for different business flows, enhancing the overall transmission performance of DetSTIN.

[0157] Example 4

[0158] This embodiment 4 provides a non-transitory computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, they implement the differentiated service flow transmission scheduling method based on the satellite-ground converged network as described above.

[0159] Example 5

[0160] This embodiment 5 provides a computer device, including a memory and a processor. The processor and the memory communicate with each other. The memory stores program instructions that can be executed by the processor. The processor calls the program instructions to execute the differentiated service flow transmission scheduling method based on the satellite-ground converged network as described above.

[0161] Example 6

[0162] This embodiment 6 provides an electronic device, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to execute instructions to implement the differentiated service flow transmission scheduling method based on the satellite-ground converged network as described above.

[0163] In summary, the DetSTIN architecture provided by this invention introduces deterministic concepts on top of the traditional STIN architecture, enabling each functional domain to possess independent and collaborative deterministic assurance capabilities. Supported by a deterministic technical framework, it achieves smooth interconnection and integration of heterogeneous networks, thereby ensuring differentiated service support capabilities from the underlying network infrastructure of the STIN entity to the entire architecture. The service flow transmission scheduling method provided by this invention, starting from actual business needs, satisfies the differentiated buffer capacity and time requirements of transmission scheduling through two modules: resource adaptation and differentiated scheduling, achieving end-to-end low-latency and high-reliability transmission of differentiated service flows. It differentially and in parallel supports applications and services with different deterministic requirements. While maintaining low network configuration overhead, it improves the effectiveness and efficiency of service flow transmission scheduling, enhances network operability and resource utilization, and compensates for the scalability deficiencies of strict time-based scheduling mechanisms.

[0164] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that, based on the technical solutions disclosed in the present invention, various modifications or variations that can be made by those skilled in the art without creative effort should be included within the scope of protection of the present invention.

Claims

1. A deterministic transmission scheduling method for differentiated service flows in a space-ground converged network, characterized in that, include: Based on the overall business flow requirements, a genetically-based elastic transmission algorithm for cache utilization is used to calculate and reserve complementary cross-domain peak cache resources between satellite and ground, and adaptive optimization of cache resources is performed within the domain to obtain the domain node resource allocation results. The genetically-based elastic transmission algorithm for cache utilization takes into account the cache resource configuration overhead. Throughput and resource utilization rate ,Right now: ； ； in, , , Here are the weighting coefficients for each indicator: , ; Indicates the overhead of normalized cache resource configuration; Based on the node resource allocation results, a transient routing and variable queue algorithm based on deep reinforcement learning is used to perform two-dimensional routing-queue scheduling for service flows, enabling deterministic transmission scheduling to meet the differentiated buffer capacity and time requirements of service flows. The transient routing and variable queue algorithm based on deep reinforcement learning includes: on the basis of a deep Q-network, the idea of ​​Double DQN is introduced to form a dual-network structure where action selection and target Q-value calculation are separated. These two networks have the same structure but different parameters; action selection is completed by the online network, while the target Q-value calculation is completed by the target network. In the transient routing and variable queue algorithm based on deep reinforcement learning, two state matrices are constructed according to the environmental input. The first state matrix describes the transmission requirements of the service flow, including source address, destination address, flow size, and transmission latency requirements. The second state matrix records the queue buffer capacity of each node across domains. Based on these two state matrices, decisions are made to be executed in two phases.

2. The deterministic transmission scheduling method for differentiated service flows in a space-ground converged network according to claim 1, characterized in that, The genetically based caching algorithm utilizes an elastic transfer algorithm to eliminate node degree constraints during initialization by randomly generating the initial population within a reasonable range. During crossover, a reflexive operation is introduced for individuals that do not undergo crossover, further enriching population diversity. In the mutation phase, the mutation magnitude is adaptively adjusted according to the node's degree value, ensuring that nodes with higher degree values ​​are allocated to higher resource allocations, as nodes with higher degree values ​​typically require more resources due to their central role in the network. Conversely, nodes with lower degree values ​​require fewer resources and are allocated to lower resource allocations.

3. The deterministic transmission scheduling method for differentiated service flows in a space-ground converged network according to claim 1, characterized in that, The first stage involves global routing decisions, which select a suitable end-to-end transmission path based on the transmission latency requirements of the service flow. This path selection is carried out in two steps: first, a suitable transmission medium is selected, and then a specific route is determined based on the selected medium to ensure that the path meets the latency and reliability requirements of the service flow.

4. The deterministic transmission scheduling method for differentiated service flows in a space-ground converged network according to claim 3, characterized in that, The second phase involves local queue scheduling decisions; selecting an appropriate sending queue at the source node to determine the sending time of the service stream; for AVB and BE streams, selecting a maximum waiting delay to achieve time-isolated queuing of differentiated service streams; If the local scheduling decision cannot meet the latency requirements or the node's cache capacity is insufficient, a compensation strategy is triggered to attempt rescheduling. Within the acceptable latency range, business flows that fail to be scheduled according to the initial decision due to upstream gating failure or arrive early are rescheduled to a relatively suboptimal queue and suffer additional latency penalties.

5. The deterministic transmission scheduling method for differentiated service flows in a space-ground converged network according to claim 4, characterized in that, After completing the scheduling decision, the system receives the corresponding reward based on the decision; the current state, action, and reward information are stored in the experience replay pool, and then the system transitions to the next state to continue the decision-making process.

6. A deterministic transmission scheduling system for differentiated service flows in a space-ground converged network, characterized in that, include: The resource adaptation module is used to calculate the reserved complementary cross-domain peak cache resources based on the total demand of the business flow and a genetic cache utilization elastic transmission algorithm. It then performs adaptive optimization of cache resources within the domain to obtain the domain-wide node resource allocation results. The genetic cache utilization elastic transmission algorithm considers the cache resource configuration overhead. Throughput and resource utilization rate ,Right now: ； ； in, , , Here are the weighting coefficients for each indicator: , ; Indicates the overhead of normalized cache resource configuration; The differentiated scheduling module, based on node resource allocation results and using a deep reinforcement learning-based transient routing and variable queue algorithm, performs two-dimensional routing-queue scheduling for service flows, enabling deterministic transmission scheduling to meet the differentiated buffer capacity and time requirements of service flows. The deep reinforcement learning-based transient routing and variable queue algorithm incorporates the concept of Double DQN on top of a deep Q-network, forming a dual-network structure where action selection and target Q-value calculation are separate. These two networks have the same structure but different parameters; action selection is performed by the online network, while the target Q-value calculation is performed by the target network. Within this algorithm, two state matrices are constructed based on the environmental input. The first state matrix describes the transmission requirements of the service flow, including source address, destination address, flow size, and transmission latency requirements. The second state matrix records the queue buffer capacity of each node across domains. Based on these two state matrices, decisions are made to execute the process in two phases.

7. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the deterministic transmission scheduling method for differentiated service flows in a space-ground converged network as described in any one of claims 1-5.

8. A computer device, characterized in that, The system includes a memory and a processor, which communicate with each other. The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the deterministic transmission scheduling method for differentiated service flows in a space-ground converged network as described in any one of claims 1-5.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions that implement the deterministic transmission scheduling method for differentiated service flows in a space-ground converged network as described in any one of claims 1-5.

Images