Underwater acoustic network resource scheduling method for multi-task quality of service guarantee

Abstract

This invention provides a resource scheduling method for underwater acoustic networks with multi-task quality of service assurance, belonging to the field of underwater acoustic communication network technology. Addressing the problem that existing resource scheduling schemes cannot simultaneously consider node survivability and the latency and reliability requirements of critical services, this invention constructs a two-level scheduling architecture that coordinates node-level local scheduling and network-level central scheduling. In node-level scheduling, each node determines the optimal amount of data to be sent by querying an optimal decision table generated offline based on reinforcement learning, and selectively sends data packets based on dynamic priority calculation using multiple QoS parameters. In network-level scheduling, the central node aims to maximize network weighted throughput by jointly optimizing routing, time slot allocation, link power, and communication systems. This invention can intelligently and dynamically provide differentiated quality of service assurance for multiple concurrent heterogeneous services such as communication, navigation, and sensing, significantly improving network throughput and node survivability, and is suitable for resource-constrained underwater acoustic network environments.

Description

Technical Field

[0001] This invention belongs to the field of underwater acoustic communication network technology, and is mainly a resource scheduling method for underwater acoustic networks oriented towards multi-task quality of service assurance. Specifically, it relates to a resource scheduling method applicable to integrated underwater information networks that can provide differentiated quality of service (QoS) assurance for multiple concurrent heterogeneous services such as communication, navigation, and sensing. Background Technology

[0002] Integrated underwater information networks need to simultaneously support multiple services such as control commands, environmental awareness, and file transfer. These services differ significantly in terms of data priority, latency tolerance, and reliability requirements. Traditional underwater acoustic network resource scheduling methods (such as "first-in, first-out" or simple round-robin) cannot perceive or adapt to these differentiated QoS requirements, leading to potential blocking of high-priority services, low overall network resource utilization efficiency, and difficulty in meeting the quality of service assurance requirements in multi-task concurrent scenarios.

[0003] Existing resource scheduling schemes typically focus on optimizing a single metric or making decisions at a single level, lacking effective coordination between node-local intelligence and global network optimization. Nodes may prematurely exhaust their energy due to over-transmission, or fail to meet the latency and reliability requirements of critical services due to improper scheduling. Therefore, there is an urgent need for a scheduling method that can intelligently and dynamically allocate resources to tasks with different QoS requirements, thereby maximizing the overall service performance of the network. Summary of the Invention

[0004] The technical problem this invention aims to solve is to overcome the shortcomings of existing technologies and provide a resource scheduling method for underwater acoustic networks oriented towards multi-task quality of service assurance. This method utilizes an intelligent scheduling mechanism that coordinates local nodes and the network center at two levels to achieve differentiated resource allocation for tasks with different priorities and QoS requirements. This maximizes the overall service efficiency of the network in resource-constrained underwater acoustic network environments and ensures the overall service quality of multiple services.

[0005] The technical solution of this invention is to provide a method for scheduling underwater acoustic network resources for multi-task quality of service assurance, comprising the following steps: S1: Construct a two-level scheduling architecture that coordinates node-level local scheduling and network-level central scheduling; S2: Perform node-level local scheduling. This step includes the following two parallel sub-processes: S2.1: Determining the Optimal Data Volume to be Sent by Nodes Based on Reinforcement Learning. Each node models its scheduling decision as a Markov Decision Process (MDP), with its state space including the node's remaining energy level and the length of its transmit buffer queue, and its action space representing the number of data packets to be sent in the next scheduling cycle. By minimizing a cost function that comprehensively considers energy consumption, queuing delay, and node failure risk, the optimal strategy is solved offline using a reinforcement learning algorithm, generating a "energy level - queue length - optimal data volume" lookup decision table. When a node is online, it only needs to perceive its own state and look up the table to obtain the optimal data volume to be sent in the current cycle. That is, each network node determines the optimal data volume to be sent in the current scheduling cycle based on its remaining energy state and transmit buffer queue length by querying the optimal decision table generated offline based on reinforcement learning. S2.2: Each network node calculates a dynamic transmission priority for each data packet in its buffer, where the priority is the maximum allowable packet loss rate. Maximum tolerable end-to-end delay Queuing delay in the cache and inherent business priorities The function; and under the constraint of the optimal amount of data to be sent, select a set of data packets to maximize the total priority value of the selected data packets; S3: Execute network-level central scheduling. The central scheduling node collects the optimal amount of data to be sent and high-priority data stream information reported by each node. With the goal of maximizing the network weighted throughput, it jointly optimizes the routing of data streams, the allocation of conflict-free transmission time slots, and the transmission power and communication system of links, generates a global scheduling scheme, and distributes it. S4: Each network node sends data according to the global scheduling scheme.

[0006] Furthermore, the two-tier scheduling architecture includes a node-level local scheduler and a network-level central scheduler. The node-level local scheduler runs on each network node and is responsible for determining the optimal amount of data to be sent and prioritizing data packets based on its own state and the QoS attributes of cached data. The network-level central scheduler runs on a central node (such as a cluster head) and is responsible for aggregating information from the entire network, performing cross-layer joint optimization, calculating and distributing the globally optimal transmission scheduling scheme.

[0007] Furthermore, in step S2.1, the optimal decision table is obtained by: modeling the sending decision of a single node as a Markov decision process (MDP), where the state space includes the node's remaining energy level and the length of the sending buffer queue, and the action space is the number of data packets requested to be sent; by minimizing a comprehensive cost function. To find the optimal policy, where The energy consumption cost is proportional to the energy consumption of this transmission. The queuing latency cost is positively correlated with the length of the cache queue, and this cost increases sharply when the queue exceeds a safe threshold; The penalty for node failure is specifically defined as a high penalty for node failure triggered when an action causes the remaining energy of a node to drop below a preset safety threshold (such as zero). The weight coefficients for each cost term are used; finally, the optimal strategy obtained by solving is transformed into a lookup table indexed by the discretized energy level and the queue length.

[0008] Furthermore, step S2.2: Prioritizing packet transmission based on quality of service. Each node calculates a dynamic priority for each packet within its buffer. This priority is the maximum allowable packet loss rate. Maximum tolerable end-to-end delay Queued delay in the cache and inherent business priorities The function, i.e. Among them, the more stringent the packet loss rate requirements, the more urgent the latency constraints, the longer the queuing latency, or the higher the inherent priority of the service, the higher the calculated data packet. The larger the value, the better. After calculating the priority, the node needs to send the optimal amount of data determined in step S2.1. Under the constraints, solve an optimization problem to select a set of data packets such that the total priority value of the selected data packets is maximized.

[0009] Furthermore, in step S2.2, the packet transmission priority sorting is an adaptive improvement to the traditional Maximum Weight Delay First (M-LWDF) algorithm. A maximum tolerable delay is set. Only when the packet queuing delay This allows for a more flexible scheduling window for data packets, reducing unnecessary packet loss caused by temporary congestion.

[0010] Furthermore, in step S2.2, the step of selecting data packets under the constraint of the optimal amount of data to be sent is modeled as the following optimization problem: in, This represents the total number of data packets currently in the buffer. and They represent the first Priority value of each data packet Its data length, binary decision variables ( Indicates the first (Number of data packets selected for transmission in this round) This represents the maximum amount of data that can be sent in the current period.

[0011] Furthermore, step S3: Perform network-level central scheduling and cross-layer joint optimization. Each node reports the optimal amount of data to be sent and high-priority data stream information determined in step S2 to the central scheduling node. The joint optimization: The central scheduling node aims to maximize the network weighted throughput, constrained by node energy and network time slot resources (constraints include node remaining energy constraints and conflict-free time slot allocation constraints, wherein the conflict-free time slot allocation constraints include: at any given time, any node can participate in data transmission or reception on at most one link; high-priority data streams are given higher weights in the objective function), jointly optimizing data stream routing, conflict-free transmission time slot allocation, and adapting transmission power and communication system for each link. Finally, the calculated globally optimal scheduling scheme is distributed to each node for execution.

[0012] Furthermore, in step S4: each network node, according to the scheme issued by the central scheduling node, sends a data packet with a specified priority to the specified next-hop node in the specified time slot with a specified power, thereby completing this round of resource scheduling.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. Differentiated quality of service assurance is achieved: Through node-level dynamic priority calculation based on multiple QoS parameters, combined with network-level optimization with weighted throughput as the target, the urgency and importance of different services can be effectively distinguished, and data transmission of high-priority, low-latency requirements can be prioritized.

[0014] 2. Balancing node survival and network performance: The node-level reinforcement learning-based transmission volume control method enables nodes to intelligently balance the immediate benefits of "sending more data" with the long-term benefits of "conserving energy," preventing premature node death and ensuring the long-term stable operation of the network.

[0015] 3. Improved global network resource utilization efficiency: Through two-level collaboration of local node optimization and cross-layer joint optimization at the network center, the real-time nature of local information is utilized while taking into account the constraints and interference of global resources. This enables data to be guided to high-quality links for transmission, significantly improving the total throughput of the network under medium and high loads.

[0016] 4. High robustness and smooth degradation characteristics: When the network is congested, the method can ensure the performance of critical services through intelligent data discarding and scheduling, so that the overall system performance declines smoothly rather than deteriorates across the board. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the two-level collaborative scheduling architecture described in this invention.

[0018] Figure 2 In the reinforcement learning-based method for determining the optimal amount of data to be sent by a node, the optimal decision representation generated by simulation is used to demonstrate the optimal sending action (data amount) of the node under different energy and buffer states.

[0019] Figure 3 This is a schematic diagram showing the simulation results comparing the performance of the method of this invention with that of the traditional unoptimized strategy in terms of end-to-end transmission delay.

[0020] Figure 4 This is a schematic diagram showing the simulation results comparing the performance of the method of this invention with that of the traditional non-optimization strategy in terms of total network throughput. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0022] This invention discloses a resource scheduling method for underwater acoustic networks with multi-task quality of service assurance. This embodiment uses an underwater acoustic network comprising one central cluster head node and six member nodes as an example. The member nodes randomly generate three types of data tasks with different priorities, latency tolerances, and reliability. The process is as follows: Step S1: Architecture Initialization. Deploy the following in the network: Figure 1 The diagram illustrates a two-tier scheduling architecture that combines node-level local scheduling with network-level central scheduling. Each member node has a built-in node-level local scheduler, while the central cluster head node runs the network-level central scheduler. The node-level local scheduler runs on each network node and is responsible for determining the optimal amount of data to be sent and prioritizing data packets based on its own state and the QoS attributes of cached data. The network-level central scheduler runs on the central node (such as the cluster head) and is responsible for aggregating information across the entire network, performing cross-layer joint optimization, calculating and distributing the globally optimal transmission scheduling scheme.

[0023] Step S2: Node-level local scheduling. At the start of each scheduling cycle, each member node performs the following operations in parallel: 1. State Awareness and Table Lookup: Nodes detect their remaining energy (discretized into several levels) and the current length of the sending buffer queue. Based on these two state values, they query the optimal decision table, which was pre-generated through offline reinforcement learning training, to immediately obtain the optimal amount of data to be sent in this cycle. The optimal decision table is obtained by modeling the sending decision of a single node as a Markov Decision Process (MDP), where the state space includes the node's remaining energy level and the length of the sending buffer queue, and the action space is the number of data packets requested to be sent; by minimizing a comprehensive cost function. To find the optimal policy, where The energy consumption cost is proportional to the energy consumption of this transmission. The queuing latency cost is positively correlated with the length of the cache queue, and this cost increases sharply when the queue exceeds a safe threshold; The penalty for node failure is specifically defined as a high penalty for node failure triggered when an action causes the remaining energy of a node to drop below a preset safety threshold (such as zero). The weight coefficients for each cost term are used; finally, the optimal strategy obtained by solving is transformed into a lookup table indexed by the discretized energy level and the queue length.

[0024] This decision table demonstrates intelligence: when energy is high and the queue is long, the decision is to send more data (e.g., 10 packets); when energy is low, even with a long queue, the decision will reduce the number of packets sent (e.g., 1-2 packets) to prioritize node survival. Figure 2 The diagram illustrates the optimal actions (data transmission volume) of a node under different energy and buffer states. Figure 2 (a) is a schematic diagram of the optimal action of a node under different energy and buffer states. Figure 2 (b) is a schematic diagram of the local magnification result at a lower energy level.

[0025] 2. Packet Priority Calculation and Selection: The node traverses the packets in the buffer queue and selects the packet priority based on its priority. (Maximum allowable packet loss rate) (Maximum tolerable delay) (Queueing time has been delayed) Parameters such as (inherent business priority) are assigned according to a priority function. Calculate its dynamic priority. Then, within a total data length not exceeding... Under the constraints, solve the optimization model to select the set of high-priority data packets to be sent in this round.

[0026] Specifically: Packet delivery prioritization is based on Quality of Service (QoS). Each node calculates a dynamic priority for each packet within its buffer. This priority is the maximum allowable packet loss rate. Maximum tolerable end-to-end delay Queued delay in the cache and inherent business priorities The function, i.e. Among them, the more stringent the packet loss rate requirements, the more urgent the latency constraints, the longer the queuing latency, or the higher the inherent priority of the service, the higher the calculated data packet. The larger the value, the better. After calculating the priority, the node needs to send the optimal amount of data determined in step S2.1. Under the constraints, solve an optimization problem to select a set of data packets that maximizes the total priority value of the selected data packets. 3. Information reporting: The node will send the optimal amount of data to be sent. The selected high-priority data streams, along with their destination nodes and QoS requirements, are packaged and reported to the central cluster head node.

[0027] The packet transmission priority sorting described herein is an adaptive improvement upon the traditional Maximum Weight Delay First (M-LWDF) algorithm. A maximum expected tolerable delay is set. Only when the packet queuing delay This allows for a more flexible scheduling window for data packets, reducing unnecessary packet loss caused by temporary congestion.

[0028] The step of selecting data packets under the constraint of the optimal amount of data to be sent is modeled as the following optimization problem: in, This represents the total number of data packets currently in the buffer. and They represent the first Priority value of each data packet Its data length, binary decision variables ( Indicates the first (Number of data packets selected for transmission in this round) This represents the maximum amount of data that can be sent in the current period.

[0029] Step S3: Network-level Central Scheduling. The central cluster head node collects the reported information from all member nodes and, combined with network-wide Channel State Information (CSI), link load, and other information, runs a cross-layer joint optimization algorithm. This algorithm aims to maximize the network-wide weighted throughput (higher priority data flows have higher weights) while simultaneously satisfying node energy constraints and avoiding inter-link communication conflicts. It jointly solves for the globally optimal solution for this round: 1) the routing of each data flow; 2) the transmission time slot allocation scheme for each link; and 3) the transmission power and communication system for each link. Finally, this scheduling scheme is distributed to all relevant nodes.

[0030] Step S4: Scheduling scheme execution. After receiving the central scheduling instruction, each member node strictly sends the data packet selected in step S2 to the designated next-hop node within the specified time slot and at the specified power.

[0031] Performance verification: To verify the effectiveness of the method of the present invention, simulations were performed in the 7-node network and compared with a simple "no-optimization strategy" (the central node allocates time slots only according to the proportion of each node's cache utilization).

[0032] Delay performance: such as Figure 3 The end-to-end transmission delay simulation results shown are as follows: Figure 3 (a) is a schematic diagram of the method without optimization strategy. Figure 3 (b) is a schematic diagram of the optimization strategy method. Without an optimization strategy, there is no difference in end-to-end latency between tasks of different priorities. However, the method of this invention can provide differentiated latency guarantees for different services. Task 3, with the lowest latency tolerance, achieves the minimum transmission latency, demonstrating its QoS guarantee capability. Under high load, the method of this invention can ensure the latency stability of high-priority services through intelligent scheduling, achieving smooth performance degradation.

[0033] Throughput performance: such as Figure 4 The diagram illustrates a comparison of network throughput obtained by the two methods. Under light load, the throughput of both methods is comparable. As the load increases, the throughput without optimization quickly reaches a bottleneck of approximately 25 packets / min. In contrast, the throughput of the method described in this invention continues to increase and stabilizes at a higher level. This is because its cross-layer optimization directs data to higher-quality links, making fuller use of network resources.

[0034] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for scheduling underwater acoustic network resources with multi-task quality of service assurance, characterized in that: Includes the following steps S1: Construct a two-level scheduling architecture that coordinates node-level local scheduling and network-level central scheduling; S2: Execute node-level local scheduling, which includes the following parallel sub-steps: S2.1: Each network node determines the optimal amount of data to be sent in the current scheduling cycle by querying the optimal decision table generated offline based on reinforcement learning, according to its own remaining energy status and the length of the sending buffer queue. S2.2: Each network node calculates a dynamic transmission priority for each data packet in its buffer, where the priority is the maximum allowable packet loss rate. Maximum tolerable end-to-end delay Queuing delay in the cache and inherent business priorities The function; and under the constraint of the optimal amount of data to be sent, select a set of data packets to maximize the total priority value of the selected data packets; S3: Execute network-level central scheduling. The central scheduling node collects the optimal amount of data to be sent and high-priority data stream information reported by each node. With the goal of maximizing the network weighted throughput, it jointly optimizes the routing of data streams, the allocation of conflict-free transmission time slots, and the transmission power and communication system of links, generates a global scheduling scheme, and distributes it. S4: Each network node sends data according to the global scheduling scheme.

2. The underwater acoustic network resource scheduling method for multi-task quality of service assurance as described in claim 1, characterized in that: The node-level local scheduler runs on each network node and is responsible for determining the optimal amount of data to be sent and prioritizing data packets based on its own status and the QoS attributes of the cached data. The network-level central scheduler runs at the central node and is responsible for aggregating information from the entire network, performing cross-layer joint optimization, and calculating and issuing the globally optimal transmission scheduling scheme.

3. The underwater acoustic network resource scheduling method for multi-task quality of service assurance according to claim 1, characterized in that: In step S2.1, the optimal decision table is obtained as follows: the sending decision of a single node is modeled as a Markov decision process, where the state space includes the node's remaining energy level and the length of the sending buffer queue, and the action space is the number of data packets requested to be sent; by minimizing a comprehensive cost function To find the optimal policy, where As a result of energy consumption, To pay the price of queuing delay, This is the penalty for node failure, triggered when an action causes the node's remaining energy to fall below a preset safety threshold. The weighting coefficients are used; finally, the optimal strategy obtained by solving is transformed into a lookup table indexed by the discretized energy level and the queue length.

4. The underwater acoustic network resource scheduling method for multi-task quality of service assurance according to claim 1, characterized in that: In step S2.2, the dynamic transmission priority of data packets is calculated. The function is Among them, the more stringent the packet loss rate requirements, the more urgent the latency constraints, the longer the queuing latency, or the higher the inherent priority of the service, the higher the calculated data packet. The larger the value.

5. The underwater acoustic network resource scheduling method for multi-task quality of service assurance according to claim 4, characterized in that: In step S2.2, an improved maximum weight delay priority rule is adopted for packet dropping management: a maximum expected tolerable delay is set. Only when the packet queuing delay It is only discarded from the cache at that time.

6. The underwater acoustic network resource scheduling method for multi-task quality of service assurance according to claim 1, characterized in that: In step S2.2, the step of selecting data packets under the constraint of the optimal amount of data to be sent is modeled as the following optimization problem: in, The total number of data packets in the buffer. and The first The priority value and data length of each data packet. For binary decision variables, This represents the optimal amount of data to be sent in the current period.

7. The underwater acoustic network resource scheduling method for multi-task quality of service assurance according to claim 1, characterized in that: In step S3, the joint optimization aims to maximize the network weighted throughput, and its constraints include node remaining energy constraints and conflict-free time slot allocation constraints; wherein the conflict-free time slot allocation constraints include: at any given time, any node can participate in data transmission or reception on at most one link; and high-priority data streams are given higher weights in the objective function.

Images