CN-121985036-A - Cross-domain marine network task unloading and computing resource allocation method based on TD3

CN121985036ACN 121985036 ACN121985036 ACN 121985036ACN-121985036-A

Abstract

The invention discloses a TD 3-based cross-domain marine network task unloading and computing resource allocation method. The method is suitable for a cross-domain ocean network consisting of an autonomous underwater vehicle, a water surface buoy node and a low-orbit satellite node. Aiming at three processing modes of local execution, cooperative execution of the water surface buoy nodes and star-end execution, the comprehensive processing cost comprising underwater light transmission, transcoding processing of the water surface buoy nodes and radio frequency transmission is constructed. Further provides a service perception executable mapping mechanism, performs service availability screening on an unloading preference vector output by the strategy network, combines with a TD3 algorithm, generates a task unloading decision in an executable mode set through maximum component selection and single-hot coding, and outputs a computing resource allocation result. The method can effectively realize the joint optimization of task unloading decision and computing resource allocation under the cross-domain ocean heterogeneous channel, thereby reducing the network task unloading delay and the energy consumption.

Inventors

LI WENFENG
LIU SHUAI
CHEN HONGYAN
ZHAO KANGLIAN

Assignees

南京大学

Dates

Publication Date: 20260505
Application Date: 20260409

Claims (4)

1. The TD 3-based cross-domain marine network task unloading and computing resource allocation method is characterized by comprising the following steps of: the method comprises the steps of S1, constructing a cross-domain ocean network consisting of an autonomous underwater vehicle, a water surface buoy node and a low-orbit satellite node, and building a calculation task model and a service availability state model, wherein the water surface buoy node is deployed on the water surface and is used for connecting heterogeneous links between the autonomous underwater vehicle and the low-orbit satellite node, and the heterogeneous links comprise an underwater optical communication link from the autonomous underwater vehicle to the water surface buoy node and a radio frequency link from the water surface buoy node to the low-orbit satellite node; s2, acquiring task information, service availability status, available computing resource status of each node and heterogeneous link transmission status generated by an autonomous underwater vehicle in each time slot, and constructing a local execution mode, a water surface buoy node cooperative execution mode and a star end execution mode, wherein the star end execution mode at least comprises underwater optical communication transmission from the autonomous underwater vehicle to the water surface buoy node, water surface buoy node transcoding processing and radio frequency transmission from the water surface buoy node to a low-orbit satellite node; S3, constructing a service perception executable mapping mechanism, screening service availability of the unloading preference vector according to the service availability state, and performing maximum component selection and independent heat coding in an executable mode set formed by a local execution mode, a water surface buoy node cooperative execution mode and a star end execution mode to generate a task unloading decision vector, wherein the star end execution mode is kept optional; And S4, calculating comprehensive processing cost according to the task unloading decision vector and the calculation resource allocation proportion, outputting unloading preference vector and the calculation resource allocation proportion by a current action network of the TD3 algorithm with the aim of minimizing long-term average comprehensive processing cost, and carrying out iterative solution on the task unloading and calculation resource allocation process to output task unloading decision and calculation resource allocation result.
2. The method for task offloading and computing resource allocation of a TD 3-based cross-domain marine network according to claim 1, wherein: Set autonomous underwater vehicle as The water surface buoy nodes are assembled into The low orbit satellite nodes are gathered into The service type set is The time slot set is Wherein, the method comprises the steps of, Representing an autonomous underwater vehicle index; representing the index of the water surface buoy node; representing a low-orbit satellite node index; representing a service type index; representing a slot index; representing the total number of time slots; at any time slot Inner autonomous underwater vehicle Generating computing tasks, definitions Respectively represent service types Autonomous underwater vehicle Water surface buoy node The system comprises a service availability status, a value 1, a low-orbit satellite node, an autonomous underwater vehicle, a water surface buoy node, a service availability status, a value 0, a service type deployment part and a service type management part, wherein the value 1 represents availability and the value 0 represents unavailability; definition autonomous underwater vehicle In time slot Is: Wherein, the 、 And The mode preference values of the current action network output the local execution mode, the water surface buoy node cooperative execution mode and the star-end execution mode are respectively represented; definition autonomous underwater vehicle In time slot The task offloading decision vector of (1) is: Wherein, the 、 And Binary decision variables respectively representing a local execution mode, a water surface buoy node cooperative execution mode and a star-end execution mode and meeting the requirements ; Definition of the definition And Respectively time slots Assigned to autonomous underwater vehicle The buoy side computing resource allocation proportion and the satellite side computing resource allocation proportion, and satisfy the following: 。
3. the method for task offloading and computing resource allocation of a TD 3-based cross-domain marine network according to claim 2, wherein: service availability screening is carried out on the unloading preference vector, and the screened executable preference vector is obtained: Wherein, the Wherein, the 、 And Respectively representing the mode preference value of the current action network after screening the local execution mode, the water surface buoy node cooperative execution mode and the star-end execution mode When the local execution mode is masked, when The star-end execution mode is always kept selectable as a backup execution mode because the low-orbit satellite nodes deploy all service types; On the basis, a task offloading decision vector is generated through maximum component selection and single-hot encoding: Wherein, the In order to perform the mode indexing, Representing a local execution mode, Represents a water surface buoy node cooperative execution mode, Representing a star-end execution mode; Representing a one-hot encoding function that maps the selected execution mode index into a three-dimensional binary vector; The comprehensive processing cost of the local execution mode is formed by the joint of task time delay and task energy consumption generated by local calculation processing of an autonomous underwater vehicle, the comprehensive processing cost of the water surface buoy node cooperative execution mode is formed by the joint of underwater optical communication transmission cost from the autonomous underwater vehicle to the water surface buoy node and the water surface buoy node calculation processing cost, and the comprehensive processing cost of the star-end execution mode is formed by the joint of underwater optical communication transmission cost from the autonomous underwater vehicle to the water surface buoy node, the water surface buoy node transcoding processing cost, the radio frequency transmission cost from the water surface buoy node to the low-orbit satellite node and the low-orbit satellite node calculation processing cost; And constructing comprehensive processing cost represented by weighted sum of task time delay and task energy consumption according to the task time delay and task energy consumption corresponding to the three processing modes, and optimizing the target by minimizing long-term average comprehensive processing cost.
4. A method for task offloading and computing resource allocation across a TD 3-based marine network according to claim 3, wherein: solving the task unloading and computing resource allocation process by adopting TD3 algorithm, and obtaining the time slot System input There are seven pieces of information, respectively, being task information Autonomous underwater vehicle Buoy node to water surface Is the uplink rate of underwater optical communication Water surface buoy node To low orbit satellite node Radio frequency uplink rate of (2) Autonomous underwater vehicle Available calculation rate of (a) Water surface buoy node Available calculation rate of (a) Low orbit satellite node Available calculation rate of (a) Service availability status set ; System output There are and only three items of information, respectively, of offload preference vectors Calculating resource allocation proportion on buoy side And satellite side calculation of resource allocation ratio ; Instant rewarding function The TD3 algorithm comprises a current action network, a target action network, two current evaluation networks and corresponding target evaluation networks, wherein the current action network outputs an unloading preference vector and a calculated resource allocation proportion after seven pieces of information are input into the system, and the unloading preference vector is processed by a service perception executable mapping mechanism to generate a task unloading decision consistent with service availability status; In the network updating stage, a target action network generates target actions, time sequence difference error updating is carried out after loss calculation, a current evaluation network is updated according to smaller evaluation results in two target evaluation networks, a strategy gradient updating strategy is adopted to update the current action network, a soft updating mode is adopted to update parameters of the target action network and the target evaluation network, and after network training is converged, task unloading decisions and corresponding calculation resource allocation results of the autonomous underwater vehicle under each time slot are output.

Description

Cross-domain marine network task unloading and computing resource allocation method based on TD3 Technical Field The invention belongs to the technical field of ocean edge calculation and task collaborative optimization, and particularly relates to a cross-domain ocean network task unloading and calculation resource allocation method based on a dual-delay depth deterministic strategy Gradient (TWIN DELAYED DEEP DETERMINISTIC Policy Gradient, TD 3). Background With the continuous development of intelligent ocean and ocean Internet of things (Ocean of Things, ooT) applications, the real-time performance, reliability and energy efficiency of task processing are required by the services of ocean environment monitoring, underwater target identification, resource exploration, offshore emergency guarantee and the like. Autonomous underwater vehicles (Autonomous Underwater Vehicle, AUV) typically require computationally intensive and time-delay sensitive processing of the perceived data during performance of the actual mission, but are limited by local computing power, energy supply, and underwater communication conditions, relying on local execution alone often makes complex business requirements difficult. In order to improve the processing capacity of marine tasks, a cross-domain marine network consisting of an autonomous underwater vehicle, a water surface buoy node and a low-orbit satellite node is attracting attention. The water surface buoy node can be connected with an underwater optical communication link from the autonomous underwater vehicle to the water surface buoy node and a radio frequency link from the water surface buoy node to the low-orbit satellite node, and can be used as an edge node to participate in task processing; the low-orbit satellite node has stronger computing resources and wider coverage range, and can form a cross-domain cooperative task processing system with the autonomous underwater vehicle and the water surface buoy node. By performing task offloading and computing resource allocation within the system, performance bottlenecks caused by local resource limitations of autonomous underwater vehicles can be alleviated to some extent. In the prior art, various task offloading and resource allocation methods have been proposed for space-to-ground networks or space Taiwan Strait Exchange Association on-network. For example, patent literature (application day: 2025, 7, 1, 202510901684.0, application publication day: CN120751445 a, application publication day: 2025, 10, 3, the content of which can be cited) provides an implementation idea of constructing tasks, communication and calculation models and making joint decisions in an air-to-air edge calculation network, but the scheme is mainly oriented to an air-to-air scenario, and does not construct a unified processing mechanism adaptive to a local execution mode, a buoy collaborative execution mode and a star-end execution mode according to the characteristics that an underwater optical link and an air radio frequency link coexist in a cross-domain marine network and the link state becomes obvious. For another example, in the patent literature, "offshore edge computing task offloading and resource allocation method based on air-sea collaboration" (application date: 2025, 7, 31, 202511069018.1, application publication date: CN120812662 a, application publication date: 2025, 10, 17, and the content of this application may still be cited), research is conducted on an air-sea collaborative offshore edge computing system, but the solution is more focused on performance modeling such as throughput and queuing under the same architecture of air Taiwan Strait Exchange Association, and on an online optimization mechanism for mode screening under service availability constraint, collaborative generation of discrete task offloading decisions and continuous computing resource allocation in a cross-domain marine network, and long-term average comprehensive processing cost guidance, a targeted solution is still lacking. Therefore, in the prior art, under a cross-domain marine network scene, the problems of constructing a unified comprehensive processing cost representation aiming at a local execution mode, a water surface buoy node cooperative execution mode and a star-end execution mode under a heterogeneous time-varying link condition, ensuring the executable performance of a task unloading decision under the constraint of service availability, and realizing the online cooperative optimization of a discrete task unloading decision and continuous computing resource allocation are difficult to solve simultaneously. The existence of the problems makes it difficult for the existing method to achieve the effects of decision-making executable performance, optimizing stability and effectively reducing task processing time delay and energy consumption. Disclosure of Invention Aiming at the problems of insufficient comprehensive processing cost characteriza