Joint service caching and task unloading method for air-ground integrated edge computing in ocean Internet of Things

Abstract

The invention discloses a joint service caching and task unloading method for air-ground integrated edge computing in the ocean Internet of Things. The method comprises the following steps: S1, constructing an air-ground-sea integrated network system model; s2, constructing a communication model; s3, formulating a service caching and task unloading strategy; s4, constructing a calculation model for calculating the total delay and the total energy consumption required for completing the task; formulating an optimization model for jointly optimizing a user association decision, a service cache decision and a task unloading proportion problem; s6, designing an iterative algorithm based on block coordinate descent; and S7, solving through multiple iterations to obtain an optimal user association decision result, a service cache decision result and a task unloading proportion, so that the total delay of the air-ground-sea integrated network system model is minimum. According to the invention, the communication efficiency is improved, the UAV position can be dynamically adjusted in the three-dimensional space based on the MIOTD distribution, and the communication quality is ensured.

Description

Technical Field

[0001] The present invention relates to the field of mobile computing technology, and in particular, to a joint service caching and task offloading method for air-ground integrated edge computing in the marine Internet of Things. Background Art

[0002] With the rapid development of the marine economy and marine research, the marine Internet of Things has become a promising paradigm for realizing marine intelligent monitoring and marine resource management. By deploying various marine Internet of Things devices (MIoTDs), marine parameters (such as temperature, salinity, flow velocity) and ship traffic information are collected in real time, which plays an important role in applications such as marine environmental protection, maritime safety, and fishery management. However, due to the limited size, power supply, and hardware capacity of these MIoTDs, they often face severe computational resource limitations. The growing volume of collected data and increasingly complex processing requirements pose great challenges to these resource-constrained devices, which may lead to a large amount of processing delay and system performance degradation. Mobile edge computing (MEC) provides computing resources close to terminal devices and has become an effective way to alleviate these problems.

[0003] However, the direct application of traditional MEC systems in the marine environment faces unique obstacles and requires innovative solutions. First, different from the ground scenario, the direct communication link between MIoTDs and shore-based edge servers is often blocked by port facilities, anchored ships, and complex marine structures, resulting in unstable connections and significant transmission delays. Second, the wide distribution of MIoTDs in the vast ocean area makes it impractical to deploy a fixed edge computing infrastructure with comprehensive coverage.

[0004] Unmanned aerial vehicles (UAVs) have been recognized as a promising MEC platform due to their flexibility and line-of-sight communication capabilities. By deploying UAVs as edge servers, the system can dynamically adjust computing resources according to the spatio-temporal distribution of task requirements while maintaining a reliable air-sea communication link with MIoTDs. In addition, the superior mobility of UAVs enables them to act as relay nodes between shore base stations and MIoTDs, effectively extending the coverage of edge computing services to remote offshore areas.

[0005] Although UAV-assisted MEC has its advantages, current research faces several key limitations in marine scenarios. Most studies focus on single-UAV systems, and it is difficult to meet the requirements of large-scale MIoT deployments due to limited computing power and coverage. Although multiple UAVs have the potential to address these limitations, in a marine environment, the long distance between UAVs and shore base stations can result in a large amount of backhaul transmission overhead when requesting services. This challenge has stimulated the need for UAVs to adopt an efficient service caching strategy. In addition, existing work usually restricts UAVs to a fixed altitude, ignoring the impact of three-dimensional (3D) deployment on system performance. This simplification is particularly problematic in a marine environment because the sea-air channel has unique characteristics such as sea surface reflection and atmospheric ducts, which can significantly affect communication quality. Summary of the Invention

[0006] The present invention provides a joint service caching and task offloading method for air-ground integrated edge computing in a marine Internet of Things to overcome the technical problems in the prior art that service caching and task offloading cannot be efficiently performed under limited computing resources and long-distance communication conditions of shore base stations, and at the same time, ignoring the three-dimensional deployment of UAVs on the sea surface, which affects communication quality.

[0007] To achieve the above object, the technical solution of the present invention is as follows:

[0008] A joint service caching and task offloading method for air-ground integrated edge computing in a marine Internet of Things, the specific steps include:

[0009] S1: Construct an air-ground-sea integrated network system model capable of deploying UAVs in three-dimensional space;

[0010] S2: Construct a communication model based on the air-ground-sea integrated network system model;

[0011] S3: Develop a service caching and task offloading strategy based on the air-ground-sea integrated network system;

[0012] S4: Construct a computing model according to the communication model and the service caching and task offloading strategy, and the computing model is used to calculate the total delay and total energy consumption required to complete the task;

[0013] S5: Based on the computing model, develop an optimization model for jointly optimizing user association decision-making, service caching decision-making, and task offloading ratio problems with the objective of minimizing the total system delay under the constraints of UAV cache capacity, available energy of marine Internet of Things devices and UAVs;

[0014] S6: Design an iterative algorithm based on block coordinate descent to decompose the optimization model into several sub-models, and select the coalition game, linear relaxation, Lagrangian duality method, and convex optimization algorithm for solution according to the convexity and structure of each sub-model;

[0015] S7: Obtain the optimal user association decision result, service caching decision result, and task offloading ratio through multiple iterative solutions, minimizing the total delay of the air-sea-land integrated network system model.

[0016] Furthermore, in S1, the constructed air-sea-land integrated network system model includes:

[0017] Set K MIoTDs randomly distributed on the sea surface, use M UAVs equipped with edge servers as airborne computing nodes with caching capabilities, and use the shore base station as a data center to maintain a complete service program library; the sets of UAVs and MIoTDs are respectively denoted as and

[0018] Assume that MioTDk has a computing task W k , W k =(L k , C k ), where L k represents the amount of computing task data; C k represents the number of CPU cycles required to compute each bit; for each computing task W k , the air-sea-land integrated network system model has pre-determined the services it requires, and the set of all services is denoted as where the storage space size occupied by each service is l j ,

[0019] Establish a three-dimensional Cartesian coordinate system with the sea level as the reference plane, where the xoy plane coincides with the sea level; define the position of MIoTDk as q k =(x k , y k , 0) T , and the initial deployment position of UAVm is where, represents the initial height of UAVm; for each computing task, there are M + 1 computing nodes, forming a set When m = 0, it means that the computing task is processed locally by the MIoTD. When 0 < m ≤ M, it means that the computing task is offloaded to UAVm for execution;

[0020] Through the binary variable a k,mIndicates user association decision. If MIoTDk selects node m to process the task, then a k,m = 1; otherwise, a k,m = 0. To ensure the unique allocation of computing tasks, there are

[0021] Define as the set of MIoTDs served by UAVm, Set the final position q of UAVm m to be located at the geometric center of the MIoTDs it serves, q m = (x m , y m , h m ); Set the calculation formulas for the horizontal and vertical coordinates of UAVm on the horizontal plane to be:

[0022]

[0023] Assume that all UAVs have the same coverage angle θ, and define the service coverage radius R of UAVm m as the projected radius of its effective service range on the horizontal plane, and the formula is:

[0024] R m = H m tan(θ) (3)

[0025] To ensure that all associated MIoTDs are within the effective coverage range of the UAV, the flight altitude of UAVm needs to satisfy:

[0026]

[0027] where μ is the safety margin to avoid MIoTDs being located at the coverage boundary.

[0028] Furthermore, in S2, the communication model constructed based on the air-sea-land integrated network system model includes:

[0029] Calculate the channel gain between MIoTDk and UAVm based on the two-ray channel propagation model, and the formula is:

[0030]

[0031] where λ is the carrier wavelength, H1 is the height of the MIoTD antenna from the sea level, and d k,m is the Euclidean distance between MIoTDk and UAVm, expressed as

[0032]

[0033] Based on the Shannon channel capacity theory, the data rate between MIoTDk and UAVm is as follows:

[0034]

[0035] where B is the bandwidth allocated to MIoTDk, P k,m is the transmit power, and σ 2 is the power of additive white Gaussian noise at the receiving end.

[0036] Furthermore, in S3, the service caching and task offloading strategy formulated based on the air-sea-land integrated network system includes:

[0037] Define a binary variable b j,m , where the b j,m represents the caching status of service j on UAVm, b j,m ∈{0, 1}. When service j is cached on UAVm, b j,m = 1; otherwise, b j,m = 0. Then the caching capacity constraint of the UAV is expressed as:

[0038]

[0039] where S m is the maximum storage capacity of UAVm;

[0040] For each computing task, MIoTD divides it into two parts based on the partial offloading mode. One part is computed locally, and the other part is offloaded through the wireless link to the UAV for computing;

[0041] It is assumed that each UAV is equipped with a storage unit with a limited capacity and some services are pre-loaded in the local storage; after receiving the offloaded computing task, the UAV first checks whether the services required for processing the task are included in the local storage. If the corresponding services exist, it directly performs the computing; otherwise, the UAV will request the required services from the shore base station through the wireless backhaul link and then perform the computing after downloading the required services.

[0042] Furthermore, in S4, the computing model constructed according to the communication model and the service caching and task offloading strategy includes:

[0043] 1) Local computing: When MIoTDk completely relies on local computing to process tasks, the air-sea-land integrated network system model does not involve any task offloading. Define the local computing ability of MIoTD as F D , then the total delay and total energy consumption for completing the computing task W k are respectively expressed as:

[0044]

[0045] Among them, γ D represents the effective capacitance coefficient of MIoTD;

[0046] 2) Offloading calculation: When MIoTDk selects to establish an association with UAVm, a partial offloading strategy is adopted. Define the continuous variable β k,m to represent the task offloading ratio, β k,m ∈[0,1]. Then the local calculation delay and energy consumption of MIoTDk are respectively expressed as:

[0047]

[0048] The transmission delay and energy consumption of MIoTDk offloading tasks to UAVm are respectively expressed as:

[0049]

[0050] Suppose the computing power of UAVm is F U . Then the computing delay and energy consumption required for UAVm to process the offloading tasks of MIoTDk are respectively expressed as:

[0051]

[0052] Among them, γ U represents the effective capacitance coefficient of UAV;

[0053] When the required service program is not cached on UAVm, the download delay for UAVm to download this service from the shore base station is expressed as:

[0054]

[0055] Among them, R b is the download rate of the wireless backhaul link;

[0056] The total delay of the air-sea-land integrated network system model based on computing offloading to complete the computing task W k is expressed as:

[0057]

[0058] The total delay of the air-sea-land integrated network system model is expressed as:

[0059]

[0060] Furthermore, in S5, the optimization model for jointly optimizing user association decision, service caching decision, and task offloading ratio problem is expressed as:

[0061]

[0062] Among them, C1 and C2 represent that the user association and service caching decisions are binary variables; C3 indicates that the value range of the task offloading ratio is [0, 1]; C4 shows that each MIoTD must select a computing destination, either using pure local computing or establishing an association with a certain UAV and adopting a partial offloading mode; C5 means that the space occupied by all services cached on each UAV cannot exceed its maximum storage capacity; C6 and C7 are used to ensure that the energy consumed by each MIoTD and UAV does not exceed its energy limit, and are the maximum available energies of the MIoTD and UAV, respectively.

[0063] Furthermore, in S6, through an efficient iterative algorithm based on block coordinate descent, the optimization problem is decomposed into several sub-models and solved, including:

[0064] For a given service caching decision B and task offloading ratio β, the optimization model regarding the user association decision A is formulated as:

[0065]

[0066] The MIoTD and UAV are jointly modeled as a coalition game problem with K + M participants, and by constructing a utility function and game rules, the participants form a stable coalition structure to solve

[0067] Furthermore, in S6, through an efficient iterative algorithm based on block coordinate descent, the optimization problem is decomposed into several sub-models and solved, also including:

[0068] For a given task offloading ratio β and the optimized user association decision A, the optimization model regarding the service caching decision B is formulated as:

[0069]

[0070] For perform linear relaxation, and relax the binary variable condition in C2 to obtain a linear programming model expressed as:

[0071]

[0072] Use the Lagrangian dual method to solve the model

[0073] Furthermore, in S6, through an efficient iterative algorithm based on block coordinate descent, the optimization problem is decomposed into several sub-models and solved, also including:

[0074] For the optimized user association decision A and service caching decision B, the optimization model regarding the task offloading ratio β is formulated as follows:

[0075]

[0076] The convex optimization toolbox is used to solve the model.

[0077] Beneficial effects: The present invention constructs an air-sea-land integrated network system model, constructs an ocean communication model based on the air-sea-land integrated network system model, and proposes a service caching and task offloading strategy. Based on the calculation model, an optimization model for jointly optimizing the user association decision, service caching decision, and task offloading ratio is formulated under the constraints of the UAV caching capacity, the available energy of the marine Internet of Things devices and UAVs, with the goal of minimizing the total system delay. Thus, it jointly optimizes the computational offloading, service caching, and three-dimensional UAV deployment, and correspondingly designs a low-complexity iterative algorithm to solve this optimization model, obtaining the optimal user association decision result, service caching decision result, and task offloading ratio, minimizing the total delay of the air-sea-land integrated network system model. The present invention improves the communication efficiency, and at the same time, based on the MIoTD distribution in three-dimensional space, it can dynamically adjust the UAV position to ensure the communication quality. Description of the Drawings

[0078] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0079] Figure 1 It is a flowchart of the joint service caching and task offloading method for air-land integrated edge computing in the marine Internet of Things of the present invention;

[0080] Figure 2 It is a schematic diagram of the air-sea-land integrated network system model in the embodiments of the present invention;

[0081] Figure 3 It is a schematic diagram of the UAV position adjustment process in the embodiments of the present invention;

[0082] Figure 4 It is a flowchart of the edge computing iterative optimization algorithm for the air-sea-land integrated network system model in the embodiments of the present invention;

[0083] Figure 5 It is a two-dimensional visualization diagram of the UAV deployment scheme in the embodiments of the present invention;

[0084] Figure 6 It is a 3D visualization diagram of the UAV deployment scheme in the embodiments of the present invention;

[0085] Figure 7 It is a diagram of the algorithm convergence in different scenarios in the embodiments of the present invention;

[0086] Figure 8 It is a performance comparison diagram of the algorithm in the embodiments of the present invention and the benchmark scheme with the change of local computing power;

[0087] Figure 9 It is a performance comparison diagram of the algorithm and the benchmark scheme in the embodiments of the present invention with the change of the number of MIoTDs. Detailed implementation manners

[0088] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0089] This embodiment provides a joint service caching and task offloading method for air-ground integrated edge computing in the marine Internet of Things, as Figure 1 shown, and the specific steps include:

[0090] S1: Construct an air-ground-sea integrated network system model capable of deploying UAVs in three-dimensional space;

[0091] S2: Construct a communication model based on the air-ground-sea integrated network system model;

[0092] S3: Develop a service caching and task offloading strategy based on the air-ground-sea integrated network system;

[0093] S4: Construct a computing model according to the communication model and the service caching and task offloading strategy, and the computing model is used to calculate the total delay and total energy consumption required to complete the task;

[0094] S5: Based on the computing model, develop an optimization model for jointly optimizing user association decisions, service caching decisions, and task offloading ratios with the objective of minimizing the total system delay under the constraints of the UAV cache capacity, the available energy of marine Internet of Things devices and UAVs;

[0095] S6: As Figure 4As shown in the figure, an efficient iterative algorithm based on block coordinate descent is designed to decompose the optimization model into several sub-models, and according to the convexity, concavity and structure of each sub-model, the coalition game, linear relaxation, Lagrangian dual method and convex optimization algorithm are correspondingly selected for solution;

[0096] S7: Through multiple iterative solutions, the optimal user association decision result, service caching decision result and task offloading ratio are obtained, so that the total delay of the air-sea-land integrated network system model is minimized.

[0097] This embodiment proposes a Figure 2 shown air-sea-land integrated network system model for multi-UAV collaborative edge computing and caching. This model integrates multiple UAVs with an onshore data center to provide flexible computing and caching services for MIoTD, and at the same time adjusts the positions of UAVs in 3D space to enhance service coverage. First, a three-layer system model is established, in which multiple UAVs form an aerial computing cluster and coordinate with the onshore base station. To minimize the total delay of the entire system, a joint optimization model for user association, service caching and task offloading decisions is proposed, while adapting the UAV deployment to the MIoTD distribution. Combining coalition game theory, linear relaxation and Lagrangian dual decomposition, an effective iterative algorithm is developed to obtain near-optimal performance while ensuring convergence.

[0098] In a specific embodiment, in S1, constructing the air-sea-land integrated network system model includes:

[0099] The constructed air-sea-land integrated network system model is as Figure 2 shown. This model aims to provide flexible edge services for resource-constrained MIoTD. In this model, K MIoTDs are randomly distributed on the sea surface. M UAVs equipped with edge servers are used as aerial computing nodes with caching capabilities, and the onshore base station is used as a data center to maintain a complete service program library. The sets of UAVs and MIoTDs are respectively represented by and respectively.

[0100] Due to the occlusion of obstacles such as port facilities and moored ships, the direct communication link between MIoTD and the onshore base station is often blocked. Assume that MIoTDk has a computing task W k , W k =(L k , C k ), where L k represents the amount of computing task data, in bits; C kDenotes the number of CPU cycles required for each bit, with the unit of cycles / bit. For each computing task W k , the air-sea-land integrated network system model has pre-determined the specific service programs it requires, hereinafter referred to as services; the set of all services is denoted as where the storage space size occupied by each service is l j ,

[0101] A three-dimensional Cartesian coordinate system is established with the sea level as the reference plane, where the xoy plane coincides with the sea level; the position of MIoTDk is defined as q k =(x k ,y k ,0) T , as Figure 3 shown, the initial deployment position of UAVm is wherein, represents the initial height of UAVm; for each computing task, there are M + 1 computing nodes, forming a set When m = 0, it means that the computing task is processed locally at MIoTD, and when 0 < m ≤ M, it means that the computing task is offloaded to UAVm for execution; through the binary variable a k,m represents the user association decision. If MIoTDk selects node m to process the task, then a k,m = 1; otherwise, a k,m = 0. To ensure the unique allocation of computing tasks, there is

[0102] Define as the set of MIoTDs served by UAVm, To optimize the communication performance and ensure service fairness, the final position q m of UAVm is located at the geometric center of the MIoTDs it serves, q m =(x m ,y m ,H m ); The process of UAV position adjustment is as Figure 2 shown. Specifically, the calculation formulas for the horizontal and vertical coordinates of UAVm on the horizontal plane are respectively:

[0103]

[0104] Assume that all UAVs have the same coverage angle θ, and define the service coverage radius R m of UAVm as the projection radius of its effective service range on the horizontal plane, and the formula is:

[0105] R m = Hm tan(θ) (3)

[0106] To ensure that all associated MIoTDs are within the effective coverage range of the UAV, the flight altitude of UAVm needs to satisfy:

[0107]

[0108] where μ is the safety margin to avoid MIoTDs being located at the coverage boundary.

[0109] In a specific embodiment, in S2, the communication model constructed based on the air-sea-land integrated network system model includes:

[0110] To achieve multi-device concurrent communication and improve spectral efficiency, MIoTDs use orthogonal frequency division multiple access (OFDMA) to offload task data to the UAV. In the marine environment, due to the reflection and scattering of electromagnetic waves caused by sea surface fluctuations, the channel exhibits significant multipath characteristics. Considering that the sea surface reflection path dominates,

[0111] This embodiment uses a two-path channel propagation model to characterize the unique characteristics of air-to-sea communication. This two-path channel propagation model includes a direct path and a sea surface reflection path, and can effectively characterize the wireless propagation characteristics in open sea areas. Based on the two-path channel propagation model, the channel gain between MIoTDk and UAVm is calculated by the formula:

[0112]

[0113] where λ is the carrier wavelength, H1 is the height of the MIoTD antenna from the sea level, and d k,m is the Euclidean distance between MIoTDk and UAVm, expressed as

[0114]

[0115] Based on the Shannon channel capacity theory, the data rate between MIoTDk and UAVm is:

[0116]

[0117] where B is the bandwidth allocated to MIoTDk, P k,m is the transmit power, and σ 2 is the additive white Gaussian noise power at the receiving end.

[0118] In a specific embodiment, in S3, the service caching and task offloading strategy formulated based on the air-sea-land integrated network system includes:

[0119] Due to the limited storage space of the UAV, it is impossible to deploy all possible services simultaneously. Therefore, it is necessary to reasonably plan the caching strategy. In this embodiment, a binary variable b is defined. j,m , where the b j,m represents the caching status of service j on UAV m. b j,m ∈ {0, 1}. When service j is cached on UAV m, b j,m = 1; otherwise, b j,m = 0. Then, the caching capacity constraint of the UAV is expressed as:

[0120]

[0121] where S m is the maximum storage capacity of UAV m;

[0122] For each computing task, MIoTD divides it into two parts based on the partial offloading mode. One part is computed locally, and the other part is offloaded via a wireless link to the UAV for execution. It is assumed that each UAV is equipped with a storage unit of limited capacity and some services are pre-loaded in the local storage; after receiving the offloaded computing task, the UAV first checks whether the local storage contains the services required for processing the task. If the corresponding services exist, the computation is directly executed; otherwise, the UAV will request the required services from the shore base station via a wireless backhaul link and then execute the computation after downloading the required services, which will also result in additional communication overhead and processing delay.

[0123] Specifically, this embodiment proposes a novel three-tier MEC architecture, where multiple UAVs act as computing servers with caching capabilities, and the shore base station maintains a complete service library. Tasks from MIoTD are either processed locally or offloaded to the UAVs. For offloaded tasks, if the required services are cached, the UAV directly executes the computation; otherwise, the services must be retrieved from the shore base center before execution.

[0124] In a specific embodiment, in S4, the computing model constructed according to the communication model and the service caching and task offloading strategy includes:

[0125] 1) Local computing: When MIoTDk completely relies on local computing to process tasks, that is, a k,0 = 1, the air-sea-land integrated network system model does not involve any task offloading. Define the local computing capacity of MIoTD as F D , then the total delay and total energy consumption for completing the computing task W k are respectively expressed as:

[0126]

[0127] where γ DDenote the effective capacitance coefficient of MIoTD, which is related to the hardware architecture of the device. Considering that MIoTD has a fixed position and relatively stable functions, assume that their common service programs have been pre - installed at the time of deployment. Therefore, the program acquisition overhead is not considered in the local computing mode.

[0128] 2) Offloading computation: When MIoTDk chooses to establish an association with UAVm, that is, when a k,m = 1, a partial offloading strategy is adopted. Define the continuous variable β k,m to represent the task offloading ratio, β k,m ∈[0,1]. Then the local computing delay and energy consumption of MIoTDk are respectively expressed as:

[0129]

[0130] The transmission delay and energy consumption of MIoTDk offloading tasks to UAVm are respectively expressed as:

[0131]

[0132] Let the computing power of UAVm be F U . Then the computing delay and energy consumption required for UAVm to process the offloaded tasks of MIoTDk are respectively expressed as:

[0133]

[0134] Among them, γ U represents the effective capacitance coefficient of UAV;

[0135] When the required service program is not cached on UAVm, UAVm needs to download the service from the shore - side base station, and the download delay is expressed as:

[0136]

[0137] Among them, R b is the download rate of the wireless backhaul link. Since the amount of data of the computing task result is usually much smaller than the amount of input data, the overhead generated by the result return is ignored. Therefore, the total delay of the air - sea - land integrated network system model based on computing offloading to complete the computing task W k is expressed as:

[0138]

[0139] Finally, the total delay of the air - sea - land integrated network system model is expressed as:

[0140]

[0141] In a specific embodiment, in S5, based on the calculation model, under the conditions of the UAV cache capacity constraint, the Marine Internet of Things devices, and the UAV available energy constraint, an optimization model for jointly optimizing the user association decision, the service caching decision, and the task offloading ratio with the goal of minimizing the total system delay is as follows:

[0142] For the air-ground-sea integrated network system model, the goal of this embodiment is to jointly optimize the user association decision under the conditions of the UAV cache capacity constraint, the MIoTD, and the UAV available energy constraint by service caching decision and task offloading ratio

[0143]

[0144] such that the total system delay is minimized. Therefore, the optimization model is formulated as: and are the maximum available energies of the MIoTD and the UAV, respectively.

[0145] Specifically, in this embodiment, since the MIoTD and the UAV are usually powered by batteries with limited capacity, energy limitations must be imposed to prevent battery depletion, that is, energy constraints are imposed through C6 and C7. At the same time, since the purpose of this embodiment is to optimize the computing energy consumption of the UAV, the hovering power is not considered for optimization, and it is assumed that the hovering power is constant.

[0146] In a specific embodiment, in S6, the specific steps of designing an efficient iterative algorithm based on Block Coordinate Descent (BCD) to decompose the optimization problem into several sub-models and selecting the coalition game, linear relaxation, or Lagrangian dual method for solution according to the convexity, concavity, and structure of each sub-model are as follows:

[0147] Specifically, in this embodiment, the joint optimization model belongs to the mixed integer nonlinear programming (MINLP) problem and is subject to practical constraints such as the energy of MIoTD, the energy of UAV, and the cache capacity. To effectively solve the NP-hard MINLP problem, the block coordinate descent method is adopted to decompose it into three tractable sub-models, and then techniques suitable for their mathematical structures are used to solve each sub-model. The coalition game is applied to user association to form a stable MIoTD-UAV partnership. Linear relaxation and the Lagrangian dual method are used for service cache optimization, and convex optimization is used for task offloading ratio optimization.

[0148] For the given service cache decision B and task offloading ratio β, the optimization model regarding the user association decision A is formulated as:

[0149]

[0150] Each MIoTD needs to decide whether to offload and to which UAV to offload the computing task, while each UAV can serve multiple MIoTDs. This many-to-many cooperative decision-making relationship naturally fits the characteristics of the coalition game. Therefore, in this embodiment, MIoTD and UAV are jointly modeled as a coalition game problem with K + M players. By constructing appropriate utility functions (Definition 4) and game rules (Definition 6), the players can autonomously form a stable coalition structure, thus solving it efficiently. Solve The process of

[0151] Definition 1 (Coalition game modeling): The air-sea-land integrated network system model is modeled as a triple where represents the set of all game players, represents the coalition characteristic function, which is used to measure the coalition utility; represents the set of all possible formed coalitions, To ensure the effectiveness of the coalition structure, it is necessary to satisfy:

[0152] (1) For any two different coalitions there is That is, each player can only belong to one coalition at the same time;

[0153] (2) That is, all players must join a certain coalition.

[0154] Coalition The physical meaning of changes with the value of c. When 1 ≤ c ≤ K, Denote the set of MIoTDs that choose to execute tasks completely locally; when \(K + 1\leq c\leq K + M\), denote the coalition composed of UAVm = c - K and all MIoTDs it serves.

[0155] Definition 2 (Coalition Delay Cost): To accurately describe the payoffs of participants in a coalition game, it is necessary to define an appropriate delay cost. For any coalition define its delay cost as:

[0156]

[0157] Definition 3 (Delay Allocation Value): For MIoTDk in the coalition its fair delay allocation value is calculated using the Shapley value and is given by:

[0158]

[0159] where S represents any subset of the coalition and T(S) represents the delay cost of subset S; considering the marginal contribution of each device to the coalition through the Shapley value ensures the fairness of the allocation.

[0160] Definition 4 (Coalition Characteristic Function): Based on the delay allocation value, for any coalition its characteristic function is defined as:

[0161]

[0162] where the characteristic function represents the delay savings brought by the coalition to its members. When an MIoTD chooses to execute locally, there is no delay savings and the characteristic function value is 0; when an MIoTD joins the UAV service coalition, the characteristic function value is the sum of the delay savings obtained by all MIoTDs through task offloading.

[0163] Definition 5 (Preference Relationship): To achieve distributed coalition formation, it is necessary to define the preference relationship and transfer rules of participants. For any MIoTDk and two different coalitions define the preference relationship as:

[0164]

[0165] Definition 6 (Coalition Transfer Rule): Given the current coalition structure U, when MIoTDk transfers from coalition to coalition , the updated coalition structure is:

[0166]

[0167] For a given task offloading ratio β and the optimized user association decision A, the optimization problem regarding the service caching decision B is formulated as:

[0168]

[0169] To solve the computational complexity, this embodiment considers linear relaxation for it, relaxing the binary variable condition in C2 to At this time, the original problem will be transformed into a linear programming problem Expressed as:

[0170]

[0171] For considering that its constraint conditions and objective function have good separable structure characteristics, the Lagrangian dual method is used for solution. This method transforms the original constrained optimization model into an unconstrained optimization form for solving the dual problem by introducing Lagrange multipliers. Since the linear property of this model guarantees the establishment of strong duality, the optimal solution of the original model can be obtained by solving the dual problem.

[0172] Specifically, first introduce the Lagrange multiplier λ m ≥0 for the constraint C5, and introduce the multipliers μ j,m ≥0 and v j,m ≥0 for the upper and lower bound constraints of C2 respectively. Correspondingly, the Lagrangian function of

[0173]

[0174] can be constructed as:

[0175]

[0176] The dual problem of

[0177]

[0178] The process of using the Lagrangian dual method to solve the optimization model includes:

[0179] In the t-th iteration, it is necessary to solve the dual function g(λ (t) , μ (t) , v (t) ), since the Lagrangian function is separable with respect to the original variables therefore It can be decomposed into M×J independent sub-models. For each pair (j,m), the optimality condition for analysis is as follows:

[0180]

[0181] Denote c j,m as the delay cost when service j is not cached on UAV m. Then The optimal update of is expressed as:

[0182]

[0183] where θ is the step size, θ>0; [x] [0,1] represents the operation of projecting the variable onto the interval [0,1]. This update form has a clear physical meaning, that is: when the caching cost is less than the delay benefit c j,m , it tends to increase the caching ratio of the corresponding service, otherwise it tends to reduce the caching.

[0184] According to the constraint violation situation, the subgradient method is used to update the Lagrange multiplier, and the update rule is:

[0185]

[0186]

[0187] where τ (t) is the step size sequence that satisfies the convergence condition of the subgradient method.

[0188] To ensure that the Lagrangian dual algorithm converges to the optimal solution, in this embodiment, the primal-dual gap and the degree of constraint violation are used as the stopping criteria of the algorithm. Specifically, in the t-th iteration, the primal feasibility residual vector is defined as:

[0189]

[0190] where,

[0191]

[0192] The dual gap is defined as:

[0193]

[0194] When both and are satisfied, the algorithm stops iterating, where ∈ p >0 and ∈ d >0 are the preset tolerance thresholds respectively.

[0195] Since The optimal solution is a continuous value. Therefore, it is necessary to convert it into a binary solution that satisfies the constraints of the original problem through an appropriate rounding strategy. Considering the special structure of the problem, a deterministic rounding algorithm based on the greedy idea is adopted in this embodiment. For each UAV, its service cache priority is defined. where ρ j,m comprehensively considers the value of the linear relaxation solution and the storage overhead of the service. The higher the value, the higher the cache cost performance of the service.

[0196] For the optimized user association decision A and service cache decision B, the optimization model regarding the task offloading ratio β is formulated as:

[0197]

[0198] Lemma 1: Given A and B, is a standard convex optimization problem.

[0199] Specifically, the objective function T total is a linear weighted sum of various transmission and computing delays, where is proportional to 1 - β k,m while and are proportional to β k,m and the others are constants. Therefore, T total is convex with respect to β k,m Regarding the constraints, C7 is a convex set, C6 can be reorganized into a linear inequality form of β k,m while still maintaining convexity, and C3 constitutes a convex feasible region. Therefore, this problem has a convex objective function and a convex constraint set, and it is a standard convex optimization problem. In this embodiment, a mature convex optimization toolbox such as CVX is used to efficiently solve the problem.

[0200] Specifically, in this embodiment, to solve the joint optimization model we propose an iterative optimization framework, as shown in Figure 4 This algorithm systematically decomposes into manageable subproblems. In each iteration, the algorithm uses the coalition game method to solve the sub-model for optimizing the user association decision, uses the linear relaxation and Lagrangian duality method to solve the sub-model for optimizing the service cache decision, and uses the CVX toolbox to solve the sub-model for optimizing the task offloading ratio. When the number of iterations reaches the maximum value R max or the relative change of the objective value is less than ∈, the algorithm terminates.

[0201] In this embodiment, in order to comprehensively evaluate the algorithm performance, the following four benchmark schemes are designed for comparison:

[0202] Local only: All computing tasks are executed locally on the MIoTD without task offloading or service caching optimization;

[0203] Random caching: Service caching is randomly performed under the condition of meeting the UAV storage capacity constraint, and the user association decision and task offloading ratio are optimized;

[0204] Fixed offloading: When selecting offloading calculation, a preset fixed task offloading ratio is adopted to optimize the user association and service caching decision;

[0205] Nearby offloading: Users can choose local calculation, but must be associated with the nearest UAV when offloading, and the task offloading ratio and service caching decision are optimized.

[0206] Figure 5 and Figure 6 is the optimization result of the proposed UAV deployment scheme. Figure 5 The two-dimensional visualization display shows that the optimized UAVs maintain a reasonable distance from the initial uniform positions, verifying the rationality of the initial deployment strategy. The coverage areas (represented by red dashed circles) form well-separated service cells with little overlap, effectively reducing the interference between UAVs. Each UAV evenly serves 5 - 7 neighboring MIoTDs, demonstrating good load distribution. In addition, all MIoTDs are within the service range of at least one UAV, successfully avoiding coverage blind spots. As Figure 6 shown, the deployment heights of the UAVs are distributed in the range of 400 - 600 meters, strictly following the safety height constraint. For UAVs that need to cover a large service area (such as the lower left area), their operating heights are relatively high (about 560 meters) to obtain sufficient coverage radius; while UAVs serving dense areas (such as the upper left area) maintain a lower flight height (about 400 meters) to achieve lower communication latency.

[0207] Figure 7Shows the convergence performance of the iterative algorithm (Proposed) in this embodiment under different scenarios. First, in all scenarios, the proposed algorithm exhibits good convergence. The total latency curve rapidly decreases in the first 4 - 5 iterations and then levels off. Second, by comparing the performance of different parameter configurations, the following conclusions can be drawn: (1) With the same number of UAVs, a larger task data volume leads to a higher total latency. This is because a larger data volume not only increases the transmission overhead but also raises the computational load. With a fixed task data volume, increasing the number of MIoTDs results in an increase in the total latency. This is mainly due to more MIoTDs competing for limited communication and computational resources; (2) The number of UAVs has a significant impact on system performance. When M decreases from 4 to 2, there is a significant increase in the total latency in all scenarios, indicating that increasing the number of UAVs can effectively disperse the system load and improve the quality of service. Finally, it is worth noting that the convergence speed of the curve is closely related to the system scale. In larger-scale scenarios, the proposed algorithm in this embodiment requires more iterations to reach a steady state. Nevertheless, even in the most complex scenario (K = 30, L k = 0.6 Mbit, M = 4), the proposed algorithm in this embodiment can still converge to a stable solution within six iterations, demonstrating good robustness.

[0208] Figure 8 Shows the performance comparison of different schemes under varying local computation capacity. As F D increases, the total latency of all schemes shows a downward trend. For the local computation-only scheme, its performance is the most sensitive to changes in computation capacity. This is because this scheme relies entirely on local processing, and an increase in computation capacity directly translates into performance improvement. In contrast, although the fixed offloading and random caching schemes also improve as F D increases, the improvement amplitude is relatively small. The proposed scheme maintains an obvious performance advantage throughout the range of computation capacity, reducing the total latency by approximately 10.53% on average compared to the nearest offloading scheme. This advantage mainly benefits from the joint optimization of user association, service caching decision, and task offloading ratio by the proposed scheme. When the local computation capacity is limited, the algorithm tends to offload more computational load to the UAV; as the computation capacity increases, the algorithm will correspondingly increase the proportion of local computation. Although the nearest offloading scheme also considers service caching and task offloading, its performance always lags behind the proposed scheme due to the lack of optimization of user association. It is worth noting that when the local computation capacity exceeds 1.2 GHz, the performance improvement of all schemes tends to level off, indicating that the system bottleneck may have shifted to energy consumption.

[0209] Figure 9The trend of the total delay varying with the Number of MioTDs is analyzed. When the number of MioTDs increases from 15 to 40, the total delay of all schemes shows an approximately linear upward trend, which reflects the basic characteristic that the system load continuously increases with the increase in the number of MioTDs. Only the local computing scheme shows the steepest growth trend, with the total delay increasing sharply from 6.43 seconds to 16.29 seconds, an increase of 153.3%. Among the benchmark schemes, the random caching and fixed offloading schemes perform similarly. When the number of MioTDs reaches 40, their total delays are 14.84 seconds and 15.10 seconds respectively. The nearest offloading scheme is slightly better than these two schemes, with a final delay of 14.70 seconds. This improvement benefits from its distance-based task allocation strategy, which can balance the communication overhead to a certain extent. The proposed scheme maintains the best performance throughout the test range. When the number of MioTDs is 40, the total delay is 14.03 seconds, which is about 4.6% lower than that of the nearest offloading scheme. It is worth noting that, as shown in the enlarged middle part, when the number of MioTDs is relatively large (35 - 40), the advantage of this scheme is more obvious. This shows that the joint optimization scheme can still maintain good system performance under high load conditions. Figure 9 As shown in the enlarged middle part

[0210] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A joint service caching and task offloading method for air-ground integrated edge computing in marine Internet of Things, characterized in that: The specific steps include: S1: Construct an air-sea-land integrated network system model capable of deploying UAVs in three-dimensional space; S2: Construct a communication model based on the air-sea-land integrated network system model; S3: Develop a service caching and task offloading strategy based on the air-sea-land integrated network system; S4: Construct a computing model according to the communication model and the service caching and task offloading strategy, where the computing model is used to calculate the total delay and total energy consumption required to complete the task; S5: Develop an optimization model for jointly optimizing user association decisions, service caching decisions, and task offloading ratios with the goal of minimizing the total system delay under the constraints of UAV cache capacity, marine Internet of Things device, and available energy of UAVs; S6: Design an iterative algorithm based on block coordinate descent to decompose the optimization model into several submodels, and select coalition game, linear relaxation, Lagrangian duality method, and convex optimization algorithm for solution according to the convexity and structure of each submodel; S7: Obtain the optimal user association decision results, service caching decision results, and task offloading ratios through multiple iterative solutions, minimizing the total delay of the air-sea-land integrated network system model.

2. The joint service caching and task offloading method for air-ground integrated edge computing in the marine Internet of Things according to claim 1, characterized in that In S1, the constructed air-sea-land integrated network system model includes: Set \(K\) MIoTDs randomly distributed on the sea surface, use \(M\) UAVs equipped with edge servers as aerial computing nodes with caching capabilities, and use the shore base station as the data center to maintain a complete service program library; the sets of UAVs and MIoTDs are respectively denoted as and Suppose MioTDk has a computing task to be executed W k =(L k , C k ), where L k represents the amount of computing task data; C k represents the number of CPU cycles required for each bit of calculation; for each computing task W k , the air-sea-land integrated network system model has pre-determined the services it requires, and the set of all services is represented as where the storage space size occupied by each service is A three-dimensional Cartesian coordinate system is established with the sea level as the reference plane, where the xoy plane coincides with the sea level; the position of MIoTDk is defined as q k =(x k , y k , 0) T , and the initial deployment position of UAVm is where represents the initial altitude of UAVm; for each computing task, there are M + 1 computing nodes, forming a set When m = 0, it means that the computing task is processed locally at MIoTD. When 0 < m ≤ M, it means that the computing task is offloaded to UAVm for execution; Through the binary variable a k,m represents the user association decision. If MIoTDk selects node m to process the task, then a k,m = 1; otherwise, a k,m = 0. To ensure the unique allocation of computing tasks, there is Definition as the set of MIoTD services for UAVm, set the final position q of UAVm m to be at the geometric center of the MIoTD it serves, q m =(x m , y m , H m ); Set the calculation formulas for the horizontal and vertical coordinates of UAVm on the horizontal plane as follows: Assume that all UAVs have the same coverage angle θ, and define the service coverage radius R of UAVm m as the projection radius of its effective service range on the horizontal plane, and the formula is: R m = H m tan(θ) (3) To ensure that all associated MIoTDs are within the effective coverage range of the UAV, the flight altitude of UAVm needs to satisfy: where μ is the safety margin to avoid MIoTDs being located at the coverage boundary.

3. The joint service caching and task offloading method for air-ground integrated edge computing in the marine Internet of Things according to claim 2, characterized in that, In S2, the communication model constructed based on the air-sea-land integrated network system model includes: Calculate the channel gain between MIoTDk and UAVm based on the two-path channel propagation model, and the formula is: where λ is the carrier wavelength, H1 is the height of the MIoTD antenna from the sea level, and d k,m is the Euclidean distance between MIoTDk and UAVm, expressed as Based on the Shannon channel capacity theory, the data rate between MIoTDk and UAVm is: Among them, B is the bandwidth allocated to MIoTDk, and P k,m is the transmit power, and σ 2 is the additive white Gaussian noise power at the receiving end.

4. The joint service caching and task offloading method for air-ground integrated edge computing in the marine Internet of Things according to claim 3, wherein, In S3, the service caching and task offloading strategy developed based on the air-sea-land integrated network system includes: Define the binary variable b j,m , where the b j,m represents the caching status of service j on UAV m. b j,m ∈{0,1}. When service j is cached on UAV m, b j,m =1; otherwise, b j,m =0. Then the caching capacity constraint of the UAV is expressed as: Among them, S m is the maximum storage capacity of UAVm; For each computing task, the MIoTD divides it into two parts based on the partial offloading mode, one part is computed locally, and the other part is offloaded via a wireless link to the UAV for computing; It is assumed that each UAV is equipped with a storage unit with a limited capacity and some services are pre-loaded in the local storage; after receiving the offloaded computing task, the UAV first checks whether the local storage contains the services required for processing the task. If the corresponding services exist, the computing is directly executed; otherwise, the UAV will request the required services from the shore base station via a wireless backhaul link and then execute the computing after downloading the required services.

5. The joint service caching and task offloading method for air-ground integrated edge computing in the marine Internet of Things according to claim 4, wherein In S4, the computing model constructed according to the communication model and the service caching and task offloading strategy includes: 1) Local computing: When the MIoTDk completely relies on local computing to process tasks, the air-sea-land integrated network system model does not involve any task offloading. Define the local computing capacity of the MIoTD as F D , then the total delay and total energy consumption for completing the computing task W k are respectively expressed as: Among them, γ D represents the effective capacitance coefficient of MIoTD; 2) Offloading Computation: When MIoTDk chooses to establish an association with UAVm, a partial offloading strategy is adopted, and a continuous variable β is defined k,m to represent the task offloading ratio, where β k,m ∈[0,1]. Then, the local computation delay and energy consumption of MIoTDk are respectively expressed as: The transmission delay and energy consumption of MIoTDk offloading tasks to UAVm are respectively expressed as: Let the computing power of UAVm be F U , then the computing delay and energy consumption required for UAVm to process the offloading task of MIoTDk are respectively expressed as: Among them, γ U represents the UAV effective capacitance coefficient; When the required service program is not cached on UAVm, the download delay of UAVm downloading the service from the shore base station is expressed as: where R b is the download rate of the wireless backhaul link; The total latency of the air-sea-land integrated network system model for completing the computing task W based on computing offloading is expressed as: k ​ The total delay of the air-sea-land integrated network system model is expressed as:

6. The joint service caching and task offloading method for air-ground integrated edge computing in the marine Internet of Things according to claim 5, characterized in that In S5, the optimization model for jointly optimizing user association decisions, service caching decisions, and task offloading ratios is expressed as: Among them, C1 and C2 represent user association and service caching decisions as binary variables; C3 represents that the value range of the task offloading ratio is [0,1]; C4 indicates that each MIoTD must select a computing destination, either adopting pure local computing or establishing an association with a certain UAV and adopting a partial offloading mode; C5 represents that the space occupied by all services cached on each UAV cannot exceed its maximum storage capacity; C6 and C7 are used to ensure that the energy consumed by each MIoTD and UAV does not exceed its energy limit, and are the maximum available energies of the MIoTD and UAV respectively.

7. The joint service caching and task offloading method for air-ground integrated edge computing in the marine Internet of Things according to claim 6, wherein In S6, through an efficient iterative algorithm based on block coordinate descent, the optimization problem is decomposed into several sub-models, and the solution includes: For a given service caching decision B and task offloading ratio β, the optimization model regarding the user association decision A is formulated as: Model MIoTD and UAV jointly as a coalition game problem with K+M participants, and by constructing a utility function and game rules, enable the participants to form a stable coalition structure to solve 8. The joint service caching and task offloading method for air-ground integrated edge computing in the marine Internet of Things according to claim 7, wherein In S6, through an efficient iterative algorithm based on block coordinate descent, the optimization problem is decomposed into several sub-models, and the solution also includes: For a given task offloading ratio β and the optimized user association decision A, the optimization model regarding the service caching decision B is formulated as: Pair Perform linear relaxation and relax the binary variable condition in C2 to Obtain a linear programming model Expressed as: Solve the model using the Lagrangian duality method 9. The joint service caching and task offloading method for air-ground integrated edge computing in the marine Internet of Things according to claim 8, wherein In S6, through an efficient iterative algorithm based on block coordinate descent, the optimization problem is decomposed into several sub-models, and the solution also includes: For the optimized user association decision A and service caching decision B, the optimization model regarding the task offloading ratio β is formulated as: Use the convex optimization toolbox to solve the model.

Images