A mobile edge task offloading method of a sand cat swarm cooperative computing strategy

Abstract

A mobile edge task offloading method based on a sandcat-based collaborative computing strategy is proposed. This method proposes a two-step offloading computing strategy combining user collaboration and UAV assistance: First, considering the user communication coverage, an improved K-Means clustering strategy is used to cluster maritime users, achieving collaborative offloading and task integration among users. Second, a mobile edge computing architecture assisted by a single UAV is constructed. An optimization model is established with task processing latency as a constraint, aiming to minimize the weighted total energy consumption of the UAV and users. Based on this, a joint trajectory optimization and task offloading algorithm are used to solve the model. This invention can effectively alleviate maritime communication network congestion, significantly reduce system weighted energy consumption while meeting latency requirements, and achieve full utilization of communication and computing resources.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication and edge computing technology, specifically relating to a mobile edge task offloading method based on a sandcat swarm intelligence collaborative computing strategy. Background Technology

[0002] With the deepening of the integrated "air, land, sea" communication network, applications such as marine environmental monitoring and emergency search and rescue are experiencing explosive growth. These computationally intensive and latency-sensitive tasks place extremely high demands on the bandwidth and computing power of communication networks. However, the harsh marine environment, difficulties in infrastructure deployment, and sparse distribution of user nodes (such as unmanned surface vessels) limit the ability of terminals to independently process large amounts of data due to limitations in battery and computing resources. Furthermore, traditional satellite or shore-based communications face high cost and high latency bottlenecks.

[0003] Against this backdrop, utilizing drones to carry mobile edge computing (MEC) servers has become a research hotspot. While this can alleviate pressure on core networks, it still has limitations in wide-area sparse scenarios: First, drones traversing dispersed nodes lead to excessively long paths and high energy consumption; second, existing research has not fully utilized the collaborative communication capabilities between users; finally, facing complex joint trajectory optimization and task offloading problems, traditional swarm intelligence algorithms converge slowly and are prone to getting trapped in local optima, making it difficult to minimize energy consumption under latency constraints. Therefore, there is an urgent need to develop an efficient MEC task offloading method that adapts to the special marine environment and integrates user collaboration mechanisms. Summary of the Invention

[0004] This invention addresses the problems of low utilization of communication and computing resources, high system energy consumption, and large task processing latency in existing marine mobile edge computing networks due to the harsh marine environment, sparse and widespread distribution of users (such as unmanned surface vessels, USVs), and lack of communication infrastructure. Furthermore, it addresses the shortcomings of existing swarm intelligence optimization algorithms in handling complex joint trajectory optimization and task offloading problems, such as getting trapped in local optima, slow convergence speed, and uneven initial population distribution. Therefore, this invention proposes a task offloading and trajectory optimization method for marine mobile edge computing based on user collaboration and UAV assistance.

[0005] This invention provides a mobile edge task offloading method for a sandcat swarm intelligence collaborative computing strategy, comprising the following key steps:

[0006] First, the construction of the UAV-assisted MEC system model:

[0007] 1.1 Establish a communication model;

[0008] Step 1.2: Establish a computational model;

[0009] Section 1.3, Modeling the Joint Trajectory and Unloading Problem;

[0010] Second, based on the improved K-Means unmanned surface vessel clustering strategy;

[0011] Section 2.1, K-Means clustering algorithm;

[0012] Section 2.2, USV criticality calculation;

[0013] Third, joint trajectory optimization and task unloading are performed based on an improved sand cat swarm optimization algorithm:

[0014] Section 3.1, Population initialization based on Piecewise chaotic mapping;

[0015] 3.2 Nonlinear adjustment mechanism;

[0016] 3.3 Spiral Search Strategy.

[0017] Furthermore, the method for establishing the communication model in step 1.1 is as follows: assuming the communication link between the UAV and USV is controlled by a line-of-sight link, the Doppler frequency shift during communication can be compensated by the receiver, the channel quality depends on the line-of-sight link, and the channel gain follows the free-space loss mode, i.e., the communication between the UAV and the cluster-head unmanned surface vessel... The channel gain between is ,in It is a cluster-headed unmanned surface vessel Between with UAV in the 1st Frame communication distance, As a reference distance of 1m, the channel gain at the th... Frame, cluster-headed unmanned surface vessel The size of the unloading task to UAV is

[0018] (1)

[0019] in It is the channel bandwidth. It is in the Frame Cluster Head Unmanned Surface Vessel The transmit power of the offloading task. It is the received noise power of the UAV, at the... Frame, cluster-headed unmanned surface vessel The energy consumption of the unloading task is

[0020] (2)

[0021] set up For the maximum transmit power of the USV, there is , For the first Frame Cluster Head Unmanned Surface Vessel The maximum amount of data that can be unloaded from the UAV is [number]. , The relationship with Pmax is

[0022] (3)

[0023] Based on different mission characteristics and application scenarios, a fixed-wing UAV is used. During flight, the UAV provides offloading computation services to the USV, optimizing the UAV's flight trajectory. This method employs a classic aircraft power consumption model known in aerodynamic theory, specifically in the [missing information - likely a specific phase or stage]. Frames model the energy consumption of drone flight as

[0024] (4)

[0025] in For the current frame's flight speed of the UAV, there is , The duration of each frame, and For two parameters related to UAV weight, wing area, and air density, in In extreme cases, the energy consumption of a fixed-wing drone is infinite, which confirms that a fixed-wing drone must maintain a minimum forward speed to stay airborne.

[0026] The method for establishing the computational model in step 1.2 is as follows: Dynamic Voltage Frequency Scaling (DVFS) technology is used to dynamically adjust the CPU frequency of each time slot USV and UAV for task computation, and it is assumed that USV can perform local computation and task offloading simultaneously.

[0027] (1) Computational model of cluster-headed unmanned surface vessel

[0028] In the Frame, cluster-headed unmanned surface vessel The local computing task data volume is

[0029] (5)

[0030] in, Computing cluster-head unmanned surface vessels for drones In the Frame CPU frequency, The highest CPU frequency for drones, To calculate the 1-bit cluster head of an unmanned surface vessel The number of CPU cycles required for the task, in the first... Frame, cluster-headed unmanned surface vessel Local computing power consumption is

[0031] (6)

[0032] in Cluster-headed unmanned surface vessel The processor's effective switched capacitors;

[0033] (2) UAV computational model

[0034] Assuming the drone can simultaneously calculate the task of each cluster-head unmanned surface vessel, in the... Frame-based unmanned aerial vehicle (UAV) computing cluster head unmanned surface vessel The amount of data is

[0035] (7)

[0036] in, Computing cluster-head unmanned surface vessels for drones In the Frame CPU frequency, The highest CPU frequency for drones, In the Frame-based unmanned aerial vehicle (UAV) computing cluster head unmanned surface vessel The energy consumption of the task is

[0037] (8)

[0038] in For the effective switched capacitor of the drone processor;

[0039] The method for modeling the joint trajectory and unloading problem in step 1.3 is as follows: the total energy consumption of the cluster-headed unmanned surface vessel consists of local computation energy consumption and task unloading energy consumption, while the total energy consumption of the UAV consists of flight energy consumption and computation energy consumption. In the The total power consumption of a frame is

[0040] (9)

[0041] Drones in The total power consumption of a frame is

[0042] (10)

[0043] The optimization objective is to minimize the weighted total energy consumption E of the cluster-head unmanned surface vessel and the unmanned aerial vehicle, expressed as:

[0044] (11)

[0045] in and These are weighting factors for the energy consumption of unmanned surface vessels and drones, respectively. Their values ​​can be preset. The drone flight trajectory is jointly optimized using Q={q[ Task data volume and CPU frequency The problem is described as follows

[0046] (12)

[0047] Among them, constraint (12)a guarantees that the UAV can only calculate the tasks received before the given time slot. Constraint (12)b guarantees that the UAV can completely calculate all the tasks unloaded by the cluster head UAV with a time delay of T. Constraint (12)c guarantees that all tasks of the cluster head UAV can be completely calculated with a time delay of T. Constraint (12)d guarantees that the UAV flies from the starting position to the ending position. Constraint (12)e guarantees that the UAV cannot exceed the maximum flight speed. Constraint (12)f guarantees that the cluster head UAV does not unload the tasks of the last time slot. Constraint (12)g guarantees that the UAV does not calculate the tasks of the first time slot. Constraint (12)h guarantees that the data volume of the tasks selected for unloading by the cluster head UAV in each frame is non-negative. Constraint (12)i guarantees that the cluster head UAV in the first time slot... The CPU's computation frequency in frame n does not exceed the maximum value. Constraint (12)j guarantees that the CPU's computation frequency in frame n does not exceed the maximum value. Constraint (12)k guarantees that the cluster-head unmanned surface vessel's computation frequency in frame n does not exceed the maximum value. The transmit power of the frame offloading task shall not exceed the maximum value.

[0048] Furthermore, the K-Means clustering algorithm in step 2.1 is as follows: Given A user nodes and M cluster heads, M nodes are randomly selected from the A user nodes as initial cluster heads. Let the cluster be... , each node Assign it to the cluster with the nearest cluster head, and determine the distance to the node. The cluster marker of the most recent cluster head is ,Will Assigned to the corresponding cluster Then, for each cluster, its optimal cluster head is recalculated. Update the cluster head, repeat the allocation and update steps until the cluster head no longer changes, and obtain the clustering result. };

[0049] The method for calculating USV criticality in step 2.2 is as follows: In the scenario of maritime mission integration and offloading, computing and energy resources are extremely valuable. The selection of cluster heads needs to consider both communication coverage and the cost-effectiveness of transmission energy consumption caused by the size of mission data. During the task integration process within a cluster, if a USV with a small amount of data is selected as the cluster head, then large-volume tasks within the cluster need to be offloaded to the cluster head, which leads to energy waste. Therefore, USVs with larger amounts of data should be selected as cluster heads whenever possible. Thus, the size of the mission data volume needs to be an important basis for cluster head selection. In addition, considering mission latency constraints, timely processing of tasks with faster data generation rates is also crucial. Data generation rate should also be a necessary principle for cluster head selection. Based on the above factors, a USV criticality is proposed. As a new indicator for measuring the importance of a task The keyness of the i-th USV within cluster j is expressed as:

[0050] (13)

[0051] in Let $i$ be the average distance between the $i$-th USV and all other USVs in cluster $j$. Let i be the data volume of the i-th USV task. The data generation rate for the i-th USV task. Generate rate weights for the data.

[0052] Furthermore, in step 3.1, the population initialization method based on Piecewise chaotic mapping is as follows: The convergence and search accuracy of the sand cat swarm algorithm are affected by the initial spatial distribution of the population. This makes it possible to effectively improve the optimization efficiency when the initial sand cat population is evenly distributed in a region, while the opposite may reduce the optimization efficiency of the algorithm. Piecewise chaotic mapping is selected to initialize the sand cat population, so that it has the characteristics of even distribution and wide traversal. The effectiveness of the improvement strategy is demonstrated by comparing the results generated by Piecewise chaotic mapping population initialization and random population initialization.

[0053] The nonlinear adjustment mechanism in step 3.2 is as follows: the balance between searching for and attacking prey in the SCSO algorithm depends on the value of the balance parameter R. As shown in formula (6), the value of R depends on auditory sensitivity. The changes. And the big ones. It is beneficial for global search, small It is advantageous for local attacks, as is the case with the traditional SCSO algorithm. As the number of iterations decreases linearly from 2 to 0, its R value does not accurately reflect the search and attack process of the SCSO algorithm. The update formula is:

[0054] (14);

[0055] The spiral search strategy in step 3.3 is as follows: During the prey-hunting phase of the SCSO algorithm, the spiral search strategy of the whale algorithm is introduced for position updates. This allows individual sand cats to have more paths to find prey, thus achieving the goal of updating their positions. This process improves both the algorithm's local exploitation and global search capabilities, thereby enhancing its optimization performance. The update formula for the standard sand cat swarm algorithm during the prey-hunting phase is:

[0056] (15)

[0057] Add the spiral search strategy factor as shown in formula (10).

[0058] (16)

[0059] Where k is a constant;

[0060] After incorporating the spiral search factor, the formula for updating the optimal position of a sand cat individual becomes:

[0061] (17)

[0062] After incorporating a spiral search strategy into the position update expression during the search phase, the sand cat swarm will search in space in a spiral form. This not only enhances the sand cat swarm's ability to explore unknown areas but also increases the probability of the SCSO algorithm escaping local optima, effectively improving the optimization performance of the SCSO algorithm. In formula (16), l is a random parameter, and the value of l changes with the increase of the number of iterations. Its update formula is:

[0063] (18).

[0064] The advantages and positive effects of this invention are:

[0065] This invention employs a two-step offloading strategy combining user collaboration and UAV assistance, significantly improving the energy efficiency of maritime communication systems. First, an improved K-Means clustering strategy is proposed, incorporating a user criticality index to effectively integrate sparsely distributed maritime tasks and shorten the UAV service flight path, thereby substantially saving UAV flight energy consumption. Second, an improved Sand Cat Group Optimization (ISCSO) algorithm is designed, utilizing Piecewise chaotic mapping initialization, nonlinear parameter adjustment, and a spiral search strategy to enhance the algorithm's global optimization capability and convergence speed, effectively solving the problem of easily getting trapped in local optima under complex constraints. Simulation results show that, while satisfying time delay constraints, this invention can significantly reduce the system's weighted total energy consumption, outperforming existing mainstream algorithms. Attached Figure Description

[0066] Figure 1 This is a model diagram of a single unmanned aerial vehicle (UAV) assisted mobile edge unloading system for offshore operations;

[0067] Figure 2 This is a timeframe diagram for task unloading and computation;

[0068] Figure 3 These are screenshots showing the effects of different initialization methods;

[0069] Figure 4 This is the ISCSO algorithm flowchart.

[0070] Figure 5 This is a schematic diagram of the USV distribution;

[0071] Figure 6 This is a clustering result diagram of the improved K-Means algorithm;

[0072] Figure 7 These are flight trajectory diagrams of UAVs at different mission completion times;

[0073] Figures 8-11 This relates the data volume of the cluster head USV task to its weighted energy consumption. Figure 8 Is Relationship diagram Figure 9 Is , Figure 10 Is Relationship diagram Figure 11 Is Relationship diagram;

[0074] Figure 12 This is a flowchart of the mobile edge task unloading method of the Sand Cat Crowd Collaborative Computing Strategy of the present invention. Detailed Implementation

[0075] This embodiment provides a mobile edge task offloading method based on the Sand Cat swarm intelligence collaborative computing strategy. This method addresses scenarios with harsh maritime communication environments and scarce infrastructure by constructing a single unmanned aerial vehicle (UAV)-assisted mobile edge computing (MEC) system. For example... Figure 1 As shown, the system comprises one fixed-wing UAV equipped with an MEC server and A unmanned surface vessels (USVs). The UAV flies at a constant altitude H, providing periodic edge computing services to a group of cluster-head USVs.

[0076] The specific implementation steps of the mobile edge task offloading method of the sand cat swarm intelligence collaborative computing strategy provided by this invention are described in detail. See appendix. Figure 9 The mobile edge task offloading method of the Sand Cat swarm intelligence collaborative computing strategy in this embodiment includes the following key steps:

[0077] First, the construction of the UAV-assisted MEC system model:

[0078] 1.1 Establish a communication model;

[0079] Step 1.2: Establish a computational model;

[0080] Section 1.3, Modeling the Joint Trajectory and Unloading Problem;

[0081] Second, based on the improved K-Means unmanned surface vessel clustering strategy;

[0082] Section 2.1, K-Means clustering algorithm;

[0083] Section 2.2, USV criticality calculation;

[0084] Third, joint trajectory optimization and task unloading are performed based on an improved sand cat swarm optimization algorithm:

[0085] Section 3.1 Population initialization based on Piecewise chaotic mapping;

[0086] 3.2 Nonlinear adjustment mechanism;

[0087] 3.3 Spiral Search Strategy.

[0088] Further details regarding the system model can be found in the appendix. Figure 1 The method for establishing the communication model in step 1.1 is as follows; for the timeframe of task unloading and computation, please refer to the appendix. Figure 2 Assuming the communication link between the UAV and USV is controlled by a line-of-sight link, the Doppler shift during communication can be compensated by the receiver, the channel quality depends on the line-of-sight link, and the channel gain follows the free-space loss mode, i.e., the communication between the UAV and the cluster-head unmanned surface vessel... The channel gain between is ,in It is a cluster-headed unmanned surface vessel Between with UAV in the 1st Frame communication distance, As a reference distance of 1m, the channel gain at the th... Frame, cluster-headed unmanned surface vessel The size of the unloading task to UAV is

[0089] (1)

[0090] in It is the channel bandwidth. It is in the Frame Cluster Head Unmanned Surface Vessel The transmit power of the offloading task. It is the received noise power of the UAV, at the... Frame, cluster-headed unmanned surface vessel The energy consumption of the unloading task is

[0091] (2)

[0092] set up For the maximum transmit power of the USV, there is , For the first Frame Cluster Head Unmanned Surface Vessel The maximum amount of data that can be unloaded from the UAV is [number]. . The relationship with Pmax is

[0093] (3)

[0094] Based on different mission characteristics and application scenarios, a fixed-wing UAV is used. During flight, the UAV provides unloading computation services to the USV, optimizing the UAV's flight trajectory. This method employs a classic aircraft power consumption model known in aerodynamic theory. Frames model the energy consumption of drone flight as

[0095] (4)

[0096] in For the UAV's current frame flight speed, there is... , The duration of each frame, and For two parameters related to UAV weight, wing area, and air density, in In extreme cases, the energy consumption of a fixed-wing drone is infinite, which confirms that a fixed-wing drone must maintain a minimum forward speed to stay airborne.

[0097] The method for establishing the computational model in step 1.2 is as follows: Dynamic Voltage Frequency Scaling (DVFS) technology is used to dynamically adjust the CPU frequency of each time slot USV and UAV for task computation, and it is assumed that USV can perform local computation and task offloading simultaneously.

[0098] (1) Computational model of cluster-headed unmanned surface vessel

[0099] In the Frame, cluster-headed unmanned surface vessel The local computing task data volume is

[0100] (5)

[0101] in, Computing cluster-head unmanned surface vessels for drones In the Frame CPU frequency, This is the highest CPU frequency for the drone. , To calculate a 1-bit cluster head unmanned surface vessel The number of CPU cycles required for the task, in the... Frame, cluster-headed unmanned surface vessel Local computing power consumption is

[0102] (6)

[0103] in Cluster-headed unmanned surface vessel The processor's effective switched capacitors;

[0104] (2) UAV computational model

[0105] Assuming the drone can simultaneously calculate the task of each cluster-head unmanned surface vessel, in the... Frame-based unmanned aerial vehicle (UAV) computing cluster head unmanned surface vessel The amount of data is

[0106] (7)

[0107] in, Computing cluster-head unmanned surface vessels for drones In the Frame CPU frequency, This is the highest CPU frequency for the drone. In the Frame-based unmanned aerial vehicle (UAV) computing cluster head unmanned surface vessel The energy consumption of the task is

[0108] (8)

[0109] in For the effective switched capacitor of the drone processor;

[0110] The modeling method for the joint trajectory and unloading problem in step 1.3 is as follows: the total energy consumption of the cluster-headed unmanned surface vessel consists of local computation energy consumption and task unloading energy consumption, while the total energy consumption of the UAV consists of flight energy consumption and computation energy consumption. In the The total power consumption of a frame is

[0111] (9)

[0112] Drones in The total power consumption of a frame is

[0113] (10)

[0114] The optimization objective is to minimize the weighted total energy consumption E of the cluster-head unmanned surface vessel and the unmanned aerial vehicle, expressed as:

[0115] (11)

[0116] in and These are weighting factors for the energy consumption of unmanned surface vessels and drones, respectively. Their values ​​can be preset. The drone flight trajectory is jointly optimized using Q={q[ Task data volume and CPU frequency The problem is described as follows

[0117] (12)

[0118] Among them, constraint (12)a guarantees that the UAV can only calculate the tasks received before the given time slot. Constraint (12)b guarantees that the UAV can completely calculate all the tasks unloaded by the cluster head UAV with a time delay of T. Constraint (12)c guarantees that all tasks of the cluster head UAV can be completely calculated with a time delay of T. Constraint (12)d guarantees that the UAV flies from the starting position to the ending position. Constraint (12)e guarantees that the UAV cannot exceed the maximum flight speed. Constraint (12)f guarantees that the cluster head UAV does not unload the tasks of the last time slot. Constraint (12)g guarantees that the UAV does not calculate the tasks of the first time slot. Constraint (12)h guarantees that the data volume of the tasks selected for unloading by the cluster head UAV in each frame is non-negative. Constraint (12)i guarantees that the cluster head UAV in the first time slot... The CPU's computation frequency in frame n does not exceed the maximum value. Constraint (12)j guarantees that the CPU's computation frequency in frame n does not exceed the maximum value. Constraint (12)k guarantees that the cluster-head unmanned surface vessel's computation frequency in frame n does not exceed the maximum value. The transmit power of the frame offloading task shall not exceed the maximum value.

[0119] Furthermore, the K-Means clustering algorithm in step 2.1 is as follows: Given A user nodes and M cluster heads, M nodes are randomly selected from the A user nodes as initial cluster heads. Let the cluster be... , each node Assign it to the cluster with the nearest cluster head, and determine the distance to the node. The cluster marker of the most recent cluster head is ,Will Assigned to the corresponding cluster Then, for each cluster, its optimal cluster head is recalculated. Update the cluster head, repeat the allocation and update steps until the cluster head no longer changes, and obtain the clustering result. };

[0120] The method for calculating the criticality of USVs in step 2.2 is as follows: In the scenario of maritime mission integration and offloading, computing and energy resources are extremely valuable. The selection of cluster heads needs to consider both communication coverage and the cost-effectiveness of transmission energy consumption caused by the size of mission data. During the task integration process within a cluster, if a USV with a small amount of data is selected as the cluster head, then large-volume tasks within the cluster need to be offloaded to the cluster head, which leads to energy waste. Therefore, USVs with larger amounts of data should be selected as cluster heads whenever possible. Thus, the size of the mission data volume needs to be an important basis for cluster head selection. In addition, considering mission latency constraints, timely processing of tasks with faster data generation rates is also crucial. Data generation rate should also be a necessary principle for cluster head selection. Based on the above factors, a USV criticality is proposed. As a new indicator for measuring the importance of a task The keyness of the i-th USV within cluster j is expressed as:

[0121] (13)

[0122] in Let $i$ be the average distance between the $i$-th USV and all other USVs in cluster $j$. Let i be the data volume of the i-th USV task. The data generation rate for the i-th USV task. Generate rate weights for the data.

[0123] The improved K-Means algorithm, which introduces the criticality of USVs and considers communication coverage and the number of cluster heads, proceeds as follows: First, the USVs are clustered using the K-Means algorithm. The USV with the highest criticality within a cluster, obtained from equation (13), is redefined as the cluster head. If the distance between the cluster head and each USV within the cluster is less than the communication distance, the clustering ends. Otherwise, the number of cluster heads is increased, and the above steps are repeated until all USVs meet the communication range requirements. The flowchart of the improved K-Means algorithm is shown in Table 1.

[0124] Table 1 Improved K-Means Algorithm

[0125]

[0126] Furthermore, the population initialization method based on Piecewise chaotic mapping in step 3.1 is as follows: The convergence and search accuracy of the sand cat swarm optimization algorithm are affected by the initial spatial distribution of the population. A uniformly distributed initial sand cat population can effectively improve the optimization efficiency, while a non-uniformly distributed population may reduce the algorithm's optimization efficiency. Piecewise chaotic mapping is chosen to initialize the sand cat population, giving it the characteristics of uniform distribution and wide traversal. The effectiveness of the improved strategy is demonstrated by comparing the results generated by Piecewise chaotic mapping population initialization and random population initialization. See the appendix for the effect diagrams of different initialization methods. Figure 3 ;

[0127] In step 3.2, the balance between searching for and attacking prey in the SCSO algorithm depends on the value of the balance parameter R. As shown in formula (6), the value of R depends on auditory sensitivity. The changes. And the big ones. It is beneficial for global search, small It is advantageous for local attacks, as is the case with the traditional SCSO algorithm. As the number of iterations decreases linearly from 2 to 0, its R value does not accurately reflect the search and attack process of the SCSO algorithm. The update formula is:

[0128] (14);

[0129] The spiral search strategy in step 3.3 is as follows: During the prey-hunting phase of the SCSO algorithm, the spiral search strategy of the whale algorithm is introduced for position updates. This allows individual sand cats to have more paths to find prey, thus achieving the goal of updating their positions. This process improves both the algorithm's local exploitation and global search capabilities, thereby enhancing its optimization performance. The update formula for the standard sand cat swarm algorithm during the prey-hunting phase is:

[0130] (15)

[0131] Add the spiral search strategy factor as shown in formula (16).

[0132] (16)

[0133] Where k is a constant;

[0134] After incorporating the spiral search factor, the formula for updating the optimal position of a sand cat individual becomes:

[0135] (17)

[0136] After incorporating a spiral search strategy into the position update expression during the search phase, the sand cat swarm will search in space in a spiral form. This not only enhances the sand cat swarm's ability to explore unknown areas but also increases the probability of the SCSO algorithm escaping local optima, effectively improving the optimization performance of the SCSO algorithm. In formula (16), l is a random parameter, and the value of l changes with the increase of the number of iterations. Its update formula is:

[0137] (18)

[0138] Based on the two improvement strategies given above, the traditional Sand Cat Group Algorithm (SCSO) is improved, and the Improved Sand Cat Group Algorithm (ISCSO) is constructed. Its specific implementation steps are as follows:

[0139] (1) Parameter initialization: Set the number of sand cats N, search dimension D, activity range [lb, ub] of each dimension, maximum number of iterations Tmax, and fitness function selection, etc.

[0140] (2) Sand cat position initialization: Piecewise chaotic mapping initialization method is adopted;

[0141] (3) Calculate the individual fitness value: Calculate the individual fitness value based on the initial individual position in steps (1) and (2) and the individual position updated in steps (5) and (6) as the number of iterations increases;

[0142] (4) Update parameters: Calculate the nonlinear factor according to formula (8). The value of is then updated, thereby updating the parameter r. The value of R;

[0143] (5) Conditional judgment: The next action of the sand cat group is determined based on the value of R. The size of |R| and 1 is judged, the position of the individual sand cat is updated, and then the size of the current iteration number t and T is judged. If t < T, then the parameters r are updated according to step (4). The value of R;

[0144] (6) Sand cat individual position update: When |R|>1, the sand cat individual position is updated according to formula (11); when |R|≤1, the sand cat individual position is updated according to formula (5).

[0145] (7) Algorithm iteration stop judgment: If the current iteration number t < the maximum iteration number T, then return to step (3) and then jump to steps (4), (5), and (6) in sequence until the algorithm reaches the maximum number of iterations and outputs the global optimal solution. For the ISCSO algorithm flowchart, please refer to the appendix. Figure 4 .

[0146] The main experimental design of this invention is described in detail below:

[0147] First, the experimental environment

[0148] All experiments in this invention were conducted under the same hardware and software environment to ensure the comparability of the results. The hardware environment included a computer equipped with an Intel Core i5-12400F @ 2.50GHz processor, 16GB of RAM, and an NVIDIA GeForce RTX 4060 dedicated graphics card (6GB of VRAM), running Windows 11 Professional 64-bit operating system. For the software environment, Python 3.10 was used as the programming language, with the primary scientific computing and visualization libraries being NumPy 1.26.4 and Matplotlib 3.8.2. The algorithm implementation and result plotting were completed within the PyCharm 2024.1 integrated development environment.

[0149] This invention compares the ISCSO algorithm with Particle Swarm Optimization (PSO), Grey Wolf Optimizer (GWO), Sparrow Search Algorithm (SSA), Bat Algorithm (BA), and Whale Optimization Algorithm (WOA) to verify the feasibility of the improved SCSO algorithm. This invention analyzes the performance of ISCSO by examining the impact of different weighting coefficients on the final unloading scheme's time and energy consumption, the algorithm's computation time consumption under different numbers of terminal devices, the relationship between different iteration numbers and the algorithm's fitness under different numbers of terminal devices, and the relationship between system cost and UAV cache capacity under different tolerable latency.

[0150] Second, experimental parameter settings

[0151] Table 2 Parameter List

[0152]

[0153] Third, simulation results and analysis

[0154] 1. Appendix Figure 5 This diagram illustrates the large-scale, widely distributed marine USVs in this system. One hundred USVs are randomly distributed within a designated area, with each USV having a communication coverage radius of 20m and a task data volume of 80Mbit. Pink dots represent USVs with a data generation rate of 20Kbit / s, blue dots represent USVs with a data generation rate of 40Kbit / s, and green dots represent USVs with a data generation rate of 100Kbit / s.

[0155] 2. From the appendix Figure 6 As can be seen, the improved K-Means algorithm divides USVs into 15 clusters, with red squares representing cluster head nodes. It is evident that all nodes within a cluster are within the communication range of the cluster head node, and the USV with the higher data generation rate has a relatively higher probability of being selected as the cluster head node.

[0156] 3. Appendix Figure 7 The relationship between mission completion time T and UAV flight trajectory is illustrated, with K=4, and the coordinates of the UAV's starting and ending points set to (-30, -30) and (0, -30), respectively. It can be seen that the larger T is, the more time the UAV has to fly close to the cluster head USV to provide unloading computation services, thus reducing the energy consumption of the cluster head USV unloading mission. However, when T is small, due to flight speed limitations, the UAV's flight path length must be shortened, sacrificing mission unloading energy consumption to meet latency requirements.

[0157] 4. Appendix Figures 8-11 The relationship between the weighted energy consumption E of different algorithm systems and the data volume L of the cluster head USV task is shown. It can be seen that E increases with the increase of L, because the more data there is, the more energy is consumed for offloading or computation. Compared with other algorithms, the ISCSO algorithm proposed in this paper has the lowest energy consumption, and its superiority is more obvious when the task data volume is large.

Claims

1. A method for offloading mobile edge tasks using a sandcat swarm intelligence collaborative computing strategy, characterized in that, The method includes the following steps: First, the construction of the UAV-assisted MEC system model: 1.1 Establish a communication model; Step 1.2: Establish a computational model; Section 1.3, Modeling the Joint Trajectory and Unloading Problem; Second, based on the improved K-Means unmanned surface vessel clustering strategy; Section 2.1, K-Means clustering algorithm; Section 2.2, USV criticality calculation; Third, joint trajectory optimization and task unloading are performed based on an improved sand cat swarm optimization algorithm: Section 3.1 Population initialization based on Piecewise chaotic mapping; 3.2 Nonlinear adjustment mechanism; 3.3 Spiral Search Strategy.

2. The mobile edge task offloading method of the sand cat swarm intelligence collaborative computing strategy as described in claim 1, characterized in that, The method for establishing the communication model in step 1.1 is as follows: It is assumed that the communication link between the UAV and USV is controlled by a line-of-sight link, the Doppler frequency shift in communication can be compensated by the receiver, the channel quality depends on the line-of-sight link, and the channel gain follows the free-space loss mode, i.e., the communication between the UAV and the cluster-head unmanned surface vessel... The channel gain between them is ,in It is a cluster-headed unmanned surface vessel Between with UAV in the 1st Frame communication distance, As a reference distance of 1m, the channel gain at the th... Frame, cluster-headed unmanned surface vessel The size of the unloading task to UAV is (1) in It is the channel bandwidth. It is in the Frame Cluster Head Unmanned Surface Vessel The transmit power of the offloading task. It is the received noise power of the UAV, at the... Frame, cluster-headed unmanned surface vessel The energy consumption of the unloading task is (2) set up For the maximum transmit power of the USV, there is , For the first Frame Cluster Head Unmanned Surface Vessel The maximum amount of data that can be unloaded from the UAV is [number]. , The relationship with Pmax is (3) Based on different mission characteristics and application scenarios, fixed-wing UAVs are used. During flight, the UAVs provide offloading computation services to the USV, optimizing the UAV's flight trajectory. Frames model the energy consumption of drone flight as (4) in For the current frame's flight speed of the UAV, there is , The duration of each frame, and For two parameters related to UAV weight, wing area, and air density, in In extreme cases, the energy consumption of a fixed-wing drone is infinite, which confirms that a fixed-wing drone must maintain a minimum forward speed to stay airborne. The method for establishing the computational model in step 1.2 is as follows: Dynamic Voltage Frequency Scaling (DVFS) technology is used to dynamically adjust the CPU frequency of each time slot USV and UAV for task computation, and it is assumed that USV can perform local computation and task offloading simultaneously. (1) Computational model of cluster-headed unmanned surface vessel In the Frame, cluster-headed unmanned surface vessel The local computing task data volume is (5) in, Computing cluster-head unmanned surface vessels for drones In the Frame CPU frequency, The highest CPU frequency for drones, , To calculate a 1-bit cluster head unmanned surface vessel The number of CPU cycles required for the task, in the first... Frame, cluster-headed unmanned surface vessel Local computing power consumption is (6) in Cluster-headed unmanned surface vessel The processor's effective switched capacitors; (2) UAV computational model Assuming the drone can simultaneously calculate the task of each cluster-head unmanned surface vessel, in the... Frame-based unmanned aerial vehicle (UAV) computing cluster head unmanned surface vessel The amount of data is (7) in, Computing cluster-head unmanned surface vessels for drones In the Frame CPU frequency, The highest CPU frequency for drones, In the Frame-based unmanned aerial vehicle (UAV) computing cluster head unmanned surface vessel The energy consumption of the task is (8) in For the effective switched capacitor of the drone processor; The method for modeling the joint trajectory and unloading problem in step 1.3 is as follows: the total energy consumption of the cluster-headed unmanned surface vessel consists of local computation energy consumption and task unloading energy consumption, while the total energy consumption of the UAV consists of flight energy consumption and computation energy consumption. In the The total power consumption of a frame is (9) Drones in The total power consumption of a frame is (10) The optimization objective is to minimize the weighted total energy consumption E of the cluster-head unmanned surface vessel and the unmanned aerial vehicle, expressed as: (11) in and These are weighting factors for the energy consumption of unmanned surface vessels and drones, respectively. Their values ​​can be preset. The drone flight trajectory is jointly optimized using Q={q[ Task data volume and CPU frequency The problem is described as follows (12) Among them, constraint (12)a guarantees that the UAV can only calculate the tasks received before the given time slot. Constraint (12)b guarantees that the UAV can completely calculate all the tasks unloaded by the cluster head UAV with a time delay of T. Constraint (12)c guarantees that all tasks of the cluster head UAV can be completely calculated with a time delay of T. Constraint (12)d guarantees that the UAV flies from the starting position to the ending position. Constraint (12)e guarantees that the UAV cannot exceed the maximum flight speed. Constraint (12)f guarantees that the cluster head UAV does not unload the tasks of the last time slot. Constraint (12)g guarantees that the UAV does not calculate the tasks of the first time slot. Constraint (12)h guarantees that the data volume of the tasks selected for unloading by the cluster head UAV in each frame is non-negative. Constraint (12)i guarantees that the cluster head UAV in the first time slot... The CPU's computation frequency in frame n does not exceed the maximum value. Constraint (12)j guarantees that the CPU's computation frequency in frame n does not exceed the maximum value. Constraint (12)k guarantees that the cluster-head unmanned surface vessel's computation frequency in frame n does not exceed the maximum value. The transmit power of the frame offloading task shall not exceed the maximum value.

3. The mobile edge task offloading method of the sand cat swarm intelligence collaborative computing strategy as described in claim 1, characterized in that, The K-Means clustering algorithm in step 2.1 is as follows: Given A user nodes and M cluster heads, M nodes are randomly selected from the A user nodes as initial cluster heads. Let the cluster be... , each node Assign it to the cluster with the nearest cluster head, and determine the distance to the node. The cluster marker of the most recent cluster head is ,Will Assigned to the corresponding cluster Then, for each cluster, its optimal cluster head is recalculated. Update the cluster head, repeat the allocation and update steps until the cluster head no longer changes, and obtain the clustering result. }; The method for calculating USV criticality in step 2.2 is as follows: In the scenario of maritime mission integration and offloading, computing and energy resources are extremely valuable. The selection of cluster heads needs to consider both communication coverage and the cost-effectiveness of transmission energy consumption caused by the size of mission data. During the task integration process within a cluster, if a USV with a small amount of data is selected as the cluster head, then large-volume tasks within the cluster need to be offloaded to the cluster head, which leads to energy waste. Therefore, USVs with larger amounts of data should be selected as cluster heads whenever possible. Thus, the size of the mission data volume needs to be an important basis for cluster head selection. In addition, considering mission latency constraints, timely processing of tasks with faster data generation rates is also crucial. Data generation rate should also be a necessary principle for cluster head selection. Based on the above factors, a USV criticality is proposed. As a new indicator for measuring the importance of a task The keyness of the i-th USV within cluster j is expressed as: (13) in Let $i$ be the average distance between the $i$-th USV and all other USVs in cluster $j$. Let i be the data volume of the i-th USV task. The data generation rate for the i-th USV task. Generate rate weights for the data.

4. The mobile edge task offloading method of the sand cat swarm intelligence collaborative computing strategy as described in claim 1, characterized in that, In step 3.1, the population initialization method based on Piecewise chaotic mapping is as follows: The convergence and search accuracy of the sand cat swarm algorithm are affected by the initial spatial distribution of the population. This makes it possible to effectively improve the optimization efficiency when the initial sand cat population is evenly distributed in a region, while the opposite may reduce the optimization efficiency of the algorithm. Piecewise chaotic mapping is selected to initialize the sand cat population, so that it has the characteristics of even distribution and wide traversal. The effectiveness of the improvement strategy is demonstrated by comparing the results generated by Piecewise chaotic mapping population initialization and random population initialization. The nonlinear adjustment mechanism in step 3.2 is as follows: the balance between searching for and attacking prey in the SCSO algorithm depends on the value of the balance parameter R. As shown in formula (6), the value of R depends on auditory sensitivity. The changes. And the big ones. It is beneficial for global search, small It is advantageous for local attacks, as is the case with the traditional SCSO algorithm. As the number of iterations decreases linearly from 2 to 0, its R value does not accurately reflect the search and attack process of the SCSO algorithm. The update formula is: (14)； The spiral search strategy in step 3.3 is as follows: During the prey-hunting phase of the SCSO algorithm, the spiral search strategy of the whale algorithm is introduced for position updates. This allows individual sand cats to have more paths to find prey, thus achieving the goal of updating their positions. This process improves both the algorithm's local exploitation and global search capabilities, thereby enhancing its optimization performance. The update formula for the standard sand cat swarm algorithm during the prey-hunting phase is: (15) Add the spiral search strategy factor as shown in formula (10). (16) Where k is a constant; After incorporating the spiral search factor, the formula for updating the optimal position of a sand cat individual becomes: (17) After incorporating a spiral search strategy into the position update expression during the search phase, the sand cat swarm will search in space in a spiral form. This not only enhances the sand cat swarm's ability to explore unknown areas but also increases the probability of the SCSO algorithm escaping local optima, effectively improving the optimization performance of the SCSO algorithm. In formula (16), l is a random parameter, and the value of l changes with the increase of the number of iterations. Its update formula is: (18)。

Images