Unmanned aerial vehicle intelligent unloading strategy based on lyapunov combined with potential game algorithm

Abstract

The application discloses an unmanned aerial vehicle intelligent unloading strategy based on a Lyapunov combined Potential Game algorithm. First, a heterogeneous computing architecture composed of mobile users, unmanned aerial vehicles and edge servers is established. Second, a LyMARLPG algorithm based on the Lyapunov combined Potential Game is proposed for the unmanned aerial vehicle edge intelligent unloading strategy. Finally, the intelligent unloading strategy problem is solved based on the LyMARLPG algorithm, so that the unmanned aerial vehicle intelligent unloading strategy is obtained.

Description

Technical Field

[0001] This invention relates to the problem of intelligent offloading strategies for applications in edge network environments of mobile devices and drones, and specifically to an intelligent offloading strategy for drones based on Lyapunov combined with the potential game algorithm. Background Technology

[0002] As the Internet of Things (IoT) becomes deeply integrated into digital life, the interconnection of massive numbers of terminal devices has spurred an explosive growth in demand for computationally intensive and low-latency-sensitive applications. To meet these stringent application requirements, it is essential to focus on reducing communication latency, optimizing energy consumption, and balancing load pressure—these core indicators are not only hot topics in academic research but also critical bottlenecks that industry applications urgently need to overcome. Although the impact of each factor on system performance varies, the degree of optimization of their synergistic effect directly determines the balance between service quality and resource utilization within the IoT ecosystem.

[0003] Communication latency, a core indicator of Quality of Service (QoS), directly impacts the real-time performance and reliability of critical applications such as intelligent traffic scheduling and industrial automation control. Mobile edge computing (MEC) effectively shortens data transmission paths and significantly reduces latency by offloading computing resources to nodes near terminals, such as base stations and edge servers, greatly improving system response efficiency. However, this architectural innovation also exposes a complex contradiction between wireless channel quality and task processing efficiency: unstable wireless connections lead to data retransmissions and a surge in energy consumption, while prioritizing high-quality channels, although reducing transmission costs, may exacerbate task queue backlogs and exceed latency tolerance limits. Meanwhile, the limited computing resources of edge servers are prone to queuing delays when faced with sudden, computationally intensive tasks, and factors such as dynamic changes in device location and network congestion further increase the uncertainty of network quality. Furthermore, resource competition among multiple devices intensifies the difficulty of coordinating interests; how to achieve efficient allocation of global resources while ensuring individual needs has become a key challenge restricting the performance improvement of MEC systems.

[0004] Against this backdrop, in-depth research into how to optimize mobile edge computing systems in complex and ever-changing environments, balancing factors such as communication latency, energy consumption, and load pressure, is particularly urgent and necessary. This is not only related to the further development and application expansion of IoT technology, but also has significant practical implications for promoting the digital transformation and intelligent upgrading of various industries. Summary of the Invention

[0005] This invention addresses the problem of intelligent offloading strategies for applications in edge network environments of mobile devices and drones, and provides an intelligent offloading strategy for drones based on Lyapunov combined with the potential game algorithm.

[0006] This invention is achieved using the following technical solution:

[0007] A drone intelligent unloading strategy based on Lyapunov combined with the potential game algorithm includes the following steps:

[0008] 1) Establish a heterogeneous computing architecture consisting of mobile users, drones, and edge servers;

[0009] 2) A LyMARLPG algorithm based on Lyapunov and potential game is proposed for intelligent offloading strategies at the edge of UAVs.

[0010] 3) Obtain the intelligent unloading strategy for drones.

[0011] In the above technical solution, further, the heterogeneous computing architecture consisting of mobile users, drones, and edge servers described in step 1) comprises three parts:

[0012] (1) Mobile user layer: Mobile users generate service requests and communicate with nearby drones to obtain network connectivity;

[0013] (2) Unmanned aerial vehicle layer: flying in the network, collecting tasks, and providing computing and storage resources;

[0014] (3) Edge server layer: provides massive computing and storage resources, can handle complex tasks, and coordinates the allocation of tasks and resources through communication with the drone layer;

[0015] Furthermore, the LyMARLPG algorithm for the intelligent unloading strategy described in step 2) is as follows:

[0016] The intelligent offloading strategy is based on the information of task ψ and the game principle between UAVs and edge servers. It aims to minimize the time and energy consumption of UAVs in processing tasks and maximize the benefits of each UAV participating in the game, while maintaining the stability of the system. The proposed LyMARLPG algorithm uses data such as channel gain between UAVs and edge servers, task queue status of UAVs, computing resources of UAVs, and interference between UAVs as inputs, and generates an intelligent offloading scheme for UAVs as output.

[0017] Furthermore, the objective function of the UAV intelligent unloading strategy based on Lyapunov combined with the potential game algorithm described in step 3) is:

[0018] The problem of minimizing latency and energy consumption during drone mission execution can be expressed as:

[0019]

[0020] in,

[0021] 1) A penalty is added to Lyapunov drift, where V is the Lyapunov penalty coefficient. It is the total cost for drone i to complete the mission, consisting of the weighted sum of latency and energy consumption. The calculation formula is as follows:

[0022]

[0023] φ{·} is an indicator function; φ{·} = 1 when the condition within the parentheses is met, and 0 otherwise. n is the number of edge servers. Here, m represents the number of connectable edge servers. This indicates a connectable drone. When the tasks collected by the drone are processed locally, d t =0, when the drone offloads the mission to ESS processing, d t =n, when the drone offloads the task to other drones for processing via V2V communication, d t =m.

[0024] 2) When the drone collects task ψ, the task needs to be processed promptly. This indicates the energy consumption for the drone to process tasks locally. If the drone cannot handle the task, it can be offloaded to an adjacent drone or edge server (ESS), meaning task ψ is executed on an adjacent drone or ESS. The energy consumption for the drone to upload mission ψ to the ESS for execution. This represents the energy consumption required for a drone to upload task ψ to a neighboring drone for execution. The formula for calculating this energy consumption is as follows:

[0025]

[0026]

[0027]

[0028] Where k is a coefficient related to the drone architecture. For the computing power of drone m, c t This indicates the number of CPU cycles required to complete the task. It is the transmission power of the drone. It refers to the transmission latency of the drone unloading the mission to the ESS. It is the transmission delay of drone m unloading the task to other drones. This refers to the latency of processing tasks received from other drones for unloading. The formula for calculating the above transmission latency is as follows:

[0029]

[0030]

[0031] Among them, S t The size of the task collected by the UAV in time slot t. It refers to the transmission rate between drones. This refers to the transmission rate between the drone and the edge server.

[0032] L t This refers to the latency of the drone processing tasks according to the offloading strategy. The processing latency includes the drone's local processing latency, task transmission latency, and the processing latency of the drone object or edge server after offloading. The calculation formula is as follows:

[0033]

[0034]

[0035]

[0036]

[0037] in It is the latency of task ψ being processed locally on the drone. This refers to the latency after the drone offloads the task to the edge server for processing. s For the computing power of edge servers, This represents the processing delay of a drone receiving unloading tasks from other drones. To ensure a balance in unloading benefits among all parties, i.e., that the unloading strategies of each drone do not affect each other's benefits, a Potential Game is used to solve for the Nash equilibrium among the drones. The latent function introduced for this purpose is:

[0038]

[0039] Where O(d) m ,d -m ) indicates that the drone m adopts strategy d m Other drones employ strategy d -m Based on the objective function (1), the utility function of the UAV can be obtained as follows:

[0040]

[0041] As defined by the potential game definition, when the potential function changes with the change in utility, the drone game is a potential game, and there must exist a Nash equilibrium point, which is the optimal unloading strategy for the drone. Meanwhile, the change in the utility function is as follows:

[0042]

[0043] It meets the definition and properties of a Potential Game, thus enabling the implementation of an unloading strategy that balances the benefits for all parties while minimizing latency and energy consumption.

[0044] The present invention also provides a process for intelligent uninstallation using the above method, as detailed below:

[0045] 1) Lyapunov drift penalty: Introducing a virtual queue Q m (t) is used to simulate the actual queue length q. m (t) and threshold ε m The deviation is minimized by minimizing Lyapunov drift plus penalty to ensure the stability of the task queue while minimizing latency and energy consumption;

[0046] 2) Potential Game: Introducing a potential function, based on the definition of EPG, to ensure that there is a Nash equilibrium point in the potential game between drones and between drones and edge servers, thus ensuring the balance of benefits for all parties;

[0047] 3) Repeat steps 1) and 2) until a task offloading strategy with minimal latency and energy consumption, stable task queue, and balanced benefits for all parties is obtained.

[0048] The inventive principle of this invention:

[0049] This invention primarily addresses the intelligent offloading strategy problem in edge network environments for mobile devices and drones. It designs an intelligent offloading strategy aimed at reducing the waiting latency and energy consumption of mobile devices in a two-layer IoT architecture, while ensuring the stability of the task queue and the balance of benefits among drones. Lyapunov optimization theory and Potential Game theory are introduced to effectively and rationally determine the offloading strategy and promote cooperation among drones. Based on this, the LyMARLPG algorithm is proposed. This algorithm innovatively proposes an intelligent drone offloading strategy based on Lyapunov combined with the Potential Game algorithm, according to the task queue status and the benefit function among drones.

[0050] The beneficial effects of this invention are as follows:

[0051] This invention proposes for the first time an intelligent unloading strategy for drones based on Lyapunov and the Potential Game algorithm. This method can effectively help drones execute unloading strategies, aiming to reduce the waiting latency and energy consumption of mobile devices in a two-layer IoT architecture. By introducing Lyapunov optimization theory and Potential Game theory, the stability of the task queue and the balance of benefits among drones are guaranteed. Based on this, the LyMARLPG algorithm is proposed. This algorithm innovatively obtains the optimal intelligent unloading strategy based on the status of the task queue and the benefit function among drones. Attached Figure Description

[0052] Figure 1 A heterogeneous computing architecture consisting of mobile users, drones, and edge servers;

[0053] Figure 2 Here is a flowchart example of the LyMARLPG algorithm;

[0054] Figure 3 To understand the impact of integrating Potential Game on the efficiency of drones;

[0055] Figure 4 The impact of different environmental conditions on algorithm performance and a performance comparison with four other algorithms;

[0056] Figure 5 The impact of different environmental conditions on the latency and energy consumption of UAV processing tasks, and a performance comparison with four other algorithms.

[0057] Figure 6 The impact of different factors on the task queue; Detailed Implementation

[0058] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0059] A drone intelligent unloading strategy based on Lyapunov combined with the potential game algorithm includes the following steps:

[0060] 1) Establish a heterogeneous computing architecture consisting of mobile users, drones, and edge servers;

[0061] 2) A LyMARLPG algorithm based on Lyapunov and potential game is proposed for intelligent offloading strategies at the edge of UAVs.

[0062] 3) Obtain the intelligent unloading strategy for drones.

[0063] The heterogeneous computing architecture proposed in this invention, consisting of mobile users, drones, and edge servers, is as follows: Figure 1 As shown, it mainly includes three parts:

[0064] (1) Mobile user layer: Mobile users generate service requests and communicate with nearby drones to obtain network connectivity;

[0065] (2) Unmanned aerial vehicle layer: flying in the network, collecting tasks, and providing computing and storage resources;

[0066] (3) Edge server layer: provides massive computing and storage resources, can handle complex tasks, and coordinates the allocation of tasks and resources through communication with the drone layer;

[0067] Figure 2 The process of the LyMARLPG algorithm is given as an example. First, the interference information between UAVs, the task queue status, the computing power of the UAVs, and the task information are taken as input. The greedy algorithm is used to solve for a suitable offloading strategy. Then, the overall benefit is solved by Lyapunov drift plus penalty. It is judged whether the change trend of the latent function meets the termination condition. The network parameters are updated according to the experience replay pool. The process is iterated continuously and finally the optimal intelligent offloading strategy is obtained.

[0068] Figure 3 The impact of incorporating the Potential Game on the efficiency of drones is presented. As the potential function increases, the overall average cost of drones decreases, which means that as the potential function increases, the latency and energy consumption of drones in processing tasks also decrease synchronously. Finally, a stable state is reached, indicating that the efficiency of drones has reached an equilibrium.

[0069] The problem of minimizing latency and energy consumption during drone mission execution can be expressed as:

[0070]

[0071] in,

[0072] 3) A penalty is added to Lyapunov drift, where V is the Lyapunov penalty coefficient. It is the total cost for drone i to complete the mission, consisting of the weighted sum of latency and energy consumption. The calculation formula is as follows:

[0073]

[0074]

[0075]

[0076] φ{·} is an indicator function; φ{·} = 1 when the condition within the parentheses is met, and 0 otherwise. n is the number of edge servers. Here, m represents the number of connectable edge servers. This indicates a connectable drone. When the tasks collected by the drone are processed locally, d t =0, when the drone offloads the mission to ESS processing, d t =n, when the drone offloads the task to other drones for processing via V2V communication, d t =m.

[0077] 4) When the drone collects task ψ, the task must be processed promptly. This indicates the energy consumption for the drone to process tasks locally. If the drone cannot handle the task, it can be offloaded to an adjacent drone or edge server (ESS), meaning task ψ is executed on an adjacent drone or ESS. The energy consumption for the drone to upload mission ψ to the ESS for execution. This represents the energy consumption required for a drone to upload task ψ to a neighboring drone for execution. The formula for calculating this energy consumption is as follows:

[0078]

[0079]

[0080]

[0081] Where k is a coefficient related to the drone architecture. For the computing power of drone m, c t This indicates the number of CPU cycles required to complete the task. It is the transmission power of the drone. It refers to the transmission latency of the drone unloading the mission to the ESS. It is the transmission delay of drone m unloading the task to other drones. This refers to the latency of processing tasks received from other drones for unloading. The formula for calculating the above transmission latency is as follows:

[0082]

[0083]

[0084] Among them, S t The size of the task collected by the UAV in time slot t. It refers to the transmission rate between drones. This refers to the transmission rate between the drone and the edge server.

[0085] L t This refers to the latency of the drone processing tasks according to the offloading strategy. The processing latency includes the drone's local processing latency, task transmission latency, and the processing latency of the drone object or edge server after offloading. The calculation formula is as follows:

[0086]

[0087]

[0088]

[0089]

[0090] in It is the latency of task ψ being processed locally on the drone. This refers to the latency after the drone offloads the task to the edge server for processing. s For the computing power of edge servers, This refers to the delay in drone processing tasks that receive unloading tasks from other drones.

[0091] To ensure that the unloading benefits of each party are balanced, i.e., that the unloading strategies of each drone do not affect each other's benefits, a Potential Game is used to solve for the Nash equilibrium among the drones. The latent function introduced for this purpose is:

[0092]

[0093] Where O(d) m ,d -m ) indicates that the drone m adopts strategy d m Other drones employ strategy d -m Based on the objective function (1), the utility function of the UAV can be obtained as follows:

[0094]

[0095] As defined by the potential game definition, when the potential function changes with the change in utility, the drone game is a potential game, and there must exist a Nash equilibrium point, which is the optimal unloading strategy for the drone. Meanwhile, the change in the utility function is as follows:

[0096]

[0097] It meets the definition and properties of a Potential Game, thus enabling the implementation of an unloading strategy that balances the benefits for all parties while minimizing latency and energy consumption.

[0098] The present invention also provides a process for intelligent uninstallation using the above method, as detailed below:

[0099] 1) Lyapunov drift penalty: Introducing a virtual queue Q m (t) is used to simulate the actual queue length q. m (t) and threshold ε m The deviation is minimized by minimizing Lyapunov drift plus penalty to ensure the stability of the task queue while minimizing latency and energy consumption;

[0100] 2) Potential Game: Introducing a potential function, based on the definition of EPG, to ensure that there is a Nash equilibrium point in the potential game between drones and between drones and edge servers, thus ensuring the balance of benefits for all parties;

[0101] 3) Repeat steps 1) and 2) until a task offloading strategy with minimal latency and energy consumption, stable task queue, and balanced benefits for all parties is obtained.

[0102] 4) Simulation Results

[0103] The method of this invention (LyMARLPG) is compared with three different conventional algorithms: Local, Edge, and Random, as well as the LyMARLNPG algorithm without incorporating Potential Game. Specifically, the comparison criteria are latency, energy consumption, and overall average efficiency across drones.

[0104] Figure 4 and Figure 5 The comparison shows that the grouping basis proposed in this invention is effective, and the optimization effect of SCCS is significantly better than the other three algorithms.

[0105] Figure 4 The comparison shows that the intelligent unloading algorithm proposed in this invention is effective. Under the influence of different factors, the overall average cost of the UAV tends to decrease and gradually converges. At the same time, the LyMARLPG algorithm is significantly better than the other four algorithms.

[0106] Figure 5 The comparison shows that the LyMARLPG algorithm achieves superior optimization in terms of average latency, energy consumption, and overall average cost.

[0107] Figure 6The comparison shows that the LyMARLPG algorithm can maintain the stability of the task queue while optimizing average latency, energy consumption, and overall average cost. This is attributed to the LyMARLPG algorithm's innovative proposal of an intelligent drone unloading strategy based on Lyapunov combined with the potential game algorithm, which effectively ensures system stability and a balance of benefits among drones, according to the status of the task queue and the benefit function among drones.

Claims

1. A drone intelligent unloading strategy based on Lyapunov combined with the potential game algorithm, characterized in that, Includes the following steps: 1) Establish a heterogeneous computing architecture consisting of mobile users, drones, and edge servers; 2) A LyMARLPG algorithm based on Lyapunov and potential game is proposed for intelligent offloading strategies at the edge of UAVs. 3) Obtain the intelligent unloading strategy for drones.

2. The intelligent unloading strategy for drones based on Lyapunov combined with the potential game algorithm as described in claim 1, characterized in that, The heterogeneous computing architecture consisting of mobile users, drones, and edge servers described in step 1) comprises three parts: 1) Mobile user layer: Mobile users generate service requests and communicate with nearby drones to obtain network connectivity; 2) Unmanned Aerial Vehicle (UAV) Layer: Flies within the network, collects task data, and provides computing and storage resources; 3) Edge server layer: Provides massive computing and storage resources, can handle complex tasks, and coordinates task and resource allocation through communication with the drone layer.

3. The intelligent unloading strategy for drones based on Lyapunov combined with the potential game algorithm as described in claim 2, characterized in that, The LyMARLPG algorithm for the edge intelligent offloading strategy of UAVs described in step 2) is as follows: The intelligent offloading strategy is based on the information of task ψ and the game principle between UAVs and edge servers. It aims to minimize the time and energy consumption of UAVs in processing tasks and maximize the benefits of each UAV participating in the game, while maintaining the stability of the system. The proposed LyMARLPG algorithm uses data such as channel gain between UAVs and edge servers, task queue status of UAVs, computing resources of UAVs, and interference between UAVs as inputs, and generates an intelligent offloading scheme for UAVs as output.

4. The UAV edge intelligent unloading strategy as described in claim 3, characterized in that, The objective function of the UAV intelligent unloading strategy based on Lyapunov combined with the potential game algorithm described in step 3) is: The problem of minimizing latency and energy consumption during drone mission execution can be expressed as: in, 1) A penalty is added to Lyapunov drift, where V is the Lyapunov penalty coefficient. It is the total cost for drone i to complete the mission, consisting of the weighted sum of latency and energy consumption. The calculation formula is as follows: φ{·} is an indicator function; φ{·} = 1 when the condition within the parentheses is met, and 0 otherwise. n is the number of edge servers. Here, m represents the number of connectable edge servers. This indicates a connectable drone. When the tasks collected by the drone are processed locally, d t =0, when the drone offloads the mission to ESS processing, d t =n, when the drone offloads the task to other drones for processing via V2V communication, d t =m. 2) When the drone collects task ψ, the task needs to be processed promptly. This indicates the energy consumption for the drone to process tasks locally. If the drone cannot handle the task, it can be offloaded to an adjacent drone or edge server (ESS), meaning task ψ is executed on an adjacent drone or ESS. The energy consumption for the drone to upload mission ψ to the ESS for execution. This represents the energy consumption required for a drone to upload task ψ to a neighboring drone for execution. The formula for calculating this energy consumption is as follows: Where k is a coefficient related to the drone architecture. For the computing power of drone m, c t This indicates the number of CPU cycles required to complete the task. It is the transmission power of the drone. It refers to the transmission latency of the drone unloading the mission to the ESS. It is the transmission delay of drone m unloading the task to other drones. This refers to the latency of processing tasks received from other drones for unloading. The formula for calculating the above transmission latency is as follows: Among them, S t The size of the task collected by the UAV in time slot t. It refers to the transmission rate between drones. This refers to the transmission rate between the drone and the edge server. L t This refers to the latency of the drone processing tasks according to the offloading strategy. The processing latency includes the drone's local processing latency, task transmission latency, and the processing latency of the drone object or edge server after offloading. The calculation formula is as follows: in It is the latency of task ψ being processed locally on the drone. This refers to the latency after the drone offloads the task to the edge server for processing. s For the computing power of edge servers, This refers to the delay in drone processing tasks that receive unloading tasks from other drones. To ensure that the unloading benefits of each party are balanced, i.e., that the unloading strategies of each drone do not affect each other's benefits, a Potential Game is used to solve for the Nash equilibrium among the drones. The latent function introduced for this purpose is: Where O(d) m ,d -m ) indicates that the drone m adopts strategy d m Other drones employ strategy d -m Based on the objective function (1), the utility function of the UAV can be obtained as follows: As defined by the potential game definition, when the potential function changes with the change in utility, the drone game is a potential game, and there must exist a Nash equilibrium point, which is the optimal unloading strategy for the drone. Meanwhile, the change in the utility function is as follows: It meets the definition and properties of a Potential Game, thus enabling the implementation of an unloading strategy that balances the benefits for all parties while minimizing latency and energy consumption.

5. The intelligent unloading strategy for drones based on Lyapunov combined with the potential game algorithm as described in any one of claims 1-4, characterized in that, The process of unloading tasks based on this strategy is as follows: 1) Lyapunov drift penalty: Introducing a virtual queue Q m (t) is used to simulate the actual queue length q. m (t) and threshold ε m The deviation is minimized by minimizing Lyapunov drift plus penalty to ensure the stability of the task queue while minimizing latency and energy consumption. 2) Potential Game: Introduce a potential function. According to the definition of EPG, ensure that there is a Nash equilibrium point in the potential game between drones and between drones and edge servers, and ensure the balance of benefits for all parties. 3) Repeat steps 1) and 2) until a task offloading strategy with minimal latency and energy consumption, stable task queue, and balanced benefits for all parties is obtained.

Images