Unmanned aerial vehicle intelligent service combination caching strategy based on density peak clustering algorithm

Abstract

The application discloses a UAV edge computing intelligent service combination caching strategy based on a density peak clustering algorithm. Firstly, a heterogeneous computing architecture composed of vehicles, UAVs and cloud servers based on software defined network technology is constructed. Secondly, a density peak clustering algorithm for the intelligent service combination caching strategy is proposed. Finally, the intelligent service combination caching strategy problem is solved based on the SCCS algorithm, so that the best service combination caching strategy is obtained.

Description

Technical Field

[0001] This invention relates to the problem of service composition caching strategies for applications in vehicle and drone edge network environments, and specifically to an intelligent service composition caching strategy for drone edge computing based on density peak clustering algorithm. Background Technology

[0002] As Vehicle Edge Networks (VENs) are integrated into modern transportation systems, intelligent vehicle systems are tasked with processing massive amounts of data to enable advanced features such as autonomous driving, real-time navigation, and environmental perception. Meeting the demands of these demanding applications requires meticulous focus on minimizing service latency, reducing transmission costs, and conserving caching expenses—key factors in both academic research and industry applications. While each parameter has a different impact on system performance, their synergistic effect is crucial for achieving the optimal balance between efficiency and economic viability within the VENs ecosystem.

[0003] Service latency is particularly critical, as it directly impacts Quality of Service (QoS), affecting the responsiveness of safety assistance features, the accuracy of navigation services, and the timeliness of vehicle data communication. Strategically placing computing resources at the network edge can significantly reduce service latency, thereby improving the immediacy and reliability of vehicle operations. However, this configuration also highlights the complex trade-off between data transmission costs and resource caching; the overhead associated with data routing between drones requires sophisticated optimization schemes to reduce transmission costs. Meanwhile, caching within the drone is crucial for reducing latency and avoiding redundant data processing. However, with the escalating service requests and the ever-expanding volume of data generated, drone caching devices are under increasing pressure. Therefore, skillful coordination of cache management is essential to ensure the effective allocation and utilization of these limited resources, thereby driving the advancement of intelligent in-vehicle systems. Summary of the Invention

[0004] This invention addresses the service composition caching strategy problem in vehicle and drone edge network environments by providing an intelligent service composition caching strategy for drone edge computing based on the density peak clustering algorithm.

[0005] This invention is achieved using the following technical solution:

[0006] A caching strategy for intelligent edge computing services for drones based on density peak clustering algorithm includes the following steps:

[0007] 1) Establish a heterogeneous computing architecture based on software-defined networking technology, consisting of vehicles, drones, and cloud servers;

[0008] 2) Propose density peak clustering algorithm and SCCS algorithm for intelligent service composition caching strategy;

[0009] 3) Obtain the intelligent service combination caching strategy.

[0010] In the above technical solution, the heterogeneous computing architecture consisting of vehicles, drones, and cloud servers described in step 1) further comprises four parts:

[0011] (1) Vehicle layer: including vehicles and onboard systems, vehicles generate service requests and communicate with nearby drones to obtain network connectivity;

[0012] (2) Unmanned Aerial Vehicle (UAV) Layer: Provides computing and storage resources at the network edge;

[0013] (3) Cloud layer: provides massive computing and storage resources, can handle complex processing tasks, and coordinates resource allocation and service delivery through communication with the unmanned aerial vehicle layer;

[0014] (4) SDN controller: It can obtain vehicle motion information and service requirements, such as vehicle location, speed, driving direction and request frequency. This information can be used to adjust caching strategies.

[0015] Furthermore, the density peak clustering algorithm for the intelligent service composition caching strategy described in step 2) is defined as follows:

[0016] The Intelligent Service Composition Caching Strategy (SCCS) is based on the principles of density region partitioning of task requests R, UAV grouping, and storage space partitioning. It aims to optimize the utilization of UAV cache resources. The SCCS algorithm is proposed, using data such as vehicle location, task requests R, service request frequency, and service composition frequency as input, and generating a caching scheme for the UAV as output. The popularity of user-requested services and vehicle distribution follow a Zipf distribution, which can be expressed as:

[0017]

[0018] Where P(z) represents the probability of ranking a service request, and l is the Zipf distribution index;

[0019] Based on the drone classification, the preferences of service request R, and the location of drone E, we can divide the drones into multiple groups. The matching degree between the drones is as follows:

[0020]

[0021] Here, D, N, and S represent dense, normal, and sparse drones, respectively, and are categorized based on the number of vehicles covered by the drone. The closer the value is to 1, the higher the compatibility between the two drones.

[0022] Density peak clustering is used to group drones. For any drone En, density peak clustering requires calculating two variables:

[0023] (1) Local density ρ of UAV En n Defined as:

[0024]

[0025] Where, φ c The cutoff value, χ(x), is defined as follows: when x < 0, χ(x) = 1, and vice versa. For different datasets, different values ​​of φ need to be set based on experience. c value;

[0026] (2) The relative value δ of the drone En n Defined as having higher local density and correlation Higher drones The distance is expressed as follows:

[0027]

[0028] The density peak clustering algorithm constructs a two-dimensional decision graph, with the horizontal axis representing the local density ρ and the vertical axis representing the relative value δ. Based on two assumptions about ideal cluster centers in the density peak clustering algorithm, ρ... n and δ n Sample points with higher values ​​are identified as cluster centers. In the decision graph, points located in the upper right quadrant are selected as cluster centers.

[0029] Based on the grouping of drones, the intelligent service composition caching strategy divides the storage space of each server into three parts: service instances required for local execution within the region, service instances required for execution by other servers within the group, and global service instances that may be requested by other servers within the group.

[0030] Furthermore, the objective function of the UAV edge computing intelligent service composition caching strategy based on density peak clustering algorithm described in step 3) is:

[0031] The problem of minimizing the cost and latency of service caching can be expressed as:

[0032]

[0033] in,

[0034] 1)C all This is the total transmission cost for all task requests using the service cache, consisting of transmission cost and update cost, C. allThe calculation formula is as follows:

[0035]

[0036] T all It is the total latency for executing all task requests using the service cache, consisting of execution latency, transmission latency, and queuing latency, T. all The calculation formula is as follows:

[0037]

[0038] K is the number of task requests;

[0039] 2) This represents the transmission cost of the task request, when the task request r k Send to drone cluster E n The cost of task requests generated when using cloud cluster C; in each service r k Upon completion, the results are transmitted to the cached facilities for subsequent services. For example, if task request r k Includes services k,h Among them, service s h Cached in drone E n Up, i.e., service k,h In drone E n If executed above, output results will be generated during execution. However, services k,h Subsequent services cached on drones middle, Must be from drone E n Transmit to In order to execute the next service, the transmission cost of this process can be expressed as... When request r k The last service Once completed, the current task is considered fully executed, and the resulting... The transmission cost of sending the result back to the user can be expressed as follows: To ensure accuracy in service processing, the state information generated by the cache instance must be updated to the corresponding original service instance in a timely manner. Therefore, the data traffic between cache service instances in the cloud and drones will be updated, and the service data update transmission cost will be incurred. The above cost calculation formula is as follows:

[0040]

[0041]

[0042] Among them, when request r kSent to drone E n When cloud C executes, then α k,n =1 or α k,C =1, when request r k When running locally, α k,m =1, P m,j This indicates that a request will be made to r. k From vehicle V m The lowest cost path for transmission to the drone Ej or the cloud C. It is the cost of transmitting a unit of data via link O. It is task r k Size, ρ k It is a task request r k The data packet rate, j∈1,2,...,N,C, if task request Rk is served by vehicle Vm, α k,m =1, then the requested transmission cost is It is the data packet rate of surrounding information. If This indicates that the transmission cost of surrounding information or execution results is 0, where P n,m Indicates from drone E n To vehicle V m Transmit execution results The path with the lowest cost It is the last service The packet rate is θ, which is the ratio of the packet rate of the microservice before and after the update.

[0043] This refers to transmission latency, which includes request transmission latency, service result transmission latency, and final task execution result transmission latency. The calculation formula is as follows:

[0044]

[0045] in It is task r k Task request transmission latency, P m,j This indicates that a request will be made to r. k From vehicle V m The lowest cost path for transmission to the drone Ej or the cloud C. It is task r k Size, ρ k It is a task request r k data packet rate, It will be in the drone E n Executed services k,h The result Transferred to cache s k,h The next service drone Transmission delay, In completing request r k The last service in At that time, the result of task execution Need to be transferred to cover V m The drone, and then from that drone to V m Transmission delay;

[0046] This refers to execution latency, which is calculated using the following formula:

[0047]

[0048]

[0049] Where, β h,j Indicates service s h Whether to cache in drone Ej, j = 1, 2, ..., M, N, C, u j This indicates the CPU execution cycle of drone Ej;

[0050] This is queuing delay. When multiple vehicles generate task requests, many services queue up for execution, resulting in queuing delay. Assume the drone E arrives... n The service follows the parameter λ n The Poisson process, ξ n This is the average service rate. The mission execution of drones can be modeled using an M / M / 1 queue model, and the formula for calculating queuing latency can be expressed as:

[0051]

[0052] Where, β h,h Indicates service s h Is it cached in drone E? n In this context, due to the superior capabilities of cloud computing, queuing delays during service operation in the cloud are not considered.

[0053] The present invention also provides a process for performing service composition caching using the above method, as detailed below:

[0054] 1) Drone Classification: Drones are classified according to the vehicles they cover;

[0055] 2) Drone grouping: Introducing density peak clustering algorithm to group drones based on service similarity, distance between drones, and regional differences;

[0056] 3) Storage space grouping: The storage space is grouped using the SCCS algorithm. Each server's storage space is divided into four parts (local (2 units), group (1 unit), and global (1 unit)), prioritizing the caching of high-frequency service combinations;

[0057] 4) Repeat steps 2)-3) until you obtain the service caching strategy with the lowest cost and latency.

[0058] The inventive principle of this invention:

[0059] This invention primarily addresses the service composition caching strategy problem in vehicle edge network environments. It designs an intelligent service composition caching strategy aimed at reducing user latency and data transmission costs in a three-layer vehicle-to-everything (V2X) architecture. A density peak clustering algorithm is introduced to effectively group drones, thereby promoting enhanced cooperation. Based on this, the SCCS algorithm is proposed. This algorithm innovatively caches service instances based on service instance type segmentation and strategic partitioning of server storage space, thus proposing an intelligent service composition caching strategy based on the density peak clustering algorithm.

[0060] The beneficial effects of this invention are as follows:

[0061] This invention proposes for the first time a UAV edge computing intelligent service composition caching strategy based on the density peak clustering algorithm. This method effectively groups UAVs and comprehensively manages network resources by leveraging data associations and overlapping service demands between different services. Aiming to reduce user latency and data transmission costs in a three-layer vehicular network architecture, the density peak clustering algorithm is introduced to effectively group UAVs, thereby promoting enhanced cooperation. Based on this, the SCCS algorithm is proposed. This algorithm innovatively caches service instances based on service instance type segmentation and strategic partitioning of server storage space, ultimately obtaining the optimal service composition caching strategy. Attached Figure Description

[0062] Figure 1 A heterogeneous computing architecture consisting of vehicles, drones, and cloud servers based on software-defined networking technology;

[0063] Figure 2 A directed acyclic graph of task requests r1 generated for vehicles;

[0064] Figure 3 A simple example of a caching scheme for the SCCS algorithm;

[0065] Figure 4 The impact of the number of groups on algorithm performance;

[0066] Figure 5 The results of grouping 100 drones using four algorithms when the number of groups is 6;

[0067] Figure 6 The impact of the number of task requests on algorithm performance;

[0068] Figure 7 This section compares the performance of four algorithms under different task request volumes. Detailed Implementation

[0069] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0070] A caching strategy for intelligent drone services based on density peak clustering algorithm includes the following steps:

[0071] 1) Establish a heterogeneous computing architecture based on software-defined networking technology, consisting of vehicles, drones, and cloud servers;

[0072] 2) Propose density peak clustering algorithm and SCCS algorithm for intelligent service composition caching strategy;

[0073] 3) Obtain the intelligent service combination caching strategy.

[0074] The heterogeneous computing architecture proposed in this invention, consisting of vehicles, drones, and cloud servers based on software-defined networking technology, is as follows: Figure 1 As shown, it mainly includes four parts:

[0075] 1) Vehicle layer: This includes vehicles and onboard systems. Vehicles generate service requests and communicate with nearby drones to obtain network connectivity.

[0076] 2) Unmanned Aerial Vehicle (UAV) Layer: Provides computing and storage resources at the network edge;

[0077] 3) Cloud layer: Provides massive computing and storage resources, can handle complex processing tasks, and coordinates resource allocation and service delivery through communication with the unmanned aerial vehicle layer;

[0078] 4) SDN Controller: It can obtain vehicle motion information and service requirements, such as vehicle location, speed, driving direction and request frequency. This information can be used to adjust caching strategies.

[0079] Figure 2The dependencies between different services in a task request are presented, represented by directed edges in a directed acyclic graph. Services can be executed on the drone or in the cloud, depending on their characteristics. An intelligent service composition caching strategy, based on the density region partitioning of task request R, drone grouping, and storage space partitioning, aims to optimize the utilization of drone caching resources. An SCCS algorithm is proposed, using data such as vehicle location, task request R, service request frequency, and service composition frequency as input, and generating a caching scheme for the drone as output. The popularity of user-requested services and vehicle distribution follow a Zipf distribution, which can be represented as:

[0080]

[0081] Where P(z) represents the probability of ranking a service request, and l is the Zipf distribution index;

[0082] Based on the drone classification, the preferences of service request R, and the location of drone E, we can divide the drones into multiple groups. The matching degree between the drones is as follows:

[0083]

[0084] Here, D, N, and S represent dense, normal, and sparse drones, respectively, and are categorized based on the number of vehicles covered by the drone. The closer the value is to 1, the higher the compatibility between the two drones.

[0085] Density peak clustering is used to group drones. For any drone En, density peak clustering requires calculating two variables:

[0086] (2) Local density ρ of UAV EN n Defined as:

[0087]

[0088] Where, φ c The cutoff value, χ(x), is defined as follows: when x < 0, χ(x) = 1, and vice versa. For different datasets, different values ​​of φ need to be set based on experience. c value;

[0089] (2) The relative value δ of the drone En n Defined as having higher local density and correlation Higher drones The distance is expressed as follows:

[0090]

[0091] The density peak clustering algorithm constructs a two-dimensional decision graph, with the horizontal axis representing the local density ρ and the vertical axis representing the relative value δ. Based on two assumptions about ideal cluster centers in the density peak clustering algorithm, ρ... n and δ n Sample points with higher values ​​are identified as cluster centers. In the decision graph, points located in the upper right quadrant are selected as cluster centers.

[0092] Based on grouping drones, the intelligent service composition caching strategy divides the storage space of each server into three parts: service instances required for local execution within the region, service instances required for execution by other servers within the group, and global service instances that may be requested by other servers within the group. Figure 3 A simple example of the SCCS algorithm is given. Servers E1 and E2 belong to the same group G1, while E3 and E4 belong to the same group G2. The set {s1, s2, ..., s8} represents the services to be cached, C represents the popularity ranking of the service combination, and S represents the popularity ranking of an individual service. The popularity of data blocks decreases from left to right. Under the management of the SDN controller, we can obtain the popularity rankings C, GC, and AC of local, group-specific, and global service combinations, as well as the popularity rankings S, GS, and AS of individual services. We assume that the storage partitions are of uniform size. Since servers E1 and E3 are located in dense areas, they will first cache the most popular service combinations {s1, s2} and {s5, s6} in their storage partition T1. Then, E1 will select the most popular service that can be combined with {s1, s2} according to GC1. If no suitable service is found, E1 will cache the next most popular service s4 according to GC1. Similarly, E3 will cache the service s7 that can be combined with s6 according to GC2, until the space in storage partition T2 is exhausted. Server E2 resides in the normal zone. It first caches the most popular local service s1 into T1. Then, following the order of GC1, E2 caches the next most popular service that can be combined with the already cached service. If a suitable service is not found, E2 will cycle through the current group to select the most popular service that is not yet cached, until storage partition T2 runs out of space.

[0093] The problem of minimizing the cost and latency of service caching can be expressed as:

[0094]

[0095] in,

[0096] 3)C all This is the total transmission cost for all task requests using the service cache, consisting of transmission cost and update cost, C. all The calculation formula is as follows:

[0097]

[0098] T all It is the total latency for executing all task requests using the service cache, consisting of execution latency, transmission latency, and queuing latency, T. all The calculation formula is as follows:

[0099]

[0100] K is the number of task requests;

[0101] 4) This represents the transmission cost of the task request, when the task request r k Send to drone cluster E n The cost of task requests generated when using cloud cluster C; in each service r k Upon completion, the results are transmitted to the cached facilities for subsequent services. For example, if task request r k Includes services k,h Among them, service s h Cached in drone E n Up, i.e., service k,h In drone E n If executed above, output results will be generated during execution. However, services k,h Subsequent services cached on drones middle, Must be from drone E n Transmit to In order to execute the next service, the transmission cost of this process can be expressed as... When request r k The last service Once completed, the current task is considered fully executed, and the resulting... The transmission cost of sending the result back to the user can be expressed as follows: To ensure accuracy in service processing, the state information generated by the cache instance must be updated to the corresponding original service instance in a timely manner. Therefore, the data traffic between cache service instances in the cloud and drones will be updated, and the service data update transmission cost will be incurred. The above cost calculation formula is as follows:

[0102]

[0103]

[0104] Among them, when request r k Sent to drone E nWhen cloud C executes, then α k,n =1 or α k,C =1, when request r k When running locally, α k,m =1, P m,j This indicates that a request will be made to r. k From vehicle V m The lowest cost path for transmission to the drone Ej or the cloud C. It is the cost of transmitting a unit of data via link O. It is task r k Size, ρ k It is a task request r k The data packet rate, j∈1,2,...,N,C, if task request Rk is served by vehicle Vm, α k,m =1, then the requested transmission cost is It is the data packet rate of surrounding information. If This indicates that the transmission cost of surrounding information or execution results is 0, where P n,m Indicates from drone E n To vehicle V m Transmit execution results The path with the lowest cost It is the last service The packet rate is θ, which is the ratio of the packet rate of the microservice before and after the update.

[0105] This refers to transmission latency, which includes request transmission latency, service result transmission latency, and final task execution result transmission latency. The calculation formula is as follows:

[0106]

[0107] in It is task r k Task request transmission latency, P m,j This indicates that a request will be made to r. k From vehicle V m The lowest cost path for transmission to the drone Ej or the cloud C. It is task r k Size, ρ k It is a task request r k data packet rate, It will be in the drone E n Executed services k,h The result Transferred to cache s k,h The next service drone Transmission delay, In completing request r kThe last service in At that time, the result of task execution Need to be transferred to cover V m The drone, and then from that drone to V m Transmission delay;

[0108] This refers to execution latency, which is calculated using the following formula:

[0109]

[0110]

[0111] Where, β h,j Indicates service s h Whether to cache in drone Ej, j = 1, 2, ..., M, N, C, u j This indicates the CPU execution cycle of drone Ej;

[0112] This is queuing delay. When multiple vehicles generate task requests, many services queue up for execution, resulting in queuing delay. Assume the drone E arrives... n The service follows the parameter λ n The Poisson process, ξ n This is the average service rate. The mission execution of drones can be modeled using an M / M / 1 queue model, and the formula for calculating queuing latency can be expressed as:

[0113]

[0114] Where, β h,n Indicates service s h Is it cached in drone E? n In this context, due to the superior capabilities of cloud computing, queuing delays during service operation in the cloud are not considered.

[0115] The process of service composition caching using the method of this invention is as follows:

[0116] 1) Drone Classification: Drones are classified according to the vehicles they cover;

[0117] 2) Drone grouping: Introducing density peak clustering algorithm to group drones based on service similarity, distance between drones, and regional differences;

[0118] 3) Storage space grouping: The storage space is grouped using the SCCS algorithm. Each server's storage space is divided into four parts (local (2 units), group (1 unit), and global (1 unit)), prioritizing the caching of high-frequency service combinations;

[0119] 4) Repeat steps 2)-3) until you obtain the service caching strategy with the lowest cost and latency.

[0120] 5) Simulation Results

[0121] The present invention's method (SCCS) is compared with three different comparison algorithms: Dist-SCCS, Sim-SCCS, and Mat-SCCS (the grouping criteria of the SCCS algorithm include distance, service similarity, and type matching degree. Specifically, the present invention proposes three comparison algorithms based on these three grouping criteria, namely Dist-SCCS, Sim-SCCS, and Mat-SCCS). Furthermore, the present invention's method is also compared with conventional methods (CC algorithm, GC algorithm, and RC algorithm). 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. Figure 6 The comparison shows that SCCS has achieved excellent optimization in terms of average latency, transmission cost, and buffer cost. Figure 7 The comparison reveals that the SCCS algorithm exhibits significant performance advantages when handling varying request volumes. The SCCS algorithm demonstrates lower average latency and lower transmission and caching costs, especially under high request volumes, where its performance surpasses that of control algorithms. This is attributed to the SCCS algorithm's innovative grouping-based and service-combination caching strategy, which effectively improves processing efficiency and resource utilization. Furthermore, the SCCS algorithm consistently reduces average transmission and caching costs across different request volumes, proving its efficiency and adaptability compared to traditional algorithms. These advantages demonstrate that the SCCS algorithm, through its grouping and service-combination caching strategy, not only maintains low latency under high request volumes but also effectively reduces transmission and caching costs, proving its superior performance as a cache management strategy.

Claims

1. A caching strategy for intelligent unmanned aerial vehicle (UAV) services based on density peak clustering algorithm, characterized in that, Includes the following steps: 1) Establish a heterogeneous computing architecture based on software-defined networking technology, consisting of vehicles, drones, and cloud servers; 2) Propose density peak clustering algorithm and SCCS algorithm for intelligent service composition caching strategy; 3) Obtain the intelligent service combination caching strategy.

2. The UAV edge computing intelligent service combination caching strategy based on density peak clustering algorithm as described in claim 1, characterized in that, The heterogeneous computing architecture consisting of vehicles, drones, and cloud servers described in step 1) comprises four parts: (1) Vehicle layer: including vehicles and onboard systems, vehicles generate service requests and communicate with nearby drones to obtain network connectivity; (2) Unmanned Aerial Vehicle (UAV) Layer: Provides computing and storage resources at the network edge; (3) Cloud layer: provides massive computing and storage resources, can handle complex processing tasks, and coordinates resource allocation and service delivery through communication with the unmanned aerial vehicle layer; (4) SDN controller: It can obtain vehicle motion information and service requirements, such as vehicle location, speed, driving direction and request frequency. This information can be used to adjust caching strategies.

3. The UAV edge computing intelligent service combination caching strategy based on density peak clustering algorithm as described in claim 2, characterized in that, The density peak clustering algorithm and SCCS algorithm for the intelligent service composition caching strategy described in step 2) are defined as follows: The Intelligent Service Composition Caching Strategy (SCCS) is based on the principles of density region partitioning of task requests R, UAV grouping, and storage space partitioning. It aims to optimize the utilization of UAV cache resources. The SCCS algorithm is proposed, using data such as vehicle location, task requests R, service request frequency, and service composition frequency as input, and generating a caching scheme for the UAV as output. The popularity of user-requested services and vehicle distribution follow a Zipf distribution, which can be expressed as: Where P(z) represents the probability of ranking a service request, and l is the Zipf distribution index; Based on the drone classification, the preferences of service request R, and the location of drone E, we can divide the drones into multiple groups. The matching degree between the drones is as follows: Here, D, N, and S represent dense, normal, and sparse drones, respectively, and are categorized based on the number of vehicles covered by the drone. The closer the value is to 1, the higher the compatibility between the two drones. Density peak clustering is used to group drones. For any drone En, density peak clustering requires calculating two variables: (1) Local density ρ of UAV EN n Defined as: Where, φ c Indicates the cutoff value. The definition is: when x < 0, on the contrary, For different datasets, different values ​​of φ need to be set based on experience. c value; (2) The relative value δ of the drone En n Defined as having higher local density and correlation Higher drones The distance is expressed as follows: The density peak clustering algorithm constructs a two-dimensional decision graph, with the horizontal axis representing the local density ρ and the vertical axis representing the relative value δ. Based on two assumptions about ideal cluster centers in the density peak clustering algorithm, ρ... n and δ n Sample points with higher values ​​are identified as cluster centers. In the decision graph, points located in the upper right quadrant are selected as cluster centers. Based on the grouping of drones, the intelligent service composition caching strategy divides the storage space of each server into three parts: service instances required for local execution within the region, service instances required for execution by other servers within the group, and global service instances that may be requested by other servers within the group.

4. The intelligent service composition caching strategy as described in claim 3, characterized in that, The objective function of the UAV edge computing intelligent service composition caching strategy based on density peak clustering algorithm described in step 3) is: The problem of minimizing the cost and latency of service caching can be expressed as: in, 1)C all This is the total transmission cost for all task requests using the service cache, consisting of transmission cost and update cost, C. all The calculation formula is as follows: T all It is the total latency for executing all task requests using the service cache, consisting of execution latency, transmission latency, and queuing latency, T. all The calculation formula is as follows: K is the number of task requests; 2) This represents the transmission cost of the task request, when the task request r k Send to drone cluster E n The cost of task requests generated when using cloud cluster C; in each service r k Upon completion, the results are transmitted to the cached facilities for subsequent services. For example, if task request r k Includes services k,h Among them, service s h Cached in drone E n Up, i.e., service k,h In drone E n If executed above, output results will be generated during execution. However, services k,h Subsequent services cached on drones middle, Must be from drone E n Transmit to In order to execute the next service, the transmission cost of this process can be expressed as... When request r k The last service Once completed, the current task is considered fully executed, and the resulting... The transmission cost of sending the result back to the user can be expressed as follows: To ensure accuracy in service processing, the state information generated by the cache instance must be updated to the corresponding original service instance in a timely manner. Therefore, the data traffic between cache service instances in the cloud and drones will be updated, and the service data update transmission cost will be incurred. The above cost calculation formula is as follows: Among them, when request r k Sent to drone E n When cloud C executes, then α k,n =1 or α k,C =1, when request r k When running locally, α k,m =1, P m,j This indicates that a request will be made to r. k From vehicle V m The lowest cost path for transmission to the drone Ej or the cloud C. It is the cost of transmitting a unit of data via link O. It is task r k Size, ρ k It is a task request r k The data packet rate, j∈1,2,...,N,C, if task request Rk is served by vehicle Vm, α k,m =1, then the requested transmission cost is It is the data packet rate of surrounding information. If This indicates that the transmission cost of surrounding information or execution results is 0, where P n,m Indicates from drone E n To vehicle V m Transmit execution results The path with the lowest cost It is the last service The packet rate is θ, which is the ratio of the packet rate of the microservice before and after the update. This refers to transmission latency, which includes request transmission latency, service result transmission latency, and final task execution result transmission latency. The calculation formula is as follows: in It is task r k Task request transmission latency, P m,j This indicates that a request will be made to r. k From vehicle V m Transmitted to drone E j Or the lowest cost path for cloud C, It is task r k Size, ρ k It is a task request r k data packet rate, It will be in the drone E n Executed services k,h The result Transferred to cache s k,h The next service drone Transmission delay, In completing request r k The last service in At that time, the result of task execution Need to be transferred to cover V m The drone, and then from that drone to V m Transmission delay; This refers to execution latency, which is calculated using the following formula: Where, β h,j Indicates service s h Whether to cache in drone Ej, j = 1, 2, ..., M, N, C, u j This indicates the CPU execution cycle of drone Ej; This is queuing delay. When multiple vehicles generate task requests, many services queue up for execution, resulting in queuing delay. Assume the drone E arrives... n The service follows the parameter λ n The Poisson process, ξ n This is the average service rate. The mission execution of drones can be modeled using an M / M / 1 queue model, and the formula for calculating queuing latency can be expressed as: Where, β h,n Indicates service s h Is it cached in drone E? n In this context, due to the superior capabilities of cloud computing, queuing delays during service operation in the cloud are not considered.

5. The UAV edge computing intelligent service combination caching strategy based on density peak clustering algorithm as described in any one of claims 1-4, characterized in that, The process of service composition caching based on this strategy is as follows: 1) Drone Classification: Drones are classified according to the vehicles they cover; 2) Drone grouping: Introducing density peak clustering algorithm to group drones based on service similarity, distance between drones, and regional differences; 3) Storage space grouping: The storage space is grouped using the SCCS algorithm. Each server's storage space is divided into four parts (local (2 units), group (1 unit), and global (1 unit)), prioritizing the caching of high-frequency service combinations; 4) Repeat steps 2)-3) until you obtain the service caching strategy with the lowest cost and latency.

Images