In the era of rapid development of the Internet of Things and fifth-generation mobile communication technology, the application scenarios of UAV drones are rapidly expanding from military to civilian domains. However, UAV drones themselves are typically constrained by size, weight, and power consumption, with limited onboard computing capabilities, storage resources, and energy reserves. This creates a sharp contradiction with the increasing complexity of tasks they undertake. How to efficiently handle these compute-intensive and latency-sensitive tasks has become a bottleneck restricting the effectiveness of UAV drone applications.
In this context, mobile edge computing technology provides an effective approach to resolve this contradiction. This technology下沉s cloud computing capabilities to the network edge, enabling UAV drones to offload part or all of computing tasks to nearby edge servers for execution, thereby significantly alleviating local computational pressure, reducing task processing energy consumption and latency, and effectively extending their endurance. However, task offloading decision-making itself is a complex optimization problem that requires comprehensive consideration of multidimensional dynamic factors such as network status, server load, task characteristics, and remaining energy of UAV drones, seeking an optimal trade-off among multiple objectives like energy consumption and delay.
To address this, researchers have approached from different technical paths,致力于 optimizing the task offloading efficiency of UAV drones. Existing research can be broadly categorized into three types: first, focusing on improving core efficiency indicators, for example, optimizing latency and energy consumption through UAV drone swarm coordination and trajectory planning; second,着眼于 system reliability and security, such as introducing blockchain and game theory to build trusted offloading mechanisms; third, exploring new collaborative computing models, such as designing adaptive switching protocols in IRS-assisted MEC systems. While these studies have advanced the field, they often focus on optimization algorithms with simplified communication and concurrency models in simulations, lacking a system implementation载体面向实际 complex operational environments.
Therefore, I design a distributed task offloading monitoring system based on the KBEngine framework. This system addresses the gap by providing an integrated platform for algorithm validation and management. The core advantage lies in its high scalability, capable of integrating and comparing various optimization algorithms. In this paper, I detail the system design, optimization problem formulation,代理模型 integration, and experimental validation, demonstrating its effectiveness as a comprehensive solution for UAV drone edge computing research.
The optimization problem solved by the system is as follows: the system contains M UAV drones, N edge servers, and K communication channels. Each UAV drone can offload tasks via a channel to a specified edge server. Each UAV drone divides its task into several dependent task nodes. These nodes can be processed locally or offloaded to edge servers. When offloading to edge servers, communication delay for data transmission from the terminal to the edge server must be considered. There are three special types of task nodes: (a) nodes requiring UAV drone-specific information, such as velocity or position, which must be processed locally; (b) nodes requiring information from other UAV drones, which must be offloaded to edge servers; (c) the first and last task nodes involving task initiation and final decision, which must be processed locally. To represent task dependencies, a directed acyclic graph is used to describe computational tasks.
For the task offloading optimization of UAV drone m, both computational delay and energy consumption need to be minimized, forming a dual-objective optimization problem. The objective functions are:
$$ \min( f_z^m(\mathbf{x}_m)), \quad z = 1,2 $$
where \( f_1^m(\mathbf{x}_m) \) represents the computational delay of UAV drone m, i.e., \( f_1^m(\mathbf{x}_m) = T_m \), and \( f_2^m(\mathbf{x}_m) \) represents the energy consumption of UAV drone m, i.e., \( f_2^m(\mathbf{x}_m) = E_m \). \( \mathbf{x}_m \) is the decision vector for all task node scheduling relations of UAV drone m. Converting \( \mathbf{x}_m \) to binary, from lowest to highest bits, are the decision factors \( s_m^i \) for task nodes from start to end. For the three special task nodes, offloading locations are fixed, so decision factors are constants.
$$ x_m = \sum_{i=1}^{\mu_m} s_m^i \cdot 2^{(i-1)} $$
The system includes M UAV drones. Thus, the average offloading delay and consumption function for the system can be expressed as:
$$ f_z(\mathbf{x}) = \frac{1}{M} \sum_{m=1}^{M} \sum_{n=1}^{N} \sum_{k=1}^{K} \eta_{mnk} \cdot f_z^m(\mathbf{x}_m), \quad z = 1,2 $$
where \( \eta_{mnk} \in \{0,1\} \); if UAV drone m offloads task nodes to edge server n via channel k, then \( \eta_{mnk} = 1 \), otherwise \( \eta_{mnk} = 0 \). \( \mathbf{x} = [x_1, x_2, \ldots, x_M] \) is the decision vector for the optimization problem.
A surrogate model is a mathematical approximation constructed via machine learning algorithms to替代 computationally expensive complex simulations. Its core value lies in learning the mapping relationship between input parameters and output responses from historical data, forming a computationally efficient and reliable predictive model, thereby significantly improving optimization efficiency. In scenarios with high real-time requirements like UAV drone task offloading, the surrogate model enables optimization algorithms to bypass time-consuming physical simulations by快速 predicting system performance under different decisions.
The construction of surrogate models can employ various machine learning algorithms, with Gaussian process regression and neural networks being the most typical. Gaussian process regression is a probabilistic model that provides both predicted values and uncertainty estimates, making it suitable for scenarios with limited samples and requiring guided active learning. Neural networks, with their powerful nonlinear fitting capabilities, are applicable for modeling complex systems with high dimensions and large data volumes. Other methods include radial basis function interpolation, support vector regression, and polynomial regression.
After achieving efficient energy consumption and delay prediction via surrogate models, researchers can integrate various optimization algorithms on this platform to validate decision effects under different conditions. For example, algorithms like Actor-Critic with Long Short-Term Memory (AC-LSTM) and Non-dominated Sorting Genetic Algorithm II (NSGA-II) are included. AC-LSTM combines LSTM’s memory能力 with actor-critic交互 learning to handle partially observable or time-dependent environments. NSGA-II, based on Pareto optimality, searches the solution space via selection, crossover, and mutation, outputting a set of non-dominated solutions for trade-offs between conflicting objectives like energy consumption and delay. NSGA-II can easily fall into local convergence when handling high-dimensional offloading tasks, while AC-LSTM can mitigate this effectively.
The KBEngine engine employs a layered architecture design, distributing entities with different functions across different logical tiers on the server side to achieve effective load separation and functional decoupling. This architecture primarily consists of three core layers: client, external logic layer, and internal spatial logic layer, each协同工作 to support complex business logic and high-performance real-time computing.
| Entity | Client | BaseApp (External Logic Layer) | CellApp (Internal Spatial Logic Layer) |
|---|---|---|---|
| Account Management Entity | Exists | Exists | Does Not Exist |
| Current Task Entity | Exists | Exists | Exists |
| Spatial Entity | Does Not Exist | Exists | Exists |
| Spatial Management Entity | Does Not Exist | Exists | Does Not Exist |
The client serves as the front-end for direct user interaction, providing interfaces for task selection, task management, and data display. It receives user input, sends requests to the server, and实时展示 status updates and task execution data returned from the server. The external logic layer is the core process handling non-real-time, global logic, responsible for account management, persistent data storage, and low-frequency business processing. The internal spatial logic layer focuses on high-frequency real-time computing and spatial logic processing,承载 task offloading, resource scheduling, and other high-performance computation demands.
In the system, entity classes are distributed across different layers based on functional requirements, as shown in Table 1. The account management entity handles account operations like login/logout and task selection, distributed across the client and external logic layer. When a user selects a task, the account management entity in the external logic层 selects the corresponding entity from the task entity storage area, synchronizes it to the current task entity, and initializes it. Initialization involves three stages: first, applying to the spatial management entity for a corresponding spatial entity based on task collaboration needs; second, initializing the current task entity in the internal spatial logic layer after obtaining the space; third, creating the corresponding current task entity on the client, establishing bidirectional communication links with the external and internal spatial logic layers for three-tier data互通.
During task offloading decision-making, client entities send requests to the internal spatial layer, where the task offloading management entity executes offloading decision algorithms and feeds results back to entities across layers. This manager generates offloading strategies based on optimization algorithms, coordinates task allocation between terminal devices and edge servers, and统计 historical offloading data for algorithm accuracy verification. If a user selects management tasks via the account management entity, the current task entity monitors algorithm execution efficiency across spaces, providing data display and interaction for management interfaces, enabling comprehensive supervision of system operation status.
The optimization core of the system revolves around the协同工作 of the task offloading management entity and the current task entity. The overall process is a closed-loop system integrating data-driven, machine learning, and intelligent decision-making. The具体流程 is as follows:
1. Initialization and Data Collection Phase: After system startup, the decision space is configured in the task offloading management entity, defining the dimensions of the optimization problem and the value range for each dimension. Then, the manager automatically generates multiple sets of representative decision vector samples using sampling strategies. These samples are transmitted to the current task entity via internal interfaces. The current task entity, as the task execution entity, receives each set of decision vectors, drives the UAV drone to complete a full task offloading simulation, and accurately collects performance result data (e.g., task delay, energy consumption, cost). These result data are实时 returned to the data collection module of the task offloading management entity,构建 an initial dataset for model training.
2. Surrogate Model Training and Construction Phase: After accumulating sufficient sample data, the process enters the machine learning phase. The surrogate model training module of the task offloading management entity is activated, which can employ algorithms like Gaussian process regression (GPR) or neural networks (NN) to learn from the historical dataset. This process optimizes model parameters to minimize prediction error, constructing a mathematical surrogate model that accurately reflects the complex mapping from “decision vector” to “performance results”. This model is characterized by极低 computational cost but高精度 approximating the original time-consuming simulation process.
3. Model Application and Optimization Solving Phase: The trained surrogate model is immediately integrated into the algorithm optimization solving module, becoming the engine for快速 optimization. When the system runs and new tasks arrive requiring immediate offloading decisions, the current task entity uploads real-time environmental parameters (e.g., network bandwidth, UAV drone position, server load status) to the task offloading management entity. The manager then invokes the embedded optimization algorithm to perform rapid search and solving in the approximate space constructed by the surrogate model, ultimately obtaining a high-quality optimal decision vector.
4. Decision Execution and Closed-Loop Feedback Phase: The obtained optimal decision vector is sent back to the current task entity, guiding the UAV drone to execute the actual task offloading. Simultaneously, the real results of this offloading are recorded and fed back to the offloading result statistics module of the task offloading management entity. These新增的高价值 data are not only used for performance monitoring and analysis but can also be regularly used to update and retrain the surrogate model, enabling it to continuously adapt to environmental changes, forming a self-evolving, continuously optimizing intelligent closed loop.
Additionally, as a监管单元, the current task entity has access permissions to obtain historical operation data and statistical information from the offloading result statistics module of the task offloading management entity, and can transmit it to the client interface for visualization, providing intuitive system status monitoring and decision support for managers.
To validate the effectiveness of the system, I design corresponding experimental environments. The experimental hardware configuration主要包括模拟 UAV drone clusters的 clients, edge server nodes, and central management nodes. The software environment is based on the KBEngine server framework and Unity front-end. Key experimental parameters are configured as shown in Table 2.
| Configuration Category | Parameter Item | Parameter Value |
|---|---|---|
| Network Topology | Architecture | Star Topology |
| Number of Base Stations | 2 | |
| Bandwidth | 20 Mbps | |
| Edge Servers | Number | 2 |
| CPU | Intel Xeon E5-2680 v4 | |
| Memory | 32 GB | |
| UAV Drones | Number | 30 (adjustable) |
| CPU | ARM Cortex-A72 | |
| Battery Capacity | 10000 mAh | |
| Tasks | Task Type | Path Planning |
| Task Input | Local Environment Grid Map |
The AC-LSTM algorithm network policy structure and value network structure both include an input layer, LSTM layer, three fully connected layers, and an output layer, with 600 training epochs. The NSGA-II algorithm has a population size of 40 and 600 generations.
The system provides complete strategy configuration and operation monitoring functions for the UAV drone edge computing task offloading control system. The interface uses a three-zone layout design, corresponding to environment configuration, cluster efficiency overview, and single-machine detailed data query functions. The system initialization state shows all data indicators as zero values in standby状态; the actual operation state displays真实监控 data generated.
The operation flow is as follows: users first need to complete the working environment setup in the left configuration area. After entering the space number (default 1) in the ROOM_ID field, click the “Enter” button to activate the workspace. Then, select the surrogate model (e.g., model1), optimization algorithm, and set multi-objective decision weight parameters via drop-down menus. After completing all parameter configurations, click the “Deploy” button to submit to the server for strategy environment initialization. When needing to change the surrogate model or optimization algorithm, one must first click the “Unload” button to clear the existing optimization environment, wait for system清理完成, then重新选择 parameters and click the “Deploy” button to construct a new environment.
Cluster efficiency monitoring is located in the middle area of the interface. Users can select a specific task (e.g., “Offload Task 1”) via the top drop-down menu and click the “Query” button to obtain aggregated data. The system returns key indicators such as the number of UAV drones executing the task, total number of task offloadings completed, average energy consumption (in Joules), and average delay (in seconds).实际运行 data shows: 6 UAV drones complete 6 offloading tasks, with average energy consumption of 18.30 J and average delay of 1.63 s,直观 reflecting overall cluster performance.
Single UAV drone detailed query is located on the right side of the interface, supporting细粒度 task analysis. Users need to specify the target UAV drone number in the “UAV ID” input box and click the “Query” button. The system returns the cumulative offloading count for that UAV drone. Users can select to view records for a specific count, and the system displays the task name, specific energy consumption, and delay data for that execution. For example, data shows for UAV drone ID=2, the first offloading record: executing Offload Task 1, energy consumption 17.12 J, delay 1.71 s, cumulative completion of 3 offloading tasks.
The system, through layered visualization design, meets both macro态势监控 needs and supports micro problem定位分析, providing a complete visual monitoring solution for UAV drone swarm task offloading strategy optimization. All data are实时 obtained from the computing engine, ensuring timeliness and accuracy of monitoring data.
In the monitoring terminal, under the same surrogate model, I configure the AC-LSTM algorithm and NSGA-II algorithm separately. Weight parameters are randomly generated in the range of 0 to 1 according to a uniform distribution, with 40 data points for configuration. The smaller this parameter, the higher the weight for energy consumption optimization; the larger, the higher the weight for delay optimization. When the weight is set to 0, it相当于 tasks processed locally by the UAV drone; when set to 1, it相当于 tasks全部 offloaded to servers for processing. Both algorithms统计 energy consumption and delay under different weights,最终 obtaining Pareto fronts for the two algorithms.
The Pareto front obtained via AC-LSTM has the leftmost point with energy consumption of 17.18 J and delay of 1.91 s, and the rightmost point with energy consumption of 27.53 J and delay of 0.97 s. The front obtained via NSGA-II has the leftmost point with energy consumption of 21.83 J and delay of 1.63 s, and the rightmost point with energy consumption of 24.32 J and delay of 1.40 s. It is evident that AC-LSTM outperforms NSGA-II in both front width and convergence, verifying that NSGA-II容易 enters local convergence when handling high-dimensional tasks.
The main contributions of this study are: first,构建 a scalable algorithm integration and validation platform, addressing the痛点 of theoretical algorithms lacking practical载体; second,通过 comparing AC-LSTM and NSGA-II algorithms on this platform via experiments,量化验证 the superiority of AC-LSTM, whose obtained Pareto front significantly outperforms the对比 algorithm in width and convergence, providing a better solution set for intelligent offloading decisions; third, the system’s provided visualization monitoring interface实现 algorithm integration, performance validation, and real-time operation.
In summary, this study not only provides an efficient experimental platform for UAV drone task offloading research but also通过 rigorous experiments验证 the巨大潜力 of intelligent algorithms in practical applications, offering support for the engineering落地 of related technologies. Future work will focus on introducing more complex dynamic channel models and privacy protection mechanisms to further enhance the practicality and security of the system.
Throughout this design, the UAV drone task offloading monitoring system demonstrates robust capabilities in handling complex edge computing scenarios. The integration of surrogate models with optimization algorithms like AC-LSTM ensures efficient and effective decision-making for UAV drone fleets. The KBEngine framework’s distributed architecture facilitates scalable management of multiple UAV drones and edge servers, making it a versatile platform for both research and practical deployments. As UAV drone applications continue to expand, such systems will play a crucial role in optimizing resource utilization and task performance in dynamic environments.