Multi-Edge Offloading and Resource Allocation with Personalized Federated DRL
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computation offloading and resource allocation methods in mobile edge computing (MEC) systems face challenges in handling complex and dynamic environments due to limited prior knowledge, network congestion, privacy issues, and inability to adapt to personalized demands of smart communities, leading to degraded Quality-of-Service (QoS) and excessive system overheads.
Innovation Solution
A method of joint computation offloading and resource allocation using personalized federated deep reinforcement learning (PFR-OA) with a twin-delayed DRL-based algorithm for single-edge scenarios and a novel personalized FL-based training framework for multi-edge scenarios, incorporating discrete-time models, energy harvesting, and proximal terms to optimize task execution delay and energy consumption under constraints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If tasks are offloaded to remote cloud for execution, then computing capacity is sufficient, but transmission delay becomes excessive
Solution Approach 1:
The patent segments the centralized cloud computing system into distributed mobile edge computing servers deployed at multiple network edges. Each edge server handles local task execution, reducing transmission distance and delay while maintaining sufficient computing capacity through distributed resource allocation across multiple nodes.
Solution Approach 2:
The patent transitions from a single centralized cloud dimension to a multi-dimensional edge computing architecture. By deploying computing resources across multiple spatial dimensions (different edge locations closer to end devices), the system reduces transmission delay while preserving computing capacity through parallel distributed processing.
2Power
If more tasks are offloaded to MEC servers, then computing capacity is extended, but delay increases due to limited MEC resources
Solution Approach 1:
The patent implements dynamic resource allocation and task scheduling at edge servers using reinforcement learning. The system continuously adapts resource allocation based on real-time system state, task characteristics, and server load conditions, optimizing the balance between computing capacity utilization and execution delay minimization.
Solution Approach 2:
The patent changes key system parameters dynamically including offloading decisions, resource allocation ratios, and scheduling policies based on learned optimal strategies. The reinforcement learning agent adjusts these parameters in response to changing environmental conditions to maintain low delay while maximizing computing capacity utilization.
3Reliability
If centralized training is used for DRL, then model performance is improved, but network congestion and privacy leakage occur
Solution Approach 1:
The patent extracts the training data from the centralized environment and keeps it distributed at local edge servers. Only model parameters (gradients) are transmitted to the central server for aggregation, while raw task data remains localized. This extraction approach maintains model performance through collaborative learning while preventing privacy leakage by never centralizing sensitive data.
Solution Approach 2:
The patent introduces federated learning as an intermediary training mechanism between centralized and fully distributed approaches. The central server acts as a coordinator that aggregates model updates without accessing raw data, enabling collaborative model improvement while preserving data privacy through the intermediary aggregation process.
4Productivity
If classic FRL is used for multi-edge scenarios, then training efficiency is improved, but personalized community demands cannot be met
Solution Approach 1:
The patent applies local quality by allowing each edge server to maintain personalized model adaptations for its local community while participating in federated training. Each server can incorporate local community characteristics and personalized demands into its local model updates, which are then aggregated to improve the global model's adaptability to diverse community requirements.
Data Source
AI summary
A new multi-edge smart community system consisting of communication, computing, and energy harvesting models, where the task execution delay and energy consumption are formalized as the optimization objectives under multiple constraints. For single-edge scenarios, we propose an improved twin-delayed DRL-based algorithm. For multi-edge scenarios, we develop a novel personalized FL-based training framework for DRL. Using the real-world settings and testbed, extensive experiments are conducted to validate the effectiveness of the proposed PFR-OA. The results show that the PFR-OA achieves better trade-offs between delay and energy consumption and exhibits higher task execution success rates than benchmark methods under different scenarios. Notably, the PFR-OA reaches a faster convergence speed compared to advanced DRL-based and FRL-based methods. Moreover, we further verify the practicality and superiority of the PFR-OA via real-world testbed experiments.