Reinforcement Learning Scheduling for Mobile Edge Computing Workflows
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The joint optimization problem of online unloading decision and resource allocation in Mobile Edge Computing (MEC) networks, characterized by Directed Acyclic Graphs (DAG), is a non-convex NP-hard problem, leading to high computational complexity and burden, which conventional mathematical methods struggle to efficiently address.
Innovation Solution
An online multi-workflow scheduling method based on reinforcement learning is proposed, utilizing a graph convolution neural network to extract structural features and a gradient policy for decision-making, which selects nodes and allocates resources between mobile devices and edge servers to minimize average completion time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional mathematical methods are used to solve the joint optimization problem of online unloading decision and resource allocation, then the optimization accuracy is improved, but the computational complexity increases significantly
Solution Approach 1:
The patent replaces conventional mathematical optimization methods with a reinforcement learning-based scheduling algorithm. The RL agent learns optimal scheduling policies through interaction with the MEC environment, substituting the need for complex mathematical solvers while achieving comparable or superior optimization performance with lower computational overhead.
Solution Approach 2:
The scheduling system uses self-learning reinforcement learning algorithms that automatically adapt to changing workloads and resource conditions without requiring manual intervention or complex real-time mathematical optimization. The agent continuously improves its scheduling decisions through experience, reducing the need for computationally intensive external optimization methods.
2Manufacturing precision
If conventional mathematical methods are used to solve the joint optimization problem, then the unloading decision quality is improved, but the processing time increases
Solution Approach 1:
The reinforcement learning agent pre-learns optimal scheduling patterns and decision-making strategies during training phases. When faced with actual scheduling decisions, the agent can quickly retrieve and apply pre-learned policies, avoiding the need for time-consuming real-time mathematical optimization while maintaining high decision quality.
Solution Approach 2:
The patent substitutes traditional real-time mathematical optimization with reinforcement learning-based decision making. The RL approach provides faster response times by leveraging pre-learned policies and neural network inference, which is computationally more efficient than solving complex optimization problems in real-time.
3Productivity
If more computing resources are allocated to solve the optimization problem, then the scheduling performance is improved, but the burden on MEC network increases
Solution Approach 1:
The MEC network uses self-learning reinforcement learning algorithms that automatically optimize resource allocation without requiring extensive external computational assistance. The distributed RL agents at edge servers make scheduling decisions locally, reducing the need for centralized high-performance computing and minimizing network burden while maintaining scheduling effectiveness.
Data Source
AI summary
System model is established to characterize mobile devices, edge servers, tasks and nodes. Node unloading rule is established, and the mobile device can be selected to unload the nodes to the edge server or leave the nodes to be executed locally. Timeline model is established to record arrival events of all tasks and execution completion events of the nodes. Online multi-workflow scheduling policy based on reinforcement learning is established, state space and action space of scheduling problem are defined, and reward function of the scheduling problem is designed. Algorithm based on policy gradient is designed to solve online multi-workflow scheduling problem for implementing the scheduling policy. Unloading decision and resource allocation are performed based on features extracted by graph convolution neural network. Current workflow and state of the server can be analyzed in real time, thereby reducing complexity and average completion time of all workflows.