Deep Reinforcement Learning-Based One-to-Multiple Cooperative Computing in Large-Scale Event-Driven Wireless Sensor...

Area of Science:

• Computer Science
• Electrical Engineering
• Network Engineering

Background:

• Wireless Sensor Networks (WSNs) are increasingly used for emergency event monitoring.
• Micro-Electro-Mechanical System (MEMS) technology enables local event processing in WSNs.
• Resource scheduling and computation offloading are challenging in large-scale, event-driven WSN environments.

Purpose of the Study:

• To propose a cooperative computing strategy for large-scale WSNs facing emergency events.
• To address the challenges of resource scheduling and computation offloading in dynamic WSN environments.
• To enhance the efficiency of emergency event monitoring in WSNs.

Main Methods:

• An equal-size K-means clustering algorithm for dynamic node activation and clustering around event locations.
• Inter-cluster task assignment where computation tasks are allocated to cluster heads.
• A Deep Deterministic Policy Gradient (DDPG)-based intra-cluster one-to-multiple cooperative computing algorithm for computation offloading strategy.

Main Results:

• The proposed algorithm demonstrates performance close to exhaustive methods.
• The algorithm outperforms classical algorithms and the Deep Q Network (DQN) algorithm.
• Effective computation offloading and task completion within deadlines were achieved.

Conclusions:

• The developed cooperative computing strategy significantly improves emergency event monitoring in WSNs.
• The DDPG-based approach provides an efficient solution for resource scheduling and computation offloading.
• This work offers a robust framework for handling dynamic and large-scale WSNs during emergency events.