Federated Deep Reinforcement Learning Approach for Real-Time Target Coverage in Wireless Sensor Networks

Maximizing target coverage is crucial in wireless sensor networks (WSN), where sensor nodes collect data to monitor events or targets in specific areas. Efficient coverage optimization involves deploying sensors to cover the entire monitored region, addressing the NP-hard nature of the problem. Various algorithms, including learning-based automata, genetic algorithms, and probabilistic approaches, have been employed. Specifically, deep reinforcement learning (DRL), using deep neural network (DNN) as a predictor, has shown promise. However, the DRL models require a large amount of data for training, which may limit their adaptability to the ever-evolving statistics of the underlying environment. Moreover, wireless networks are inherently distributed systems, where centralized DRL approaches would require excessive data exchange, while fully distributed approaches may result in slower convergence rates and performance degradation. To overcome these challenges, we propose a novel federated learning (FL) approach to Deep Q-Network (DQN), called Federated Reinforcement Deep Q-Network (FEDQN), which allows sensors to collaboratively train the DQN by sharing only model weights, not training data. This paper demonstrates how FEDQN maximizes target coverage in a continuous, partially observable visual sensor network environment, simulating traffic flow with mobile targets.