SIMULTANEOUS DESIGN AND OPTIMIZATION OF AGE ENERGY EFFICIENT AND AGE OF INCORRECT INFORMATION IN WIRELESS POWERED SENSOR NETWORKS

The evolution of Wireless Powered Communication Networks (WPCN) and the growing diffusion of IoT devices require communication policies capable of balancing information freshness and energy sustainability. The traditional Age of Information (AoI) metric, although widespread, does not capture the semantic dimension of data and is insufficient in scenarios independent of information anomalies. To address this limitation, this thesis integrates the Age of Incorrect Information (AoII) metric into the network model and analyzes the behavior of a WPCN system through a Markov decision-making process solved using Deep Reinforcement Learning techniques. Three architectures are developed and compared: Deep Q-Network (DQN), Double DQN, and Dueling DQN. The objective is to evaluate how different hyperparameters (learning rate, discount factor, transmitted power, batch size, and update frequency) influence the agent's ability to balance energy consumption, semantic accuracy, and update timeliness. The model is tested in three scenarios based on different probabilities of information anomalies.