Comparative Study Between Deep Learning and Reinforcement Learning in Multiple Interactions and Large Network

This article provides a comparison of classic reinforcement learning (RL) and deep reinforcement learning (DRL) algorithms in order to optimize traffic signal control in a vast network. The goal is to decrease three primary variables: queue length, average waiting time, and throughput. Conventional reinforcement learning (RL) uses tabular representations and value iteration algorithms, whereas deep reinforcement learning (DRL) leverages deep neural networks to approximate value functions. The experimental results indicate that Deep Reinforcement Learning (DRL) performs better than classical Reinforcement Learning (RL) in the domain of traffic signal control. DRL demonstrates exceptional performance by efficiently collecting intricate traffic patterns and adjusting signal timings using input data with a high number of dimensions. It effectively decreases the length of queues and the average waiting time, while also enhancing throughput. This study emphasizes the capacity of Deep Reinforcement Learning (DRL) algorithms to tackle traffic congestion issues and optimize traffic signal control in extensive networks.