Abstract:
Traffic monitoring and control, as well as traffic simulation, still have major unsolved challenges, despite extensive research efforts, especially on artificial intelligence approaches to overcome these problems. Introduces a reinforcement learning approach to traffic Lighting control that uses Deep Q Learning algorithms to make a smart traffic signal control. Reinforcement learning: states, actions, rewards. RL state captures environment information
for decision making. Actions are agent’s decisions based on observed conditions. Rewards are feedback after performing actions in a state. For Training our agent We use data about 1,000 vehicles to train our agent. It also builds communication between each intersection
of road what contains 200 edges..Training uses 30 episodes, which determines the number
of iterations in the RL training process, and 240 max-step, which refers to the maximum
number of time intervals or steps within each episode. Set the duration of each episode and the number of steps an RL agent can take to make a decision and observe its impact. Deep Q Learning gives Q value, By estimating and updating Q-values, the agent can learn which actions are more beneficial in different traffic conditions. Here, we cut the queue length by 9.7% to shorten wait times and relieve traffic. Get a happy outcome of traffic signal control and improve rewards of 9.7%.
