The project focuses on controlling outlet flow temperatures in heat exchangers, particularly under transient heat sources, using Reinforcement Learning (RL). The Q-learning algorithm is employed to regulate the outlet flow temperature of a co-current flow heat exchanger in dynamic situations.
In this problem, a co-current heat exchanger serves as the environment, dictating the thermodynamic states of flows. The agent's decision-making involves adjusting the outlet mass flow rate of the pump, either increasing or decreasing it. The reward function is defined based on the variance between the outlet flow temperature and a reference temperature. During each step of system operation, the environment provides feedback on the outlet flow temperature relative to the agent's actions via the reward function. Subsequently, upon policy improvement, the agent selects the most advantageous action, aiming for higher returns in the next step.
matrix of state-actions
Energy Equation in Heat Exchanger
Controlling Method
The findings demonstrate that utilizing the Q-learning algorithm can effectively maintain the outlet flow temperature of a heat exchanger with an error margin of just 1 degree.
