 
Summary: Synthesis of Reinforcement Learning, Neural Networks, and PI Control
Applied to a Simulated Heating Coil
Charles W. Anderson1, Douglas C. Hittle2, Alon D. Katz2, and R. Matt Kretchmar1
1Department of Computer Science
Colorado State University
Fort Collins, CO 80523
fanderson,kretchmag@cs.colostate.edu
2Department of Mechanical Engineering
Colorado State University
Fort Collins, CO 80523
fhittle,along@lamar.colostate.edu
Abstract
An accurate simulation of a heating coil is used to compare the performance of a proportional plus
integral (PI) controller, a neural network trained to predict the steadystate output of the PI controller, a
neural network trained to minimize the nstep ahead error between the coil output and the set point, and
a reinforcement learning agent trained to minimize the sum of the squared error over time. Although the
PI controller works very well for this task, the neural networks produce improved performance. The rein
forcement learning agent, when combined with a PI controller, learned to augment the PI control output
for a small number of states for which control can be improved.
Keywords: neural networks, reinforcement learning, PI control, HVAC
