Reinforcement Learning environment for temperature control in naturally ventilated calf barns

Abstract

The health and well-being of calves strongly depend on barn climate conditions. Ventilation is regulated by conventional control strategies, leading to suboptimal thermal conditions. Reinforcement learning (RL) polices promise a better solution but require realistic and computationally efficient environments. We present a fast and accurate Gymnasium environment for temperature control in calf barns. A reduced-order thermal network was coupled to an airflow network and simulated in an integrated manner using Differential Algebraic Equation solvers and packaged in the Gymnasium interface. The presented models were validated against measurements made on a case study calf barn, achieving a RootMean Squared Error of 1.1 °C.