Groundhog Day: Iterative Learning for Building Temperature Control

As the cost of energy continues to grow, there is an increasing need for more effective building control, particularly regarding heating, ventilation, and air conditioning (HVAC) systems. Existing HVAC control systems are primarily based on measured temperature feedback, and typically do not utilize temperature forecasts and historical data. As a result, building room temperatures tend to fluctuate with the outside temperature, compromising occupant comfort and energy efficiency (as users overcompensate in setting the desired temperature). This paper proposes a feedforward scheme for building temperature control based on iterative learning control (ILC) to extract information from historical data with similar temperature patterns and preemptively account for expected future error. We apply the ILC strategy to building temperature control by considering a 24-hour period as one iteration. The weather forecast is used to find the historical record best matched with the predicted outside temperature and initial condition (room temperature). The recorded heat input and room temperature data is then used to generate a feedforward update of the heat input based on an ILC update. This method allows anticipatory feedforward control (on top of the feedback control) to prepare the room condition for the upcoming weather conditions, instead of only reacting to the current condition. We use a 4-room simulation to illustrate our approach. The result shows that the iterative learning controller produces substantially less error and oscillation as compared to the feedback control alone. However, the scheme consumes more energy as the temperature is more tightly regulated around the desired setting. This issue is addressed by relaxing the temperature learning criterion to reduce the control effort.