In a cold climate, the electrical power demand for space conditioning during certain periods of the day becomes a critical issue for utility companies from an environmental and financial point of view. Shifting a portion or all of this demand to off-peak periods can reduce peak demand and stress on the electrical grid. One possibility is to use an electrically heated floor as a storage system in residential houses. 
To shift a significant part of the consumption while maintaining occupants’ thermal comfort, predictive supervisory control strategies such as Model Predictive Control (MPC) have been developed for forecasting future energy demand. However, MPC requires a building model and an optimization algorithm. Their development is time-consuming, leading to a high implementation cost. This thesis reports the development of a new simplified predictive controller to control an electrically heated floor in order to shift and/or shave the building peak energy demand.
First, a method to model an EHF in TRNSYS was proposed in order to study the potential of using an electrically heated floor (EHF) in terms of load management without predictive control. Some parametric studies on the floor assembly and its impact on the thermal comfort were conducted. Results showed that a complete night-running control strategy cannot maintain an acceptable thermal comfort in all rooms. Therefore, it is required to predict the future demand of the building in order to anticipate the charging/discharging process of the storage system. 
Therefore, a simplified self-learning predictive controller was proposed. The function of the proposed simplified predictive controller is to increase the rate of stored energy during off-peak periods and to decrease it during peak periods, while maintaining thermal comfort. To achieve this goal without using a detailed building model, a simplified solar prediction model using available online weather conditions forecast was proposed. The controller approach is based on a learning process; it takes building responses of previous days into consideration. The developed algorithm was applied to a single-storey building with and without basement. Results show a significant decrease in thermal discomfort, average applied powers during peak and mid-peak periods. The approach has also proven to be financially attractive to both supplier and owner.