Buildings represent more than 40 % of world-wide energy consumption. Even if several control strategies have been proposed to enhance energy management systems in buildings, this issue remains essentially open. This thesis is concerned with the development and assessment of Model Predictive Control (MPC) algorithms for energy management in buildings. In this work, a study of implementability of the control algorithm on a real-time hardware target is conducted beside yearly simulations showing a substantial energy saving potential. The thesis explores also the ability of MPC to deal with the diversity of complex situations that could be encountered (varying energy price, power limitations, local storage capability, large scale buildings). This thesis proposes the design of a distributed predictive control scheme to control the indoor conditions in each zone of the building and manage resource constraints in the context of multi-source buildings. This CIFRE Ph.D. thesis was prepared within the Gipsa-lab laboratory in partnership with Schneider-Electric in the scope of the HOMES program (www.homesprogramme.com).