Two additive thermal sources are generally not simultaneously distinguishable from the only observation of their effect on the heat balance. However, there are cases where information about the variation regularity of these sources is known. This is typically the case of convective internal gains in the building, for which the use scenarios create discontinuous inputs while heat gains relating to the air leakage are regular in time. In the present paper, we introduce a method aiming to distinguish heat sources using this a priori knowledge about their dynamics. We provide numerical and experimental evidence that the method succeeds in separating/distinguishing these kind of sources. This method could be applied to the identification of the occupancy rate for measurement and verification (M&V) plans or smart home systems such as learning thermostats.