Building thermal modelling plays an important role in managing the thermal comfort and the energy consumption of buildings. A major challenge for modellers is how to deal with uncertainty problems in order to have a robust model with an acceptable computational time for the improvement of predictive control. This paper presents a methodology which allows obtaining the good model of a controllable thermal zone able to adapt regularly to the measurements by a robust identification procedure. Its input data are achieved by the modelling simplification of adjacent zones under uncontrollable uncertainties. This method is applied for a multi-zone positive energy building in south of France to validate our approach.