This paper presents an approach which is called meta-optimization combining with scattering analysis used to enhance on-site real-time temperature anticipation for energy management. The aim of this approach is to analyse the sensitivity of the parameters in order to simplify, and then attain, the best reduced model able to match with measurements regularly in a robust manner. This will be done with the effort of keeping their physical properties. Indeed, parameter identification is a key challenge for modelling system that are used with many uncertainties such as construction and material quality, weather conditions, and occupant behaviour that are changing during building life. This method is applied for a nearly-zero energy building in France to validate our approach.