A critical review on forward and data driven methods for building energy analysis

The classic approach to evaluate the building-HVAC system energy use is based on the application of a thermal model with known system structure and proprieties as well as forcing variables (forward approach). This model can be more or less complex depending on the requested result accuracy and output time step. For a tailored analysis the forward approach requires a detailed knowledge of the physical phenomena (and their relative magnitude and interactions) affecting the system behavior, and the building system operating mode. ESP-r, BLAST, DOE-2 and EnergyPlus are the most widespread simulation codes based on forward simulation models. A more recent approach for building energy analysis is based on the so called inverse or data-driven models. In this case, the input (regressor variables) and output variables (response) are known and measured and the objective is to estimate the system parameters and to describe the mathematical model. By a data driven approach is possible to evaluate the as-built system performance (the model parameters are calculated on the actual building energy use) allowing often a more accurate energy savings prediction of retrofit actions respect to the forward approach. In this paper, a description of the progress evolution of the two approaches, from the beginning to nowadays, is presented. A critical analysis is carried out to evaluate the advantages and disadvantages of every single method with respect to the energy analysis purposes.