Learning Models - A New approch To Simulation

In conducting and teaching Building Simulation, we often find two main disadvantages of conventional models: inconsistency of simulation results obtained by different users of the same model, and long machine times required for annual simulations of relatively simple buildings. In searching for better simulation methods, we decided to depart from the conventional method and to introduce machine learning into mathematical modelling of buildings. This resulted with a new model, based on learning of building energy properties from monitored data. The model was named LEARNSIM LEARNing SImulation Model. A comparative testing of LEARNSIM and of a conventional model confirmed higher accuracy and faster operation of the former. The aim of the paper is to present the main features of learning simulation models, to demonstrate how they can be used at present, and in the future.