Improvement of design model of radiator heating area in low-temperature heating:Application of machine learning techniques

This study proposes a so-called BPNN-based design model for radiator sizing in low-temperature heating by combining machine learning techniques with building simulation. Representative conditions are identified through an orthogonal experimental design to encompass different building characteristics, radiator specifications and heating systems’ operational features in different climatic zones. Simulations conducted under these representative conditions using DeST software provide a comprehensive database detailing minimum indoor temperatures during heating and various influencing factors, including radiator heating area. Simulation results accurately capture the coupled heat transfer between radiators and rooms, as well as the decay and delay effects of thermal disturbances on indoor temperatures due to thermal inertia. Moreover, the inclusion of heating load parameter improves the database structure. BP neural network is used to extract inherent patterns from the database and the BPNN based design model of radiator heating area is established. The error characteristics of this model are evaluated to demonstrate its predictive performance. Furthermore, application analyses are conducted on the actual heating buildings under different climatic zones. Results show that this BPNN-based design model achieves good control accuracy over a range of heating buildings under different climatic zones. Again, the limitations of this model regarding the mass flow parameter are analyzed.