Performance of ANN-based predictive and adaptive thermal-control methods for disturbances in and around residential buildings

This study aimed at investigating an advanced residential thermal-control method through developing ANN (Artificial Neural Network)-based thermal-control logics and investigating their adaptabilities to diverse variables occurring in/around residential buildings. Via performance tests for variables that residential buildings may experience during their life span such as variation in infiltration rate, variation in internal loads, and variation in climate conditions, proposed logics proved their adaptabilities to the disturbances of building-related backgrounds and stabilities to create comfortable thermal conditions. For this purpose, four different types of control logic were developed: temperature and humidity control without ANNs, as with conventional strategy; temperature and humidity control with ANNs; PMV (Predicted Mean Vote) control without ANN; and PMV control with ANN. A sliding-window method was adopted as a training data-managing technique for conducting iterative self-tuning process. Incorporating simulation tools IBPT (International Building Physics Toolbox) and MATLAB (matrix laboratory), performances of developed logics were tested in a typical U.S. detached single-family house. Analysis proved the ANN models’ adaptabilities to disturbances with the accuracy of prediction; improved thermal comfort; and decreased over- and undershoots of thermal factors. Based on this improved performances, the proposed ANN-based predictive and adaptive residential thermal-control methods have shown their potentials as advanced methods for a detached, single-family house.

► Adaptabilities of ANN-based residential thermal-control logics were studied. ► Logic performance was tested for disturbances occurring around residential buildings. ► Computer simulation methods (IBPT and MATLAB) were incorporated for testing. ► ANN models proved their adaptabilities by thermal comfort, accuracy and stability. ► By this, ANN-based methods show potentials as an advanced thermal-control method.