Learning based personalized energy management systems for residential buildings

Development of a Grey Box Thermal Dynamic Model without Building Construction Knowledge

“…HVAC systems are one of the largest contributors to the overall energy consumption of commercial building's [2]. The implementation of advanced and predictive control systems has significant potential in reducing energy consumption, with energy savings as high as 25% [13].…”

“…Similarly, a single room test bed study conducted by Soudari et al [13] investigated the use of personalized energy management systems for HVAC control based on economic MPC in residential buildings. Their research concluded energy savings between 9.7% and 25% and cost savings between 8.2% and 18.2% can be achieved using this system, depending on the occupant behavior and external conditions [13]. Economic MPC systems have also been researched within commercial buildings, as shown in the study of a single storey commercial building in Chicago Illinois by Ma et al [27].…”

“…In addition to the ambient temperature measurements, grey box models also collect surface temperature measurements for parameter estimation or tuning purposes [13,18]. This is seen in the work of Terés-Zubiaga et al [18] where the surface temperature measurements are used to adjust parameters in the grey box model to achieve more accurate surface temperature results.…”

“…1-2005 standards [10]. More complex occupancy data collection can be seen in [13] with limit switches measuring door-firings, along with, retro-reflective sensors and door-mat sensors measuring occupant movement to infer room occupancy. In other studies, carbon dioxide sensors were used to determine the occupancy based on the carbon dioxide ppm concentrations within the room [23,31,32].…”

“…Previous research has solved this issue by using parameter estimation techniques to determine the resistance and capacitance values [5,6,8,10,11,12,13,14,15,16,17]. The parameter estimation techniques used within these studies include Matlab Optimization Toolbox using hill climbing algorithm [10], Matlab Optimization Toolbox using interior point algorithm [11], Matlab greyest function [12], and linear fitting using least squares method [13]. The study conducted by Berthou et al [11] used an interior point algorithm to solve for the RC parameters in the grey box model, resulting in an ambient temperature prediction error of less than 2% for both heating and cooling seasons.…”