Abstract

This work proposes a data-driven decision-making approach to develop a smart avatar that allows for evaluating the thermal comfort experienced by a user in Chile. The ANSI/ASHRAE 55-2020 standard is the basis for the predicted mean vote (PMV) comfort index, which is calculated by a random forest (RF) regressor using temperature, humidity, airspeed, metabolic rate, and clothing as inputs. To generate data from four cities with different climates, a 3.0 m x 3.0 m x 2.4 m shoe box with two adiabatic walls was modeled in Rhino and evaluated using Grasshopper's ClimateStudio plugin based on Energy Plus+. Long short-term memory (LSTM) was used to forecast the PMV for the next hour and inform decisions. A rule-based decision-making algorithm was implemented to emulate user behavior, which included turning the air conditioner (AC) or heater ON/OFF, recommendations such as dressing/undressing, opening/closing the window, and doing nothing in the case of neutral thermal comfort. The RF regressor achieved a root mean square error (RMSE) of 0.54 and a mean absolute error (MAE) of 0.28, while the LSTM had an RMSE of 0.051 and an MAE of 0.025. The proposed system was successful in saving energy in Calama (31.2%), Valparaiso (69.2%), and the southern cities of Puerto Montt and Punta Arena (23.6%), despite the increased energy consumption needed to maintain thermal comfort.