Uncertainty Aware Learning Model for Thermal Comfort in Smart Residential Buildings

Smart energy management for homes is a potential area of research. Establishing a trade-off between occupant comfort and energy saving is quite a crucial issue to handle. To address this issue, the paper presents a multi-objective optimisation routine-based approach for handling this conflicting trade-off. Air quality, visual, and thermal comforts are important factors when considering occupant comfort. Heating, ventilation, and air conditioning systems employ the predicted values for comfort temperature and air quality comfort inside the room. The lighting system is used to maintain visual comfort inside the room. A data-driven strategy is put out in this research to forecast user thermal comfort for each occupant in residential homes. Six deep-learning approaches are applied to estimate each occupant's interior comfort temperature. These approaches are evaluated with the help of mean square error (MSE), mean absolute percentage error (MAPE), root mean square error (RMSE), normalised mean square error (NMSE), and R-squared score (R2 score). Bi-directional gated recurrent units (Bi-GRU) have shown to be the best. Bi-GRU outperforms in the following metrics: MSE, MAPE, RMSE, NMSE, and normalised R2 score values are 0.0104, 8.1768, 0.1022, 7.5121, and 1, respectively; these metrics are found optimal when compared with other techniques. The overall system is optimised with the application of particle swarm optimisation. Summer, winter, and monsoon seasons weather data in India are used to test the proposed model. The influence of temperature uncertainty on total energy consumption, overall comfort, and energy consumption cost is also being examined in all three seasons.