Metamodel Development to Predict Thermal Loads for Single-family Residential Buildings.

Several equitable approaches have been proposed to reduce world energy consumption against a backdrop of a growing global climate crisis. Among these, we can mention the attempts to improve the energy use of household appliances and utilities, such as air conditioners. One of the strategies used to reduce these devices’ unnecessary energy consumption is estimating the thermal variation in the environments, especially still during their design phase. One of the most advanced methods for this estimation uses computer simulations, which require a high level of technical knowledge. For that, a relatively simple alternative is the creation of metamodels. This work compares two machine learning approaches for developing a metamodel capable of estimating the thermal load in single-family buildings. The metamodels evaluated were the Artificial Neural Networks and the Gradient Boosting Machine. The results obtained made it possible to observe a better performance in the Gradient Boosting Machine approach indicators in relation to Artificial Neural Networks. The negative point is that Gradient Boosting Machine requires a relatively long training time, making its use in routine projects less feasible.