Trombe wall thermal performance: Data mining techniques for indoor temperatures and heat flux forecasting

•Trombe wall thermal performance was predicted using indoor temperature and heat flux as outputs.•Data mining process was performed applying ANN, SVM and MLR models.•The results revealed high accuracy by the three models for Ti and HF forecasting.•The capacity of ANN and SVM to predict Ti and HF is very similar while MLR presents more adequacy for Ti forecasting.•More input variables are required for HF prediction. Building sector is responsible for the majority of energy consumption in the world, becoming priority target in energy efficiency policies. The integration of bioclimatic solutions combined with energy use prediction models will allow to achieve more energy efficient and sustainable buildings. Trombe wall is a passive solar system that uses a renewable energy source to improve building's energy efficiency by reducing heating demand. Although prediction models of energy use in buildings have received a remarkable attention from the scientific community as an approach to reduce energy consumption and environmental impacts, no similar applications were identified for the particular case of Trombe walls. In this work, Trombe wall thermal performance was predicted for different data set combinations, considering indoor temperature (Ti) and heat flux (HF) as output variables. Data mining process was performed applying artificial neural networks (ANN), support vector machines (SVM) and multiple linear regressions (MLR) algorithms. The results revealed high accuracy by the three models for Ti and HF forecasting. The capacity of ANN and SVM models to predict Ti and HF is very similar while MLR model presents more adequacy in the case of Ti forecasting. It was also concluded that a high number of input variables will improve the model’s prediction capacity. However, more input variables are required for HF than to Ti prediction. Furthermore, the inclusion of air layer temperature (Tca) or the massive wall outer surface temperature (Tsupe) as input variables strongly improves the capacity of Ti predictors, especially ANN and SVM models, while the massive wall inner surface temperature (Tsupi) will lead to a better accuracy of MLR model for HF forecasting. The interconnections established between the input and output variables for different data set combinations will contribute to optimize the Trombe wall thermal performance and to define the algorithms that will support the operating modes of an automation and control system.