Artificial intelligence approaches to predict thermal behavior of light earth cell incorporating PCMs: Experimental CNN and LSTM validation

The dynamic modeling of heat transfer in bio-based materials is a challenging endeavor due to the wide variety of factors that influence the thermal behavior of bio-based materials with Phase Change Materials (PCMs). There are a lot of factors related to heat transfer with phase change that are sometimes beyond our control. Since the phase change response of bio-based walls with PCM is significantly non-linear, in this paper, we present the thermal behavior prediction using artificial neural networks. Based on in-situ study of light earth incorporating PCMs, the effectiveness and training of Long-Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) are assessed. According to the findings, the results showed that these models can predict precisely the heat transfers in light earth with or without phase change materials. Compared to the LSTM, it was found that the CNN model is more accurate in predicting processes involving phase shift, such as the one presented by the light earth incorporating PCMs. •Hygrothermal behavior prediction of bio-based materials incorporating phase change materials.•Comparing Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) for hygrothermal behavior prediction.•CNN and LSTM are able to predict the combined heat (phase change material) and mass transfer.•CNN model outperforms LSTM considerably when predicting PCM- hygroscopic materials behavior.