Assessing the performance of a Trombe wall enhanced with phase change material using deep learning

Nowadays, proper determination of the thermal efficiency of new building envelope solutions focusing on energy efficiency is vital for effective energy management. Determining the thermal efficiency of thermal storage (Trombe) wall modified with phase change material (TWPCM) is challenging, and its inaccurate estimation may lead to unnecessary waste of resources, failures, and financial losses. The aim of this work is to develop a reliable deep learning prediction model to determine the thermal efficiency of the TWPCM. The performance of the proposed Convolutional Neural Network combined with Long Short-Term Memory (CNN + LSTM) was compared with seven other developed machine learning models. Eight input variables were used: outdoor air-dry bulb temperature, relative humidity, wind speed, wind direction, total solar radiation intensity on the horizontal surface, direct solar radiation intensity on the horizontal surface, and time of day and year. Input variables from the last 240 h were input data for the models. A model consisting of 4 LSTM layers, 5 CNN layers joined together with fully connected layers was used. The models were trained, tested, and validated in the data set from real-world energy performance data. The CNN + LSTM model was found to outperform the other models with the highest determination coefficient (0.99891) and the lowest mean absolute error (0.19188 W/m²) and root mean square error (0.26324 W/m²). The results show that the proposed deep learning model (1) effectively predicts the thermal behavior of TWPCMs by taking into account heat storage capacity of phase change materials, (2) has very good generalization ability verified on a new data set, (3) enables comparison of results with other building envelopes under typical conditions, e.g. in relation to a Typical Meteorological Year (TMY), by forecasting using weather data from a TMY, and (4) enables a reduction in the time required for direct testing, thus reducing the cost of the analysis.