A deep learning method to predict outdoor temperature profile of pavement bricks with phase change materials

Urban heat islands are one of the common issues of the 21st century. Phase change materials could potentially minimise the impact of the urban heat island. There is growing interest in using phase change material (PCM) incorporated bricks for building materials and the results so far have been promising. The objective of this study was to evaluate the temperature profiles of phase change materials, develop a corresponding machine learning dataset, and design a deep learning method to predict temperature profiles from PCM variables. The phase change materials selected in this study were lauric acids, stearic acids, and paraffin wax. The normalized data was split into a training set and a test set with a ratio of 9:1. Further to this, for the purpose of training the model, the original heatmap images were resized to a dimension of 160 x 120 pixels. The mean squared error between heat maps from the test set and corresponding predictions was computed, yielding an error of 0.0465, which is consistent with the viability of the approach. This work demonstrates that deep learning models based on fully connected and convolutional layers are capable of accurately predicting heat maps from PCM variables, even when relatively small datasets are used, and therefore that they are a good place to start for authors conducting similar work.