Design and selection of suitable sustainable phase change materials for latent heat thermal energy storage system using data-driven machine learning models

The present study aims to develop and implement data-driven machine learning (ML) models for performance prediction of heat flow and specific heat of sustainable composite phase change materials (SCPCMs). The implementation of ML models is being investigated for the first time, though the usage of PCMs has been studied in many applications. In this work, five ML models, namely decision tree regression (DTR), k-nearest neighbour (k-NN), random forest regression (RFR), extreme gradient boosting regression (XGBR), and cat boost regression (CBR), are considered for predicting the heat flow and specific heat of SCPCMs. A total of 14,303 data points for heat flow and 9059 data points for specific heat are considered. Five input parameters are considered: concentration of PCM, the concentration of biochar, concentration of multi-walled carbon nanotubes (MWCNT), heating rate of the sample, and temperature of the sample. The output parameters are heat flow (mW mg(-1)) and specific heat (J g(-1) & DEG;C-1). From the results of performance predictions, the k-NN model exhibited the best coefficient of determination of 0.997 and 0.994 for heat flow and specific heat, respectively, among its peers. A model sensitivity analysis for heat flow prediction is performed and found that the errors between the actual and predicted values are 1.79%, 3.41%, 1.16%, 14.95%, and 1.66% for RFR, DTR, k-NN, XGBR, and CBR, respectively. Similarly, for the specific heat prediction, the error between the actual and predicted values is 0.687%, 0.99%, 0.37%, 10%, and 0.44% for RFR, DTR, k-NN, XGBR, and CBR, respectively. Thus, the developed data-driven machine learning models can be graded as k-NN > CBR > RFR > DTR > XGBR, based on their prediction accuracy and are found to be helpful in the selection of suitable sustainable PCMs for latent heat thermal energy storage systems.