Modeling of water-PCM solar thermal storage system for domestic hot water application using Artificial neural networks

•Artificial Neural Network model of solar thermal storage for hot water application.•Framework to develop and optimize the artificial neural network model parameters.•Validated numerical model used to train and test the artificial neural network model.•Artificial neural network model outperforms traditional regression modeling.•Neural network model yields five orders improvement in the computational time.•Optimized model-generated design map for the solar thermal storage system. Numerical modeling of solar thermal storage systems is often challenged with limitations on the computational effort due to their transient non-linear behavior that dictates accurate modeling of the physics over long-term operations (i.e., annual). Such challenges result into scarcity of literature on comprehensive design guidelines for solar thermal storage systems. This study presents a framework through which the potential of artificial neural network (ANN) modeling of a hybrid solar thermal storage system involving phase change materials is extensively investigated. An experimentally validated numerical model for the system is used to generate the training and testing datasets for the ANN model. The effect of changing the sampling method and the number of training samples is studied on the ANN model prediction. The results show that Sobol sequence sampling is superior to other sampling methods especially for low number of samples. The best sampling method is utilized to generate the training dataset with which the hyperparameters of the learning algorithm are optimized. The optimized ANN model is ultimately used to predict the system solar fraction under various design conditions to develop design maps that offer better visualization and sizing guidelines for the hybrid solar thermal storage systems. ANN is shown to offer a potential candidate for accurate and computationally efficient modeling of complex thermal systems. Upon proper configuration and training, the ANN model can accurately (i.e., a coefficient of determination of 0.9999) predict the performance of the hybrid thermal storage system with approximately five orders of magnitude reduction in computational time compared to conventional numerical models.