Implementation of adaptive neuro-fuzzy inference system in design and process optimization of latent heat storage system

Abstract

This study focuses on implementing an Adaptive Neuro-Fuzzy Inference System (ANFIS) as an artificial neural network model to optimize latent heat thermal energy storage (LHTES) systems. Experimental data comprising seven input parameters (length, shell diameter, tube diameter, mass flow rate, difference in temperature, latent heat, and density) and two output parameters (charging and partial charging time) are collected and used to train the ANFIS model. Initially, the model is developed using a grid partition method, employing Triangular and Gbell membership functions and assigning diverse membership rules. This initial model demonstrated satisfactory accuracy based on quality metrics (R2, mean absolute, and root mean square error). Then, a subtractive clustering approach is applied to further enhance the model's accuracy, which groups different data clusters together. This enhanced model exhibited superior performance compared to the grid partition method, making it the optimal choice. In addition, an experimentally validated numerical model is developed using COMSOL Multiphysics and is utilized to crosscheck the predicted output of the ANFIS model. From a set of 100 combinations of storage capacity between 0.45 and 0.55 MJ, a fixed latent heat storage capacity of 0.456 MJ was identified by considering the combination with the least charging time based on the input parameters. Using this storage capacity, a minimum charging time of 223 min and a partial charging time of 180 min was determined. The numerical model also confirmed these optimal values.