Deploying Machine Learning (ML) for Improving Reliability and Resiliency of Thermal Energy Storage (TES) Platforms by Leveraging Phase Change Materials (PCM) for Sustainability Applications and Mitigating Food-Energy-Water (FEW) Nexus

In this paper, machine learning (ML) techniques, more specifically artificial neural networks (ANN), are utilized to enhance the efficacy of Cold Finger Technique (CFT). Experiments were conducted by melting the PCM at different values of power input to an electrical heater (mounted at the base of the container and immersed in PCM). Temperature transients were recorded by three thermocouples that were mounted at locations corresponding to liquid-meniscus heights for melt fraction values of 30%, 60% and 85%. The surface temperature transients were measured using thermocouples mounted on the exterior of the container surface that were mounted at locations corresponding to liquid-meniscus heights for melt fraction values of 30%, 60% and 90%. The surface temperature transients afford a cheap, reliable and cost-effective option for predicting the required values in real-time (i.e., the time remaining to attain a desired melt fraction, say 85%, at any particular instant during the melting cycle). These results validated the approach reported by (Chuttar et al. 2022). The average prediction error in the last half hour (before reaching a target melt fraction of 85%) was less than 10 minutes for all but one of the datasets. The Mean Absolute Percentage Error (MAPE) was as low as 11% for some of the predicted values of the datasets.