Second law performance prediction of heat pump integrated stratified thermal energy storage system using long short-term memory neural networks

Abstract

Thermal energy storages (TES) are transient state energy devices. These devices are used in renewable energy systems as a buffer for non-coincidence in heat supply and demand. TESs use thermal stratification to ensure high efficiency in heat storage and acquisition. This article is focused on predicting the performance of thermal energy storage (TES) integrated with heat pump using neural networks. In addition, exergy and entropy equations were derived for the calculation and prediction of the stratification efficiency in storage systems and of the performance factor (PF) of renewable energy systems (RES). As for data analytics, real time data-streaming edge devices were customized. The model fitting and prediction were done directly on the edge devices. The key objectives and findings are:•To demonstrate stream-data processing framework which can graphically represent the stratification decay of an active Thermal Energy Storage (TES) charge/discharge process in real time.•Derivation of a custom exergy equation for stratification efficiency and streaming it graphically in real time. The optimized key performance index (KPI) at the heat pump end i.e. coefficient of performance (COP) or performance of factor (PF) was 3.3, and at charge and discharge end, in terms of efficiency was 83 % and 84 % respectively.•A deep neuronal network applying a long short-term memory (LSTM) architecture for predicting stratification deterioration in the charge/discharge cycle with a prediction error below 5 %.