Artificial neural network prediction models of stratified thermal energy storage system and borehole heat exchanger for model predictive control

Abstract

We present a method for constructing artificial neural network (ANN) models of the stratified chilled water thermal energy storage (TES) system and borehole heat exchanger (BHE) of a ground-source heat pump (GSHP) system to assess the feasibility of using ANNs in model predictive control (MPC) applications. In the MPC technique, prediction models are required to describe the system being studied, and ANNs have been used to emulate the system of late. However, the training dataset and structure for ANNs should be constructed with care since incorrect training may lead to prediction errors. This work involved performing case studies on different combination of input parameters of training dataset and the ANN structure for modeling the stratified TES tank and BHE. The suitability of the ANNs of the TES system trained using the simulation results of a physical model and that of the model of the BHE trained using the results of a numerical simulation were assessed. The trained ANNs were evaluated based on the coefficient of determination (R2), root mean square error, and coefficient of variation of the root mean square error. Selected ANNs showed a high prediction accuracy for both systems, and the speed of model run was significantly improved.