A data‐driven approach for simulating the energy performance of closed‐loop hydronic heating systems

Water supplies in closed-loop hydronic heating systems might contain varying amounts of impurities which can affect system’s energy performance. Additionally, these systems are often poorly maintained, resulting in degraded performance due to problems like corrosion. Traditional first-principle models used to simulate the energy performance of such systems inherently do not account for such problems. This study aims to develop a data-driven model to predict the impact of low-quality water supplies on the energy performance of closed-loop hydronic heating systems.

Controlled experiments in dedicated testing facilities were conducted to collect energy-performance-related data to train four different ML models. For this purpose, an experimental rig was installed in a laboratorial facility to replicate the operation of a closed-loop hydronic heating system and control the parameters that influence its energy performance. The data retrieved were used to train an Artificial Neural Network (ANN), a Random Forest (RF), a Gradient Boosting Machine (GBM), and a Support Vector Machine (SVM) model to benchmark their predictive accuracies.

The RF model demonstrates the lowest errors in predictions with an MSE of 0.65 and 2.36 in the training and validation sets, respectively, followed closely by the ANN with a training MSE of 1.64 and a validation MSE of 2.62. The ANN showed more balanced performance across the training, validation, and testing sets, suggesting better generalization capabilities.