Optimal Selection of Features for Heat Pump Models based on artificial neural networks

Building Simulation Conference proceedings, vol. 18

Abstract

Exploiting the energy-saving potential of heat pump technology, services like operational optimization or fault detection are subject to current research. In order to optimize the operation of the heat pump, model-based control, which typically utilizes grey- or white-box models, has proven its potential. Disadvantages of physical models, such as the high modeling effort of the refrigeration cycle and the high calculation time during the simulation, can be addressed by methods such as Artificial Neural Networks. However, this black-box modeling approach is based entirely on statistical correlations of measurement data and thus neglects any physical behavior of the plant. Although ANN approaches for heat pumps have already been investigated in the literature, the influence of feature and signal selection on the results still needs to be addressed. Therefore, this paper investigates an ANN model of an air source heat pump. The model is trained and tested via different simulation data sets while the hyper parameters remain constant. With an optimized selection, we can increase the accuracy of ANN-based heat pump models for hourly prediction of COP up to about 45%. At the same time, the reference system represents a neural network with common features. Hence, this work demonstrated that an optimal selection could increase the reliability of ANN-based models for heat pumps and can be used to improve model-based control.

Authors 4

1. Sebastian Borges Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen , Aachen , Germany

   RWTH Aachen, Aachen, Germany

2. Lasse Jöhnk Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen , Aachen , Germany

   RWTH Aachen, Aachen, Germany

3. Christian Vering Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen , Aachen , Germany

   RWTH Aachen, Aachen, Germany

4. Dirk Müller Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen , Aachen , Germany

   RWTH Aachen, Aachen, Germany