Transactions of FAMENA, Vol. 50 No. 2, 2026.
Original scientific paper
https://doi.org/10.21278/TOF.502080025
Application of Artificial Neural Networks to Assess the Trend of Energy Storage Control Variables in a Multi-Storage Energy System
Filip Milešević
orcid.org/0009-0008-4197-1395
; University of Zagreb, Faculty of Mining, Geology and Petroleum Engineering, Zagreb, Croatia
Luka Perković
orcid.org/0000-0002-3273-6333
; University of Zagreb, Faculty of Mining, Geology and Petroleum Engineering, Zagreb, Croatia
*
* Corresponding author.
Abstract
The aim of this paper is to present the possibilities of using Artificial Neural Networks (ANNs) for the smart management of energy storage in systems with a high share of renewable energy sources. The overall method consists of two separate parts. First, optimal energy flows are simulated in EnergyPLAN, where the management of the energy system, including five energy storage systems, is optimised for daily and seasonal management. EnergyPLAN here represents an expert system from which the ANN model learns optimal energy system management. The outputs are then used in the second part which is training, validating, and testing ANNs in PyTorch for the reproduction of energy storage management with respect to the same set of input data. Input variables are related to intermittent, but predictable, supply from the wind and solar insolation, seasonal demand for grid gas, district heating demand, as well as electricity demand for BEVs. Output variables are the charge and discharge signals of all five energy storage systems, as well as the control variables of other components of the energy system. Several configurations of ANNs were tested, showing that ANNs can obtain statistically significant results, achieving an overall R2 score of 0.8 in the prediction of energy flows in all five storage systems. The results show that better performance is achieved if residual connection blocks are included in the ANN architecture and better agreement if physical constraints are integrated in the model training loss.
Keywords
Artificial Neural Network (ANN); EnergyPLAN; smart energy; energy storage; renewable energy
Hrčak ID:
346077
URI
Publication date:
9.4.2026.
Visits: 316 *