Author(s):
1. Miloš Simonović, Faculty of Mechanical Engineering in Niš, Serbia
2. Vlastimir Nikolic, Faculty of Mechanical Engineering in Niš, Serbia
3. Ivan Ćirić, Faculty of Mechanical Engineering in Niš, Serbia
4. Emina Petrovic, Faculty of Mechanical Engineering in Niš, Serbia

Abstract:
District heating companies have growing and significant need for improving economic and energy efficiency. Also, they have a challenge to keep the cost of produced and delivered heating energy as lower as possible. That is why it is very important to optimize production of heating energy using better prediction and control of customer needs. In this paper, the focus is on short-term prediction. Real historical data are used from city of Nis, southeastern Serbia, heating plant Krivi vir, 128 MW installed power. This prediction is particularly important for heating in transient regimes which unlike the standard heating regime does not have continuous supply of heating energy throughout the specified heating time period. An application of neural networks is realized based on original historical data of heating source by using recurrent neural network to fulfill demands on variation in ambient temperature during a heating day and satisfied results are obtained.

Key words:
district heating systems, recurrent neural network, short-term prediction, energy efficiency

Thematic field:
Energy and Thermal Engineering

Date of abstract submission:
16.03.2015.

Conference:
12th International conference on accomplishments in Electrical and Mechanical Engineering and Information Technology (DEMI 2015)

Final paper