Stefan Čubonović
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Stefan Čubonović has a total of 5 co-authors. Here are the top co-authors by number of shared publications:
- Aleksandar Ranković
Faculty of Technical Sciences Čačak, University of Kragujevac, 32102 Čačak, Serbia
4 shared publications - Marko Krstić
University of Belgrade - School of Electrical Engineering
2 shared publications - Dragan Ćetenović
Faculty of Technical Sciences, Čačak
1 shared publication Fakultet Tehničkih Nauka Univerzitet U Kragujevcu
1 shared publicationPublic Enterprise „Elektroprivreda Srbije“
1 shared publication
Latest Publications
- 5 December 2025
- Published by University Library in Kragujevac in Energija Ekonomija Ekologija
- Vol. 27 (4)
Improvement of Power Output Prediction in Hydroelectric Plants through Cluster Analysis of Data and a Combined Artificial Neural Network Model
This research continues previous efforts invested in the application of Artificial Neural Networks (ANN) for predicting the output power of hydroelectric power plants, aiming for further improvements in the accuracy and robustness of the models. The paper focuses on the implementation and evaluation of a combined model that utilizes cluster analysis of water flow through the turbine data with a set threshold and two ANN models: Feedforward Artificial Neural Network (FF-ANN) and Recurrent Neural Network (RNN). Cluster analysis of data with a set threshold, defined based on water flow stability, was applied to select one of the two proposed ANN models. This approach ensured confidence in the model's accuracy and identified specific operational conditions of the hydroelectric power plant. The analysis includes data collected during various hydrological cycles in one year, testing the adaptability of the proposed models. The results of this research promise significant improvements in the efficiency of the previously created model for predicting the power output of hydroelectric power plants.
- 5 December 2025
- Published by University Library in Kragujevac in Energija Ekonomija Ekologija
- Vol. 27 (4)
Improvement of Power Output Prediction in Hydroelectric Plants through Cluster Analysis of Data and a Combined Artificial Neural Network Model
This research continues previous efforts invested in the application of Artificial Neural Networks (ANN) for predicting the output power of hydroelectric power plants, aiming for further improvements in the accuracy and robustness of the models. The paper focuses on the implementation and evaluation of a combined model that utilizes cluster analysis of water flow through the turbine data with a set threshold and two ANN models: Feedforward Artificial Neural Network (FF-ANN) and Recurrent Neural Network (RNN). Cluster analysis of data with a set threshold, defined based on water flow stability, was applied to select one of the two proposed ANN models. This approach ensured confidence in the model's accuracy and identified specific operational conditions of the hydroelectric power plant. The analysis includes data collected during various hydrological cycles in one year, testing the adaptability of the proposed models. The results of this research promise significant improvements in the efficiency of the previously created model for predicting the power output of hydroelectric power plants.
- 1 January 2025
- Published by Akademska Misao
- p. 91-96
Unapređenje predikcije snage na izlazu hidroelektrane primenom sezonske segmentacije i SOM-metoda klasterizacije podataka
Ovaj rad predstavlja nastavak prethodnog istraživanja u primeni veštačkih neuronskih mreža (Artificial Neural Networks – ANN) za predikciju snage na izlazu hidroelektrane (HE), korišćenjem dva tipa ANN: veštačke neuronske mreže sa jednosmernim prostiranjem signala (Feedforward Artificial Neural Networks – FF-ANN) i rekurentne neuronske mreže (Recurrent Neural Network – RNN). U novom pristupu povećan je obim podataka za testiranje modela na celu godinu. Godišnji skup podataka podeljen je u četiri sezonske celine po meteorološkom kriterijumu – proleće, leto, jesen i zima – pri čemu je za svaku sezonu razvijen poseban adaptivni model ANN, sa dodatnom podelom podataka na osnovu dinamike protoka. Kako bi se unapredila preciznost segmentacije radnih režima, u radu je takođe primenjena Self-Organizing Map (SOM) neuronska mreža za klasterizaciju podataka, čime je omogućeno bolje prilagođavanje ANN modela različitim uslovima rada. Cilj rada je dodatno unapređenje preciznosti i robusnosti modela kako bi se omogućila bolja prilagodljivost specifičnim hidrološkim i operativnim uslovima svake sezone. Rezultati testiranja pokazuju da sezonski i SOM modeli nadmašuju model treniran na ograničenom periodu, po pitanju preciznosti predviđanja. Na ovaj način doprinosi se poverenju u kreirane modele i otvaraju se mogućnosti za njihovo unapređenje.
- 1 January 2025
- Published by Akademska Misao
- p. 91-96
Unapređenje predikcije snage na izlazu hidroelektrane primenom sezonske segmentacije i SOM-metoda klasterizacije podataka
Ovaj rad predstavlja nastavak prethodnog istraživanja u primeni veštačkih neuronskih mreža (Artificial Neural Networks – ANN) za predikciju snage na izlazu hidroelektrane (HE), korišćenjem dva tipa ANN: veštačke neuronske mreže sa jednosmernim prostiranjem signala (Feedforward Artificial Neural Networks – FF-ANN) i rekurentne neuronske mreže (Recurrent Neural Network – RNN). U novom pristupu povećan je obim podataka za testiranje modela na celu godinu. Godišnji skup podataka podeljen je u četiri sezonske celine po meteorološkom kriterijumu – proleće, leto, jesen i zima – pri čemu je za svaku sezonu razvijen poseban adaptivni model ANN, sa dodatnom podelom podataka na osnovu dinamike protoka. Kako bi se unapredila preciznost segmentacije radnih režima, u radu je takođe primenjena Self-Organizing Map (SOM) neuronska mreža za klasterizaciju podataka, čime je omogućeno bolje prilagođavanje ANN modela različitim uslovima rada. Cilj rada je dodatno unapređenje preciznosti i robusnosti modela kako bi se omogućila bolja prilagodljivost specifičnim hidrološkim i operativnim uslovima svake sezone. Rezultati testiranja pokazuju da sezonski i SOM modeli nadmašuju model treniran na ograničenom periodu, po pitanju preciznosti predviđanja. Na ovaj način doprinosi se poverenju u kreirane modele i otvaraju se mogućnosti za njihovo unapređenje.
- 1 January 2024
- Published by University Library in Kragujevac in Energija Ekonomija Ekologija
- Vol. XXVI (2) , 47-54
The Application of Artificial Neural Networks for Predicting the Power Output of a Hydroelectric Power Plant
The research paper analysed two types of artificial neural networks (ANN) for predicting the power output of a hydroelectric power plant. The first type is the feedforward artificial neural network (FF-ANN), while the second type is the recurrent neural network (RNN). Authors thoroughly analysed the steps taken in the process of implementing neural networks for this purpose, from data collection and preparation to training, evaluation, and results analysis. The input data includes net head, water flow through the turbine, upper water level, lower water level, and the temperature of the inlet cooling water. Based on the correlation coefficient of individual variables from the input layer with the output power, new ANN implementations were conducted. The results obtained from these variations were systematically analysed to achieve precise modelling, focusing on dynamic changes in water flow through the turbine.
- 1 January 2024
- Published by University Library in Kragujevac in Energija Ekonomija Ekologija
- Vol. XXVI (2) , 47-54
The Application of Artificial Neural Networks for Predicting the Power Output of a Hydroelectric Power Plant
The research paper analysed two types of artificial neural networks (ANN) for predicting the power output of a hydroelectric power plant. The first type is the feedforward artificial neural network (FF-ANN), while the second type is the recurrent neural network (RNN). Authors thoroughly analysed the steps taken in the process of implementing neural networks for this purpose, from data collection and preparation to training, evaluation, and results analysis. The input data includes net head, water flow through the turbine, upper water level, lower water level, and the temperature of the inlet cooling water. Based on the correlation coefficient of individual variables from the input layer with the output power, new ANN implementations were conducted. The results obtained from these variations were systematically analysed to achieve precise modelling, focusing on dynamic changes in water flow through the turbine.
- Open Access
- 1 January 2024
- Published by National Library of Serbia in Serbian Journal of Electrical Engineering
- Vol. 21 (1) , 113-133
Impact of the non-Gaussian measurement noise on the performance of state-of-the-art state estimators for distribution systems
This paper aims to investigate the impact of non-Gaussian measurement noise on state estimation (SE) results in distribution systems. To this end, the measurement noise is assumed to be distributed according to Gaussian or one of the following non-Gaussian probability distribution functions: Uniform, Laplace, Weibull and Gaussian mixture of two Gaussian components. The influence is investigated on three different state-of-the-art SE methods: weighted least squares (WLS) based static SE method, and two Kalman filter based forecasting-aided SE methods, namely extended Kalman filter (EKF) and unscented Kalman filter (UKF). Analyses are conducted on modified IEEE 37-bus system under different operating conditions, including quasi-steady state, sudden state changes and bad data. Performance of the methods in the presence of non-Gaussian measurement noise is compared against their performance when measurement noise is Gaussian distributed. The main conclusions were drawn, summarizing the impacts non-Gaussian measurement noise has on SE and proposing the solutions for overcoming some of the negative impacts.
- 1 January 2024
- Published by National Library of Serbia in Serbian Journal of Electrical Engineering
- Vol. 21 (1) , 113-133
Open AccessImpact of the non-Gaussian measurement noise on the performance of state-of-the-art state estimators for distribution systems
This paper aims to investigate the impact of non-Gaussian measurement noise on state estimation (SE) results in distribution systems. To this end, the measurement noise is assumed to be distributed according to Gaussian or one of the following non-Gaussian probability distribution functions: Uniform, Laplace, Weibull and Gaussian mixture of two Gaussian components. The influence is investigated on three different state-of-the-art SE methods: weighted least squares (WLS) based static SE method, and two Kalman filter based forecasting-aided SE methods, namely extended Kalman filter (EKF) and unscented Kalman filter (UKF). Analyses are conducted on modified IEEE 37-bus system under different operating conditions, including quasi-steady state, sudden state changes and bad data. Performance of the methods in the presence of non-Gaussian measurement noise is compared against their performance when measurement noise is Gaussian distributed. The main conclusions were drawn, summarizing the impacts non-Gaussian measurement noise has on SE and proposing the solutions for overcoming some of the negative impacts.