Sławomir Drapik

The article presents some of the results of measurements of loads for a selected tram traction substation for a continuous period of time covering sixteen weeks (112 days). Particular attention was paid to overloads occurring in consequent days. The second part of the article presents the analysis of overloads relating to the time interval of 60 minutes in successive days. This analysis was implemented in Matlab using a two-layer feedforward artificial neural network (ANN). The results of the approximation of the analyzed overloads are promising. A continuation of research may lead to the formulation of mathematical equations that might be useful for designers in terms of sufficiently precise calculations of overloads of rectifier units of DC traction substations.

The paper presents the load currents of two tram traction substations. These substations power lines with different characteristics of tram traffic. Loads were analysed for one working day and one public holiday. Attention was paid to the differences and similarities in the current waveforms. The demand level was assessed, with respect to the rectifier units in the substations. The time series method was used, inter alia, and chiefly an analysis of the characteristics of autoregression and partial autoregression was carried out, to act as a diversity index with respect to the nature of variability in the analysed traction load currents.

W artykule przedstawiono kolejne wyniki badań zmienności obciążeń zespołów prostownikowych wybranej tramwajowej podstacji trakcyjnej. Do analiz wykorzystano rzeczywiste wyniki pomiarów. Tym razem zwrócono uwagę na specyfikę maksymalnych obciążeń i przeciążeń w odcinkach czasowych 5 min i 60 min – dla kilku wybranych przypadków. W drugiej części artykułu przedstawiono efektywność wykorzystania sztucznych sieci neuronowych, typu feedforward o jednej warstwie ukrytej z ośmioma neuronowami, do analizy przeciążeń dla eksploatacji podstacji trakcyjnej w dłuższej skali czasowej. Uzyskane pozytywne wyniki wskazują na konieczność kontynuowania tego typu badań, m.in. wykorzystując inne warianty sztucznych sieci neuronowych. 

This paper presents further results of research on the load variability of rectifier units for the selected tram traction substation. Actual measurements were used in the performed analysis. This time, the analysis was focused on the characteristics of maximum loads and overloads for time periods of five minutes and sixty minutes, for a number of selected cases. The second part of the article discusses the effectiveness of the use of artificial neural networks of the feedforward type with one hidden layer with eight neurons to analyse the overloads of the traction substation over a longer time scale. The obtained positive results indicate that this type of research should be continued, using different variants of artificial neural networks.