Railway Power Supply Models and Methods for Long-term Investment Analysis

Abstract: The aim of the project is to suggest an investment planning programwhere the welfare of the society is to be maximized. In order to beable to decide on a wise investment plan, one needs to know theconsequences of different choices of power system configurations.Therefore the impacts of different future traffic demands are ofinterest for a railway power system owner.Since investments are supposed to last a long time, their futureusage has to be considered. Moreover, the lead times of investmentscan be of considerable duration lengths. Because of the uncertaintyof the future, deterministic case studies might not be suitable andthen a large number of outcomes are to be studied, probable outcomesas well as outcomes with a high level of impact.In order to be able to make a valid long-term investment analysis ofthe railway power supply system, one needs to use proper railwaypower supply models and methods. The aim of this thesis is topresent a stable modeling and methodological basis for the cominginvestment planning phase of this PhD research project. The focus isset on studying the consequences of a railway power supply systemwhich is too weak.The thesis contains an overview of models of some electrical andmechanical relations important for electric traction systems. Someof these models are further developed, and some are modified forimproved computational properties. A flexible electric tractionsystem simulator based on the above mentioned models has beendeveloped and the applied methods and resulting abilities arepresented.The main scientific contribution of this thesis is that a fast andapproximative neural network model, which calculates some importantaggregated results of the interaction between the railway powersystem and the train traffic, has been developed. This approximativemodel was developed in order to reduce computation times. Reductionof computation times is very important when a huge number ofoutcomes are studied. A complete simulation of a train power systemin operation takes a long time, often not less than about a tenth ofthe simulated traffic time. The neural network is trained with someselected aggregated results extracted from a wide set of railwayoperation simulation cases. The choices of network inputs andoutputs are motivated in the thesis. The performance of thesimulator as well as the approximator are visualized in casestudies.