The task of resource saving is relevant for all transport companies and many world communities of scientists and engineers are engaged in search for ways to solve it. World railway companies, and especially large ones such as Russian Railways, are large consumers of energy resources and the problem of saving is the most urgent for them. One of directions for solving this task can be the use of an optimal energy regulation system for traction of each train – an operational rationing system. Such a task can only be solved by modelling the process of movement through the dynamic programming method. In modern conditions of development of engineering and technology, it has become possible to develop such operational standardization systems endowed with important properties: high performance, multitasking, solution accuracy, ease of use and maintenance. These requirements impose certain restrictions on the architecture of the operational rationing system. Typical system architecture should be built around a centralized node, which will act as a solver and storage, nodes for input and output of information can be geographically separated. The method of dynamic programming can be improved by using it in the process of training artificial neural networks, which will form not only a priori estimates of energy consumption for traction, but also an a posteriori estimate of train control (by a train driver or auto-driving system). Also, the use of artificial neural networks will allow us to continuously improve the method due to training using the accumulated amount of data from real trips, which will allow us to clarify the norms of energy consumption and to plan our costs in the future. The prototype of the operational standardization system was developed at the department of traction rolling stock of Russian Open Academy of Transport of Russian University of Transport and the results obtained allow us to state that the chosen approach to solving the problem of energy saving has been chosen correctly.
Control of a Buck DC/DC Converter Using Approximate Dynamic Programming and Artificial Neural Networks
