OPTIMAL ROUTE DEFINITION IN THE RAILWAY INFORMATION NETWORK USING NEURAL-FUZZY MODELS

Abstract

Purpose. Modern algorithms for choosing the shortest route, for example, the Bellman-Ford and Dijkstra algorithms, which are currently widely used in existing routing protocols (RIP, OSPF), do not always lead to an effective result. Therefore, there is a need to study the possibility of organizing routing in in the railway network of information and telecommunication system (ITS) using the methods of artificial intelligence. Methodology. On the basis of the simulation model created in the OPNET modeling system a fragment of the ITS railway network was considered and the following samples were formed: training, testing, and control one. For modeling a neural-fuzzy network (hybrid system) in the the MatLAB system the following parameters are input: packet length (three term sets), traffic intensity (five term sets), and the number of intermediate routers that make up the route (four term sets). As the resulting characteristic, the time spent by the packet in the routers along its route in the ITS network (four term sets) was taken. On the basis of a certain time of packet residence in the routers and queue delays on the routers making up different paths (with the same number of the routers) the optimal route was determined. Findings. For the railway ITS fragment under consideration, a forecast was made of the packet residence time in the routers along its route based on the neural-fuzzy network created in the MatLAB system. The authors conducted the study of the average error of the neural-fuzzy network`s training with various membership functions and according to the different methods of training optimization. It was found that the smallest value of the average learning error is provided by the neuro-fuzzy network configuration 3–12–60–60–1 when using the symmetric Gaussian membership function according to the hybrid optimization method. Originality. According to the RIP and OSPF scenarios, the following characteristics were obtained on the simulation model created in the OPNET simulation system: average server load, average packet processing time by the router, average waiting time for packets in the queue, average number of lost packets, and network convergence time. It was determined that the best results are achieved by the simulation network model according to the OSPF scenario. The proposed integrated routing system in the ITS network of railway transport, which is based on the neural-fuzzy networks created, determines the optimal route in the network faster than the existing OSPF routing protocol. Practical value. An integrated routing system in the ITS system of railway transport will make it possible to determine the optimal route in the network with the same number of the routers that make up the packet path in real time.