Search academic texts

Information × Registration Number 0220U101941, 0119U000258 , R & D reports Title Research of united computer network of railway transport using simulation and neural model popup.stage_title Head Pakhomova Viktoriia M, Registration Date 28-02-2020 Organization Dnipro National University of Railway Transport named after Academician V. Lazaryan popup.description2  OpNet Modeler Modeling System is a simulation model of the unified computer network of the Ukrainian rail transport (simplified fragment) based on Ethernet family technologies, which consists of 44 main stations, which are connected into six subnets: Southwestern; Southern; Lviv; Odessa; Dnieper and Donetsk. The created simulation model conducted research on RIP and OSPF scenarios of the following characteristics: server load; waiting time for packages in the queue; packet processing time by the router; packet loss on the router due to packet length change and traffic type. To determine the optimal route for the transmission of control messages in the integrated rail network (trunk level) using the Neural Network Toolbox MatLAB environment created a neural model "21-1-X-21", which is fed by an array of delays on routers; as a resultant vector of the entry of the network links to the route. The neural network investigated the rms error and learning time from the number of hidden neurons (10, 45, and 90), from the length of the training sample (50, 100, and 152 examples) according to different learning algorithms: Levenberg-Marquardt; Bayesian Regularization; Scaled Conjugate Gradient. It is determined that the configuration is 21–1–45–21 according to the Levenberg-Marquardt algorithm, for which 100 examples are sufficient to teach. Scientific novelty. Dependences of the mean square error and time of learning of the neural network (number of epochs) on the number of hidden neurons according to the algorithms of learning Levenberg-Marquardt, Bayesian Regularization, Scaled Conjugate Gradient on samples of different length. Practical importance. The use of a multilayer neural model, which is inputted by the latency on the routers, will allow real-time identification of the corresponding routes of transmission of the guide messages in the network of ITS of the railway transport at the backbone level. Product Description popup.authors Pakhomova Victoria M popup.nrat_date 2020-04-02 Close