The paper proposes the algorithm for solutions of railway safety task by multiple-criteria decision making. The goal of the algorithm is to prevent clashes and collision of trains at railway stations by reducing the human factor and using of genetic algorithm for multi-objective optimization of the interaction between various moving railway vehicles. The main challenge of this research is a real-time optimization using embedded devices taking in account dynamics and stochastics of the real railway transport system. The main reason of this research is various train accidents that happed mainly at railway stations, when two or more trains have moved on parallel tracks and collided at the switches. Some of the causes are delay of trains and infringements of the planned schedule. Current railway traffic control systems do not allow forecasting the probability of collision and preventing it. Therefore, the main goal of the research is to investigate and develop the algorithms of MCDM for real-time railway transport scheduling and control. The combination of two solutions is proposed. On one side, it is possible to stop the train just before the fact of collision. On the other side, it is more effective to foreseen the probability of collision by real time control of the actual situation and then just minor reduction of the speed of change of the schedule may prevent the crash. The research includes the analysis of existing railway transport system, the development of the mathematical models and multiple criteria schedule and speed fitness function of the genetic algorithm for train anti-collision task. The computer model of the train movement including stochastic parameters, such as traffic, technical condition, delays, and weather conditions is created. In addition, prototypes of embedded intelligent devices are developed to test the algorithm in real conditions. During the optimization process the information about all trains approaching the station are collected and the control system is self-trained to minimize the risk of the collision. The real system is based on microprocessors and using RF wireless communication. The system is able to evaluate the situation and to propose to change the speed of the trains to avoid the dangerous situation. Set of more than thousands experiments have been performed to get statistics for analysis. The developed models and algorithms may improve the safety level of transport system control. The schedules developed by the algorithm provide better results than the original schedule by all target function criteria. Smart networks of the embedded devices for railway transport may be used to prevent collisions, and can be integrated in existing working infrastructure.