Application Of Fuzzy Neural Network On Modeling Train Braking

Safety and efficiency are eternal themes for railway transportation, especially as the rapid development of China's railway. In recent years, the railway transportation has shown heavy, high-speed, high-density characteristics, all which need a high reliable train control system as a guarantee. Braking control, as a part of the train control, plays a key role in train safety. But there are some errors between the braking mode curve based on train-traction-calculation theory and the actual operation of the train, which results in that the effect of the control is not good. Therefore, how to construct a more accurate train braking control model becomes a hot research in recent years.Complex systems modeling and simulation is a key issue in systems modeling field. Fuzzy Neural Network(FNN) integrated both advantages of fuzzy systems and neural networks, can better realize complex systems modeling. In this paper, the method of FNN modeling, FNN structure which is suitable for modeling of the train braking control and its applications have been studied, which is based on the background of complex systems modeling.This paper, based on the research at home and abroad, is on the issue of train braking control modeling, and the main research contents are as follows:1. The fuzzy problems existed in the train braking control and its causes have been analyzed from the perspective of complex systems modeling, and it is drawn that the FNN used in the train braking modeling is feasible.2. In this paper, it is focused on how to construct a suitable FNN model for train braking control under the premise that the characteristics of the braking control system is not clear enough. Then the standard FNN has been improved and an improved four layers fuzzy neural network was acquired with its learning algorithm deduced, which is proved that this model has the characteristics of the overall approach in theory .Afterwards, we used the improved model to the train braking process and determine the input and output variables from practical point of view on the manipulation, which made the model is more adapted to the actual operating environment. This is different from the previous model structure and the variables selected are also different.3. Traditionally, too many language variables often lead too many fuzzy rules, and they have impact on network learning speed and accuracy. To avoid it, K-means clustering algorithm has been introduced to the FNN structure identification and been used for the data classification. At the same time, the actual manipulation is also considered combined with the classification to determine the necessary rules. It both takes into account the actual data modeling and the characteristics of the manipulation.4. In order to verify the validity of the improved network structure used in the train braking control modeling, in this paper, a fright train was used as an example for modeling and calculation. The numerical results show that improved fuzzy neural network model has the characteristics of high speed and precision. Then we used the Matlab module- Simulink to simulate the train braking process, and it proved that the improved model used in the train braking control is feasible.