Fault Prognosis Of High-Voltage Asymmetric Pulse Track Circuits Based On Fuzzy Neural Network

Faults of track circuits will not only affect the efficiency of railway transportation, but also cause major accidents. Therefore, fault diagnosis of track circuits is of important practical significance. Following the principle of track circuit, we build an asymmetrical high voltage impulse track circuit system model. By simulating different kinds of fault states in this model, we get several sets of fault data samples. And on this basis we conduct the thorough research of the fuzzy neural network and genetic algorithm. We then establish two fault diagnosis models separately, one is based on generalized weighted average fuzzy neural network, and the other is based on genetic algorithm fuzzy neural network. The simulation results show that these two models both have good fault diagnosis accuracy and generalization ability. The innovation points of this paper include:(1) Fault prognosis of Track Circuit based on GWA Fuzzy Neural Network.Fuzzy operator plays a key role in Fuzzy Neural Network (FNN), in which it factors the process of fuzzification on input layer and the defuzzification on output layer. Although there have been some related research applying FNN in track circuits fault diagnosis system, they only choose no-argument fuzzy operators. The functions of fuzzy operators have not been adequately considered.After in-depth study of the aggregation ability of typical fuzzy operators, we proposed an improved model by employing the fuzzy operator which contains compensation degrees parameters. We use Generalized Weighted Average (GWA) operator to replace the transfer functions of neurons in rule and output layers so as to build GWA Fuzzy Neural Network. Through simulation of the GWA-FNN model and the no-argument fuzzy operators-base model, the result shows that the fault prognosis of track circuit based on GWA-FNN has better accuracy and generalization ability.(2) Fault prognosis of Track Circuit based on GA Fuzzy Neural Network.There are many factors influence the performance of fuzzy neural network, among which the network topology determines the generalization ability of FNN. However, present researches of applying FNN in track circuits fault diagnosis system do not take account of the dynamic change of network topology. We introduce the Genetic Algorithm (GA) into fault diagnosis model, and trained network parameters in two phases to build GA-GWA-FNN model. First, according to the fault data samples, we use genetic algorithm’s ability of searching in whole range to add or remove rule reasoning layer neurons automatically. Second, we use the same parameter optimization algorithm in GWA-FNN to train parameters in GA-GWA-FNN model.Then we carry out both GWA-FNN and GA-GWA-FNN model in experiment, the result shows that the fault prognosis of track circuit based on GA-GWA-FNN has better generalization ability and faster convergence speed compared GWA-FNN model.