一种基于竞争神经网络的铁路UM71信号频谱识别方法  

The UM71 Signal Spectrum Identification Method Based on the Competitive Neural Network

作　　者：赵林海[1,2] 邱宽民[1,2] 穆建成 

机构地区：[1]北京交通大学轨道交通控制与安全国家重点实验室,北京100044 [2]北京交通大学电子信息工程学院,北京100044 [3]铁道部科技司,北京100844

出　　处：《铁道学报》2009年第1期98-102,共5页Journal of the China Railway Society

摘　　要：机车信号设备译码的抗干扰能力直接影响设备显示的准确性和稳定性,同时也关系到铁路的行车安全。本文将神经网络模式识别技术引入到铁路车载信号的解调译码过程中,利用该技术非线性处理能力强和性能稳定等特点,提出基于竞争神经网络的铁路UM71信号特征频谱的识别方法。该方法根据不同调制低频下的UM71信号频谱的谱线结构规律建立起相应的神经网络结构模型,通过计算实际输入信号的频谱与神经网络各向量间的曼哈顿距离而产生竞争,再根据竞争结果与UM71信号的映射关系实现译码功能。实验证明:该方法具有学习算法简单、迭代步数少和运行速度快等优点,在提高UM71信号译码的准确性和抗干扰能力等方面效果明显,能够满足铁路现场实际运用的要求。The anti-interference performance of the cab signaling device in signal decoding has direct influence on the accuracy and stability of onboard signal indication, and is therefore related to train operation safety. This paper introduces the neural network based pattern recognition technique into the demodulating and decoding process of the cab signaling device, and by utilizing the strong nonlinear processing ability and stable performance of this technique, proposes a method to identify the characteristic spectrums of the UM71 signals based on the competitive neural network. This paper establishes the neural network structure model according to the spectrums of the UM71 signals under different low modulating frequencies, produces competition by computing the Manhattan distances between the spectrums of the real input signals and the various vectors of the neural network, and then implements the decoding function on the basis of the mapping relationship between the competition results and the UM71 signals. The experiment results show that this method is characterized by the easy learning algorithm, fewer iteration steps, fast running speed and obvious effect on enhancing the accuracy and anti-interFerence ability in decoding of the UM71 signals, and it is suitable for practical uses on railway sites.

关 键 词：竞争神经网络 频谱识别 机车信号 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] U284.237[自动化与计算机技术—计算机科学与技术]