摘要
针对现有铁路信号设备故障识别算法特征提取不准确导致正确率偏低的问题,提出了深度信念网络(DBN)的故障识别模型。该模型首先利用无监督训练方法对DBN的多个堆叠受限玻尔兹曼机(RBM)进行预训练,获得网络初始参数;然后,结合铁路信号设备识别问题,构建BP神经网络,利用有标签样本进行反向传播训练,实现网络参数微调。实验结果表明,该模型避免特征提取的人工操作,能够有效实现铁路信号设备故障的准确智能识别。
In the existing recognition algorithms for existing railway signal equipment faults,the accuracy of recognition will be low due to inaccurate feature extraction.The fault recognition model of deep belief network(DBN) is proposed in this paper.An unsupervised training method is used to pertain multiple stacked restricted Boltzmann machine(RBM) of DBN.In this way,initial network parameters are obtained.Then,combined with the railway signal equipment recognition problem,a BP neural network is constructed.Back-propagation training is performed using labeled samples to achieve fine-tuning of network parameters.Experimental results show that the model avoids the manual operation of feature extraction and can effectively realize the accurate and intelligent identification of railway signal equipment faults.
作者
张玉霞
ZHANG Yu-xia(Shaanxi Communications Vocational and Technical College,Xi’an 710010,China)
出处
《信息技术》
2020年第5期150-154,164,共6页
Information Technology
关键词
铁路信号设备
故障识别
深度信念网络
受限玻尔兹曼机
railway signal equipment
fault identification
deep belief network
restricted Boltzmann machine
