基于RFID传感器标签与深度学习的变压器状态监测方法研究

Research on monitoring technology of transformer based on RFID sensor tag and deep learning

变压器的状态直接决定了电网运行的可靠性。在现有的变压器状态监测方法的基础上,提出基于射频识别(RFID)传感器标签和深度学习的变压器状态监测手段。首先提出了一种基于自取能RFID传感器标签的变压器振动信号采集方法,该方法具有成本低,便利性强,寿命长,可实现快速定位等优点。针对于变压器信号成分复杂,信噪比低等特点,利用深度学习技术在复杂数据处理方面的优势,采用堆叠去噪自编码器(SDA)信号进行特征提取,并采用相关向量机(RVM)对提取的特征数据进行故障诊断,最后采用量子粒子群算法(QPSO)对SDA与RVM进行参数寻优。实验结果表明,方法能够可靠地获取变压器振动信号,同时,能够取得99. 75%的故障诊断准确度,且诊断时间仅需要0. 98 s。

The condition of transformer determines the reliability of power grid,this paper presents a monitoring approach for transformer based on self-powered radio frequency identification( RFID) sensor tag and deep learning method. A data collection technology based on self-powered RFID sensor tag,which has advantages of convenience,low cost,long life and rapid positioning. The transformer vibration signal has features of complex and low signal to noise ratio,and the deep learning technology has merits in dealing with complex signal,therefore,the stacked denoising autoencoder( SDA) is employed to learn features of the vibration signal. Moreover,relevance vector machine( RVM) is employed to realize fault diagnosis for the transformer utilizing extracted feature data. The quantum-behaved particle swarm optimization( QPSO) algorithm is employed to search parameters for the SDA and RVM. The experimental results show that the proposed method can reliably collect the vibration signal of transformer,and it can realize a diagnosis accuracy of 99. 75% with diagnosis time of 0. 98 s.