基于改进的粗糙集和神经网络的WSN故障诊断

WSN Fault Diagnosis with Improved Rough Set and Neural Network

综合粗糙集理论和人工神经网络的优点,提出了改进的粗糙集理论算法,并结合人工神经网络,实现了一种无线传感器网络(Wireless Sensor Network,WSN)节点智能故障诊断方法。首先基于WSN的应用环境和故障特征的分析,通过数据采集、数据预处理和数据压缩来获得诊断决策表,并利用粗糙集中改进的归纳属性约简算法(Improved Inductive Attribute Reduction Algorithm,IIARA)对决策表进行属性约简,从而提取对故障诊断贡献最大的最小故障诊断特征集合,进而确定后端径向基函数神经网络(Radial Basis Function Neural Network,RBFNN)的拓扑结构。最后通过网络训练建立故障征兆与故障类型之间的非线性映射关系,得到诊断结果。仿真实验结果显示,该诊断算法在对WSN节点进行故障诊断时,可以有效地减少网络输入层个数,简化神经网络结构,减少网络的训练时间,提高模型的诊断准确性。

Integrating the advantages of rough set theory and artificial neural network,improved rough set theory algorithm was proposed,and combined with artificial neural network,an intelligent fault diagnosis method of wireless sensor network(WSN)node was achieved.First,based on the analysis of application environment of WSN and fault characteristics,diagnosis decision table is obtained through data acquisition,data pretreatment and data compression,and is reduced by the improved inductive attribute reduction algorithm(IIARA)of rough set,thus the minimum fault diagnosis feature set is extracted which contributed most to the fault diagnosis,and then the topology of the radial basis function neural network(RBFNN)is determined.Finally,diagnosis results are obtained through the nonlinear mapping relationship between fault symptoms and fault types established by the network training.Simulation results show that at the time of fault diagnosis of WSN nodes,the diagnosis algorithm can effectively reduce network input layer,simplify the structure of neural network,reduce the training time of the network and improve the diagnosis accuracy of the models.