关键词: 无线传感器网络, 可辨识矩阵, 属性约简, 概率神经网络, 故障诊断

Abstract: In light of redundant data, noisy data and data reliability existing in fault diagnosis of node in wireless sensor network, this paper proposes a fault diagnosis algorithm（RSOPNN）of wireless sensor network nodes based on the Rough Set theory and Optimized Probabilistic Neural Network. The method uses rough set theory to get reductions of fault diagnosis from the collection of samples’ properties, so that it can decrease the impact of redundant attributes, noisy data on fault diagnosis, and save energy. Then, it chooses the best reduction from multiple reductions above by using the correlation between properties of reduction instead of subjective choice to solve the irrationality of subjective choice. Finally, the method reconstructs the fault samples with the best reduction of fault diagnosis as the input of the optimized probabilistic neural network, builds classification model to diagnose faults. Experimental results show that in the case of different data reliabilities, the RSOPNN proposed in this paper can eliminate redundant and noisy data in the original data effectively with higher diagnosis rate, which meets the demand of wireless sensor network.

Key words: wireless sensor network, discernibility matrix, attribute reduction, probabilistic neural network, fault diagnosis