The Research And Validation Of Data Fault Detection Schemes In Wireless Sensor Network

Wireless sensor network is one of the most important technologies including data acquisition,data processing,data transmission and data storage.Due to its flexible,low energy consumption characteristics,it is one of the hot research spots of experts and scholars in recent years.As an important technology of WSN to ensure the reliability of data,data fault detection technologies are widely received extensive attention of academic circles.There have been many intelligent algorithms applied to the data fault detection in WSN.These methods not only extend the diversity of the existing fault detection algorithm,but also improve the performance of fault detection for a part of application scenarios.This paper analyzes the common types of data fault,data model and the cause of fault in WSN,and researches the spatial correlation of sensor network data.The sensor data fault leads to the differences in the cumulative probability distribution between the sensor and the data of its neighbor node.Therefore,real immune genetic fault detection algorithm(RIGFDA)based on KS test and voting algorithm is designed in this paper.Immune genetic algorithm is rarely used for fault detection in existing literature.The innovation of this algorithm lies in using the real immune genetic algorithm for fault detection without building a model of the sensor data or setting the threshold for fault detection.On the other hand,the different characteristics of KS test and absolute same index are applied to detecting different types of sensor data fault.Experimental results show that RIGFDA can achieve a higher detection rate and low false alarm rate,and adapt to a variety of complex environment and it has a certain commonality.On the other hand,there is not only spatial correlation but also temporal correlation between sensor data.The temporal correlation of sensor data can be used for fault detection on a single node and it can reduce the communication overhead of the network.Most of the existing fault detection algorithm based on temporal correlation uses only correlation of continuous sampling point but ignores the periodicity of the data.Fault detection algorithm based on the ensemble of deep belief network is designed in this paper.This proposed method is based on the research of temporal correlation and the periodicity of the sensor data.The innovation of this algorithm lies in using neural network integration to achieve a variety of fault types of detection.The more important feature is that the proposed method decreases the complexity of network training,and simultaneously improves the detection accuracy.At the same time,this proposed method has been implemented on the MCU and FPGA to demonstate the feasibility in the real applications.Experimental results show that,compared with the traditional machine learning fault detection algorithm,deep learning algorithm provides the better performance of fault detection.In this paper,these two kinds of fault detection algorithm are validated in IBRL data set and GStB data set.Experiments show that the algorithm can detect a variety of fault types.They get better detection accuracy than the existing algorithm in IBRL and GStB data set and have lower false alarm rate.Simultaneously,these two methods are implemented and verified in the real WSN platforms.