Sensor Failure Detection Mechanism Based On Support Vector Regression

The main role of wireless sensor networks is to collect the environmental data. However, as the sensor nodes are very vulnerable and always work in unpredictable and bad environments, sensors are possible to fail and return unexpected response. Therefore, fault detection and recovery are important in wireless sensor networks. In this paper, we propose a fault detection algorithm based on support vector regression, which predicts the future measurements of sensor nodes by using local data and historical data. Sensor nodes’credit levels will be determined by the contrast between predicted values and actual measured values. Then the dependable data set which is constructed by cluster head and high credit level measurements will be used to further detection of sensor faults. In this paper we also propose a fault recovery algorithm according to the node credit levels combined with genetic algorithm and principal component analysis, which reduce sensor network recovery sharply.The simulation results demonstrate that the fault detection algorithm we propose works well in raising failure detection rate and conserving energy. The fault detection rate of SFDA remain above94%in different network scales. Besides, the algorithm results will not be affected when the network size increases. It can be obtained that SFDA algorithm is much better than the contrastive algorithm in fault detection rate. Meanwhile, SFDA obtains the best simulation result in energy consumption.What’s more, the fault recovery algorithm we propose works well in conserving energy and accelerating sensor network recovery. CL-GA does well in reducing the recovery sensors with only60%～70%sensors to be replaced. Meanwhile, the energy consumption curve is low and nearly linear, which means CL-GA will not degenerate when the network size increases.