Research On Distributed Fault Detection For Meteorological Sensor Networks Based On Neighbor-coordination

Meteorological sensor network is a high integration of network system which has many characteristics, such as dynamic topology, self-organized. As a result of nodes in the meteorological sensor network are resource-constrained and work in unpredictable environment for a long time, the meteorological sensor network failure occurred frequently during runtime and shortened the service life of the network.Hence, the proposed fault diagnosis algorithm must take advantage of network information interaction to improve the performance of diagnosis and to avoid the disadvantage of the large redundant information and high complexity. It must make accurate, real-time and low power consumption decisions in the operation management to meteorological sensor network. The main work of this article are based on the characteristics of meteorological sensor network, was given a full play to the superiority of the meteorological sensor node space and time correlation,combined with the fault diagnosis methods of neighbor cooperation, deeply discussed on the fault diagnosis algorithm of meteorological sensor network:The network structure and characteristics of Meteorological sensor network were introduced in this paper. The meteorological sensor network causes of failure were analyzed and the faults are classified. Meteorological sensor network fault diagnosis technology are outlined, and main problems of centralized fault diagnosis methods and distributed diagnosis methods are elaborate, and gives the performance evaluation criteria of the fault diagnosis algorithm is given. These studies provide preparation for the next step research on the fault diagnosis of meteorological sensor network preparation.To solve the problem that current fault diagnosis methods for meteorological sensor network not only produce a high false alarm ratio, but also incur huge overhead of communication and resources on nodes. To address the above issues, a new distributed fault diagnosis scheme on the basis of Cumulative Sum Chart was proposed. First, for purpose of increasing the sensitivity of fault diagnosis and locating the change point, the historical data on a single node were analyzed by Cumulative Sum Chart. Then the proposed algorithm determined the status of nodes by the collaborative operation of neighboring peers. Simulations show that even if the fault probabilities are very high, the proposed algorithm has a low false alarm ratio and high detection accuracy. The influence of failure rate on this algorithm is very little. The proposed algorithm in this paper required a little amount of traffic which had a good scalability and stability. It could adapt to the fault diagnosis of large-scale meteorological sensor network.Concern the problem that current distributed fault diagnosis methods in meteorological sensor networks with low node density and high failure ratio lead to a significant decline in the accuracy of fault detection, a new fault detection mechanism based on support vector regression and neighbor-coordination was proposed. According to the redundant information of meteorological elements which collected by multi-sensor, the fault prediction model was built by support vector regression algorithm and achieve residual sequences. Then the status of nodes was identified by mutual testing among reliable neighbor nodes. Simulations show that the proposed algorithm not only reduces communication to sensor nodes, but also has a better fault diagnosis performance in the meteorological sensor networks.