Abstract:

We present a novel sensor node fault detection method for wireless sensor network (WSN). Modified Recurrent Neural Network (MRNN) is used to model sensor nodes, the nodes’ dynamics, and the interconnections with other sensor network nodes. An MRNN modeling approach is used for sensor node identification and fault detection in WSN. The input to the MRNN chooses those that include previous output samples of the modeling sensor nodes, and the current and previous output samples of the neighboring sensors. The model is based on a new structure of a backpropagation type neural network. The input to the MRNN and the topology of the network are based on a general nonlinear sensor model. Simulation results demonstrate the effectiveness of the proposed scheme and the MRNN method has higher failure detection accuracy in the case with smaller confidence factors  compared with the Kalman filter method.

Key words: fault detection;modeling;recurrent neural networks;wireless sensor network