Multidimensional Data Outlier Detection For Wireless Sensor Network

Multidimensional data outlier detection is a serious problem in wireless sensor network. For the limitation of power and compute ability, energy consumption becomes the most important factor for algorithm. The way of data preprocessing, the way of transmitting, and the use of temporal and spatial correlation, can all effect the energy consumption. Based on all these reasons, this paper propose two kinds of detection algorithm.Multidimensional data detection algorithm based on trajectory. First, cluster the WSN according to the node similarity; then, based on the dimension reduction of data, the multidimensional data detection algorithm based on trajectory is presented. Stimulation shows outlier detection method based on trajectory has the ability to deal with multiple data in continuous time span, and the detection effectiveness is good.Multidimensional data outlier detection method based on the improved k-means algorithm. First of all, improve the traditional k-means; Then, cluster the network with the improved k-means algorithm; Last, judge whether the cluster normal or abnormal. Stimulation shows this algorithm is effective, and can guarantee the real-time detection.Multidimensional data detection algorithm based on improved k-means and evolutionary strategy. First, improved k-means method is used to determine the k initial cluster center; Then, based on the evolutionary strategy get the offspring and pick over. Last, judge the cluster. Stimulation shows this algorithm is effective, and it is real-time detection.