WSN-oriented Stream Data Clustering Algorithm Research

As wireless sensor network (WSN) is used widely, the action of access to massive data is presented frequently. However, the traditional technology of data processing such as data query and statistics method has not met the requirements of human societies about more complex information implicated in the data stream. The way of data mining just deal with this problem, and WSN-oriented data mining has become a vibrant area of research among which the task of monitor relevant environment is very important and has dept in people's daily and production life by using WSN. As a branch of data stream mining, data stream clustering is a useful technology to monitor WSN environment, because the technology of data clustering is applicable to cope with the problem of monitoring with which priori knowledge about data distribution is not easy to capture. To attempt to process WSN intrusion monitoring, we take the way of data stream clustering.As the development of technology about WSN, the features of data stream from network has presented, and people need extract complex information from data stream in the limit of strictly linear time complexity, but the process of tradition algorithms such intrusion monitoring system based on BP Artificial Neural Networks whose computing time complexity is significantly higher than the linear growth rate, are difficult to used to online monitor WSN environment. In essence, some previous methods of intrusion monitoring attempt to learn the invasive behavior directly. However, the model of unknown network intrusion behavior cannot be created, and thus the systems fail to react to the unknown network intrusion in time. For this, researchers have already designed some algorithms trying to deal with the problem of unknown network intrusion, but too many data points are classed incorrectly using these methods. For this reason, we design a novel method of WSN-oriented data stream clustering which monitor effectively the behavior of the unknown network intrusion, and attempt to extend the technology as distributed monitoring algorithm to settle the problem of data block in the way of centralized computing and reduce the consumption of power and bandwith from WSN.Now, we will discuss our main study in this article:Firstly, we design a new network flow-based data clustering algorithm for known intrusion monitoring which is called DOExMiCluster. The method adapt from the classic data stream clustering algorithm CLUSTREAM, and abandon time windows and snapshot model which is helpful to meet the requirement about history data information from users. To do this, our algorithm's time and space complexity can be lower than CLUSTREAM. we also design numerous new micro-cluster structures for clustering normal network behavior models recorded in the micro-clustering eigenvectors which are used to distinguish the normal network behavior and unknown network intrusion behavior.Secondly, as the scale of WSN is being expanded, the collection of mass data stream everywhere will not go beyond the limit of power and bandwidth of WSN. At the same time competing center is probably blocked and fail to scan each data point because of access of mass data, causing that a network intrusion behavior is missed. Obviously, monitoring network intrusion behaviors is just a task in the WSN, so only few sensors can be assigned to inspect data stream in the respective local area. We usually consider distributed approach to cope with problem centralized algorithms difficultly deal with. In WSN environment, we must free most sensors to process other daily task, meanwhile the scarce resources such as power and bandwidth can be saved and network congestion is eased. Therefore we build a DOExMiCluster-based distributed network intrusion monitoring system which is adapted from system topology structure based on cluster routing and includes some computing units. Moreover, the computing unit is composed by cluster, secondary computing node, flow-oriented module forwarding data stream. Finally, we need organize these computing units and plan the process of network intrusion monitoring under distributred condition.The experiment presents the technology of DOExMiCluster-based WSN intrusion monitoring achieves the high accuracy and the low error rate about data detection. Our monitoring system can be extended to identify exactly known intrusion behaviors, and meet the linear time complexity. Finally, the data processing in the way of assembly line increase throughput of data stream in our system.However, our methods don't consider the problem of heterogeneous data, noise and fault tolerance from loss of function of a computing unit. These problems need to be research in future work.