Research On Association Rules Mining And Graph Clustering Method In Sensor Network

With the mass deployment of sensor devices, wireless sensor networks that can reflect and detect human environment are established and constructed. Massive sensor data analysis, processing and transmission have become the research focus in the field of public safety. In wireless sensor networks, multi-scale and multi-modal information has make it convenient to monitor the environment and detect potential anomalies and dangerous condition. However, with the increase of the number and types of sensor devices, the amount of data detected in the same environment becomes very large. How to filter out effective nodes from vast amounts of data that can efficiently report the incident, and reduce information redundancy node acquisition and transmission, and to achieve efficient detection of abnormal events in public environment, are serious problems to be solved.In this thesis, graph clustering will be applied to the study of a wireless sensor network, which can solve the problem of redundant data producing at the time of event detection in sensor networks, and increase the network's traffic overhead and other issues. This thesis describes the sensor network to multi-attribute undirected weighted graph model of graph theory, and stimulates the formation of an effective sensor network topology and property relations by dynamic evolution of the model. Based on this model, analysis the similarity between the sensor nodes from multi-angle, and establishing the graph clustering algorithm based on similarity and analyze screening and scheduling policy of cluster nodes. Finally, based on graph clustering on the establishment of screening and scheduling frame sensor nodes, under the premise of complete event detection, different sensor nodes transmit on similar data are removed to save network overhead. The main contents are as follows:First, a multi-attribute undirected weighted graph model was defined to describe the sensor network, with the model, the relationship between non-adjacent properties are analyzed from respectively three aspects-direct adjacent, indirect adjacent properties and topology, and then to come to the similarity calculation between nodes, and then come to the similarity calculation between nodes.Second, based on the quantized graph model of the sensor network and the similarity computing method between sensor nodes, a graph clustering method was studied. That is the selection algorithm of head nodes and the graph clustering process of remaining nodes.Third, without sacrificing network coverage and connectivity, the study of graph clustering based on cluster nodes in high-throughput screening events and scheduling strategies for removing different sensor nodes transmit on similar data, saving the overhead of network-based.Through a comprehensive analysis of a simulation experiment on NS2 platform, the results showed that graph clustering method can be well applied in sensor networks among research, the theoretical method and frame proposed acquisition and transmission networks can reduce redundant data to reduce network traffic overhead goal.