The Research Of Outlier Detection Based On Compressed Data In WSN

With the rapid development of sensor technology, embedded computing technology, distributed information processing technology and communication technology, wireless sensor networks (WSN) came into being. Wireless sensor network consists of a large number of sensor nodes with communication of cheap computational nodes, and connected to each other forming a multi-hop self-organizing network wirelessly. As the wireless sensor network has broad application prospects, it has become a new frontier hot research field in the 21st century. And it has been used in military, aviation, terrorism, explosion, disaster relief, environment, medical health, industrial, commercial and other related fields.Real-time monitoring of data and analysis of the data is the main purpose of wireless sensor network applications. In practice, however, the raw data collected by the nodes are not very accurate and reliable. On the one hand, cheap and poor quality nodes have limited resource conditions, such as restrictions on battery power, memory, computing power and communication bandwidth, etc., these will result in the unreliable data collection; on the other hand, wireless sensor nodes are often randomly deployed in harsh conditions in the natural environment, which may make nodes fail caused by sudden noise data, error data, data loss, redundant data. Also sensor nodes may face enemy malicious attacks, such as a black hole attacks, eavesdropping, and interference. Therefore, how to effectively ensure the reliability and accuracy of the WSN node data becomes an urgent problem to be solved. The anomaly detection technology provides effective means for solving problems. At present, a large number of researchers have proposed a variety of effective anomaly detection algorithms in WSN. However, due to the variability of their resource-constrained WSN nodes and environment, the existing anomaly detection algorithms still have many problems to solve and improve.In particular, the vast majority of approach ignores the effects of the anomaly detection node energy and treats the energy-saving problem and anomaly detection as two separate issues. However in the practice, when the sensor nodes in the harsh environment tend to run out of energy, the accuracy of the data will decrease. This requires us to consider energy issues in the process of anomaly detection. For WSN energy saving issues, related studies show data transfer process consumes the most of the node energy, data fusion technology has been introduced into the WSN applications. Therefore, in order to establish an effective anomaly detection model, and to consider the energy of the nodes, the paper focuses on the anomaly detection algorithm based on compressed data. The research results have high academic value and broad application prospects. Thesis research content and results can be summarized as follows:(1) In practical applications of WSN, sensor nodes are often arranged in a harsh environment and the battery is limited, so how to effectively carry out energy-saving process to extend the life of the WSN node applications become an urgent problem. Since the energy consumed by the data transfer node accounted for the vast majority of network energy consumption, a variety of data fusion technology is introduced into the application of which wireless sensor networks. In order to effectively implement WSN energy saving, we studied a SAX-based space-time compression algorithm. The algorithm is based on SAX efficient data compression algorithm and makes full use of the relevant characteristics of their own data between nodes to achieve further compression for a greater degree of reduction in the amount of data transfer node.(2) In order to achieve effective energy conservation network, data fusion algorithm is introduced into the WSN applications. However, most of these algorithms are assumed that the current data can be transferred directly from the node to the sink node, ignoring the practical application of the network topology is set. In order to combine data fusion algorithms and topology, we present a sensing algorithm based on improved compression algorithm LEACH protocol. First, we propose the improved way of cluster head LEACH protocol so that it can be more reasonable to choose cluster heads. In the data transfer phase of the agreement, we take advantage of the compressed sensing algorithm. Since LEACH protocol itself can effectively implement network longevity, we use data fusion technology to compress the node data. This guarantees that the method can achieve a more effective network of energy-saving purposes.(3) In the battery power running out stage, data quality will be greatly reduced, and this may have a greater impact on the subsequent data processing. At present most of the anomaly detection algorithm ignores the node energy impact for the data quality, and treats anomaly detection and energy as two separate issues in WSN applications. In order to take account of the anomaly detection and energy, we propose an anomaly detection algorithm based on compressed data Firstly, we use segmentation aggregation approximation algorithm (PAA) for raw data compression. But in order to guarantee the accuracy of anomaly detection, we improved the PAA algorithm. Then we directly use of K-means classification algorithm for the data to classify exception information. Meanwhile, in order to protect the overall classification, we use the artificial immune algorithm to optimize K-means algorithm. Since the detection process is directly carried out based on the data compressed, this ensures that the real-time of anomaly detection process.(4) The error and abnormal events are two different exception types in WSN. Such as noise, node failure caused by erroneous data is wrong exception, fire, water pollution and other natural phenomena can be treat abnormal events. Most anomaly detection algorithms treat these two abnormal types as an exception messages. This causes the hidden event in the raw data information cannot be found. In order to effectively distinguish between errors and abnormal events, we propose an anomaly detection model based on locally linear embedding algorithm (LLE). The model is based on an error exception and abnormal events on the spatial correlation and implements the distinction between the two categories of exception. And because LLE algorithm itself is an efficient dimension reduction algorithm, it can effectively detect abnormal network while achieving energy savings.