Anomaly Detection Method Research Of Sensor Data Stream Based On Multi-dimensional Data Model

With the rapid development of sensor technology, wireless sensor networks(WSN) are increasingly being applied to industry, agriculture, medical care and hygiene service domains. Mostly, the wireless sensor nodes are often distributed in the target area to acquire the information of environmental parameters(temperature, humidity, CO2 concentration, etc.) and monitor the environmental variety. In order to find out the unexpected events in the natural environment, monitor the health status of the sensor networks, as well as improve the reliability of the wireless sensor network, it is highly significant to detect the abnormal data from the wireless sensor nodes data stream.The main work of this paper is as follows:1. A new method of WSN anomaly detection based on multi-dimensional data model is proposed, which can use the multi-dimensional data model to detect the abnormal data in the sensor data stream, and then confirm the source of abnormal data according to the temporal-spatial correlation. Then, we can monitor the work status of WSN in time.2. The detection method of sensor data stream based on multi-dimensional data model proposed in this paper is mainly consisted of three steps. Firstly, detect and identify the abnormal data in the sensor data stream by using the statistical characteristics of the multi-dimensional data model. Secondly, verify the source of abnormal data with temporal-spatial correlation of nodes and correlation of multi-dimensional data. The last but not least, confirm the work status of the wireless sensor nodes according to the results of the anomaly source verification.3. The performance of the method proposed in this paper is evaluated by the simulation experiments. The experiments on the MATLAB platform are carried out using the artificial synthetic data of sensor nodes. The performance of method proposed is tested, as well as compared with the existing methods. Experimental results show that the abnormal data detection rate of the method proposed in this paper is about 95%, which can be improved by 3% to 4% with four dimensional data sets.