Study On Sensor Data Fusion For Water Environment In Wireless Sensor Network

A wireless sensor network consists of a large number of sensors which deployedover a geographical area for monitoring physical phenomena. Sensors can collectinformation in their coverage area and sent it to staff computer. in the1990s Sensornetworks has been started in the United States. But there is a problem, each sensor hasa limited energy supply and they are often deployed in remote areas or even areaswithout accessibility, which make the replacement battery for additional powerbecome unrealistic. Moreover, due to sensors internal device aging or by the influencefrom outside, noise is recognized as essential issue. A good data fusion method cangreatly reduce the impact of noise on fusion results.There are several water environment monitoring problems as follows:1. It’sdifficult to get harsh natural condition’s information in the water environmentmonitoring.2. It can not be fast enough to deal with dynamic event or emergency inthe water monitoring.3. it’s difficult to obtain accurate and comprehensive regionaldistribution of multi-dimensional water environment information. Taking into accountthat the wireless sensor networks have the function of rapid deployment and good foreffective information from the outside world, this paper choose water environment asresearch background. Motivated by above, we proposed a new Variance weighted datafusion method, after preprocessing Gaussian noise and impulse noise. With thismethod, measurement data sample in a cycle is taken as a sliding window, and doublesample normal score test is employed into samples. Further more, a relationshipmatrix is created, according to which a maximum adjacent subgraph is obtained. T heabnormal sensors’ processing use the median K-means clustering algorithm, In themedian K-means clustering algorithm, the median value of the sensor sliding windowsamples as a collection of median K-means’ clustering, and use a reasonable methodto determine the cluster center, conditional diagnosis of abnormal sensors andexclusion. At last, estimated fusion value is achieved in terms of weighting andaveraging its normal sensors’ vertices.Simulation results show that this data fusion method is active. The accuracy ofthe results can be improved effectively when the window length is set to6. At that time, not only the result accuracy is improved, but also the communication traffic andenergy consumption are reduced in this phase, thus, It also extend the networklifetime. The median K-means clustering algorithm removed the abnormal sensorswhich damaged by the current rush of water or other damages to sensors and restrainsthe impact of sensor failure on fusion.