Energy Efficient Query Processing Techniques Based On Particle Filters In Wireless Sensor Networks

Wireless sensor networks have been widely used due to its small size and low cost. However,there is a lot of uncertain sensor data in real world, traditional database management system cann’tafford to query these data. So it is especially important to search the query system for uncertain data.At the same time, the sensor nodes are powered by battery and its energy is limited. In fact, the sensornetwork should have a lifetime long enough to fulfill the application. In many case a lifetime in theorder of several months, or even years, may be required. Therefore, in order to solve the crucialquestion in wireless sensor networks that how to manage uncertain sensor data energy-efficiently toprolong the network lifetime to such a long time, we exploit spatial-temporal correlations on uncertainsensor data to construct probabilistic models and utilize particle filtering and improved particlefilterings to do probabilistic inference for saving sensor energy. The main innovations are summarizedas follows:Firstly, particle filtering based on model for processing uncertain data is proposed. We exploitspatial-temporal correlations on uncertain sensor data of different attributes for one sensor node toconstruct probabilistic models. Then, due to the large difference of the power requirements forsampling different attributes with specific type of sensor, particle filtering techniques are utilized toinfer values of attributes which consume more energy from values of attributes which consume lessenergy for data acquisition. This will save energy of sensors efficiently.Secondly, improved particle filtering based on model for processing uncertain data is proposed.Cluster the nodes deployed in the same area according to the spatial correlation of sensor nodes.Nodes that are spatially correated are clustered together and we build Multivariate Gaussian Modelsfor nodes in the same cluster. In additon, according to the fact that the sensor data generally satisfiesGaussian distribution, Gaussian particle filtering technique and Gaussian sum particle filteringtechnique are adopted to do probabilistic inference, which improves query accuracy. We evaluate ourapproaches from accuracy and efficiency, the experimental results illustrate good performance of ourproposed particle filter techniques.Thirdly, realize query of sensor data with particle filtering in database management systemPostgreSQL. Use function、procedural language and auxiliary modules provided by PostgreSQL torealize each steps of particle filtering: sampling, importance sampling, resampling and integrate thesesteps to form a complete query mechanism, which can response user query and improve the queryefficiency.