Research Of Localization Method For Semi-free Distributed Wireless Sensor Networks Based On Learning Model

Wireless Sensor Networks are constructed by a large number of portable sensor nodes, it can provides various types of sensor data which are very valuable for environmental monitoring and weather forecasting, geological disaster monitoring and early warning, monitoring and control of pollution of rivers and lakes. Query data based on the location information is one of the basic functions of WSNs, therefore, node self-positioning technology has been paid very much attention. Most localization methods based on the wireless signal attenuation information to estimate the node location, but the attenuation mode will become very complicated in the actual network environment, a high cost of communication and computation are need to pay to guarantee the accurate results. At the same time, node density and network topology will also affect the accuracy of localization results.This paper commence from the reference point used in the localization method, let the reference point to be able to provide more effective location service through the rational planning in advance of its location distribution. Based on this, we will design the semi-free distributed WSNs and discuss a distributed parallel localization method-CTLL (Cell-based Transferable Learning method for Localization in WSNs). It is not sensitive to the node density and topology and also stable to the dynamic noise. The main research work includes the following aspects.1. We use the Entropy Reduction theory and the Cramer Rao Lower Bound to analysis the influence of the number of reference points and their geometry structure on the position estimation uncertainty. Based on the results, we study the establishment of semi-free distributed WSNs.2. We design the distributed parallel localization method CTLL to estimate the location of node:According to the measurement results, node first roughly positing itself to the local area, then, a SVR regression model on this area will be used to determine the precise location of the node.3. We analysis the sampling point set problem on the SVR model, and make the data acquisition work be independent of the network by the introduction of instance transfer learning method. Finally, CTLL localization method will be described in detail.