Research On Localization Algorithm In Wireless Sensor Network Based On Improved Locality Preserving Canonical Correlation Analysis

Wireless Sensor Network(WSN)is mainly applied in monitoring events,and the location of events is crucial to monitoring.This paper mainly studies the localization methods of directly constructing the mapping between the signal space consisting of the Received Signal Strength(RSSI)and the physical space comprised of the Location data.The Location Estimation-Locality Preserving Canonical Correlation Analysis(LE-LPCCA)method has problems such as inadequate training data information for solving the mapping,excessive reliance on a large number of RSSI data labeled by Location data(paired RSSI and Location data),etc.To solve the problems above respectively,this paper proposes two WSN localization methods using manifold learning,Canonical Correlation Analysis(CCA)and Semi-supervised learning as tools.The specific researches are as follows:1.A Location Estimation-Improved Locality Preserving Canonical Correlation Analysis(LE-ILPCCA)method for localization in WSN is proposed.The LE-LPCCA method only uses data similarity while ignoring data dependency between signal and physical spaces,and employs rough centroid method when computing coordinates of unknown nodes.As to the problems above,the LE-ILPCCA method has made corresponding improvements.In training phrase,it combines data similarity and dependency by a balance parameter to compute a more precise projection transformation of RSSI inner low-dimensional coordinates;in localization phrase,it calculates Location of unknown nodes utilizing an accurate transformational relation between Location coordinates and RSSI inner low-dimensional coordinates of known nodes.The experimental results on simulated and realistic data show that LE-ILPCCA has a higher accuracy and stability than LE-LPCCA and LE-CCA.2.A Semi-supervised Locality Preserving Canonical Correlation Analysis(SLPCCA)method for localization in WSN is proposed.As to the problem that the LE-LPCCA method has an excessive reliance on a large number of RSSI data labeled by Location data,which will increase localization cost,this method constructs the SLPCCA model combined with inner local structure information of data by introducing similarity matrices of the labeled data and the whole data in the prototype of SemiCCA.The SLPCCA model for WSN localization can fit the topology structure of network and employ unlabeled data efficiently.Then the mapping that maximizes correlation between signal and physical spaces can be calculated in condition of preserving the inner local structure of data.The experimental results on the simulated data and the realistic data show that the SLPCCA method can reduce the reliance on the labeled data under the premise of ensuring the localization accuracy.3.A localization simulation system in WSN is designed.The localization methods proposed in this paper are integrated in the simulation system to make it easier for operating.Localization process can be realized by selecting localization methods and setting relevant parameters on the system interface,and the localization results and errors will be displayed on the system interface,which is more convenient and flexible compared with separate operation.