Research On Wireless Networks Localization Algorithm Using Fingerprint

Thanks to the popularity of the intelligent terminals,such as the smart phone,the technologies of Mobile Internet and Internet of Things(IOT)has greatly influenced and changed the our lifestyle by providing the Location Based Service(LBS).Although the satellite positioning technologies like Global Positioning System(GPS)and BeiDou Navigation Satellite System(BDS)have been able to achieve the high positioning accuracy with satisfying speed,this kind of technology can hardly work well in some blind areas,where the satellite signal can not reach.While by utilizing the existing foundation of the WLAN(IEEE 802.11x)and cellular network(GSM?CDMA?WCDMA?LTE)we are able to make up the shortcomings of the satellite positioning technology with low price.The fingerprint based localization technology for wireless networks has advantage over the traditional localization technology.However to achieve a robust and high accuracy localization system with low cost,there still some problems need to be addressed.Firstly,the unavoidable interference causes fluctuation of the radio signal,which can produce noise fingerprint.It degrades performances of the fingerprint matching algorithm.Secondly in order to reduce the cost of building and maintaining fingerprint database,Semi-supervised Extreme Learning Machine(SS-ELM)has been applied to the wireless network localization.However the performances of SS-ELM are sensitive to the hyperparameters,and there is lack of systematic theoretical guidance telling us how to manually select optimal hyperparameters.To solve the problems mentioned above our work can be regarded as two part:1.In order to obtain robust fingerprint,we propose a method to extract the RSSI feature by using the mapping of Logistic function.Through the experiment we carried out,we prove that this method can improve the performance of fingerprint matching algorithms.2.We treat the second problem as a parameters optimization problem,so that we can use the evolutionary algorithm to optimize the hyperparameters of SS-ELM.Thus a SS-ELM hybrid evolutionary algorithm framework is proposed in this paper.Compared with similar frameworks that are used to optimize the supervised learning algorithm,the framework we proposed utilize both labeled and unlabeled data.This is because the SS-ELM is used in the application with few labeled data.Finally through experiment we prove that by applying our hybrid framework,the hyper parameters of SS-ELM algorithm is self-adaptive,which improves the performance of SS-ELM.