Distributed Fingerprint Matching Position Algorithm Based On Multi Feature Learning

With the rapid progress of economic and widespread of mobile device and wireless network,users can connect to the network anytime and anywhere.As a result,more and more enterprises and service providers focus on Location Based Service.Mobile location service is that the users can be provided with value-added services associated with the location of UE by the cooperation of mobile terminal and network operators(such as LTE,WCDMA and GSM).Though fingerprint matching position technology has raised widespread concern all over the word and has been applied within a certain range,there is none mature and applicable theory system as far now.In this paper,we will put forward a set of relatively completed algorithm model by data processing,data analysis,feature extraction,matching algorithm and timeliness with big data,then test it.In this paper,we focus on the problem of position algorithm with crowd sourcing data on cellular network.To improve the accuracy and the efficiency,we filter out the data with GPS offset.Then we go on analyzing data and modeling under the direction of fingerprint matching position algorithm.To train the multidimensional features,we come up with a variety of machine learning models,which constantly improve the location accuracy.At the same time,in order to improve the training efficiency,we build a set of Hadoop distributed computing framework by using some computers.Except the crowd sourcing data on cellular network,we also use the mall data from TIANCHI competition to verify the algorithm.As the result shows,the algorithm improves the efficiency greatly.In this paper,we first introduce the traditional localization algorithm with cellular network.Then we introduce widely used machine learning algorithm in recent years,especially about the model and training algorithm of Restricted Boltzmann Machine.Except this,the framework and the components of Hadoop will be introduced.Then we will put the RBM and Maximum Similarity algorithm together.We also come up with Weighted Maximum algorithm by combining KNN and Maximum Similarity algorithm.What's more,to get more reliable data set,we can divide the all area into several parts,then we filter out the singular data with error.It is proved to improve the positioning accuracy.In the end,we introduce how to build a set of Hadoop framework,and then program to realize the algorithm.Then,the Hadoop cluster is optimized to improve its fault tolerance and operating efficiency.