| With the rapid development of Internet and wireless technologies,the use of mobile devices to obtain location services is affecting people’s learning and life in all aspects.Although in the outdoor open environment,positioning systems such as GPS can effectively locate;however,in a complex indoor environment,the wireless signal will cause more serious attenuation when it penetrates the shield,resulting in low positioning accuracy of the satellite positioning system.Nowadays,locating fingerprint matching with Zig Bee network is a reasonable solution to this problem.This paper aims at providing indoor users with accurate location information,and researches and analyzes three key links in Zig Bee indoor positioning technology based on location fingerprint matching.In order to overcome the shortcomings of three aspects(fingerprint database establishment,reference node selection and location prediction)based on Zig Bee indoor location technology,a complete Zig Bee indoor location system is built.In this paper,starting from reducing the volatility of the Zig Bee signal,the signal strength is corrected,in order to improve the accuracy of indoor positioning.We analyze the influence of Zig Bee signal on the signal with the change of time,personnel movement,dynamic environment and other factors.Aiming at the characteristics of Zig Bee signal being time-varying,the improved z-score method is used to pre-process the signal intensity data set of offline phase sampling.The experimental results show that the proposed method can effectively eliminate the large variation of distortion value and establish a more accurate database of location fingerprints so as to improve the system’s positioning accuracy.Secondly,to improve the position prediction accuracy of position fingerprinting,a deep neural network algorithm based on improved stacked self-encoder is proposed.Re Lu activation function and Adam’s optimization algorithm are introduced into the stacked self-encoder.Feature extraction and parameter optimization are carried out by improving the stack self-encoder.Then the feature extracted information is predicted by nonlinear regression through deep neural network,and Zig Bee Signal characteristics and target location information model,through the model to meet the needs of positioning.The experimental results show that this method can effectively reduce the influence of the randomness caused by the fluctuation of the signalstrength on the system positioning accuracy and achieve a good balance between the positioning accuracy and the system high availability.Finally,aiming at the deficiency of all available Zig Bee reference nodes and the lack of correlation between sub-region and ZR,an algorithm based on k-means and strong correlation ZR selection is proposed.The indoor areas are clustered and divided into several sub-areas The correlation between the sub-region and ZR is calculated.ZR with strong relativity is selected as the training node in this subspace.Through the training of the localization model based on the improved deep neural network model of stacked auto-encoder,The experimental results show that this method can effectively select ZR with strong position resolution and reduce the impact of noise on the positioning accuracy. |
