Research On Indoor Location Algorithm Based On Deep Belief Network

With the development of society and technology,the indoor environment of people's living is increasing.Because the GNSS signal can not be obtained in the room accurately,mobile users can not enjoy the location service provided by GNSS system in the indoor environment.The location based service LBS in the indoor environment has become a hot spot of attention.WIFI indoor positioning technology has many advantages in many aspects,such as low cost,network coverage and high positioning accuracy.At the same time,indoor location technology based on WIFI has become the main method of indoor positioning,which has greatly promoted the development of indoor location service.In this paper,in view of the defects in the establishment of fingerprint library and the selection of AP nodes in WIFI indoor positioning technology,this paper studies the related problems of indoor location algorithm based on deep belief network.The main achievements include:(1)The interference of the time variation of WIFI signal itself,the distance between the receiving point and the AP node,the occlusion of the personnel,the movement of the personnel and the attitude of the receiving equipment,etc.,on the signal intensity data are analyzed.(2)In view of the strong time variability of signal intensity signal itself,an indoor location algorithm based on the belief network of denoising depth is proposed,which aims to reduce the redundant information in the fingerprint library and improve the efficiency and accuracy of the algorithm.The algorithm divides the target location area into several network lattice regions,simplifies and preprocesses the collected RSSI signals,uses the network frame of the stacked noise reduction self-encoder,and passes the nonlinear transformation of the multilayer neural network to input the RSSI signal value after the reconstruction.The WIFI signal fingerprint location model after the network learning and training is combined with the classifier to classify the RSSI signal values obtained online,so as to speculate the corresponding indoor location area.(3)Aiming at the indoor location environment with a large number of AP,a noise reduction depth belief network localization algorithm based on reliable AP selection is proposed.In order to solve the problem of overfitting and increasing the complexity of time and space,a large number of AP nodes can be easily solved.The method uses K-means clustering algorithm to classify the fingerprint library,select reliable Fisher-AP selection mechanism to select reliable AP nodes,and train fingerprint library by using the denoising depth belief network model to implement the fingerprint library.(4)In view of the inadequacy of the complex indoor structure,which is difficult to obtain enough effective markup data,a semi supervised location algorithm,which combines the fast search with the CFSFDP and the limit learning machine(ELM),is proposed,which uses the CFSFDP clustering data set,and marks the missing location information of the clustering center,and extends the initial mark.Data are trained by ELM,and the marking data are expanded according to the output threshold vector and "transposition" idea.The accuracy of positioning is improved and the workload of data acquisition is reduced.