Research On Indoor Positioning Algorithm Based On Integrated Learning And Location Fingerprinting

Scalable,stable and accurate indoor positioning technology is the primary goal of large location-aware services in the future.With the widespread deployment of wireless hotspot,the demand for location-based services is increasing.Among various indoor positioning technologies,the positioning technology based on location fingerprint is one of the positioning algorithms mainly studied by many researchers in the field of indoor positioning,because it does not require expensive hardware facilities and can realize the establishment of a positioning system in a pure software manner through existing resources.However,due to the inherent noise and severe time-varying of wireless signals,and the large number of APs deployed in large buildings,the collected location fingerprint database has the characteristics of high dimension and sparse values,which makes the positioning algorithm not high positioning accuracy in practical application and not robust to the dynamic changes in the environment.In this paper,the uncertainty of RSS changes when location fingerprint algorithm is located,that is,the volatility leads to low positioning accuracy,and the deployment of a large number of wireless hotspots in indoor environments across buildings and floors may cause "dimensionality disaster" to influence the positioning accuracy and efficiency.Therefore,the neural network structure based on Contractive Auto-encoders is proposed to reduce the AP characteristics to reduce the data storage space,reduce the noise impact and improve the efficiency of the algorithm.In this paper,based on the fingerprint model RSS localization algorithm in the realization of accurate positioning in the period of change lead to the uncertainty of the volatility high-dimensional redundant features,AP and may cause the "dimension disaster" influence on indoor positioning accuracy and efficiency,is proposed based on contraction since the neural network structure of the encoder AP feature reduction,to reduce the dimension of feature space so as to achieve strong robustness and precision of positioning results.By reducing AP features,it can compress data,reduce data storage space,remove noise influence in AP features and reduce algorithm running time.Aiming at the hierarchical level of location estimation across buildings,floors and multiple regions,the integrated learning technology is introduced and a positioning algorithm based on CAE-XGBoost is proposed.In practical applications,the positioning problem is transformed into a classification problem by predicting the symbol position instead of the physical coordinates,and the mapping relationship between the location fingerprint and the target building,the target floor,and the target area is established.The CAE-XGBoost model proposed in this paper integrates CAE into a trainable feature extractor,and automatically obtains the most characteristic fingerprint features of highdimensional features from the input end,and uses XGBoost as the top-level identifier of the network to provide more accurate output.This unique two-stage model guarantees high reliability of feature extraction and positioning results,robustness of the positioning system and scalability of large-scale positioning scenarios.The experiment was tested on UCI's public dataset UJIIndoorLoc dataset.The results show that the proposed method has higher positioning accuracy than other methods in the same database,and the effectiveness of the method in indoor positioning problems is verified.