DF-AUWIFI:Adaptive Indoor Localization System Based On Deep Forest

The digitalization of the physical world and the development of Internet of Things(IoT)technology have stimulated the widespread application of indoor positioning,and indoor positioning technology is booming.In recent years,WiFi-based fingerprint indoor positioning technology has attracted much attention due to its low cost and high precision.Although the existing indoor positioning scheme no longer relies on Global Navigation Satellite System(GNSS)signals,there are still shortcomings such as high cost of positioning solution or low real-time positioning.Therefore,we propose a DF-AUWIFI positioning system,aiming at Construct a universal positioning system with low cost,high precision and good real-time performance.This article has three innovations,which are described as follows:(1)Aiming at the problem that the labor cost of offline fingerprint database construction is large and the received signal strength value changes with the environment,this paper proposes an automatic construction and update scheme of fingerprint database based on RPL-MRT1 model.In this scheme,the proposed fingerprint database construction algorithm FPAC-RPL based on the regional path loss model is used to automatically build the fingerprint database,and it is executed in parallel in the construction process,which greatly saves the fingerprint database construction time.In addition,the fingerprint database automatic update algorithm FPA U-MRTI based on multi-frequency hybrid radio tomography(MRTI)technology is proposed to monitor the dynamic changes of obstacles to trigger the dynamic update of the fingerprint database.(2)In view of the problem that the existing positioning model cannot provide good support in big data scenarios,this paper proposes the LightRF algorithm.The algorithm introduces the random characteristics based on the original PV-Tree algorithm,and builds the LightRF algorithm through the ensemble learning mechanism.In the positioning system,we apply the LightRF algorithm to the fingerprint augmentation module(FA),so that the model can cope with large data volume scenarios and reduce the risk of model overfitting.Finally,experiments show that the LightRF algorithm can effectively deal with big data scenarios and improve the positioning accuracy of the positioning model.(3)Aiming at the problem that the positioning accuracy and the real-time positioning cannot be balanced,this paper proposes an adaptive WiFi fingerprint indoor positioning system based on deep fingerprint forest(DeepFPF):DF-AUWiFi,which uses cloud-based computing platform to conduct parallel model training and location prediction.The system consists of three parts,namely offline DeepFPF model training,online position decision and automatic reconstruction of the localization model.The DeepFPF localization model draws on the implementation mechanism of deep forests,and divides the model into a fingerprint augmentation module and a cascade forest module.The adaptive WiFi fingerprint indoor localization system combines the low training cost of random forest(RF)and the insensitivity to noise data,and the model has the ability of deep learning to characterize learning through the mechanism of cascading forests.We use the multi-granularity scanning mechanism of deep forest and LightRF algorithm in the fingerprint augmentation module(FA),and propose a fingerprint augmentation module FA construction algorithm.In particular,we have added an automatic reconfiguration mechanism for location models to cope with changing environments.Finally,experiments show that the DF-AUWiFi indoor positioning system can achieve higher performance in terms of localization accuracy,timeliness and cost compared with the existing neural network indoor positioning scheme.