Research On Indoor Location Algorithm For Dynamic Environment And Privacy Protection

With the continuous development of human society and various emerging technologies such as mobile communication,cloud computing,Internet of Things and autonomous driving,Location Based Services(LBS)has become a basic application requirement for these industries and technologies,and research on location-based services is gradually gaining attention.Among them,WiFi-based indoor positioning systems can make use of the WiFi network infrastructure that has been widely deployed in various places,which has become a kind of indoor positioning system with low application cost.Moreover,WiFi fingerprinting based on Received Signal Strength(RSS)values does not require knowledge of the location of the WIFI signal transmitting base station and is less affected by indoor multipath effects.The environment changes or the AP changes can lead to changes in the distribution of signal strength values,resulting in reduced localization accuracy of WiFi fingerprint-based localization methods.To improve the localization accuracy of WiFi fingerprinting in dynamic environments,this paper proposes a localization strategy based on a priori knowledge and transfer learning.The strategy determines the location of special points through the priori knowledge,uses the RSS values of special points to accurately discern which APs’ signal strength distribution has changed,divides the offline domain and online random domain data into homogeneous and heterogeneous feature spaces according to the change of APs,solves the migration matrix reflecting the mapping relationship between them through different feature space data,then migrates the offline domain data to obtain the target domain matrix The new classifier is then trained for localization.The experimental results show that the proposed strategy maintains high localization accuracy in a dynamic environment.The location information of WiFi fingerprint-based localization methods is private information,which may lead to privacy leakage when multiple users jointly build a localization model.To protect user privacy,this paper proposes a federal learning-based indoor localization strategy that can fully fuse the feature vectors of subsystems while protecting user privacy.In this strategy,each participant first uses the initial model and their own data for model training,and then transmits their respective models to the global server,which receives the models and performs weighted aggregation of the global models based on model reliability,and then sends the models to the participants after weighted aggregation,and iterates until the global models converge.In order to prevent the model parameters from being accessed by intruders during the model transmission and to infer the user privacy information based on the model parameters,we use a data protection method with differential privacy for the model parameters so that intruders cannot access sensitive information.Experiments show that our strategy can achieve protection of user privacy with acceptable loss of accuracy while making the system more adaptive.