Research On WiFi Indoor Localization Algorithm Based On Deep Learning

In daily life,Wi Fi wireless communication is widely used.The location fingerprinting algorithm based on Wi Fi technology attracts the attention of researchers.However,the RSSI of Wi Fi is easily affected by the complex indoor environmen t causing RSSI contain a lot of noise and has a strong time-varying nature.Currently,most of indoor localization methods are based on the shallow structure of machine learning algorithm,the feature express ability on nonlinear RSSI data in indoor space i s not strong,and it is difficult to structure a complex localization model.In the deep learning model,the stacked denoising auto-encoder has strong generalization ability,it can extract and transform nonlinear feature of the original input data,which can effectively overcome the noise interference of the data and the data's characteristics obtained by automatic learning are more robust.The main work of this paper is arrange d as follows:Firstly,we set up a Wi Fi signal collection platform,and research the influence of indoor environmental factors on RSSI.Based on the collected Wi Fi RSSI data,experiments are performed on the characteristics of time and location areas.In the experiment,we further study the relationship between the time regular of people indoor activities and the time-varying characteristics of RSSI.From the analysis of experimental results show that Wi Fi signal is interfered by indoor complex environmental factors in the propagation process,which is the main reason weaked the stability of RSSI.The RSSI of Wi Fi has characteristics such as time-varying,non-linear,correlation,and stationary,which makes RSSI exhibit a greater complexity and diversity in the indoor environment.Secondly,the main reasons of generated high-dimensional redundant AP features are analyzed.Point to the problem that high-dimensional redundant AP features easily cause the “dimensional disaster” of the fingerprint database,and the problems that may affect the research of indoor location,we propose an indoor localization method based on feature reduction(RFB-SDAE).ReliefF algorithm is used to simplify the AP characteristics of the fingerprint database,and then trainning and learning the database via a deep auto-encoder.Finally,experimental results on the standard database verify that RFB-SDAE not only has a better positioning effect,but also has a higher operating efficiency than the positioning method without reduction pretreatment.Finally,according to the analysis of the characteristics of the RSS I,the indoor location method is affected by the time-varying characteristics of the RSSI,a deep denoising auto-encoder localization algorithm based on RSSI time sequence is proposed.In the first place,we establish a RSSI fingerprint database that added time correlation,and then a stacked denoising auto-encoder constitutes a deep learning network,pre-training and fine-tuning the RSSI time sequence data that containing the regularity of human activity noise,and establishing a nonlinear interior positio ning model with good robustness,and lastly the model is used to position prediction in indoor environment.Experiment results show that TS-SDAE effectively reduces the impact of time-varying characteristics of RSSI on indoor positioning accuracy,and the location accuracy of TS-SDAE is better than the KNN and DBN methods with the same situation.Via two consecutive weeks of experimental results show that TS-SDAE has better robustness,stability and positioning effect in long-term regional positioning prediction.