Research On Indoor Positioning Algorithm Based On Reinforcement Learning

The indoor location fingerprint positioning technology based on iBeacon is currently the most in-depth research technology,which is mainly divided into two parts: offline sampling and online positioning.In the indoor environment,the propagation distance of iBeacon is free space.Due to artificial walking,the wall blocking will cause iBeacon Path loss model changes,offline stage and online stage RSSI changes have a great impact on positioning accuracy,here for this situation,study how to obtain a stable RSSI,and can better adapt to changes in the environment to improve indoor positioning Accuracy.The main contents are as follows:First,the basic principle and propagation model of iBeacon are studied.The iBeacon free space propagation loss formula is obtained.The corresponding distance-based indoor positioning method is studied according to the publicity.The advantages and disadvantages of various methods are summarized.algorithm.Then it briefly introduces two algorithms based on RSSI reception strength,focusing on the location fingerprint algorithm.The common KNN algorithm in the location fingerprint matching algorithm is improved,and a more comprehensive AWKNN algorithm is proposed.By constructing an experimental platform and based on the comparison of simulation data,the positioning error of the AWKNN algorithm is reduced by 55%.Then the method of establishing fingerprint database offline in location fingerprint is studied.Considering that wireless signal propagation will receive interference,and according to offline sampling data analysis,it is found that RSSI shows variability,and a critical iBeacon filtering method is proposed.The adaptive Gaussian filtering method processes the data.In the result,the variance of RSRS has been reorganized,and the fingerprint library the quality has been improved.Then,deviating from the traditional method,the latest artificial neural network technology is studied,and the iBeacon neural network positioning technology that has been implemented now is studied.A series of decreasing errors and a BP neural network structure with comparable positioning accuracy are gradually reduced by 29.5%.The error is reduced by 1.1%.Finally,the implementation of the method on the Android platform confirmed the feasibility and performance improvement of the algorithm based on the experimental data.