Indoor Positioning System Based On Improved Filtering And Improved BP Neural Network

With the development of outdoor GPS positioning and the increasing demand for indoor positioning,researchers have shifted the study from outdoor positioning to indoor positioning.At present,indoor positioning technology has Wi-Fi,Bluetooth,ultra-wideband technology(UWB),inertial navigation system(INS),ZigBee positioning technology,etc.However,indoor positioning is much more complicated than outdoor positioning,and there are factors such as variable interference,strong environmental dependence,and obstruction.These can result in inaccurate positioning to give the user the wrong location and reduce efficiency.This thesis first introduces the background of indoor positioning,current research techniques and algorithms.By comparing the advantages and disadvantages of related technologies and algorithms,a positioning system based on the ZigBee protocol and the Received Signal Strength Indication positioning algorithm is adopted and improved it.It then analyzes the defects in the RSSI positioning algorithm,uses the RSSI signal strength interference factor and the RSSI ranging formula parameter as the improvement point.Finally,the RSSI-based trilateral positioning algorithm is proposed to integrate the system.Firstly,the RSSI value is subject to burst interference factors and random interference factors,resulting in inaccurate RSSI values and low precision.It is proposed to denoise and correct RSSI values using Kalman filter.At the same time,based on the influence of initial value on Kalman filtering,a density-based clustering algorithm-DBSCAN algorithm is proposed.The bursting factor is eliminated by the fusion algorithm of DBSCAN and Kalman filtering.The fusion algorithm using the mean filtering and the limited average filtering algorithm is then used to reduce the random interference of the RSSI.Secondly,for the uncertainty of parameters in the RSSI ranging formula,BP neural network is used to establish the prediction model with RSSI value as input and distance as output.To improve the BP neural network model,this thesis uses grey theory prediction model and Adam optimization algorithm to optimize BP neural network.This model improves the accuracy and stability of ranging and improves the efficiency of the algorithm.Finally,the improved three-edge weighted centroid algorithm based on RSSI is used to calculate the coordinates of unknown nodes to obtain more accurate coordinates.According to the improvement method described above,an experimental platform is constructed,and a laboratory in the school is used as a test environment to perform actual positioning effect test.It is verified that the improved algorithm proposed in this thesis is better than the original algorithm,the positioning error is smaller,the complexity of the ranging formula is avoided,and the positioning accuracy is improved.