Research On High Precision 3D Indoor Visible Light Location Algorithm Based On Neural Network

With the rapid development of GPS,BDS,and other systems,high-precision outdoor positioning technology has made significant progress.In recent years,research on how to achieve high-precision indoor positioning technology has become the research direction of various scientific research teams,with indoor visible light positioning technology becoming a research hotspot in the field of indoor positioning technology.LED has become a signal transmitter for indoor positioning technology due to its advantages of high photoelectric conversion efficiency,energy conservation and environmental protection,and low cost.In this paper,research is conducted on the optimization of light source layout and algorithm optimization in the positioning stage after receiving light signals in indoor visible light positioning systems to improve the positioning accuracy of indoor visible light positioning systems.Indoor visible light positioning algorithms based on sparse training fingerprint library,momentum factor optimized particle swarm optimization,and limit learning machine(MMPSO-ELM)are proposed respectively to improve the positioning accuracy of the positioning system in the indoor two-dimensional plane;An indoor visible light positioning system scheme based on the Kernel Limit Learning Machine Integrated Framework(KELM-AWBagging)is proposed for light source layout fusion,which uses a weighted learning strategy combined with clustering pruning to improve positioning accuracy in three-dimensional space;Propose a technical scheme for correcting positioning errors in indoor visible light edge areas to solve the problem of large positioning errors in boundary areas.(1)Firstly,in order to solve the problem of large randomness of the model caused by the random generation of the initial training weights and threshold matrices of the limit learning machine neural network,a particle swarm optimization algorithm is used to optimize the initial training optimal matrix of the limit learning machine,and a momentum factor is added to optimize the traditional particle swarm intelligence algorithm,To avoid particle swarm optimization in the initial training weight matrix and threshold matrix optimization process of limit learning machine neural networks falling into local optimal values;At the same time,due to the characteristic of extreme learning machine neural networks that can obtain high prediction accuracy through sparse training data,data collection is conducted on the location plane,and a sparse fingerprint database based on received optical power is established for training and predicting location models;Finally,high-precision positioning in the indoor two-dimensional plane is achieved.(2)From two-dimensional planar positioning to three-dimensional spatial positioning,in terms of light source arrangement: add auxiliary light sources and optimize the layout of light source locations,enhance the intensity of received light signals in the wall edge area,improve the flatness of signal received light power,and use the optimized bat algorithm(DBA)to optimize the positions of the other four LED lights;Using an integrated learning model based on a kernel limit learning machine neural network to achieve localization: Applying the idea of integrated learning algorithms to the training process of the localization model,and adding weight strategies and clustering pruning to the Bagging algorithm to further improve its performance,using the kernel limit learning machine network as a sub learner.Experimental results show that this method greatly improves the positioning accuracy in three-dimensional space.(3)Finally,based on the analysis of the simulation experiment results of the indoor visible light positioning scheme proposed in the above two parts,aiming at the problem that the boundary area is greatly affected by multiple reflections of the light signal from the wall,and there is a multipath effect that causes the positioning accuracy of the positioning system to be low,K-means clustering algorithm is used to divide the spatial area into boundary areas and internal spatial areas,The kernel limit learning machine neural network algorithm is used for secondary training and positioning of the boundary region location model to achieve correction of the boundary region location error.