Research On Visible Light Indoor Positioning Technology Based On Biological Inspiration And Neural Network

Visible light positioning(VLP)is a new positioning technology that integrates lighting and positioning based on LED lighting technology.Due to the non-linear model of the Lambertian radiation model of LED light sources,which differs greatly from the actual application process,the property of indoor positioning systems has been seriously involved.Although artificial neural networks can effectively fit the non-linear relationship between light radiation information and position coordinates to improve positioning performance,the lack of prior knowledge of indoor environmental parameters cannot meet the data requirements of neural networks in the training phase,resulting in slow convergence speed,easy overfitting,and low system positioning performance.Therefore,the objective of this paper is to incorporate bio-inspired algorithms into artificial neural networks,in order to overcome the limitations of using these networks for indoor VLP systems,and proposes a visible light indoor positioning technology that combines bio-inspired algorithms with neural networks.Firstly,the indoor VLP system consists of both a line-of-sight(LOS)link model and a nonline-of-sight(NLOS)link mode is composed,and the light source layout is optimized to achieve international lighting standards.On this basis,the indoor illuminance is simulated,the spectral estimation detection technology is used to separate light radiation information on receiving planes at different heights,and a fingerprint database is constructed in combination with calibrated coordinates.Secondly,two types of neural networks suitable for fitting nonlinear problems,the BP neural network and the RBF neural network,are applied to the VLP system to establish a mapping relationship between indoor light radiation information and corresponding calibrated coordinates to each light source.The output data is then used to calculate the positioning coordinates through an error constraint model established based on RSS principles.Experiments were conducted to compare the performance of the two neural networks used in the indoor VLP.The consequence emerges that under the same target error and training data size,to achieve the target error,the training velocity of RBFNN is more quickly to BPNN.Furthermore,due to the performance limitations of neural network structures,a visible light indoor positioning scheme that combines bio-inspired algorithms with neural networks is designed based on the neural network.The GA,PSO,and BAS algorithms are used to optimize the RBFNN,exploring the optimization effect of bio-inspired algorithms on the performance of RBFNN applied to indoor VLP systems.According to the findings,bio-inspired algorithms can notably enhance the rate at which neural networks are trained during the training phase,prevent networks from falling into local optima and overfitting.Finally,through comparative experiments,the optimization performance of bio-inspired algorithms on neural networks in the training phase is verified.The results show that under the same indoor space,light source layout,target error,and training data size,various bio-inspired algorithms can reduce the duration of neural network training by 20% to 50%,while also ensuring that the mean error is limited to 4cm.In the actual measurement experiment,an experimental area measuring 0.8m in length,width,and height was created to conduct numerous positioning trials,resulting in an average positional error of 4.33 cm.This paper adopts a bio-inspired algorithm to optimize neural networks,significantly improving their training speed.This provides a new perspective for applying visible light indoor localization in practical scenarios.