Research On Neural Network Algorithm Of Visible Light Indoor Location Based On Lambert Model Optimization

Based on people’s demand for positioning functionality,visible light indoor positioning technology has gradually attracted the attention of experts and scholars.This technology has the advantages of low cost,high security,communication,and lighting without frequency control.Among a variety of visible-light indoor positioning systems,the most widely used method is to confirm the position of the receiving end based on the received signal strength(RSS).This method has a simple principle and strong practicability,but it is difficult to achieve accurate location of a complex and changeable indoor environment due to restriction factors such as noise interference and radiation model differences.Therefore,in this paper,the visible light radiation mechanism is integrated with the RSS positioning theory,the difference between the actual radiation model and the theoretical radiation model is reduced through Lambert radiation model optimization,and a complete fingerprint database is constructed.Then,the RBF neural network positioning system is taken as the model,and the mapping relationship between the internal data in the complete fingerprint database is fitted with the nonlinear mapping ability of the RBF neural network.To achieve accurate positioning of a complex and changeable indoor environment,the specific research content is mainly divided into the following aspects:1.The system model for visible light indoor positioning is built.The line of sight,non-line of sight,and noise in the visible light channel are analyzed,and the influence of non-visual links and noise on the accuracy of the positioning fingerprint database is studied.The theoretical model of the positioning system is also modeled and analyzed.2.Aiming at the difference between the actual radiation model of the light source and the theoretical model and the influence of a sparse fingerprint database on positioning accuracy,a Lambert model optimization algorithm is proposed.There are signal constraints such as Non Line of Sight and noise in the visible light positioning system,which lead to differences between the actual radiation model and the theoretical radiation model.The direct establishment of a fingerprint database will introduce large errors.In this paper,the collected RSS data will be modified through Lambert model optimization,with the measured data guiding the theoretical mathematical model.In order to solve the problem of sparse fingerprint databases,an Elman neural network was used to refine the parameters separated during correction and expand the parameter capacity.Finally,the parameters are converted into RSS,and a complete fingerprint database is constructed to support the RSS location technology.3.The RBF neural network algorithm is used for positioning.The complete RSS fingerprint database is integrated with the RBF neural network,and the fingerprint database is divided into training data sets and test data sets,and the two data sets do not intersect each other.The training data set was introduced into the RBF neural network,the nonlinear mapping relationship between RSS and coordinate position was established,and the final positioning model was established by simulating the indoor channel.The test data set is used to test the universality of the model.The error function was established,and the parameters of the RBF neural network were adjusted by the error of the test data set to achieve high precision indoor positioning.In order to verify the feasibility and reliability of combining the Lambert model optimization algorithm with a neural network in this paper,an indoor environment was simulated for verification,and a measurement platform was built.The results show that the average positioning error is 4.79 cm in the simulation environment of 4m×4m×3m,which verifies the feasibility of the algorithm.On the measured platform of 0.8m×0.8m×0.8m,the positioning test was carried out at the heights of H=0m,H=0.35 m,and H=0.6m.The results show that the Lambert model optimization algorithm proposed in this paper has an average positioning error of 3.67 cm when the measured data amount is reduced by half.Compared with the measured data amount at the same time,positioning accuracy improved by 48.38%.It provides a high-efficiency and highprecision positioning scheme for indoor positioning.