With the rapid development of the fifth-generation (5G) mobile communication and Internet of Things (IoT) technologies and the dramatic increase in the number of public indoor environments, research in the field of location estimation and tracking has shown great practical significance. The current mainstream global positioning system (GPS) and other various indoor wireless positioning technologies based on radio frequency communication cannot realize indoor precision positioning. Visible light communication (VLC) has attracted widespread attention from the academic community because of its performance advantages such as spectrum without authentication, high speed, environmental protection, energy saving, safety, economy. Therefore, visible light positioning technology has a broad application prospect. Specifically, the positioning method based on received signal strength indication (RSSI) is widely used for indoor visible light positioning with the advantages of high accuracy, low cost, and no clock synchronization. In recent years, research on machine learning (ML) and neural networks has developed rapidly, and various optimization algorithms and improvement schemes for indoor visible positioning have been proposed by integrating neural networks and RSSI positioning methods. Most of the previous studies on indoor visible light positioning based on neural networks only consider direct light and ignore the reflection of ceilings, walls, and indoor objects. In the actual environment, the existence of reflection will seriously affect the transmission of light signals and thus reduce the positioning accuracy, which cannot meet the actual needs when direct radiation is used to consider the positioning problem. At the same time, due to the difference in the distribution of light signals at different heights in the room, the height of a receiver will also directly affect the positioning accuracy and positioning error. In addition, the randomness of the initial weights and thresholds of the neural network can easily make the neural network fall into a local optimum. Using the intelligent search algorithm to determine the initial weights and thresholds of the neural network can both solve this problem well and accelerate the training speed of the network. In summary, this paper uses Circle chaotic mapping to improve the sparrow search algorithm (SSA) and optimize the extreme learning machine (ELM) neural network, and the paper proposes a positioning method combining ISSA-ELM neural network and RSSI based on Circle chaotic mapping optimization to achieve indoor visible light positioning with low latency and high accuracy. This method has taken into account the role of ceiling, wall, and ground reflections.
This study establishes an indoor visible light positioning model based on ISSA-ELM neural network and divides the positioning process into three stages. The first stage is the RSSI data acquisition stage, which establishes an indoor visible light positioning channel model based on the principle of VLC, arranges multiple LED light sources on the ceiling of the room, and uses multiple photoelectric detectors (PDs) in the receiving plane to receive and process RSSI signals from each LED light source respectively, and the real coordinates of the receiver are combined to determine the training set and test set of the neural network. The second stage is the neural network training stage. In this paper, we use an optimized ISSA based on Circle chaotic mapping optimization to determine the initial weights and thresholds of the ELM neural network, and the optimized SSA is not easy to fall into a local optimum and has a faster convergence speed. The neural network training uses the training set collected in the first stage, takes the RSSI data as the input of the neural network and the position coordinates corresponding to the PDs as the output of the neural network, trains the neural network, and establishes a prediction model for indoor visible light positioning. The third stage is the prediction model testing stage, in which the RSSI data are used as the input of the neural network using the test set selected in the first stage, and the predicted coordinates of the neural network are compared with the real coordinates of the test points, and the performance of the prediction model is evaluated using the positioning error and root mean square error function.
In this study, an ISSA-ELM neural network-based indoor visible light positioning method is proposed. The method firstly uses the ISSA algorithm to determine the initial weights and thresholds of the ELM neural network, which effectively avoids the problem of the weak generalization ability of the neural network brought by random initialization of weights and thresholds and speeds up the training speed of the ELM neural network to some extent. Secondly, considering the reflection of ceilings, walls, and floors inside the room, an indoor visible light positioning system based on the ISSA-ELM neural network is built. In a room of 5 m×5 m×3 m, it achieves low latency and high accuracy positioning with a neural network training time of 0.0454 s, average positioning time of 3.5 ms, and average positioning error of less than 4 cm. Finally, the ISSA-ELM method proposed in this paper is compared with seven other classical indoor visible light positioning methods. The results show that the positioning performance of the proposed method is superior, and the positioning error is reduced by 20.47%, 19.72%, 37.91%, and 42.32% in terms of four heights of 0 m, 0.5 m, 1.0 m, and 1.5 m, respectively, compared with the ELM neural network. The ISSA algorithm plays an obvious optimization role. In summary, the method proposed in this paper has fast positioning speed and high positioning accuracy, which can meet the positioning requirements of most indoor application scenarios.