Visible light positioning system based on stacking learning model

In the field of Visible Light Positioning (VLP), ensemble learning algorithms can improve positioning accuracy. Among these, stacking learning models reduce the overfitting issues that may occur with individual models by integrating multiple algorithms, thereby making the model more robust in positioning systems. This paper proposes a three-dimensional visible light positioning scheme based on a stacking ensemble learning algorithm. The scheme uses Weighted K-Nearest Neighbors (WKNN) and Extreme Learning Machine (ELM) as base learners, with Linear Regression (LR) as the meta-learner. By integrating the advantages of different algorithms, it enhances the accuracy of indoor positioning. Experimental results show that in a 2.4 m $\times$ 2.4 m $\times$ 1.5 m indoor environment, the VLP system using this scheme achieved an average positioning error of 0.021 m, which is significantly better than traditional single algorithms. Particularly, even under varying lighting conditions, the stacking algorithm maintained high accuracy, verifying its adaptability in complex environments. These results demonstrate the feasibility and advantages of the proposed scheme for practical applications.