Abstract: To address the challenges of complex network architectures, a large number of hyperparameters, and gradient explosion encountered during deep learning training and position prediction in visible light indoor positioning,which can cause the problem of poor positioning timeliness and unstable accuracy ,this paper introduces a broad-learning approach for visible light indoor position perception. First, the light radiation data acquired from the indoor visible light environment was utilized as the feature node for the network. Subsequently, a prediction model for visible light indoor positioning based on the broad learning system was established. Next, the dimensionality of the network feature nodes was expanded. Finally, the pseudo-inverse matrix was computed to train and evaluate the network model. In the laboratory experiment conducted in a 4 m×4 m×3 m environment, the proposed algorithm demonstrated an average positioning error of 11.63 cm. The cumulative probability of prediction errors less than 10 cm was 52%, and that for errors less than 20 cm was 97%. Compared to BP(back propagation) and RBF(radial basis function) neural networks, the width learning network exhibited a 55.1% and 39.9% increase in positioning speed, respectively. The method in this paper improves the accuracy and speed of visible light indoor position perception, and provides a stable and reliable method for indoor position perception.