The Extreme Learning Machine (ELM) neural network algorithm in the traditional indoor Visible Light Positioning (VLP) system suffers from unstable convergence and a tendency to get stuck in local optimal states, resulting in decreased positioning accuracy. In this paper, a Sparrow Search Algorithm (SSA)-ELM neural network algorithm was proposed by introducing the SSA to determine the initial weights and thresholds of the ELM neural network.

Firstly, the Received Signal Strength (RSS) and location information within the targeted area are collected as fingerprint data. Subsequently, the SSA-ELM neural network is trained to obtain a prediction model, and the test set data is input into this model to derive the positioning result for the location under test. Finally, simulation experiments and a testing platform are designed.

The simulation results show that at the four receiving heights of 0, 0.3, 0.6, and 0.9 m in the three-dimensional space model, the average errors are 1.73, 1.86, 2.18, and 3.47 cm, respectively. Compared with the Back Propagation (BP), SSA-BP, and ELM positioning algorithms, the positioning accuracy of the SSA-ELM algorithm is improved by 83.55%, 45.71%, and 26.26%, respectively, while the positioning time is reduced by 36.48%, 17.69%, and 6.61%, respectively. Experimental tests have shown that the average positioning error of the SSA-ELM neural network algorithm proposed in the article is 3.75 cm, representing a 16.38% improvement in positioning accuracy compared to the unoptimized ELM neural network.

The SSA-ELM neural network algorithm proposed in this paper has an obvious optimization effect on the ELM neural network, which can significantly reduce the positioning error of the system and reduce the positioning time of the system.