EPSO–CRP–ELM: LED layout and cross-LED RSS optimization for visible light positioning

Irregular room boundaries and localized background-light interference can produce spatially nonuniform received signal strength (RSS) fingerprints and degrade positioning accuracy. This paper proposes an enhanced particle swarm optimization–cross-LED relation projection–extreme learning machine (EPSO–CRP–ELM) method. EPSO optimizes the LED layout according to the mean and difficult-position errors, subject to a minimum signal-to-noise ratio (SNR) constraint and neighborhood refinement. CRP subsequently adjusts the relationships among multi-LED RSS channels, and ELM estimates the receiver position. Experimental results show that the proposed method achieves a mean absolute error (MAE) of 0.0372 m, representing a 53.56% reduction compared with fixed-layout ELM (FL–ELM). It also provides the best overall performance in MAE, root-mean-square error (RMSE), 75th- and 95th-percentile errors (P75 and P95), and the mean error of the worst 30% positions (Hard30) among the evaluated methods and demonstrates a degree of stability under receiver-height variations.