Aiming at the problems of low positioning accuracy and poor stability of traditional visible-light positioning methods based on the strength of the received signal in dynamic environments ,this paper proposes an indoor visible-light positioning system using an improved northern goshawk optimization ( NGO) algorithm fused with an optimized Elman neural network (RNGO-Elman) based on the received signalstrength ratio (RSSR) . Thisarticle proposes selecting an auxiliary reference point ,using the RSSR of the testreference pointto the auxiliary reference pointand the trueposition ofthe receiverasthe training setdatato establish a fingerprintdatabase that is not affected by a dynamic environment. Aiming at the problems of NGO algorithms such as slow convergence speed and tendency to fallinto local optimums ,the refractive reverse learning strategy was used to initialize the population ,increase its diversity ,and introduce nonlinear weighting factors to accelerate the convergence speed and avoid falling into localoptimums. The improved NGO algorithm was used to optimize the initialweights and thresholds of the Elman neural network and construct the RNGO-Elman dynamic localization prediction model. The simulation results show that under an experimental space of4 m × 4 m × 3 m ,the optimized RNGO-Elman localization modelhad an averagelocalization errorof1. 34cm ,and the localization accuracy was improved by 82% and 21% compared with the Elman localization algorithm and the NGO-Elman localization algorithm ,respectively. When the LED emission powerfluctuated ,the positioning errorsofthe RNGO-Elman modelbased on the RSSR were 1. 29 and 1. 38cm. The proposed visible lightpositioning method hasthe advantagesofhigh positioning accuracy and stable positioning performance.