Due to significant difference in precision of indoor visible light positioning caused by non-uniform light intensity distribution of the LED, a light intensity RSS fingerprint analysis based indoor visible light positioning method combined with sequence to sequence model. The light intensity RSS fingerprint vectors are set as input sequence, the three dimensional coordinates are set as output sequence, the sequence to sequence model is used to learn the non-liner mapping relationship between the RSS fingerprints and the indoor three dimensional coordinates. The self-attention mechanism is introduced to sequence to sequence model, the self-attention mechanism is used to analyze the discrimination of each element in the light intensity RSS fingerprint for different position, in the prediction phase, the importance of each RSS element is adjusted. Simulation results show that the proposed method reduces the average visible light positioning error under different signal to noise ratio conditions.