In order to overcome the impact of multipath effects and inter user interference on communication reliability in indoor multi-user scenarios faced by existing non orthogonal multiple access (NOMA) models in visible light communication(VLC) systems, improve spectral efficiency and communication rate, a NOMA-VLC system combining deep neural network (DNN) channel estimation and virtual time reversal(VTR) channel equalization is proposed. By analyzing the characteristics of multi-user NOMA-VLC channels, DNN is used for accurate channel estimation, and VTR technology is utilized to achieve channel equalization, focus energy, suppress multipath effects, enhance communication reliability and user fairness. The simulation results show that in a two user scenario, the system performance improved by 5.1 dB and 4.9 dB respectively at a bit error rate of 10^-3, with performance advantages of 2 dB and 2.4 dB compared to other algorithms.