In order to enhance the bandwidth utilization of the communication network, a service efficiency optimization algorithm founded on supermartingale theory is devised for the visible light communication （VLC） uplink system to fulfill the differentiated demands of terminal services regarding the network quality of service （QoS）. Firstly, the network queue system model based on the Markov modulated on-off （MMOO） process and the ALOHA random access process is established, and the probability of each terminal device accessing the coordinator is derived. Then, the supermartingale theory is employed to assess the delay performance of the system, and the multi-user QoS efficient guarantee issue is transformed into a service rate minimization problem constrained by the martingale domain delay bound. Finally, the honey badger algorithm （HBA） is utilized to solve the problem. The simulation results demonstrate that the algorithm can satisfy the diverse QoS requirements of terminal services and optimize service efficiency of the network.