For the problem of cross subframe interference caused by the use of dynamic subframe configuration in Long Term Evolution in Unlicensed band (LTE-U) system, a hybrid dynamic clustering algorithm which considered large-scale loss and cell traffic volume was proposed. Firstly, the information of large-scale loss and cell traffic volume was periodically measured by base station, then the corresponding relevance metric value was calculated. Secondly, according to the relevance metric value, the cells were clustered by polling to realize periodic update of clustering result. Finally, the dynamic subframe configuration was performed according to the updated cell clustering result. Simulation results showed that compared with the traditional static clustering algorithm in the medium traffic arrival rate conditions, the uplink-downlink user average throughput of hybrid dynamic clustering algorithm was increased by about 16.92% and 34.33%, the uplink-downlink user average delay was decreased by about 14.18% and 36.32%. Simulation results demonstrate that the hybrid dynamic clustering algorithm can effectively reduce the impact of cross subframe interference, and improve system throughput, which has better performance compared to the traditional static clustering algorithm.