Experiments submitted to equipment grid have quality of service (QoS) requirements, and advance reservation is used to satisfy such requirements. Due to the dynamic behaviors and fluctuations of resources in equipment grid, some previously accepted advance reservations are unable to be fulfilled. In this paper, we present a predictive admission control algorithm to decide whether new advance reservation requests can be accepted according to their QoS requirements and prediction of future resource utilization. Historical data are used in this algorithm to predict future status of resources. Experiments demonstrate that our algorithm can reduce the number of accepted advance reservations that fail to be fulfilled and keep the resource utilization ratio at an acceptable level.