Abstract: Patrol cars play an important role in emergency response of traffic safety on freeways.Existing studies are based on prior history,accident data,or structure of road network,the essence of which is static and inflexible.Based on accident data,traffic flow,occupancy,and speed,a real-time forecast model for accidents is developed using Gradient Boosting Decision Tree(GBDT)method.Entity,states,and attributes of the patrol system are analyzed,and a discrete event simulation model for emergency response of patrol cares is developed.Computer simulation technology is applied to study the influences of accuracy and error rate of forecast on emergency response time.Compared with the time in normal state,the simulation results show that the average response time decreases from 23.5 min to 17.2 min on weekdays.