<p>The coverage rate of the front vehicle cut-in scenarios in the tests of autonomous vehicles is deficient, and the scenario elements show high-dimensional characteristics. A comprehensive mode in this paper that integrated the random forest method, Monte Carlo double-layer sampling(MCDLS), K-means clustering algorithm, and simulation was proposed to generate and test high-speed front vehicle cut-in scenarios. Among these, the random forest method was used to evaluate the importance of scenario elements and the risk perception parameter. The MCDLS was applied to generate test cases for the Automatic Emergency Braking (AEB) scenarios of front vehicle cut-in scenarios under high-speed conditions, which were compared with Metropolis-Hastings sampling and importance sampling. The sampled results were categorized into four risk levels, and typical scenarios under high-speed conditions were derived via K-means clustering for simulation testing. A lateral danger state recognition module was incorporated into the simulation to trigger AEB system, when both lateral and longitudinal danger conditions were simultaneously met, preventing false or premature activation of AEB system. The result revealed that lateral speed of cut-in vehicle, longitudinal speed of host vehicle, relative longitudinal distance, and relative longitudinal speed were identified as key scenario elements. Compared with the other two sampling methods, the application of MCDLS generated more dangerous scenarios and significantly improved the danger rate. In the simulation test, all the typical test cases for the AEB scenarios of front vehicle cut-in scenarios under high-speed conditions successfully avoided collisions. The method can potentially assist researchers in conducting real vehicle tests.</p>