Optimization of wind shear escape training for civil aviation pilot trainees via reinforcement learning and sequential pattern mining

Abstract In wind shear escape training, the operation manual-based operations, which were developed by experts in the field of civil aviation, obtained via qualitative analysis would perhaps cause untimely, redundant, omitted, and out-of-sequence operations when under complex conditions. To improve windshear escape training, this study starts with an actor-separated proximal policy optimization (ASPPO) algorithm to generate optimal agents, considering that operations in flight involve both discrete and continuous actions. Further based on the action sequence generated from the optimal agent, sequential pattern mining is used for quantitative analysis to obtain the maximum frequent action sequence. The results of the wind shear escape test, as an example, revealed that the optimal agent based on ASPPO could make an aircraft escape fast with stable attitudes. Meanwhile, the comparison of action sequences obtained from the PrefixSpan algorithm and pilot trainees indicated the imperfections in pilot trainees’ operations.