This paper researches into the actual machine learning system taking advantage of the idea of evolutionary computation, and realizes the multi-goal optimization technique by applying the implicit parallelism mechanism and the parallel rule triggering strategy. In accordance with the specific task of machine learning, simulated pilot training grading standards, this paper adopts a kind of free hybrid coding scheme (multi-level and dynamic), establishes several simple and complete evolutionary operations, realizes the mechanism of competition by the credit assignment with brigade bucket algorithm. It relies on evolutionary algorithm to search for, find and select the proper rules, realizes several implicit objectives by inducting heuristic knowledge to create default rule structure, and designs a genetic evolutionary machine learning system (GEML-1) with satisfied robustness and flexibility. The method of machine learning can be generalized to apply in different kinds of expert system and military decision support system so as to provide a new kind of designing method and realizing path for machine learning.