Platform　on　probabilistic　automata　and　combined　with　evolution　thought　of　genetic　algorithm,　this　paper　constructs　a　bionic　learning　model　which　can　reflect　the　essence　of　Skinner　operant　conditioning.　The　designed　learning　model　is　named　as　genetic　algorithm-operant　conditioning　probabilistic　automaton　(GA-OCPA)　bionic　autonomous　learning　system.　After　each　learning　trial,　the　learning　system　firstly　obtains　the　information　entropy　value　based　on　operant　conditioning　(OC)　learning　result　and　uses　it　as　the　fitness　of　individual.　And　then　genetic　algorithm　is　performed　based　on　information　entropy　value　to　find　the　optimal　individual.　At　last,　the　OC　learning　algorithm　is　performed　to　learn　the　optimal　operant　action　in　optimal　individual,　and　correspondingly　a　new　information　entropy　value　will　be　obtained.　The　convergence　theorems　for　the　learning　algorithm　of　GA-OCPA　bionic　learning　system　is　presented,　and　the　simulation　analyses　in　motion　balancing　control　of　two-wheeled　robot　demonstrate　that　the　learning　of　GA-OCPA　bionic　learning　system　is　a　process　of　autonomously　acquiring　and　epurating　knowledge　and　has　high　adaptive　ability.