To　improve　the　levitation　stability　of　the　maglev　train,　an　adaptive　PID　controller　based　on　Q-learning　is　proposed.　In　particular,　the　parameters　of　the　traditional　PID　are　trained　through　the　Q-learning　algorithm　where　three　Q　tables　for　the　PID　parameters　are　obtained.　When　the　maglev　train　runs　in　different　operating　conditions,　the　controller　parameters　can　be　adaptively　and　efficiently　selected　according　to　the　Q　tables.　The　performance　of　the　proposed　Q-learning　based　PID　controller　is　verified　by　comparing　it　with　the　traditional　PID　and　the　experimental　results　show　that　the　proposed　PID　controller　via　Q-learning　has　favorable　features　on　rapidity　and　stability.　Compared　with　the　traditional　PID　controller,　its　overshoot　is　reduced　by　6.19%,　and　the　adjustment　time　is　shortened　about　32.25%　during　the　transient　levitation　process.　©　2023　IEEE.