This　paper　addresses　the　consensus　tracking　problem　of　leader-follower　discrete-time　multi-agent　systems.　To　overcome　the　shortcomings　of　existing　reinforcement　learning　(RL)　algorithms,　we　propose　a　novel　data-based　adaptive　reinforcement　learning　algorithm　by　introducing　adaptive　parameter.　Distinguishing　itself　from　most　existing　work,　the　study　assumes　that　the　dynamics　of　all　agents　are　entirely　unknown　and　the　complete　state　vector　is　not　accessible,　the　algorithm　proposed　only　requires　some　input-output　data　relevant　to　the　agents　to　be　available.　Based　on　this　representation,　the　algorithm　strikes　a　balance　between　data-based　value　iteration　and　data-based　policy　iteration,　obviates　the　need　for　the　limitation　of　initial　control,　exhibits　a　fast　rate　of　convergence,　and　achieves　the　consensus　in　multi-agent　systems.　Through　comprehensive　simulation　research,　the　effectiveness　of　the　proposed　algorithm　is　verified.　©　2024　Asian　Control　Association.