Ensemble　learning　shows　very　good　performance　when　dealing　with　the　problem　of　concept　drift.　Accuracy　and　diversity　are　two　important　characteristics　of　ensemble　learning,　and　diversity　is　the　key　factor　affecting　the　generalization　ability　of　ensemble　learning.　In　this　paper,　a　modeling　method　is　proposed　to　enhance　diversity　among　base　learners　based　on　parameters　evolution　of　base　learners.　The　method　uses　online　sequential　extreme　learning　machine　(OS_ELM)　as　base　learner.　The　base　learners　are　grouped　according　to　their　performance　on　the　sliding　window,　and　perform　evolution　operations.　At　the　same　time,　the　concept　of　gene　flow　is　introduced,　which　increases　the　diversity　among　base　learners　and　improves　the　prediction　performance　of　the　ensemble　algorithm　in　dealing　with　the　concept　drift　data　streams.　Finally,　the　rationality　and　effectiveness　of　the　proposed　algorithm　are　verified　by　using　the　synthetic　data　sets　and　real-world　data　sets.