Fuzzy　broad　learning　system　is　regarded　as　an　effective　algorithm　to　utilize　the　measured　data　for　modeling　nonlinear　systems.　However,　due　to　the　possible　existence　of　data　inadequate　or　data　loss,　it　is　a　challenge　to　design　a　suitable　fuzzy　broad　learning　system　with　the　data　shortage　issue　for　modeling.　Therefore,　a　knowledge　transfer-based　fuzzy　broad　learning　system　is　developed　in　this　paper.　First,　the　knowledge　extracted　from　the　process　is　used　to　construct　the　initial　condition.　Then,　this　model　can　obtain　the　precise　parameter　and　structure.　Second,　a　knowledge　evaluation　mechanism　is　employed　to　rebuild　the　knowledge　by　judging　the　correlation　and　discrepancy.　Then,　the　knowledge　can　be　preferably　integrated.　Third,　a　transfer　gradient　algorithm　is　employed　to　adjust　the　parameters　of　fuzzy　broad　learning　system.　Then,　the　modeling　performance　of　knowledge　transfer-based　fuzzy　broad　learning　system　can　be　improved.　Finally,　a　benchmark　problem　and　a　practical　application　are　used　to　test　the　merits　of　knowledge　transfer-based　fuzzy　broad　learning　system.　The　results　demonstrate　that　this　model　can　achieve　superior　modeling　performance.　©　2021　IEEE.