The　fuel　economy　of　a　plug-in　hybrid　electric　vehicle　is　largely　dependent　on　the　battery　energy　usage　during　various　driving　cycles.　In　this　research,　within　the　model　predictive　control　(MPC)　principle,　an　Ensemble　Learning　Velocity　Prediction　(ELVP)-based　energy　management　strategy　(EMS)　considering　the　driving　pattern　Adaptive　Reference　State　of　Charge　(AR-SOC)　is　proposed.　Firstly,　the　existing　methods　including　Markov　chain　(MC),　back　propagation　(BP)　and　radial　basis　function　(RBF)　neural　network　(NN)-based　velocity　prediction　models　are　described.　Then,　these　models　are　embedded　into　MPC-based　EMS　respectively,　and　the　validation　results　show　that　the　NN　performs　better　than　the　MC　by　comparing　the　prediction　precision,　computational　cost,　and　resultant　vehicular　fuel　economy.　By　incorporating　these　prior　knowledges,　a　novel　ensemble　learning　velocity　prediction　method　is　established　by　blending　BP-NN　and　RBF-NN.　Subsequently,　based　on　the　expected　trip　distance　and　the　velocity　prediction　results,　an　adaptive　reference　SOC　(AR-SOC)　trajectory　planning　method　is　developed　to　direct　the　distribution　of　battery　energy　for　different　driving　patterns.　Combining　with　the　ELVP　and　the　AR-SOC,　the　MPC-based　EMS　derives　the　optimal　torque-distribution　decisions.　Finally,　the　validation　results　indicate　that　the　proposed　strategy　achieves　superior　fuel　economy　under　various　driving　cycle　compared　with　the　benchmark　strategies.　(c)　2021　Elsevier　Ltd.　All　rights　reserved.