Background:　Nuclear　masses　are　of　fundamental　importance　in　both　nuclear　physics　and　astrophysics,　and　the　masses　for　most　neutron-rich　exotic　nuclei　are　still　beyond　the　experimental　capability.　The　relativistic　continuum　Hartree-Bogoliubov　(RCHB)　theory　has　achieved　great　successes　in　the　studies　of　both　stable　and　exotic　nuclei.　The　mass　table　based　on　the　RCHB　theory　has　been　constructed　with　the　assumption　of　spherical　symmetry　[Xia　et　al.,　At.　Data　Nucl.　Data　Tables　121,　1　(2018)].　The　upgraded　version　including　deformation　effects　based　on　the　deformed　relativistic　Hartree-Bogoliubov　theory　in　continuum　(DRHBc)　is　under　construction,　and　the　part　for　even-even　nuclei　has　been　finished　[Zhang　et　al.,　At.　Data　Nucl.　Data　Tables　144,　101488　(2022)].　The　kernel　ridge　regression　(KRR)　approach　is　a　useful　machine-learning　approach　in　refining　nuclear　mass　prediction,　and　is　found　to　be　reliable　in　avoiding　the　risk　of　worsening　predictions　at　large　extrapolation　distance　[Wu　and　Zhao,　Phys.　Rev.　C　101,　051301(R)　(2020)].　Purpose:　The　aim　of　this　work　is　to　combine　the　RCHB　mass　model　and　the　KRR　approach　to　construct　a　high-precision　and　reliable　nuclear　mass　model　describing　both　stable　and　weakly　bound　neutron-rich　exotic　nuclei.　Another　purpose　is　to　utilize　the　masses　of　even-even　nuclei　from　the　DRHBc　theory　to　validate　the　performance　of　the　KRR　approach.　Method:　The　KRR　approach　is　employed　to　refine　the　RCHB　mass　model　by　learning　and　representing　the　mass　residual　of　the　RCHB　mass　model　with　the　experimental　data.　The　leave-one-out　cross-validation　is　applied　to　determine　the　hyperparameters　in　the　KRR　approach.　The　DRHBc　mass　model　for　even-even　nuclei　is　employed　to　help　to　analyze　the　physical　effects　included　in　the　KRR　corrections　and　examine　the　KRR　extrapolations.　Results:　The　refined　RCHB　mass　model　with　KRR　corrections　can　achieve　an　accuracy　of　root-mean-square　deviation　385　keV　from　the　experimental　masses.　The　major　contributions　contained　in　the　KRR　corrections　are　found　to　be　the　deformation　effects.　The　KRR　corrections　also　contain　some　residual　deformation　effects　and　some　other　effects　beyond　the　scope　of　the　DRHBc　theory.　The　extrapolation　of　the　KRR　approach　in　refining　the　RCHB　predictions　is　found　to　be　very　reliable.　Conclusions:　A　mass　model　benefiting　from　the　RCHB　model　with　continuum　effects　properly　treated　and　the　KRR　approach　is　constructed.　This　model　is　demonstrated　to　be　accurate　in　reproducing　the　masses　of　experimentally　known　nuclei　and　reliable　in　extrapolating　to　the　experimentally　unknown　neutron-rich　regions.