Renewable　energy　and　large-scale　power　electronic　equipment　are　connected　to　the　grid,　which　makes　it　more　difficult　to　analyze　the　behavior　of　harmonic　sources　and　build　models.　The　existing　models　are　accurate　value　estimates　with　limited　application　scope,　which　are　difficult　to　reflect　the　uncertainty　influence　of　unconsidered　physical　factors　on　the　characteristics　of　harmonic　source.　To　solve　these　problems,　this　paper　proposed　a　modeling　method　based　on　Gaussian　processes　regression　(GPR).　Firstly,　the　linear　dependence　between　harmonic　current　and　voltage　of　harmonic　source　was　embedded　into　the　mean　function　of　Gaussian　process.　Secondly,　the　appropriate　covariance　function　was　selected　to　reflect　the　similarity　of　harmonic　characteristics　of　harmonic　source　under　different　working　conditions.　Thirdly,　the　maximum　likelihood　method　was　used　to　solve　the　model　parameters,　and　the　interval　prediction　of　harmonic　current　was　carried　out.　Finally,　aiming　at　the　problem　that　the　model　cannot　reflect　the　dynamic　change　of　harmonic　characteristics,　an　online　updating　strategy　was　proposed　to　make　the　model　accurately　track　the　harmonic　characteristics　of　the　monitored　harmonic　source.　The　feasibility　of　this　method　for　single　harmonic　source　modeling　was　verified　by　the　measured　data　of　EAF　and　the　simulation　data　of　12　pulse　rectifier,　and　the　feasibility　for　complex　multi　harmonic　source　system　modeling　was　verified　by　the　simulation　data　of　complex　network　of　multi　harmonic　sources.　The　results　show　that　the　proposed　method　can　reflect　the　uncertain　behavior　of　harmonic　characteristics,　track　the　changes　of　harmonic　characteristics,　and　has　the　characteristics　of　high　accuracy　and　strong　universality.　©　2022　Chin.　Soc.　for　Elec.　Eng.