Existing　non-intrusive　load　identification　methods　face　with　the　difficulty　of　being　applied　to　industrial　users　whose　operation　and　ON/OFF　status　are　hard　to　be　acquired.　Focusing　on　the　above　challenge,　this　paper　proposes　a　non-intrusive　identification　method　of　harmonic　source　loads　for　industrial　users.　Utilizing　Clarke　transform,　an　equation　with　respect　to　voltage　and　current　is　formulated　to　calculate　the　integrated　equivalent　impedance,　which　is　then　decomposed　with　the　improved　complex　local　mean　decomposition　(CLMD)　and　complex　fast　independent　component　analysis　(C-FastICA)　achieving　non-intrusive　separation　of　the　equivalent　impedance　signals　of　individual　branches.　The　separated　impedance　signals　will　be　used　as　the　basis　for　load　type　identification.　The　proposed　method　is　validated　by　numerical　simulations　as　well　as　actual　three-phase　voltage　and　current　monitoring　data.　The　case　study　shows　that　it　can　effectively　identify　the　types　of　harmonic　source　loads　while　being　resistant　to　noise　and　impacts　of　sampling　accuracy　and　diverse　load　capacities,　which　obviously　demonstrates　its　practical　applicability.