Most　of　the　existing　fault　arc　circuit　breaker　is　only　for　single　circuit　single　load.　Aiming　at　the　problem　of　complex　running　load　of　low-voltage　power　lines,　set　up　an　experimental　platform　to　collect　the　current　signals　of　series　fault　arcs　in　main　and　branch　lines　under　the　conditions　of　single　load　operation　and　multi-load　operation　of　typical　household　loads,　to　create　a　waveform　database.　Various　time-frequency　domain　feature　quantities　are　extracted,　and　a　subset　of　the　feature　quantity　data　is　constructed　by　random　forest　and　input　to　a　regularized　limit　learning　machine　(RELM)　for　fault　arc　detection,　and　the　most　appropriate　input　weights　and　implied　layer　thresholds　of　the　RELM　are　optimally　selected　by　a　particle　swarm　optimization　(PSO)　to　improve　the　accuracy　of　fault　arc　identification.　The　result　shows　that　the　PSO-RELM　algorithm　achieves　an　accuracy　of　98.37%　and　can　accurately　identify　fault　arcs　in　multi-load　situations.　©　Beijing　Paike　Culture　Commu.　Co.,　Ltd.　2024.