Conventional　harmonic　source　location　methods　based　on　harmonic　state　estimation　require　phasor　measurement　units，therefore　their　engineering　applications　are　limited.　Aiming　at　this　problem，a　harmonic　source　location　method　based　on　feature　set　reconstruction　and　multi-label　classification　model　is　proposed　based　on　asynchronous　measurement　data　collected　by　power　quality　monitoring　devices.　The　sufficient　statistics　of　the　monitoring　data　is　used　to　mine　the　harmonic　information　of　the　measurement　period.　Meanwhile，a　label-specific　feature　learning　algorithm　is　used　to　reconstruct　the　feature　set，so　as　to　eliminate　the　influence　of　redundant　and　irrelevant　features　on　the　accuracy　of　harmonic　sources　location.　Then　a　configuration　method　of　measurement　devices　is　proposed　based　on　the　adjacency　matrix　and　sensitivity　analysis，which　uses　circuit　network　topology　information　to　achieve　measurement　device　configuration.　An　improved　extreme　learning　machine　based　harmonic　source　location　method　is　proposed，which　uses　the　reconstructed　feature　set　as　input　and　establishes　a　multi-label　classification　model　to　achieve　harmonic　source　location.　The　feasibility　and　effectiveness　of　the　proposed　method　are　verified　by　simulation　and　arithmetic　cases.　©　2024　Electric　Power　Automation　Equipment　Press.　All　rights　reserved.