The　type　and　number　of　interference　source　users　at　the　point　of　common　coupling　(PCC)　are　increasing.　This　makes　the　index　and　spatial-temporal　characteristics　of　power　quality　(PQ)　disturbance　more　complex.　However,　it　is　difficult　for　the　PQ　monitoring　devices　to　monitor　all　terminal　users　of　each　feeder.　Recently,　some　methods　were　presented　to　identify　the　PQ　characteristics　of　multiple　users,　but　most　of　them　failed　to　accommodate　long-term　PQ　phenomena　and　uncertain　user　operating　conditions.　To　address　these　issues,　this　paper　proposes　a　industrial　users　PQ　characteristics　identification　method.　Using　derivative　dynamic　time　warping　(DDTW)　to　calculate　the　correlation　of　the　existing　multi-source　data,　it　accurately　identifies　the　PQ　characteristics　for　industrial　users.　Firstly,　the　over-limit　and　fluctuation　periods　of　PQ　monitoring　index　data　are　defined　as　disturbance　periods.　According　to　national　standard　limits　and　probability　distribution,　the　over-limit　and　fluctuation　values　of　PQ　time　series　and　their　periods　are　extracted.　Secondly,　the　screening　rules　of　key　harmonic　indicators　are　set　up　to　obtain　the　disturbance　time　series　data　of　key　harmonic　indicators,　voltage　deviation　and　negative　sequence　voltage　imbalance,　so　that　the　characteristics　identification　is　more　targeted.　Thirdly,　the　user　power　data　during　the　disturbance　period　are　extracted,　and　the　correlation　between　them　and　the　monitoring　indicators　are　calculated　by　using　DDTW.　Finally,　the　power　quality　characteristics　of　users　are　identified　according　to　the　correlation　degree　of　different　indicators.　The　simulation　results　show　that　the　proposed　method　can　select　the　user　groups　most　related　to　PQ　problem,　which　is　consistent　with　the　simulation　settings.　In　addition,　compared　with　the　traditional　DTW　algorithm,　this　shows　that　DDTW　overcomes　ill-conditioned　matching　of　time　series　data,　so　as　to　effectively　filter　out　the　data　with　strong　correlation　in　time　series.To　compare　with　the　identification　method　based　on　mutual　information,　the　power　quality　monitoring　data　with　an　interval　of　3　minutes　is　averaged　every　5　as　a　group,　which　is　aligned　with　the　power　data　of　15　minutes,　and　then　the　maximum　mutual　information　decision　method　is　used　to　identify　the　user　power　quality　characteristics.　The　result　shows　that　the　maximum　mutual　information　method　obtains　more　types　of　characteristics　and　some　errors.　The　following　conclusions　can　be　drawn　from　the　simulation　and　real　measurement　analysis:　(1)　Through　the　national　standard　limits　and　probability　distribution　method,　it　effectively　determines　key　indicators　for　long　time　scale　series.　(2)　The　DDTW　algorithm　is　used　to　match　the　power　quality　index　of　the　monitoring　point　and　the　power　time　series　data　of　the　user　side.　In　this　way,　it　is　more　practical　to　select　the　representative　indicators　that　are　greatly　affected　by　the　power　fluctuation.　(3)　The　proposed　method　only　needs　long-term　statistical　data　in　the　existing　relevant　systems.　It　is　convenient　to　realize　online　identification　of　user　power　quality　characteristics.　©　2023　Chinese　Machine　Press.　All　rights　reserved.