Aiming　at　the　problem　that　single　typical　curve　cannot　satisfy　the　demand　of　load　uncertainty　analysis,　a　typical　interval　scene　mining　method　for　consumer　load　based　on　haar　wavelet　coding　and　improved　K-medoids　algorithm　aggregation　is　proposed.　The　low　dimensional　load　approximate　sequences　are　obtained　by　haar　wavelet　transform　of　the　original　load　curve.　The　feature　set　of　load　approximate　sequence　in　each　dimension　is　clustered　separately,　the　boundary　values　of　the　features　contained　in　the　clusters　are　extracted,　and　numerical　intervals　are　obtained　and　coded.　The　non-significant　numerical　intervals　are　eliminated　according　to　the　feature　proportion,　and　the　significant　numerical　intervals　with　different　dimensions　are　combined　to　obtain　the　load　interval　sequences　represented　by　strings.　The　string　difference　is　defined　to　measure　the　similarity　of　load　interval　sequences,　the　improved　K-medoids　algorithm　aggregation　is　used　to　obtain　the　clusters　of　load　interval　sequences,　and　the　boundary　values　of　load　approximate　sequences　in　the　clusters　are　extracted　to　obtain　the　typical　interval　scenes.　The　difference　threshold　is　set　to　realize　granularity　adjustment　of　typical　interval　scenes.　The　measured　load　data　of　a　user　in　Ireland　is　used　for　verification,　and　the　experimental　results　show　that　the　proposed　method　can　realize　mining　of　typical　load　interval　scenes　with　different　granularities.　©　2022,　Electric　Power　Automation　Equipment　Press.　All　right　reserved.