As　new　energy　sources　such　as　wind　and　photovoltaic　connected　to　the　distribution　network　in　large　quantities,　how　to　describe　the　uncertainty　of　wind,　photovoltaic　and　load　(WPL)　variations　and　generate　typical　operating　scenarios　is　becoming　more　and　more　important　for　the　operation　scheduling　and　planning　of　the　distribution　network.　Aiming　at　the　problems　of　traditional　clustering　methods,　which　are　difficult　to　deal　with　high-dimensional　data,　the　separation　of　feature　extraction　process　and　clustering　process,　and　the　lack　of　clustering　performance,　this　paper　proposes　a　scenario　generation　method　based　on　deep　fuzzy　k-means　(DFKM)　for　WPL　uncertainty　sources.　First,　based　on　one-dimensional　convolutional　auto-encoder,　we　extract　the　temporal　coupling　features　of　the　WPL　data,　and　then　use　the　fuzzy　k-means　with　adaptive　loss　function　to　perform　clustering　in　the　embedded　low-dimensional　feature　space.　In　the　model　optimization　process,　the　feature　extraction　process　and　the　clustering　process　are　combined　to　obtain　the　final　clustering　results　and　generate　typical　scenarios　based　on　different　types　of　centers.　The　example　takes　the　historical　data　of　a　region　in　southeast　China　as　the　research　object,　then　analyzes　the　clustering　indexes　to　verify　the　effectiveness　and　superiority　of　the　proposed　method.　©　2023　IEEE.