The　active　distribution　network　(ADN)　integrates　a　substantial　amount　of　renewable　energy　sources　(RESs),　which　exhibits　considerable　variability　and　uncertainty.　To　solve　the　challenges　of　renewable　energy　volatility　and　the　lack　of　intelligence　and　flexibility　in　traditional　distribution　networks,　this　paper　firstly　constructs　a　bi-level　optimization　model　for　the　dynamic　reconfiguration　of　ADNs,　incorporating　soft　open　points　(SOPs)　and　energy　storage　systems　(ESSs).　The　mathematical　model　formulated　in　this　paper　is　inherently　characterized　by　its　high-dimensionality,　complexity,　non-linearity,　and　stochastic　optimization　nature.　Secondly,　a　double　deep　Q　network　algorithm　embedded　with　physical　knowledge　(PK-DDQN)　is　developed　to　resolve　the　constructed　model　accurately　and　rapidly.　The　upper-level　model　optimizes　the　topology　of　the　distribution　network　using　the　double　deep　Q　network　algorithm.　In　contrast,　the　lower-level　model　employs　second-order　cone　programming　to　optimize　the　operation　of　ADNs　incorporating　SOPs　and　ESSs.　This　divide-and-conquer　approach　enhances　the　solution＇s　efficiency.　Finally,　the　superiority　and　scalability　of　the　proposed　algorithm　are　verified　on　the　modified　IEEE　33　and　69-bus　distribution　systems.　The　simulation　results　demonstrate　that　compared　with　the　genetic　algorithm　(GA),　the　power　losses　are　reduced　by　3.77%　and　23.47%,　and　the　voltage　deviations　are　reduced　by　28.63%　and　23.41%,　respectively.　Additionally,　compared　with　the　mixed-integer　second-order　cone　programming　(MISOCP),　the　computational　efficiency　is　increased　by　21.84　times　and　36.15　times.