With　the　high　access　of　renewable　energy,　complex　and　changeable　transmission　networks,　and　frequent　load　interactions,　the　dynamic　characteristics　of　low-carbon　power　systems　have　become　more　complex　and　random.　The　10　ms-level　dynamic　measurement　data　of　Phasor　Measurement　Unit　are　the　basis　for　dynamic　awareness,　control　and　decision.　However,　the　phenomenon　of　low-quality　data　usually　exists　in　PMU　measurements.　Considering　the　spatial–temporal　correlation　reflected　by　the　random　matrix　single-ring　theorem　and　correlation　coefficients,　two　system　operating　states　and　two　types　of　low-quality　PMU　data　are　determined.　Based　on　singular　value　decomposition　and　reconstruction,　the　distribution　of　the　residuals　between　the　original　data　and　reconstructed　data　is　analyzed　to　realize　low-quality　PMU　data　identification.　To　improve　the　identification　accuracy,　a　dynamic　threshold　selection　method　of　spatial–temporal　correlation　analysis　is　proposed　for　identification　criteria.　The　feasibility　and　applicability　of　this　method　has　been　verified　in　the　simulation　data　of　IEEE　39　bus　system　and　PMU　measured　data　of　the　practical　power　grid.　©　2023　Elsevier　Ltd