With　the　growing　penetration　of　solar　photovoltaic　(PV)　generation,　advanced　data　analysis　methods　have　been　applied　to　the　smart　grid　operation.　However,　the　low-temporal-resolution　PV　generation　data　limits　the　utilization　of　the　data　analysis　methods,　because　the　low-temporal-resolution　PV　generation　data　contains　too　little　information.　On　the　other　hand,　the　existing　data　reconstruction　methods　are　less　than　satisfactory　in　reconstructing　high-temporal-resolution　PV　generation　data　from　low-temporal-resolution　data,　since　most　of　them　cannot　fully　capture　the　characteristics　of　PV　generation　data.　To　address　this　issue,　a　PV　generation　data　reconstruction　method　based　on　improved　super-resolution　generative　adversarial　network　is　proposed　in　this　paper.　First,　a　data-image　construction　method　is　proposed　to　encode　the　PV　generation　data　into　the　so-called　data-images.　Furthermore,　we　develop　a　data-image　super-resolution　generative　adversarial　network　(DISRGAN)　model,　and　the　data-images　are　used　to　train　the　DISRGAN　model.　Finally,　based　on　the　trained　DISRGAN　model,　a　general　framework　is　developed　to　reconstruct　high-temporal-resolution　PV　generation　data　from　low-temporal-resolution　data.　Numerical　experiments　have　been　carried　out　based　on　PV　generation　data　from　the　State　Grid　Corporation　of　China,　to　reconstruct　the　high-temporal-resolution　data　from　low-temporal-resolution　data.　The　results　demonstrate　the　superior　performance　of　the　proposed　framework　compared　with　a　series　of　state-of-the-art　methods.　©　2021　Elsevier　Ltd