Daily　operation　conditions　of　the　pump　turbine　(PT)　in　the　pumped　storage　power　plant　(PSPP)　are　changeable　and　complex.　The　flow　pattern　inside　the　PT　is　also　difficult　to　accurately　grasp.　In　the　vaneless　area,　unstable　flow　pattern　is　easy　to　occur,　which　will　further　cause　abnormal　pressure　fluctuation　of　the　PT　or　even　the　structure　of　the　plant.　Therefore,　it　is　necessary　to　analyze　dynamic　features　of　pressure　fluctuation　signals　(PFSs)　and　recognize　accurate　flow　patterns　in　the　vaneless　area　of　the　PT.　In　view　of　the　strong　non-stationarity　and　nonlinearity　of　the　PFSs　in　the　vaneless　area,　and　the　existing　signal　feature　extraction　methods　are　not　ideal,　a　method　called　generalized　refined　composite　multiscale　slope　entropy　(GRCMSE)　algorithm　is　proposed　to　extract　the　dynamic　features　of　PFSs　and　further　recognize　the　flow　pattern　by　combining　the　long　short-term　memory　(LSTM)　neural　network.　Verification　of　the　proposed　method　is　conducted　using　the　measured　signals　in　the　vaneless　area　of　the　PT　of　a　Chinese　PSPP.　The　results　showed　that　the　proposed　GRCMSE　algorithm　can　distinguish　the　dynamic　features　of　the　PFSs　corresponding　to　different　flow　patterns　more　accurately.　At　the　same　time,　compared　with　other　popular　recognition　methods,　the　GRCMSE-LSTM　method　has　the　best　recognition　performance　with　the　highest　recognition　accuracy　(＞95%)　and　Macro-F1　score　(＞0.95).　This　method　can　provide　effective　technical　support　for　the　intelligent　development　of　PSPP.　©　2025　Elsevier　Ltd