In　order　to　improve　the　prediction　accuracy　of　ultra-high　voltage　(UHV)　transformer　winding　temperature,　a　hybrid　model　based　on　convolutional　neural　network　(CNN)　and　long　short-term　memory　(LSTM)　network　is　proposed.　This　method　uses　a　large　number　of　historical　monitoring　data,　including　data　of　winding　temperature,　top　oil　temperature,　dissolved　gas　in　oil,　environmental　data　and　so　on.　Combined　with　the　characteristics　of　CNN　and　LSTM　networks,　CNN　convolution　layer　and　pooling　layer　are　used　to　extract　relevant　data　information　to　generate　feature　vectors,　then　the　feature　vectors　are　used　as　input　data　to　predict　winding　temperature　through　LSTM　network.　The　example　results　show　that　this　method　is　of　high　prediction　accuracy,　its　prediction　accuracy　is　better　than　that　of　single　LSTM　model,　which　can　effectively　improve　the　prediction　accuracy　of　UHV　transformer　winding　temperature.　©　2022,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.