High-voltage　switchgear　contact　overheating　is　a　critical　factor　in　power　system　failures,　requiring　effective　prediction　models　for　safe　operation　and　maintenance.　This　paper　proposes　a　two-step　prediction　method　based　on　Long　Short-Term　Memory　(LSTM)　networks　and　Partial　Least　Squares　(PLS)　regression　to　address　medium-　and　long-term　prediction　challenges.　The　LSTM　model　predicts　future　ambient　temperature　and　load　current　using　historical　data　from　a　10kV　substation　in　Fujian　Province,　China,　collected　every　5　minutes,　with　trigonometric　features　incorporated　for　periodicity.　PLS　regression　then　uses　these　predictions　to　estimate　switchgear　contact　temperature.　The　dataset　comprises　2018　data　points　spanning　seven　days.　Experimental　results　demonstrate　that　the　proposed　LSTM-PLS　model　achieves　a　37.3%　reduction　in　RMSE　and　an　8.4%　improvement　in　correlation　coefficient　compared　to　standalone　LSTM,　with　a　40.8%　reduction　in　error　against　other　combined　models.　This　method　significantly　enhances　accuracy　and　efficiency,　offering　a　practical　solution　for　monitoring　heating　in　high-voltage　switchgear　under　real-world　conditions.　Future　research　could　explore　integrating　additional　features,　such　as　environmental　factors　or　advanced　optimization　algorithms,　to　further　refine　the　prediction　performance.　©　2025　IEEE.