Modern　blast　furnace　ironmaking　technology　primarily　utilizes　the　thermal　condition　of　the　furnace　belly　to　reflect　the　furnace　temperature　status.　However,　the　complexity　of　the　smelting　process　makes　effective　modeling　and　control　extremely　challenging.　During　the　ironmaking　process,　the　control　of　furnace　temperature　directly　affects　production　efficiency　and　product　quality.　Since　the　furnace　temperature　is　difficult　to　measure　directly,　the　silicon　content　in　molten　iron　is　commonly　used　to　reflect　the　thermal　state　of　the　blast　furnace.　Traditional　methods　for　predicting　the　silicon　content　in　molten　iron　have　limitations　and　struggle　to　adapt　to　complex　and　variable　production　conditions.　With　the　advancement　of　neural　network　technology,　this　paper　constructs　a　prediction　model　for　the　silicon　content　in　blast　furnace　molten　iron　by　creating　a　hybrid　of　Convolutional　Neural　Networks　(CNN),　Long　Short-Term　Memory　Networks　(LSTM),　and　a　MultiHead　Attention　Mechanism　(MA).　Through　deploying　and　analyzing　the　CNN-LSTM-MA　model　in　a　real　production　environment,　the　superiority　of　the　CNN-LSTM-MA　model　in　silicon　content　prediction　has　been　verified.　©　2024　SPIE.