Thermal　history　during　the　additive　manufacturing　(AM)　process　significantly　influences　the　performance　of　the　produced　structures.　In　numerous　studies,　researchers　have　focused　on　developing　thermal　prediction　models　for　enhanced　quality　control　and　defect　mitigation　in　AM　processes.　In　this　study,　we　address　the　challenge　of　online　thermal　history　prediction　in　arbitrary　observation　areas　during　the　AM　process.　We　propose　an　online　prediction　method　based　on　infrared　monitoring　and　deep　learning　technology.　Specifically,　we　consider　the　impact　of　the　surface　morphology　of　deposited　parts　on　emissivity,　and　develop　a　machine　learning　model　to　identify　and　predict　emissivity　values　within　the　observation　area.　We　use　an　improved　TCN-TrF　model　for　the　long-term　temporal　prediction　of　thermal　history　trends,　and　combine　it　with　a　cross-attention　mechanism　to　better　capture　spatial　and　temporal　features.　Numerical　and　case　studies　demonstrate　that　the　proposed　emissivity　identification　method　and　the　parallel　architecture　with　a　cross-attention　mechanism　significantly　enhance　the　model＇s　predictive　accuracy,　achieving　excellent　performance　on　unseen　data.　The　results　indicate　that　the　proposed　method　can　effectively　predict　emissivity　and　accurately　capture　thermal　trends　in　observation　areas,　showing　great　potential　for　process　control　and　thermal　management　in　AM.