Some　special　manufacturing　fields　such　as　aerospace　may　encounter　mixed　production　of　multiple　research　and　development　projects　and　multiple　batch　production　projects.　Under　these　special　production　conditions　resource　conflicts　are　more　severe,　resulting　in　uncertain　operating　times　that　are　difficult　to　predict.　In　addition,　a　single　project　may　have　tens　of　thousands　of　supporting　products,　making　it　difficult　to　effectively　control　the　total　construction　process.　To　address　these　challenges　this　paper　proposes　new　methods.　A　model,　EMA-DCPM　(dynamic　critical　path　method)　incorporating　attention　mechanisms　in　Enterprise　Resource　Planning　and　Mechanical　Engineering　Society)　has　been　proposed.　This　model　predicts　product　job　time　through　machine　learning　methods　and　discovers　the　predictive　advantage　of　the　attention　mechanism　through　data　comparison.　The　CPM　control　algorithm　was　improved　to　enhance　its　robustness　and　an　efficient　modeling　method,　＂5+X＂　was　proposed.　This　new　method　is　suitable　for　mixed　line　planning　management　in　sophisticated　manufacturing　projects　and　has　value　for　practical　applications.