Due　to　the　huge　difference　of　multiphase　batch　processes　in　the　dominant　variables　and　process　characteristics　of　each　operation　phase,　meanwhile,　in　order　to　reduce　the　leaking　alarm　rate　and　false　alarm　rate　of　traditional　methods　in　phase　hard　Classifying　and　process　modeling　ignoring　dynamic,　a　multiphase　auto　regression-principal　component　analysis　(AR-PCA)　monitoring　method　for　batch　progress　based　on　the　batch　weighted　soft　classifying　is　proposed.　inverse　distance　weighted　(IDW)　and　single　variable　control　charts　are　introduced　to　improve　affinity　propagation　clustering　(AP),　which　avoids　the　limitation　of　a　single　batch　as　the　input　of　AP　cannot　represent　the　stage　characteristics　of　the　entire　production　process,　and　the　defect　of　AP　unrecognizing　the　transition　stage　can　be　addressed.　After　AR-PCA　and　PCA　models　are　established　for　the　transition　phase　and　the　stable　phase　respectively,　higher　precision　than　the　traditional　method　to　establish　a　unique　model　with　entire　batch　data　can　be　achieved,　while　eliminating　the　dynamic　of　transition　phase.　Leaking　alarm　and　false　alarm　can　be　effectively　reduced.　Design　of　experiments　is　carried　out　by　the　penicillin　fermentation　simulation　platform　and　the　actual　production　process　of　recombinant　E.　coli,　and　results　indicate　the　feasibility　and　effectiveness　of　the　proposed　method.　©,　2015,　Yi　Qi　Yi　Biao　Xue　Bao/Chinese　Journal　of　Scientific　Instrument.　All　right　reserved.