To　overcome　the　problem　of　batch　process　caused　by　the　traditional　process　dynamics　multistage　characteristic,　the　multiphase　auto　regression-principal　component　analysis　(AR-PCA)　monitoring　method　is　proposed　based　on　affine　propagation　(AP)　clustering　optimized　with　a　population　diversity-based　particle　swarm　optimization　algorithm　(PDPSO).　The　method　introduced　PDPSO　method　to　improve　the　AP　clustering.　It　avoided　the　blindness　of　common　method　that　indirectly　chose　the　preference　based　on　the　clustering　evaluation　index.　Then　we　established　the　AR-PCA　model　for　the　data　samples　of　the　multiphase　fermentation　process　to　eliminate　the　dynamic　characteristics　of　each　stage　and　the　auto-and-cross-correlation　between　variables.　Finally,　the　PCA　model　is　established　for　the　residual　of　the　AR　model　for　fault　monitoring　of　the　batch　process.　The　method　is　applied　to　the　process　of　penicillin　fermentation.　Experiments　show　that　the　method　can　effectively　divide　the　process　into　different　phases　and　reduce　the　false　and　leak　alarms.　©　All　Right　Reserved.