In　industrial　manufacturing,　most　batch　processes　are　multi-phase　and　uneven-length　batch　processes　in　nature,　phase-based　approaches　are　intuitively　well　suited　for　batch　process　monitoring　and　quality　prediction.　In　this　paper,　a　new　strategy　is　proposed　using　multi-phase　dynamic　partial　least　squares　(DPLS)　for　batch　processes　monitoring　and　quality　prediction.　Firstly,　batch　process　data　was　automatically　divided　into　several　phases　using　Gaussian　mixture　model　(GMM)　clustering　arithmetic.　Then　run-to-run　variations　among　different　instances　of　a　phase　are　synchronized　by　using　dynamic　time　warping　(DTW).　Finally,　multi-phase　DPLS　model　is　built　between　each　phase　and　the　quality　variables.　The　proposed　method　easily　handles　the　following　problems:　(1)　static　single　model;　(2)　process　and　its　model　do　not　match;　(3)　linear　method　may　not　be　efficient　in　compressing　and　extracting　dynamic　nonlinear　process　data.　The　idea　and　algorithm　are　illustrated　with　respect　to　the　typical　data　collected　from　a　benchmark　simulation　of　fed-batch　penicillin　fermentation　production.　The　simulation　results　demonstrate　the　effectiveness　of　the　proposed　method　in　comparison　to　original　DPLS.