It　takes　many　efforts　to　establish　accurate　soft　sensor　models　because　of　the　increasing　complication　of　processes.　For　the　sake　of　solving　this　problem,　a　novel　multi-kernel　extreme　learning　machine　based　on　partial　least　square　regression　(PLSR)　is　proposed.　In　the　proposed　method,　different　kernel　functions　are　used　for　mapping　the　space　of　process　data　to　highly　nonlinear　space.　The　partial　least　square　regression　is　adopted　to　obtain　the　relationship　between　the　nonlinear　space　and　output　layer.　To　validate　the　performance　of　the　proposed　model,　a　case　study　using　the　High　Density　Polyethylene　process　is　executed.　Simulation　results　confirm　the　performance　of　the　proposed　model.　©　2019　JSME.