Municipal　solid　waste　incineration　(MSWI)　is　the　most　widely　used　waste　treatment　technology　worldwide.　Dioxin　(DXN),　one　of　the　by-products　of　the　MSWI　process,　is　by　far　the　most　toxic　contaminant.　Labeled　samples　are　extremely　limited　for　constructing　its　soft　sensor　measurement　model　because　offline　DXN　detection　takes　considerable　amount　of　time　and　cost.　In　addition,　the　number　of　pseudo-label　samples　and　the　optimization　of　hyperparameters　in　semi-supervised　models　is　a　challenging　problem.　A　multi-objective　particle　swarm　optimization　(PSO)　semi-supervised　random　forest　(RF)　algorithm　is　proposed　in　this　paper　for　DXN　emission　concentration　measurement.　First,　the　coding　design　of　the　selected　hyperparameter　value　and　pseudo-labeled　samples　is　realized　for　the　semi-supervised　algorithm　oriented　to　hybrid　optimization.　Subsequently,　the　particles　are　initialized　and　decoded　to　evaluate　the　fitness　of　the　model-oriented　generalization　performance　and　the　number　of　pseudo-labeled　samples.　The　termination　condition　of　optimization　is　then　assessed.　If　the　condition　is　unsatisfied,　then　the　decision　variable　of　multi-objective　PSO　is　updated.　Otherwise,　the　Pareto　solution　set　is　used　to　determine　the　optimal　solution.　Finally,　the　RF　model　is　constructed　on　the　basis　of　optimal　mixed　samples.　The　effectiveness　of　the　proposed　method　is　verified　by　using　benchmark　and　actual　MSWI　process　datasets.