Aiming　at　the　problem　that　the　presence　of　outliers　in　sample　data　can　corrupt　the　model　performance,　which　leads　to　undesirable　results,　a　soft　sensor　modeling　method,　i.　e.　the　adaptive　weighted　least　squares　support　vector　machine　(AWLS-SVM)　regression　method　is　presented　for　the　modeling　of　wastewater　treatment　process.　Firstly,　in　AWLS-SVM,　the　least　square　support　vector　machine　regression　method　is　employed　on　the　sample　data　to　develop　the　model　and　obtain　the　sample　datum　fitting　error.　Secondly,　according　to　the　fitting　error,　a　weight　is　adaptively　assigned　to　each　modeling　sample　via　the　improved　exponential　distribution　weighting　scheme　to　reduce　the　influence　of　random　error　on　model　performance.　Then,　a　global　optimization　algorithm,　i.　e.　the　hybrid　chaos　particle　swarm　optimization　simulated　annealing　(CPSO-SA)　algorithm　is　adopted　to　select　the　optimal　model　parameters　of　the　LS-SVM　and　improve　the　generalization　capability　of　the　model.　The　simulation　experiment　results　show　that　the　influence　of　the　outliers　on　the　model　performance　is　eliminated　in　AWLS-SVM,　and　the　prediction　performance　and　robustness　of　the　AWLS-SVM　model　are　better　than　those　of　WLS-SVM　and　LS-SVM　methods.　Furthermore,　the　AWLS-SVM　method　was　applied　to　develop　the　soft　sensor　model　for　sewage　disposing　effluent　quality　in　wastewater　treatment　process,　and　satisfactory　result　is　obtained.　©,　2015,　Science　Press.　All　right　reserved.