SVI,　a　sludge　bulking　index,　is　difficult　to　be　obtained　online.　A　soft　sensor　model　of　SVI　based　on　improved　PSO-Elman　neural　network　is　proposed　in　this　paper.　First,　to　solve　the　problems　of　nonlinear,　hysteresis　characteristics　and　so　on　of　sludge　bulking　process,　an　Elman　neural　network　with　dynamic　recursive　properties　is　introduced　to　determine　the　model　structure.　Secondly,　to　improve　the　learning　ability　and　convergent　accuracy　of　the　proposed　SVI　soft　sensor　model,　an　improved　particle　swarm　algorithm　is　studied　to　optimize　the　connection　weights　of　Elman　neural　network.　Finally,　the　proposed　SVI　soft　sensor　model　is　applied　to　the　actual　process　of　wastewater　treatment　process.　The　simulation　results　show　that　the　soft　sensor　model　can　predict　the　SVI　values　online,　and　owns　better　predicting　accuracy.　©　2014　IEEE.