Data　aggregation　has　been　emerged　as　a　basic　approach　in　wireless　sensor　networks　(WSNs)　in　order　to　reduce　the　number　of　transmissions　of　sensor　nodes.This　paper　proposes　an　energy-efficient　multi-source　temporal　data　aggregation　model　called　MSTDA　in　WSNs.　In　MSTDA　model,　a　feature　selection　algorithm　using　particle　swarm　optimization　(PSO)　is　presented　to　simplify　the　historical　data　source　firstly.　And　then　a　data　prediction　algorithm　based　on　improved　BP　neural　network　with　PSO　(PSO-BPNN)　is　proposed.　This　MSTDA　model,　which　helps　to　find　out　potential　laws　according　to　historical　data　sets,　is　deployed　at　both　the　base　station　(BS)　and　the　node.　Only　when　the　deviation　between　the　actual　and　the　predicted　value　at　the　node　exceeds　a　certain　threshold,　the　sampling　value　and　new　model　are　sent　to　BS.　The　experiments　on　the　dataset　which　comes　from　the　actual　data　collected　from　54　sensors　deployed　in　the　Intel　Berkeley　Research　lab　made　a　satisfied　performance.　When　the　error　threshold　greater　than　0.15,　it　can　decrease　more　than　80%　data　transmissions.