Time　series　forecasting　is　one　of　the　important　problem　in　the　time　series　analysis.　As　one　of　the　most　powerful　analysis　tools　for　time　series　forecasting,　Neural　Network(NN)　has　been　receiving　considerable　attention　since　many　years　ago　and　a　large　number　of　improvements　of　NN-based　forecasting　on　time　series　have　appeared　in　the　relevant　literature.　This　paper　review　the　structure　improvement　of　NN　and　the　main　combination　of　NN　and　other　pop　technologies　in　the　improvement　of　algorithms.　It　is　the　summary　of　this　paper　in　the　end.　©　2005　IEEE.