In　order　to　improve　the　forecast　accuracy　and　adaptability　for　rotating　machinery　working　conditions　with　unsteady　and　nonlinear　features,　an　optimization　prediction　method　of　variable　weight　RBF　combination　model　was　suggested.　This　model　was　built　based　on　variable　weight　RBF　network.　The　samples　were　weighted　according　to　the　time　to　output　and　the　combined　models　were　selected　according　to　the　average　relative　error　while　the　model　built.　As　a　result,　the　sufficient　effective　information　was　used,　and　the　fact　that　new　and　old　information　taking　different　effect　on　the　future　state　was　stressed.　The　method　was　verified　by　measured　data.　The　accuracy　of　variable　weight　RBF　combination　forecasting　method　was　better　than　single　RBF　model　and　single　weight　combination　forecasting　methods.　This　method　is　simple　to　program　and　more　adaptable　on　prediction　with　high　farecast　accuracy.