GMDH NN ALGORITHM AND ITS APPLICATION IN DEFORMATION FORECASTING
Pan Guorong 1, 2) ; and Gu Chuan 1)
1)Department of Surveying and Geo-Informatics,Tongji University, Shanghai 2000922)Key Laboratory of Modern Engineering Surveying of SBSM, Shanghai 200092
Abstract:Aiming at several defects of GMDH NN(neural network), this paper has done some improvements on the aspects of preselection of initial variables number, constitute of partial expression, choice criterion of middle variables, stopping principle, and realized it with Matlab language. Applying the improved GMDH NN in nonlinear deformation data forecasting, and comparing the shortterm and longterm forecasting result with those obtained with BP NN and GMDH NN, in this paper, it is concluded that improved GMDH NN has preferable practicality and its forecasting precision has enhanced.