Application of Elman dynamic recurrent neural network to forecast tree growth

Fully considering the specific dynamics,randomicity and nonlinear of forest growth,and the structural characteristics of dynamically recursive Elman model,the authors set up Elman type breast-height diameter and tree-height growth of dynamic neural network models with growth data of analytical Pinus tabulaeformis in Beijing Mountainous Area of China.Research showed that it reached very good fitting and simulation in the nonlinear modeling,which was 99.45% of fitting accuracy,99.42% of simulation accuracy used in breast-height diameter growth model;97.30% of fitting accuracy,97.29% of simulation accuracy in the tree-height growth model.Moreover,the fitting and simulation S-shape curve conformed to the rule of tree growth.Further compared the Elman dynamic model with the conventional BP static model,the authors discovered that the Elman model has better fitting,forecasting and stability,which must have the good application prospect in forestry in each kind of dynamic process models.