Study On Growth Modeling With Artificial Neural Network For Masson Pine Plantation

In order to probe a set of new stand growth simulation technology, the paper regarded Masson pine plantation as the research object ,and regarded management density experiment data, fixed sample data of individual trees and routine sample plot data of stand as training samples and examining samples. Under MATLAB environment, the paper, using BP neural network modeling technology, carried on systematic research to all kinds of growth model of the masson pine plantation.Firstly, the material is discussed and is studied with the method to what full text will be used. Putting forward one kind of man-made neuron model simplifying and with this stand growth neural network model serving as the foundation constructs building the method. Putting forward one kind of simple active definite method of hiddeing layer number, putting forward the method that with the quality with the quantitative analysis each other combines at the same time carries on the model performance analysis, and defining the layer neuron number hiddeing from view, in order to surmount the neural network has appeared easily excessively to draw up to reckon up according to and reducing the major defect of model example ability in traininging, and makes the analysis kill reliably and can believe.Then, the paper constructed site index and neural network model of stand density index at first, and used a large number of routine sample plot and trunk analyzed data to train and examine the built models so that the suitable model structure was gotten The analysis results indicated that the precision of the single shape neural network site index models was very high and generalization ability was very strong; the derived formulas of single shape site index curve and calculation had very good charting and calculating result. The polymorphous site index neural network model was better simulation result than Richards'. The results of comparison and analysis indicated that the polymorphous site index model based on Richards growth function couldn't solve polymorphous shape and cross problems of site index curve in essence; while the polymorphous site index based model on neural network could do it well. At the same time, the inverse model of the polymorphous site index neural network model was successfully built and trained so that the site index value could accurately calculated Stand density index neural network model' simulation result was better than Reineke'. The derived formulas of stand density index curve and density index calculation had very good charting and calculation result.On the foundation of site index and density index model study, systematically study three kinds of stand growth models and their application in growing and yield forecasting at the stand.By the age, the site index and the land area per tree as input matrix and by the stand average breast-height diameter, the stand average height and the stock per hectare as output matrix, the paper constructed a whole-stand growth neural network model with multi-input and multi-output, and selected suitable model structure by a large number of routine plot data. Visible qualitative analysis with the simulation curved surface of three-dimensional and quantitative analysis by calculation of precision and regress showed that the model was not only fit the growth law of forest but also has very high precision and very strong generalization ability. The compatibility according to between the model at the same time by stand number density and stand diameter growth neural network model is jointly constructedbuilding the stand section to amass growth neural network model.Using the relative diameter as input variable and using the accumulating frequency in number of trees as output variable, the paper constructed the neural network model of tree diameter distribution.The optimum model training and selecting with the real data does not have the system deviation, and weibull's distribution model comparison, and what possesses better draws up closing effect. Grinding to build the neural network model of calculation...