Research On The Impact Of Climate Factors On Wood Properties Based On The Intelligent Decision

The timber production is primarily determined by properties of wood,which influenced by various factors such as genetic factor,site factor,climatic factors,and the climatic factors are the most significant.Therefore,facing with a serious of climatic and environmental problems and the sharp drop of forest resources,it is very useful for timber production to know the effect laws of wood properties by climatic factors,and for the realization of quality,efficient,high-yield de-velopment and application.According to research fruits,the climatic influencing factors of wood properties is studied with data analysis and regression methods,which could not by wood properties itself and not by climatic factors.As artificial Linda Qing Yang to be main study object,based on the studies of the wood properties influence of climatic factors,the prediction method of wood properties of artificial Linda Qing Yang by intelligent decision measure.First of all,three important properties for timber production are determined,and the climatic factors apparently unrelated to these properties are ignored.Secondly,based on the studies on the RBF neural network,the RBF neural which network structure has been improved is provided and used in the prediction model.Thirdly,based on determination of the important factors,the extreme and cumulative temperature values which could impact wood properties are studied,the inputs of model are qualified and prediction model based on fuzzy neural network is established.At last,according to the problems in the studies,a RBF neural network improved by genetic algorithm is proposed,the parameters of model are determined and the wood properties prediction model of climatic factors is established.The results obtained by above process:By studying the references,the outputs of the prediction model contain:growth ring width,latewood percentage,latewood fiber wall thickness.By using Linear Regression Analysis,the inputs of every model are determined.In witch,the inputs of growth ring prediction model contain:September sunshine percentage,November sunshine percentage,April relative humidity,July relative humidity,August relative humidity,February highest temperature.The inputs of latewood percentage contain:August average temperature,June relative humidity,June rainfall,August rainfall,August sunshine percentage,November sunshine percentage,January ground temperature,January minimum ground temperature,March maximum temperature.The inputs of latewood fiber wall thickness contain:July average temperature,May relative humidity, September relative humidity,October relative humidity,November relative humidity,May rainfall,September rainfall,October rainfall,August sunshine percentage,July surface temperature,October ground temperature, August lowest temperature. The prediction model was established by using RBF neural network which network structure has been improved and the simulation was conducted by using Matlab software.From the results we could know that the convergence speed and prediction accuracy are ideal.In the view of velocity,the convergence speed were36steps,38steps,38steps respectively,which were enough to meet online training.In the view of prediction accuracy,the average errors of three models were2.37%,0.24%,0.09%respectively,the maximum errors were2.87%,0.29%,0.12%respectively.So,the RBF neural network based on the structure improvement is suitable for the prediction for latewood rate and latewood fiber wall thickness,but has too much errors in the prediction of growth ring width witch need improve further.The inputs of latewood rate prediction model were refined,and the model was established by using the RBF neural which has been improved,the August sunshine percentage and January lowest temperature were replaced by biological temperature and warmth index which were extreme and cumulative values impact latewood rate seriously.According to simulation by Matlab software,the average error is0.21%,maximum error is0.25%,compared with the0.24%and0.29%before,the prediction accuracy has been improved in some extent.A kind of RBF neural network improved by genetic algorithm was used in the prediction of wood properties,and the growth ring width,latewood percentage,latewood fiber wall thickness were predicted.According to the results,we could know that errors of growth ring width model,latewood rate width model and latewood fiber wall thickness model were0.31%,0.15%,0.10%respectively,whose accuracy was all improved compared with the models before,and the convergence speed of latewood fiber thickness was improved,which was32steps from38steps before.The results show the accuracy and ability to learn of the prediction was improved by the modified algorithm.