Optimization Research On Model Of Ramie Yield

Model of ramie yield can not only raise the efficiency of ramie selection, speed up the breeding process, but also play a vital role in the cultivation and the field yield estimation. Hence high-precision ramie yield model is highly helpful to the ramie production. The "zhongzhul" and "xiangzhu3" are taken as test material in this research, utilizing multiple linear regression model, BP neural network and genetic algorithm BP neural network (GA-BP) to respectively establish and predict the predictive model for the yield of the two test material. Finally, compare the prediction accuracy and draw a conclusion. The main conclusions and progress are as follows:(1)In accordance with the research on ramie yield and practical experience of cultivation, the5factors that are most associated with ramie yield are identified, that is plant height, diameter of stem, thickness of cortex, fresh weight of the stem and effective plants.(2)Making use of the software SPSS to establish multiple linear regression predictive model for "Zhongzhul" and "Xiangzhu3". The multiple linear regression model for "Zhongzhul" yield is Y=-0.334+0.093X1+0.007X2-1.106X3+0.003X4+0.159X5(Y is the weight of raw linen, X1is plant height, X2is stem diameter, X3is thickness of cortex, X4is effective plants, X5is fresh weight of the stem), the maximum absolute error of the yield predicted and actual obtained from Conduct prediction using the regression model is10.89%, an average error is8.77%. Similarly, those of "xiangzhu3" are9.98%and8.84%respectively.(3)Design the parameters of BP neural network through analysis of characteristics of ramie field data and establish ramie yield predictive model. For "Zhongzhul", the maximum absolute error of the yield predicted from BP neural network ramie yield forecast model and yield actual is6.32%,an average absolute error is3.60%; Similarly, for " xiangzhu3", those are5.39%and4.57%respectively.(4)In view of the defects and shortcomings of the BP neural network itself, using genetic algorithm (GA) to optimize parameters for BP neural network, building ramie yield predictive model, carrying out sample testing similar with the BP neural network for "zhongzhul", the maximum absolute error of the yield predicted from GA-BP neural network ramie yield predictive model and yield actual is2.73%, an average absolute error is1.39%; Similarly, for "xiangzhu3", those are2.75%and2.65%respectively.(5)By comparison, the accuracy and stability of GA-BP neural network based Ramie yield predictive model of "zhongzhul" and "xiangzhu3" is the highest, followed by the BP neural network, the multiple linear regression analysis is the lowest.