Study Of Sensitivity Analysis Method For Model With Dependent Input And Its Application

There is great signaificance in sensitivity analysis for model selection,model verification and parameter screening.Most of the existing sensitivity analysis methods are limited to models with independent input variables.There are a few sensitivity analysis methods based on variance decomposition that have been proposed for cross-correlation input parameters,but the calculation cost is higher.Although this type of sensitivity analysis method can quantitatively compare the sensitivity of parameters,it is not suitable for high-dimensional complex models with cross-correlation input variables.Because of the inherent physiological pro-cesses described,plant models are essentially closely related to the epistasis effect and pleiotropy effect.Therefore,input parameters inevitably of the model have cross-correlation,and the dimensions of the parameter are high;rather,the sensi-tivity analysis method based on variance directly is not operable.The calculation cost of the traditional Morris method is low,whose qualitative analysis can re-alize the preprocessing of parameter dimensionality reduction,which reduces the amount of calculation for subsequent quantitative sensitivity analysis.However,most of the sensitivity analysis based on the Morris method does not consider cross-correlation input parameters.In view of this,this paper proposes a new improvement measure based on the thoery of the classical Morris method and Nataf transformation,which can be applied to the model with dependent input.In order to reduce the calculation cost,an efficient algorithm is designed in this paper.Compared with the im-provement measures proposed by Qiao Geu and Monica Menendez[1]:(1)higher computational efficiency;(2)less sensitivity indicators.Three classic examples are provided to verify the correctness and efficiency of the extend Morris method.Next,the extend Morris method proposed is combined with the quantitative sensi-tivity analysis method to analysis the SUNFLO model.The results show that the influence of high-sensitivity parameters mainly comes from the epistasis effect and the pleiotropy effect.The guiding suggestions for plant models in the application of virtual breeding are provided.sensitivity analysis strategy for high-dimensional complex models also is afforded,which has the advantages of low computing cost and feasibility...