Statistical Diagnosis Of Nonlinear Model With Right-censored Data

This paper considers how to solve statistical diagnosis problem of the nonlinear models with right-censored data,which is an extension of general nonlinear model.First,we reconstruct the log-likelihood function and use the Guass-Newton Iterative method to obtain the parameters.Then we use the method of case-deletion model,mean-shift outlier model and local influence analysis.Based on the idea of case-deletion model,we obtain the formula of the parameters and propose generalized Cook distance and likelihood distance to determine outliers or influential points.Then we prove the equivalence between case-deletion model and mean-shift outlier model and deduce the likelihood ratio statistic and score test statistic,which are used to check whether a point is abnormal or not.And we also use the method of local influence analysis to obtain the influence matrix and analyze the data by diagonal element method.Finally,we use numerical simulation analysis to verify the rationality and feasibility of the theory.