Nonlinear methods for inverse statistical problems

نویسندگان

  • Pierre Barbillon
  • Gilles Celeux
  • Agnès Grimaud
  • Yannick Lefebvre
  • Etienne de Rocquigny
چکیده

In the uncertainty treatment framework considered in this paper, the intrinsic variability of the inputs of a physical simulation model is modelled by a multivariate probability distribution. The objective is to identify this probability distribution the dispersion of which is independent of the sample size since intrinsic variability is at stake based on observation of some model outputs. Moreover, in order to limit to a reasonable level the number of (usually burdensome) physical model runs inside the inversion algorithm, a non linear approximation methodology making use of Kriging and stochastic EM algorithm is presented. It is compared with iterated linear approximation on the basis of numerical experiments on simulated data sets coming from a simplified but realistic modelling of a dyke overflow. Situations where this non linear approach is to be preferred to linearisation are highlighted. Key-words: Uncertainty Modelling, Non linear Approximation, Kriging, Stochastic Algorithm ∗ Université Paris-sud 11, INRIA Saclay † Institut Mathématique de Luminy ‡ Schlumberger § École Centrale Paris in ria -0 04 41 96 7, v er si on 1 17 D ec 2 00 9 Méthodes non linéaires pour des problèmes statistiques inverses Résumé : Dans le cadre du traitement des incertitudes étudié dans cet article, la variabilité intrinsèque des entrées d’un modèle physique est modélisée par une loi de probabilité multivariée. L’objectif est d’identifier cette loi de probabilité (sa dispersion est indépendante de la taille de l’échantillon puisque l’on traite de la variabilité intrinsèque) à partir d’observations des sorties du modèle. Afin de se limiter à un nombre d’appels raisonnable au code de calcul (souvent coûteux) du modèle physique dans l’algorithme d’inversion, une méthodologie d’approximation non linéaire faisant intervenir le krigeage et un algorithme EM stochastique est présentée. Elle est comparée à une méthode utilisant une approximation linéaire itérative sur la base de jeux de données simulées provenant d’un modèle de crues simplifié mais réaliste. Les cas où cette approche non linéaire est préférable seront mis en lumière. Mots-clés : Modélisation des incertitudes, Approximation non linéaire, Krigeage, Algorithme stochastique in ria -0 04 41 96 7, v er si on 1 17 D ec 2 00 9 Non linear methods for inverse statistical problems 3

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Error-propagation in weakly nonlinear inverse problems

In applications of inversion methods to real data, nonlinear inverse problems are often simpli ed to more easily solvable linearized inverse problems. By doing so one introduces an error made by the linearization. Nonlinear inverse methods are more accurate because the methods that are used are more correct from a physical point of view. However, if data are used that have a statistical error, ...

متن کامل

Application of different inverse methods for combination of vS and vGPR data to estimate porosity and water saturation

Inverse problem is one of the most important problems in geophysics as model parameters can be estimated from the measured data directly using inverse techniques. In this paper, applying different inverse methods on integration of S-wave and GPR velocities are investigated for estimation of porosity and water saturation. A combination of linear and nonlinear inverse problems are solved. Linear ...

متن کامل

Solving a nonlinear inverse system of Burgers equations

By applying finite difference formula to time discretization and the cubic B-splines for spatial variable, a numerical method for solving the inverse system of Burgers equations is presented. Also, the convergence analysis and stability for this problem are investigated and the order of convergence is obtained. By using two test problems, the accuracy of presented method is verified. Additional...

متن کامل

Probabilistic Numerical Methods for PDE-constrained Bayesian Inverse Problems

This paper develops meshless methods for probabilistically describing discretisation error in the numerical solution of partial differential equations. This construction enables the solution of Bayesian inverse problems while accounting for the impact of the discretisation of the forward problem. In particular, this drives statistical inferences to be more conservative in the presence of signif...

متن کامل

A regularization method for solving a nonlinear backward inverse heat conduction problem using discrete mollification method

The present essay scrutinizes the application of discrete mollification as a filtering procedure to solve a nonlinear backward inverse heat conduction problem in one dimensional space. These problems are seriously ill-posed. So, we combine discrete mollification and space marching method to address the ill-posedness of the proposed problem. Moreover, a proof of stability and<b...

متن کامل

A meshless technique for nonlinear Volterra-Fredholm integral equations via hybrid of radial basis functions

In this paper, an effective technique is proposed to determine thenumerical solution of nonlinear Volterra-Fredholm integralequations (VFIEs) which is based on interpolation by the hybrid ofradial basis functions (RBFs) including both inverse multiquadrics(IMQs), hyperbolic secant (Sechs) and strictly positive definitefunctions. Zeros of the shifted Legendre polynomial are used asthe collocatio...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

عنوان ژورنال:
  • Computational Statistics & Data Analysis

دوره 55  شماره 

صفحات  -

تاریخ انتشار 2011