Results for “random variables” — Page 11

NASA NTRS1971

Comments on the continuity of distribution functions obtained by superposition.

Assuming that Y is a nonnegative random variable independent of the differential process X(t), attention is given to the question of whether or not the superposition X(Y) can have a continuous probability distribution. If the process has continuous distributions, then the superposition is continuous if and only if P/Y = 0/ = 0. If the process has discontinuous distributions and no trend, then no superposition can have continuous distribution. If the process has discontinuous distributions and nonzero trend, then the superposition onto a random epoch has continuous distribution if and only if Y has continuous distribution.

Huff, B. W.↗

DOE OSTI2016

Bayesian Integration of multiscale environmental data

The software is designed for efficiently integrating large-size of multi-scale environmental data using the Bayesian framework. Suppose we need to estimate the spatial distribution of variable X with high spatial resolution. The available data include (1) direct measurements Z of the unknowns with high resolution in a subset of the spatial domain (small spatial coverage), (2) measurements C of the unknowns at the median scale, and (3) measurements A of the unknowns at the coarsest scale but with large spatial coverage. The goal is to estimate the unknowns at the fine grids by conditioning to all the available data. We first consider all the unknowns as random variables and estimate conditional probability distribution of those variables by conditioning to the limited high-resolution observations (Z). We then treat the estimated probability distribution as the prior distribution. Within the Bayesian framework, we combine the median and large-scale measurements (C and A) through likelihood functions. Since we assume that all the relevant multivariate distributions are Gaussian, the resulting posterior distribution is a multivariate Gaussian distribution. The developed software provides numerical solutions of the posterior probability distribution. The software can be extended in several different ways to solve more general multi-scale data integration problems.

CHEN, JINSONG↗

DOE OSTI2016

Bayesian Integration of multiscale environmental data

The software is designed for efficiently integrating large-size of multi-scale environmental data using the Bayesian framework. Suppose we need to estimate the spatial distribution of variable X with high spatial resolution. The available data include (1) direct measurements Z of the unknowns with high resolution in a subset of the spatial domain (small spatial coverage), (2) measurements C of the unknowns at the median scale, and (3) measurements A of the unknowns at the coarsest scale but with large spatial coverage. The goal is to estimate the unknowns at the fine grids by conditioning to all the available data. We first consider all the unknowns as random variables and estimate conditional probability distribution of those variables by conditioning to the limited high-resolution observations (Z). We then treat the estimated probability distribution as the prior distribution. Within the Bayesian framework, we combine the median and large-scale measurements (C and A) through likelihood functions. Since we assume that all the relevant multivariate distributions are Gaussian, the resulting posterior distribution is a multivariate Gaussian distribution. The developed software provides numerical solutions of the posterior probability distribution. The software can be extended in several different ways to solve more general multi-scale data integration problems.

CHEN, JINSONG↗

DOE OSTI2016

Development of a Severe Accident Analysis Engine Using Approximate Reasoning

In hypothetical severe accident scenarios, nuclear power plant safety is challenged by core degradation and release of fission products. To aid the plant operator and technical support center staff, research has been performed to develop understanding of severe accident phenomenology, assessing their risk and devise measures to mitigate accident consequences (Sehgal, 2011). Nuclear power plant accidents involve highly complex physical and chemical interactions, a large number of safety related systems, and emergency operating procedures (EOP) and severe accident management guidelines (SAMG). Large number of factors with uncertainty and stress can overwhelm the operators. The overall objective of the present research is to develop a methodology and tools that an operator can use to predict severe accident progression and make effective decisions to arrest and manage the accident under large uncertainties. Towards this goal, this study examines a novel approach that brings to bear a combined use of techniques in artificial intelligence (Russel and Norvig, 2010), including Bayesian Network (BN), Answer Set Programming (ASP) and Fuzzy Inference (FI) to achieve faster-than-real-time, yet sufficiently accurate analysis of severe accident progression. Bayesian network (BN) is a type of probabilistic graphical models that can be used to represent knowledge and conditional dependencies about an uncertain realm. It is useful for capturing the causal relationships in a complex system. It also allows prediction via Bayesian inference (Kelly and Smith, 2012). It is composed of nodes and edges, which express random variables and probabilistic relationships between the nodes, respectively. This relationship, given by conditional probability is calculated using stochastic methods. The BN is used in several systems for analysis and management of accident risk; e.g. (Knochenhauer and Frid, 2010). In those systems, the nodes or so-called 'effective factors' are presumably known within the network by system descriptions in SAMGs. There are two steps to configure the BN, namely (i) identifying compositions within the network and (ii) defining conditional dependency between nodes. The first step is performed following the spirit of ASP (Baral, 2003) using Prolog (Wielemaker, 2005). Knowledge representation is used to build databases required for processing by ASP. Knowledge representation is defined as the expression of knowledge in a way that is understandable to both human and computers. Originally knowledge indicates data, information and facts. To focus, knowledge should be formed to achieve a goal. In a severe accident, key physics and safety related components status are defined according to a failure mode. The conditional dependencies between nodes are generated by approximate reasoning using Fuzzy inference (Zadeh, 1995). Fuzzy logic is a multiple-valued logic to treat ambiguous features such as linguistic description. It indicates possibility, which is distinguished from probability. Fuzzy set consists of linguistic variables according to the fixed rule. In other words, even uncertain states could be expressed as not binary logic, but as many-value logic by rule-based technology. Fuzzy control system is comprised of Fuzzy inference engine that is performed based on the 'If-Then Rule'. The 'If-Then Rule' is of the following form. If Y1 is B11 and Y2 is B12 then Z is C1. If Y1 is B21 and Y2 is B22 then Z is C2. Y1 and Y2 are inputs, which indicate conditional variables. Z indicates decision-making and both B and C are linguistic variables. With this rule-based technique, the approximate reasoning can be realized by Fuzzy inference. (authors)

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

DOE OSTI2018

Effects of defects in composite wind turbine blades – Part 3: A framework for treating defects as uncertainty variables for blade analysis

Abstract. Given that wind turbine blades are large structures, the use of low-cost composite manufacturing processes and materials has been necessary for the industry to be cost competitive. Since these manufacturing methods can lead to the inclusion of unwanted defects, potentially reducing blade life, the Blade Reliability Collaborative tasked the Montana State University Composites Group with assessing the effects of these defects. Utilizing the results of characterization and mechanical testing studies, probabilistic models were developed to assess the reliability of a wind blade with known defects. As such, defects were found to be best assessed as design parameters in a parametric probabilistic analysis allowing for establishment of a consistent framework to validate categorization and analysis. Monte Carlo simulations were found to adequately describe the probability of failure of composite blades with included defects. By treating defects as random variables, the approaches utilized indicate the level of conservation used in blade design may be reduced when considering fatigue. In turn, safety factors may be reduced as some of the uncertainty surrounding blade failure is reduced when analyzed with application specific data. Overall, the results indicate that characterization of defects and reduction of design uncertainty is possible for wind turbine blades.

Riddle, Trey W.↗

DOE OSTI2017

Second-Order Uncertainty Analysis for Radiation Shielding Problems Using MCNP

Standard uncertainty propagation is based on a first-order Taylor expansion. This approximation is usually justified by assuming that 'uncertainties are very small numbers compared with the values of their corresponding quantities'. Furthermore, the second derivatives needed for a second-order analysis are not usually available, and a second-order analysis requires more knowledge regarding the distribution of the random variables. For radiation shielding problems, the MCNP6 Monte Carlo code computes first and second derivatives of a response with respect to nuclide density or reaction cross section. This paper explores the application of these derivatives to second-order uncertainty analysis.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

DOE OSTI2022

Behavior and mechanisms of Doppler wind lidar error in varying stability regimes

Abstract. Wind lidars are widespread and important tools in atmospheric observations. An intrinsic part of lidar measurement error is due to atmospheric variability in the remote-sensing scan volume. This study describes and quantifies the distribution of measurement error due to turbulence in varying atmospheric stability. While the lidar error model is general, we demonstrate the approach using large ensembles of virtual WindCube V2 lidar performing a profiling Doppler-beam-swinging scan in quasi-stationary large-eddy simulations (LESs) of convective and stable boundary layers. Error trends vary with the stability regime, time averaging of results, and observation height. A systematic analysis of the observation error explains dominant mechanisms and supports the findings of the empirical results. Treating the error under a random variable framework allows for informed predictions about the effect of different configurations or conditions on lidar performance. Convective conditions are most prone to large errors (up to 1.5 m s−1 in 1 Hz wind speed in strong convection), driven by the large vertical velocity variances in convective conditions and the high elevation angle of the scanning beams (62∘). Range-gate weighting induces a negative bias into the horizontal wind speeds near the surface shear layer (−0.2 m s−1 in the stable test case). Errors in the horizontal wind speed and direction computed from the wind components are sensitive to the background wind speed but have negligible dependence on the relative orientation of the instrument. Especially during low winds and in the presence of large errors in the horizontal velocity estimates, the reported wind speed is subject to a systematic positive bias (up to 0.4 m s−1 in 1 Hz measurements in strong convection). Vector time-averaged measurements can improve the behavior of the error distributions (reducing the 10 min wind speed error standard deviation to <0.3 m s−1 and the bias to <0.1 m s−1 in strong convection) with a predictable effectiveness related to the number of decorrelated samples in the time window. Hybrid schemes weighting the 10 min scalar- and vector-averaged lidar measurements are shown to be effective at reducing the wind speed biases compared to cup measurements in most of the simulated conditions, with time averages longer than 10 min recommended for best use in some unstable conditions. The approach in decomposing the error mechanisms with the help of the LES flow field could be extended to more complex measurement scenarios and scans.

17 WIND ENERGY↗

DOE OSTI2022

An efficient method to propagate model uncertainty when inverting seismic data for time domain seismic moment tensors

SUMMARY We present a computationally efficient method to approximately propagate uncertainty when linearly inverting seismic data for point source, time variable moment tensor components. The method is based on the assumption that the data residual, given by the difference between the observed seismic data and the data predicated by a linear inversion, contains the effects of both data and model uncertainty. Our method uses a distribution of data residuals, added directly to the data, in a pseudo-Monte Carlo scheme. Using the assumption that the data residual is a stochastic process, we use the well-known Karhunen–Loève (KL) theorem to construct a distribution of data residuals, where the required basis functions are constructed using Fourier series. The Fourier series are scaled by a product of a random variable and the real-valued spectral amplitudes of the original data residual’s spectrum. Thus, the Fourier series and spectral amplitudes are eigenfunction-eigenvalue pairs used in the KL-based construction of data residual distribution. Using tests with synthetic data, we show that our method compares closely with a Finite Difference Monte Carlo (FDMC) method that we presented previously. More importantly, the method presented here is computationally several orders of magnitude faster than our previous FDMC method, and requires no a priori assumptions of model and/or data uncertainty.

Poppeliers, Christian (ORCID:0000000159526849)↗

DOE OSTI2025

A Class of Sparse Johnson–Lindenstrauss Transforms and Analysis of their Extreme Singular Values

The Johnson–Lindenstrauss (JL) lemma is a powerful tool for dimensionality reduction in modern algorithm design. The lemma states that any set of high-dimensional points in a Euclidean space can be projected into lower dimensions while approximately preserving pairwise Euclidean distances. Random matrices satisfying this lemma are called JL transforms (JLTs). Inspired by existing $s$-hashing JLTs with exactly $s$ nonzero elements on each column, the present work introduces an ensemble of sparse matrices encompassing so-called $s$-hashing-like matrices whose expected number of nonzero elements on each column is $s$. The independence of the sub-Gaussian entries of these matrices and the knowledge of their exact distribution play an important role in their analyses. Using properties of independent sub-Gaussian random variables, these matrices are demonstrated to be JLTs, and their smallest nontrivial singular values and largest singular values are estimated nonasymptotically using a technique from geometric functional analysis. As the dimensions of the matrix grow to infinity, these singular values are proved to converge almost surely to fixed quantities (by using the universal Bai–Yin law) and in distribution to the Gaussian orthogonal ensemble Tracy–Widom law after proper rescalings. Understanding the behaviors of extreme singular values is important in general because they are often used to define a measure of stability of matrix algorithms. For example, JLTs were recently used in derivative-free optimization algorithmic frameworks to select random subspaces in which are constructed random models or poll directions to achieve scalability, and hence estimating their smallest singular value in particular helps determine the dimension of these subspaces.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2018

Discrete One-Dimensional Coverage Process on a Renewal Process

Consider the following coverage model on ℕ, for each site i∈ℕ associate a pair (ξ{sub i},R{sub i}) where (ξ{sub i}){sub i≥0} is a 1-dimensional undelayed discrete renewal point process and (R{sub i}){sub i≥0} is an i.i.d. sequence of ℕ-valued random variables. At each site where ξ{sub i}=1 start an interval of length R{sub i}. Coverage occurs if every site of ℕ is covered by some interval. We obtain sharp conditions for both, positive and null probability of coverage. As corollaries, we extend results of the literature of rumor processes and discrete one-dimensional Boolean percolation.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

DOE OSTI2018

Equilibration and Diffusion for a Dynamical Lorentz Gas

We consider a model of a dynamical Lorentz gaz: a single particle is moving in ℝ{sup d} through an array of fixed and soft scatterers each possessing an internal degree of freedom coupled to the particle. Assuming the initial velocity is sufficiently high and modelling the parameters of the scatterers as random variables, we describe the evolution of the kinetic energy of the particle by a Markov chain for which each step corresponds to a collision. We show that the momentum distribution of the particle approaches a Maxwell–Boltzmann distribution with effective temperature T such that k{sub B}T corresponds to an average of the scatterers’ kinetic energy.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

NASA NTRS2014

Statistical Mechanics of Turbulent Dynamos

Incompressible magnetohydrodynamic (MHD) turbulence and magnetic dynamos, which occur in magnetofluids with large fluid and magnetic Reynolds numbers, will be discussed. When Reynolds numbers are large and energy decays slowly, the distribution of energy with respect to length scale becomes quasi-stationary and MHD turbulence can be described statistically. In the limit of infinite Reynolds numbers, viscosity and resistivity become zero and if these values are used in the MHD equations ab initio, a model system called ideal MHD turbulence results. This model system is typically confined in simple geometries with some form of homogeneous boundary conditions, allowing for velocity and magnetic field to be represented by orthogonal function expansions. One advantage to this is that the coefficients of the expansions form a set of nonlinearly interacting variables whose behavior can be described by equilibrium statistical mechanics, i.e., by a canonical ensemble theory based on the global invariants (energy, cross helicity and magnetic helicity) of ideal MHD turbulence. Another advantage is that truncated expansions provide a finite dynamical system whose time evolution can be numerically simulated to test the predictions of the associated statistical mechanics. If ensemble predictions are the same as time averages, then the system is said to be ergodic; if not, the system is nonergodic. Although it had been implicitly assumed in the early days of ideal MHD statistical theory development that these finite dynamical systems were ergodic, numerical simulations provided sufficient evidence that they were, in fact, nonergodic. Specifically, while canonical ensemble theory predicted that expansion coefficients would be (i) zero-mean random variables with (ii) energy that decreased with length scale, it was found that although (ii) was correct, (i) was not and the expected ergodicity was broken. The exact cause of this broken ergodicity was explained, after much investigation, by greatly extending the statistical theory of ideal MHD turbulence. The mathematical details of broken ergodicity, in fact, give a quantitative explanation of how coherent structure, dynamic alignment and force-free states appear in turbulent magnetofluids. The relevance of these ideal results to real MHD turbulence occurs because broken ergodicity is most manifest in the ideal case at the largest length scales and it is in these largest scales that a real magnetofluid has the least dissipation, i.e., most closely approaches the behavior of an ideal magnetofluid. Furthermore, the effects grow stronger when cross and magnetic helicities grow large with respect to energy, and this is exactly what occurs with time in a real magnetofluid, where it is called selective decay. The relevance of these results found in ideal MHD turbulence theory to the real world is that they provide at least a qualitative explanation of why confined turbulent magnetofluids, such as the liquid iron that fills the Earth's outer core, produce stationary, large-scale magnetic fields, i.e., the geomagnetic field. These results should also apply to other planets as well as to plasma confinement devices on Earth and in space, and the effects should be manifest if Reynolds numbers are high enough and there is enough time for stationarity to occur, at least approximately. In the presentation, details will be given for both theoretical and numerical results, and references will be provided.

Shebalin, John V.↗

NASA NTRS2014

MHD Turbulence and Magnetic Dynamos

Incompressible magnetohydrodynamic (MHD) turbulence and magnetic dynamos, which occur in magnetofluids with large fluid and magnetic Reynolds numbers, will be discussed. When Reynolds numbers are large and energy decays slowly, the distribution of energy with respect to length scale becomes quasi-stationary and MHD turbulence can be described statistically. In the limit of infinite Reynolds numbers, viscosity and resistivity become zero and if these values are used in the MHD equations ab initio, a model system called ideal MHD turbulence results. This model system is typically confined in simple geometries with some form of homogeneous boundary conditions, allowing for velocity and magnetic field to be represented by orthogonal function expansions. One advantage to this is that the coefficients of the expansions form a set of nonlinearly interacting variables whose behavior can be described by equilibrium statistical mechanics, i.e., by a canonical ensemble theory based on the global invariants (energy, cross helicity and magnetic helicity) of ideal MHD turbulence. Another advantage is that truncated expansions provide a finite dynamical system whose time evolution can be numerically simulated to test the predictions of the associated statistical mechanics. If ensemble predictions are the same as time averages, then the system is said to be ergodic; if not, the system is nonergodic. Although it had been implicitly assumed in the early days of ideal MHD statistical theory development that these finite dynamical systems were ergodic, numerical simulations provided sufficient evidence that they were, in fact, nonergodic. Specifically, while canonical ensemble theory predicted that expansion coefficients would be (i) zero-mean random variables with (ii) energy that decreased with length scale, it was found that although (ii) was correct, (i) was not and the expected ergodicity was broken. The exact cause of this broken ergodicity was explained, after much investigation, by greatly extending the statistical theory of ideal MHD turbulence. The mathematical details of broken ergodicity, in fact, give a quantitative explanation of how coherent structure, dynamic alignment and force-free states appear in turbulent magnetofluids. The relevance of these ideal results to real MHD turbulence occurs because broken ergodicity is most manifest in the ideal case at the largest length scales and it is in these largest scales that a real magnetofluid has the least dissipation, i.e., most closely approaches the behavior of an ideal magnetofluid. Furthermore, the effects grow stronger when cross and magnetic helicities grow large with respect to energy, and this is exactly what occurs with time in a real magnetofluid, where it is called selective decay. The relevance of these results found in ideal MHD turbulence theory to the real world is that they provide at least a qualitative explanation of why confined turbulent magnetofluids, such as the liquid iron that fills the Earth's outer core, produce stationary, large-scale magnetic fields, i.e., the geomagnetic field. These results should also apply to other planets as well as to plasma confinement devices on Earth and in space, and the effects should be manifest if Reynolds numbers are high enough and there is enough time for stationarity to occur, at least approximately. In the presentation, details will be given for both theoretical and numerical results, and references will be provided.

Shebalin, John V↗

DOE OSTI2018

Asymptotic Behaviour of Ground States for Mixtures of Ferromagnetic and Antiferromagnetic Interactions in a Dilute Regime

We consider randomly distributed mixtures of bonds of ferromagnetic and antiferromagnetic type in a two-dimensional square lattice with probability 1−p and p, respectively, according to an i.i.d. random variable. We study minimizers of the corresponding nearest-neighbour spin energy on large domains in ℤ{sup 2}. We prove that there exists p{sub 0} such that for p≤p{sub 0} such minimizers are characterized by a majority phase; i.e., they take identically the value 1 or − 1 except for small disconnected sets. A deterministic analogue is also proved.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

NASA NTRS2001

Distribution functions of probabilistic automata

Each probabilistic automaton M over an alphabet A defines a probability measure Prob sub(M) on the set of all finite and infinite words over A. We can identify a k letter alphabet A with the set {0, 1,..., k-1}, and, hence, we can consider every finite or infinite word w over A as a radix k expansion of a real number X(w) in the interval [0, 1]. This makes X(w) a random variable and the distribution function of M is defined as usual: F(x) := Prob sub(M) { w: X(w) < x }. Utilizing the fixed-point semantics (denotational semantics), extended to probabilistic computations, we investigate the distribution functions of probabilistic automata in detail. Automata with continuous distribution functions are characterized. By a new, and much more easier method, it is shown that the distribution function F(x) is an analytic function if it is a polynomial. Finally, answering a question posed by D. Knuth and A. Yao, we show that a polynomial distribution function F(x) on [0, 1] can be generated by a prob abilistic automaton iff all the roots of F'(x) = 0 in this interval, if any, are rational numbers. For this, we define two dynamical systems on the set of polynomial distributions and study attracting fixed points of random composition of these two systems.

Denotational semantics↗

NASA NTRS2003

Structural Reliability Using Probability Density Estimation Methods Within NESSUS

A reliability analysis studies a mathematical model of a physical system taking into account uncertainties of design variables and common results are estimations of a response density, which also implies estimations of its parameters. Some common density parameters include the mean value, the standard deviation, and specific percentile(s) of the response, which are measures of central tendency, variation, and probability regions, respectively. Reliability analyses are important since the results can lead to different designs by calculating the probability of observing safe responses in each of the proposed designs. All of this is done at the expense of added computational time as compared to a single deterministic analysis which will result in one value of the response out of many that make up the density of the response. Sampling methods, such as monte carlo (MC) and latin hypercube sampling (LHS), can be used to perform reliability analyses and can compute nonlinear response density parameters even if the response is dependent on many random variables. Hence, both methods are very robust; however, they are computationally expensive to use in the estimation of the response density parameters. Both methods are 2 of 13 stochastic methods that are contained within the Numerical Evaluation of Stochastic Structures Under Stress (NESSUS) program. NESSUS is a probabilistic finite element analysis (FEA) program that was developed through funding from NASA Glenn Research Center (GRC). It has the additional capability of being linked to other analysis programs; therefore, probabilistic fluid dynamics, fracture mechanics, and heat transfer are only a few of what is possible with this software. The LHS method is the newest addition to the stochastic methods within NESSUS. Part of this work was to enhance NESSUS with the LHS method. The new LHS module is complete, has been successfully integrated with NESSUS, and been used to study four different test cases that have been proposed by the Society of Automotive Engineers (SAE). The test cases compare different probabilistic methods within NESSUS because it is important that a user can have confidence that estimates of stochastic parameters of a response will be within an acceptable error limit. For each response, the mean, standard deviation, and 0.99 percentile, are repeatedly estimated which allows confidence statements to be made for each parameter estimated, and for each method. Thus, the ability of several stochastic methods to efficiently and accurately estimate density parameters is compared using four valid test cases. While all of the reliability methods used performed quite well, for the new LHS module within NESSUS it was found that it had a lower estimation error than MC when they were used to estimate the mean, standard deviation, and 0.99 percentile of the four different stochastic responses. Also, LHS required a smaller amount of calculations to obtain low error answers with a high amount of confidence than MC. It can therefore be stated that NESSUS is an important reliability tool that has a variety of sound probabilistic methods a user can employ and the newest LHS module is a valuable new enhancement of the program.

Chamis, Chrisos C.↗

DOE OSTI2015

Detecting anomalies in CMB maps: a new method

Ever since WMAP announced its first results, different analyses have shown that there is weak evidence for several large-scale anomalies in the CMB data. While the evidence for each anomaly appears to be weak, the fact that there are multiple seemingly unrelated anomalies makes it difficult to account for them via a single statistical fluke. So, one is led to considering a combination of these anomalies. But, if we ''hand-pick'' the anomalies (test statistics) to consider, we are making an a posteriori choice. In this article, we propose two statistics that do not suffer from this problem. The statistics are linear and quadratic combinations of the a{sub ℓ m}'s with random co-efficients, and they test the null hypothesis that the a{sub ℓ m}'s are independent, normally-distributed, zero-mean random variables with an m-independent variance. The motivation for considering multiple modes is this: because most physical models that lead to large-scale anomalies result in coupling multiple ℓ and m modes, the ''coherence'' of this coupling should get enhanced if a combination of different modes is considered. In this sense, the statistics are thus much more generic than those that have been hitherto considered in literature. Using fiducial data, we demonstrate that the method works and discuss how it can be used with actual CMB data to make quite general statements about the incompatibility of the data with the null hypothesis.

79 ASTRONOMY AND ASTROPHYSICS↗

DOE OSTI2019

Scaling Input Distributions for Probabilistic Models - 19472

It is important to address scaling issues when building models to support risk-based decisions for radioactive waste disposal sites. Currently, probabilistic risk assessment and probabilistic performance assessment models are run for multiple realizations where each realization uses one value from each input distribution over the entire modeled time. The input distributions represent the current state of knowledge in the mean of the input variable over the spatial region of interest and over the time period of the model run. This paper focuses on the time factor in input distribution scaling because it presents more serious conceptual and distributional challenges than the spatial factor. One of the challenges in temporal scaling is the effect of non-linearity on propagation of distribution uncertainty through to uncertainty in results. Radioactive waste disposal models tend to have both linear and non-linear components in their models. Dose assessment calculations are typically represented by linear models, while contaminant transport models have linear and non-linear components. In principle, temporal scaling of input parameter distributions for a linear model is straightforward, but scaling of input distributions for non-linear models is not. This paper uses examples to investigate the effects of input distribution temporal scaling on simple linear and non-linear models. Different scaling scenarios were investigated by comparing three analysis tools: analytical mathematical representations, statistical software programming, and GoldSim Monte Carlo simulation software. R statistical software is an open source statistical programming language often used for statistical analyses and graphical representation of data. GoldSim is a robust probabilistic dynamic modeling platform used in performance assessment modeling of radioactive waste sites. A simple linear model, a quadratic model and a multiplicative model were the cases explored for each of these tools. For the three analysis tools considered and with each of the scaling cases, 1,000 realizations were implemented from distributions of the random variables of interest. This is a sufficient number of realizations for simple models to capture the model space. An annual temporal resolution of a 100-year simulation period was chosen for all models and analysis tools. The effects of temporal scaling were demonstrated by investigating two different options for time scaling. Option 1 was set up so that the input distributions represented the expected value on an annual time scale, and a new value from that distribution was chosen every year (at every time step which equaled a year) for 100 years. In Option 2, the input distributions were scaled to represent the expected mean value of the input variable over the 100 years of simulation, and one value from that distribution was chosen for the entire elapsed time. Results show agreement across the three analysis tools. As expected, the linear models have identical results for the two different scaling Options, and the non-linear models diverge in the results for different time scale implementation. The implications of this work point to the need for careful consideration of input distribution temporal scaling. (authors)

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗