Search NASA⌕ Search

SEARCH · Search NASA

Results for “Probabilistic learning”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 469 records · Page 26

A reforecasting-based dynamic reserve estimation for variable renewable generation and demand uncertainty

The installed capacity of renewables-based energy sources has been increasing in traditional power systems. In order to accommodate the increased variability and uncertainty associated with the deeper penetration of renewable sources like solar and wind, adjusted amounts of dynamic reserve are needed. Although probabilistic dynamic reserve estimation methods have been previously developed, most of them consider the uncertainty to be represented by parametric density functions that tend to perform poorly under extreme events and, moreover, neglect uncertainty introduced by the forecasting model itself. Toward addressing these limitations, this work presents, for the first time, a dynamic reserve estimation method for flexibility that incorporates nonparametric density estimation and a machine learning based reforecasting to provide a day-ahead prediction of the mean and spread of uncertainty around the base forecast. The prediction is, in turn, used to estimate the up and down reserve relative to the base forecast. Here, the present method takes various endogenous and exogenous features, including the calendar variables, as input to estimate the day-ahead reserve. Using a combination of reforecasting and dynamic reserve estimation techniques, the method is shown to adjust better to the dynamic nature of reserve requirements providing only what is needed to accommodate the expected deviations. Considering California Independent System Operator (CAISO) solar, wind and load data over an 18 month period, up to 67% reduction in the amount of reserve capacity needed for a one day reserve and reserve penalty for solar uncertainty is demonstrated. Additionally, the risk of reserve insufficiency in meeting the net demand is reduced by 20% with the proposed method.

14 SOLAR ENERGY↗

Development and testing of improved statistical wind power forecasting methods.

Wind power forecasting (WPF) provides important inputs to power system operators and electricity market participants. It is therefore not surprising that WPF has attracted increasing interest within the electric power industry. In this report, we document our research on improving statistical WPF algorithms for point, uncertainty, and ramp forecasting. Below, we provide a brief introduction to the research presented in the following chapters. For a detailed overview of the state-of-the-art in wind power forecasting, we refer to [1]. Our related work on the application of WPF in operational decisions is documented in [2]. Point forecasts of wind power are highly dependent on the training criteria used in the statistical algorithms that are used to convert weather forecasts and observational data to a power forecast. In Chapter 2, we explore the application of information theoretic learning (ITL) as opposed to the classical minimum square error (MSE) criterion for point forecasting. In contrast to the MSE criterion, ITL criteria do not assume a Gaussian distribution of the forecasting errors. We investigate to what extent ITL criteria yield better results. In addition, we analyze time-adaptive training algorithms and how they enable WPF algorithms to cope with non-stationary data and, thus, to adapt to new situations without requiring additional offline training of the model. We test the new point forecasting algorithms on two wind farms located in the U.S. Midwest. Although there have been advancements in deterministic WPF, a single-valued forecast cannot provide information on the dispersion of observations around the predicted value. We argue that it is essential to generate, together with (or as an alternative to) point forecasts, a representation of the wind power uncertainty. Wind power uncertainty representation can take the form of probabilistic forecasts (e.g., probability density function, quantiles), risk indices (e.g., prediction risk index) or scenarios (with spatial and/or temporal dependence). Statistical approaches to uncertainty forecasting basically consist of estimating the uncertainty based on observed forecasting errors. Quantile regression (QR) is currently a commonly used approach in uncertainty forecasting. In Chapter 3, we propose new statistical approaches to the uncertainty estimation problem by employing kernel density forecast (KDF) methods. We use two estimators in both offline and time-adaptive modes, namely, the Nadaraya-Watson (NW) and Quantilecopula (QC) estimators. We conduct detailed tests of the new approaches using QR as a benchmark. One of the major issues in wind power generation are sudden and large changes of wind power output over a short period of time, namely ramping events. In Chapter 4, we perform a comparative study of existing definitions and methodologies for ramp forecasting. We also introduce a new probabilistic method for ramp event detection. The method starts with a stochastic algorithm that generates wind power scenarios, which are passed through a high-pass filter for ramp detection and estimation of the likelihood of ramp events to happen. The report is organized as follows: Chapter 2 presents the results of the application of ITL training criteria to deterministic WPF; Chapter 3 reports the study on probabilistic WPF, including new contributions to wind power uncertainty forecasting; Chapter 4 presents a new method to predict and visualize ramp events, comparing it with state-of-the-art methodologies; Chapter 5 briefly summarizes the main findings and contributions of this report.

17 WIND ENERGY↗

Deep-Learning-Based Koopman Modeling for Online Control Synthesis of Nonlinear Power System Transient Dynamics

Power system stability and control have become more challenging due to the increasing uncertainty associated with renewable generation. Here, the performance of conventional control is highly driven by the physics-based offline-developed dynamic models that can deviate from the actual system characteristics under different operating conditions and/or configurations. Data-driven approaches based on online measurements can be a better solution to addressing these issues by capturing real-time operation conditions. This article describes a novel fully data-driven probabilistic framework to derive a linear representation of postcontingency grid dynamics and online prescribe control based on the derived model to enhance transient stability. The complex nonlinear power system dynamics is approximated by a linear model by using multiple neural network modules that infer distributions of the observations and introducing a Koopman layer to sample possible Koopman linear models from the inferred distributions. The trained model features linearity that can be easily incorporated into the existing linear control design paradigm and ease the controller design process. The effectiveness of Koopman-based control designs is validated through comparative case studies, which demonstrate increased prediction accuracy and control performance when applied to a power system with heterogeneous generator dynamics.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Scalable Bayesian optimization with randomized prior networks

Several fundamental problems in science and engineering consist of global optimization tasks involving unknown high-dimensional (black-box) functions that map a set of controllable variables to the outcomes of an expensive experiment. Bayesian Optimization (BO) techniques are known to be effective in tackling global optimization problems using a relatively small number objective function evaluations, but their performance suffers when dealing with high-dimensional outputs. To overcome the major challenge of dimensionality, here we propose a deep learning framework for BO and sequential decision making based on bootstrapped ensembles of neural architectures with randomized priors. Using appropriate architecture choices, we show that the proposed framework can approximate functional relationships between design variables and quantities of interest, even in cases where the latter take values in high-dimensional vector spaces or even infinite-dimensional function spaces. In the context of BO, we augmented the proposed probabilistic surrogates with re-parameterized Monte Carlo approximations of multiple-point (parallel) acquisition functions, as well as methodological extensions for accommodating black-box constraints and multi-fidelity information sources. We test the proposed framework against state-of-the-art methods for BO and demonstrate superior performance across several challenging tasks with high-dimensional outputs, including a constrained multi-fidelity optimization task involving shape optimization of rotor blades in turbo-machinery.

97 MATHEMATICS AND COMPUTING↗

Gaussian processes with built-in dimensionality reduction: Applications to high-dimensional uncertainty propagation

Uncertainty quantification (UQ) tasks, such as model calibration, uncertainty propagation, and optimization under uncertainty, typically require several thousand evaluations of the underlying computer codes. To cope with the cost of simulations, one replaces the real response surface with a cheap surrogate based, e.g., on polynomial chaos expansions, neural networks, support vector machines, or Gaussian processes (GP). However, the number of simulations required to learn a generic multivariate response grows exponentially as the input dimension increases. This curse of dimensionality can only be addressed, if the response exhibits some special structure that can be discovered and exploited. A wide range of physical responses exhibit a special structure known as an active subspace (AS). An AS is a linear manifold of the stochastic space characterized by maximal response variation. The idea is that one should first identify this low dimensional manifold, project the high-dimensional input onto it, and then link the projection to the output. If the dimensionality of the AS is low enough, then learning the link function is a much easier problem than the original problem of learning a high-dimensional function. The classic approach to discovering the AS requires gradient information, a fact that severely limits its applicability. Furthermore, and partly because of its reliance to gradients, it is not able to handle noisy observations. The latter is an essential trait if one wants to be able to propagate uncertainty through stochastic simulators, e.g., through molecular dynamics codes. In this work, we develop a probabilistic version of AS which is gradient-free and robust to observational noise. Our approach relies on a novel Gaussian process regression with built-in dimensionality reduction. In particular, the AS is represented as an orthogonal projection matrix that serves as yet another covariance function hyper-parameter to be estimated from the data. To train the model, we design a two-step maximum likelihood optimization procedure that ensures the orthogonality of the projection matrix by exploiting recent results on the Stiefel manifold, i.e., the manifold of matrices with orthogonal columns. The additional benefit of our probabilistic formulation, is that it allows us to select the dimensionality of the AS via the Bayesian information criterion. We validate our approach by showing that it can discover the right AS in synthetic examples without gradient information using both noiseless and noisy observations. We demonstrate that our method is able to discover the same AS as the classical approach in a challenging one-hundred-dimensional problem involving an elliptic stochastic partial differential equation with random conductivity. Finally, we use our approach to study the effect of geometric and material uncertainties in the propagation of solitary waves in a one dimensional granular system.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Low-Scaling Algorithm for Nudged Elastic Band Calculations Using a Surrogate Machine Learning Model

We present the incorporation of a surrogate Gaussian process regression (GPR) atomistic model to greatly accelerate the rate of convergence of classical nudged elastic band (NEB) calculations. In our surrogate model approach, the cost of converging the elastic band no longer scales with the number of moving images on the path. This provides a far more efficient and robust transition state search. In contrast to a conventional NEB calculation, the algorithm presented here eliminates any need for manipulating the number of images to obtain a converged result. This is achieved by inventing a new convergence criteria that exploits the probabilistic nature of the GPR to use uncertainty estimates of all images in combination with the force in the saddle point in the target model potential. In conclusion, our method is an order of magnitude faster in terms of function evaluations than the conventional NEB method with no accuracy loss for the converged energy barrier values.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

A Computational Approach for Probabilistic Analysis of LS-DYNA Water Impact Simulations

NASA s development of new concepts for the Crew Exploration Vehicle Orion presents many similar challenges to those worked in the sixties during the Apollo program. However, with improved modeling capabilities, new challenges arise. For example, the use of the commercial code LS-DYNA, although widely used and accepted in the technical community, often involves high-dimensional, time consuming, and computationally intensive simulations. Because of the computational cost, these tools are often used to evaluate specific conditions and rarely used for statistical analysis. The challenge is to capture what is learned from a limited number of LS-DYNA simulations to develop models that allow users to conduct interpolation of solutions at a fraction of the computational time. For this problem, response surface models are used to predict the system time responses to a water landing as a function of capsule speed, direction, attitude, water speed, and water direction. Furthermore, these models can also be used to ascertain the adequacy of the design in terms of probability measures. This paper presents a description of the LS-DYNA model, a brief summary of the response surface techniques, the analysis of variance approach used in the sensitivity studies, equations used to estimate impact parameters, results showing conditions that might cause injuries, and concluding remarks.

Horta, Lucas G.↗

CHP-PRA: Sensorimotor Countermeasures Proof of Concept

The capabilities included in Crew Health and Performance (CHP) systems are designed to keep the crew healthy, happy, and productive. In doing so, the CHP system allows reductions in human related risks to be realized. All human missions, though particularly future Artemis and Mars missions, have constraints on the mass and volume allocated to the CHP system. Thus, trades must be made on how to best buy down risk while still meeting other requirements. Understanding how different CHP capabilities influence medical, performance, and long-term health risks is key to making informed choices among these trades. To address this, an integrated CHP Probabilistic Risk Assessment (PRA) model is being developed. Much like how IMPACT is designed to allow medical resource trades informed by medical risks, CHP PRA will enable analogous trades in human system risks across all CHP functions and capabilities. There are 29 Human System Risk Board (HSRB) risk areas to consider for model implementation. This effort focused solely on the implementation of countermeasures associated with sensorimotor risk. The framework and lessons learned from this proof of concept will benefit the implementation of other risks and countermeasures. Astronauts experience sensorimotor changes when entering or exiting a microgravity environment. While most sensorimotor issues are resolved within a few days, they pose a serious risk to crew health and performance during the adaptation period. This adaptation period aligns with gravity transitions and key mission phases where the crew members may be required to complete challenging tasks ideally with an undisturbed sensorimotor system. Countermeasures such as training or pharmaceuticals can be used to mitigate risk. This effort began by identifying physiological changes to the sensorimotor system that could lead to functional performance decrements or medical conditions and potential countermeasures, so that the overall effect of sensorimotor disturbance on human risk could be captured. We then selected example countermeasures, mapped relationships between countermeasures and outcomes, and used results from existing sensorimotor investigations to define the relationship between a sensorimotor countermeasure and subsequent performance of a sensorimotor-affected task. Specifically, using shuttle landing performance data, we compared the use or absence of inflight training capabilities to yield a relative performance change that can be applied to the prediction of the risk associated with manual control performance in future missions, as depicted in the figure. This sensorimotor proof of concept demonstrates how HRP research information can be transformed into quantifiable relationships to describe changes in HSRB medical and performance risks.

Caroline R Austin↗

LLMs and GenAI Tools to Depict Contributions of Human Systems to Spaceflight Tasks Execution

Recent advancements in Artificial Intelligence and Machine Learning (AI/ML) technologies, particularly Large Language Models (LLMs) capable of sophisticated syntax analysis, offer substantial potential in automating complex processes, thereby saving time and human resources. This study explores the development of an LLM-driven model designed to analyze and categorize a diverse set of Mars mission tasks into 18 predefined Human System Task Categories (HSTCs) based on their textual descriptions. As part of developing the Crew Health and Performance – Probabilistic Risk Assessment (CHP-PRA projects Performance Risk Model (PRisM) proof-of-concept, we established a framework to project performance scores from small-scale tests onto a preliminary list of Mars tasks. The foundation of our model was a comprehensive spreadsheet populated by NASA experts and clinicians, which detailed each Mars task alongside binary indicators of HSTC involvement. This dataset enabled the initial application of supervised ML, training and testing on existing HSTC labels. The HSTCs were originally defined from a medical system perspective, focusing on task impairments due to deteriorated human health. To expand our model's scope to include categories impacting performance, we face the challenge of generating binary labels (0 or 1) for new categories without pre-existing data. We address this by employing Generative AI (GenAI) software to determine whether a given task involved a new category by asking, "Does task A involve using category B?" We validate our approach by comparing the GenAI's binary classifications with the expert-provided labels for existing HSTCs. Notably, we utilize Ollama [4], a locally hosted GenAI tool that does not require cloud access, thus safeguarding NASA's proprietary data from unauthorized exposure. This study demonstrates the feasibility of leveraging cutting-edge AI tools to advance research, paving the way for automation and rapid decision-making in space exploration.

Mona Matar↗

Live cell imaging of cellular dynamics in poplar wood using computational cannula microscopy

This study presents significant advancements in computational cannula microscopy for live imaging of cellular dynamics in poplar wood tissues. Leveraging machine-learning models such as pix2pix for image reconstruction, we achieved high-resolution imaging with a field of view of 55µm using a 50µm-core diameter probe. Our method allows for real-time image reconstruction at 0.29 s per frame with a mean absolute error of 0.07. We successfully captured cellular-level dynamics in vivo , demonstrating morphological changes at resolutions as small as 3µm. We implemented two types of probabilistic neural network models to quantify confidence levels in the reconstructed images. This approach facilitates context-aware, human-in-the-loop analysis, which is crucial for in vivo imaging where ground-truth data is unavailable. Using this approach we demonstrated deep in vivo computational imaging of living plant tissue with high confidence (disagreement score ⪅0.2). This work addresses the challenges of imaging live plant tissues, offering a practical and minimally invasive tool for plant biologists.

Ingold, Alexander (ORCID:0009000752380016)↗

Application of Fault Management Theory to the Quantitative Selection of a Launch Vehicle Abort Trigger Suite

The theory of System Health Management (SHM) and of its operational subset Fault Management (FM) states that FM is implemented as a "meta" control loop, known as an FM Control Loop (FMCL). The FMCL detects that all or part of a system is now failed, or in the future will fail (that is, cannot be controlled within acceptable limits to achieve its objectives), and takes a control action (a response) to return the system to a controllable state. In terms of control theory, the effectiveness of each FMCL is estimated based on its ability to correctly estimate the system state, and on the speed of its response to the current or impending failure effects. This paper describes how this theory has been successfully applied on the National Aeronautics and Space Administration's (NASA) Space Launch System (SLS) Program to quantitatively estimate the effectiveness of proposed abort triggers so as to select the most effective suite to protect the astronauts from catastrophic failure of the SLS. The premise behind this process is to be able to quantitatively provide the value versus risk trade‐off for any given abort trigger, allowing decision makers to make more informed decisions. All current and planned crewed launch vehicles have some form of vehicle health management system integrated with an emergency launch abort system to ensure crew safety. While the design can vary, the underlying principle is the same: detect imminent catastrophic vehicle failure, initiate launch abort, and extract the crew to safety. Abort triggers are the detection mechanisms that identify that a catastrophic launch vehicle failure is occurring or is imminent and cause the initiation of a notification to the crew vehicle that the escape system must be activated. While ensuring that the abort triggers provide this function, designers must also ensure that the abort triggers do not signal that a catastrophic failure is imminent when in fact the launch vehicle can successfully achieve orbit. That is, the abort triggers must have low false negative rates to be sure that real crew‐threatening failures are detected, and also low false positive rates to ensure that the crew does not abort from non‐crew‐threatening launch vehicle behaviors. The analysis process described in this paper is a compilation of over six years of lessons learned and refinements from experiences developing abort triggers for NASA's Constellation Program (Ares I Project) and the SLS Program, as well as the simultaneous development of SHM/FM theory. The paper will describe the abort analysis concepts and process, developed in conjunction with SLS Safety and Mission Assurance (S&MA) to define a common set of mission phase, failure scenario, and Loss of Mission Environment (LOME) combinations upon which the SLS Loss of Mission (LOM) Probabilistic Risk Assessment (PRA) models are built. This abort analysis also requires strong coordination with the Multi‐Purpose Crew Vehicle (MPCV) and SLS Structures and Environments (STE) to formulate a series of abortability tables that encapsulate explosion dynamics over the ascent mission phase. The design and assessment of abort conditions and triggers to estimate their Loss of Crew (LOC) Benefits also requires in‐depth integration with other groups, including Avionics, Guidance, Navigation and Control(GN&C), the Crew Office, Mission Operations, and Ground Systems. The outputs of this analysis are a critical input to SLS S&MA's LOC PRA models. The process described here may well be the first full quantitative application of SHM/FM theory to the selection of a sensor suite for any aerospace system.

Lo, Yunnhon↗

A Probabilistic Reasoner Based on Bayes Risk for Damage Detection in Structural Systems

Structural health monitoring (SHM) systems are used to inform operation of structural systems subject to loads and environments that may affect their integrity. SHM systems rely on continuous monitoring of the structure to determine its health state. These systems are often coupled with a model of the deployed structure to determine the consequences of changes in the system by forecasting the response to future states. These models, which may be thought of as digital twins, need to be updated to reflect the latest state of the structural system. This work makes use of an uncertainty-aware machine learning model that enforces distance preservation of the original input space to determine deviations from the training data input space distributions. This workflow enables domain shift detection to determine whether damage is present in the structure. The uncertainty metrics generated by this network are then used in a Bayes risk framework to design an optimal damage detector given cost and risk considerations. The approach is demonstrated on a computational example with simulated damage.

Najera-Flores, David [ATA Engineering, Inc.]↗

2009 Space Shuttle Probabilistic Risk Assessment Overview

Loss of a Space Shuttle during flight has severe consequences, including loss of a significant national asset; loss of national confidence and pride; and, most importantly, loss of human life. The Shuttle Probabilistic Risk Assessment (SPRA) is used to identify risk contributors and their significance; thus, assisting management in determining how to reduce risk. In 2006, an overview of the SPRA Iteration 2.1 was presented at PSAM 8 [1]. Like all successful PRAs, the SPRA is a living PRA and has undergone revisions since PSAM 8. The latest revision to the SPRA is Iteration 3. 1, and it will not be the last as the Shuttle program progresses and more is learned. This paper discusses the SPRA scope, overall methodology, and results, as well as provides risk insights. The scope, assumptions, uncertainties, and limitations of this assessment provide risk-informed perspective to aid management s decision-making process. In addition, this paper compares the Iteration 3.1 analysis and results to the Iteration 2.1 analysis and results presented at PSAM 8.

Hamlin, Teri L.↗

AI and extreme scale computing to learn and infer the physics of higher order gravitational wave modes of quasi-circular, spinning, non-precessing black hole mergers

We use artificial intelligence (AI) to learn and infer the physics of higher order gravitational wave modes of quasi-circular, spinning, non precessing binary black hole mergers. We trained AI models using 14 million waveforms, produced with the surrogate model NRHybSur3dq8, that include modes up to $\ell$ ≤ 4 and (5,5), except for (4,0) and (4,1), that describe binaries with mass-ratios $\textit{q}$ ≤ 8, individual spins $s^z_{\{1,2\}} \in$[–0.8,0.8], and inclination angle $θ \in$ [0,π]. Our probabilistic AI surrogates can accurately constrain the mass-ratio, individual spins, effective spin, and inclination angle of numerical relativity waveforms that describe such signal manifold. We compared the predictions of our AI models with Gaussian process regression, random forest, k-nearest neighbors, and linear regression, and with traditional Bayesian inference methods through the PyCBC Inference toolkit, finding that AI outperforms all these approaches in terms of accuracy, and are between three to four orders of magnitude faster than traditional Bayesian inference methods. Our AI surrogates were trained within 3.4 hours using distributed training on 1,536 NVIDIA V100 GPUs in the Summit supercomputer.

79 ASTRONOMY AND ASTROPHYSICS↗

Emulation With Uncertainty Quantification of Regional Sea‐Level Change Caused by the Antarctic Ice Sheet

Abstract Projecting regional sea‐level change under various climate‐change scenarios typically involves running forward simulations of the Earth's gravitational, rotational and deformational (GRD) response to ice‐mass change, which requires substantial computational cost if applied to probabilistic frameworks requiring thousands to millions of samples. Here we build emulators of regional sea‐level change at 27 coastal locations, due to the GRD effects associated with future Antarctic Ice Sheet mass change over the 21st century. The emulators are evaluated against a numerical sea‐level model applied to an ensemble of ice‐sheet model simulations of the Antarctic Ice Sheet through 2100. We build a physics‐based emulator using a recent sensitivity kernel approach and compare it to machine learning based emulators (neural network and conditional variational autoencoder methods). In order to quantify uncertainty, we derive well‐calibrated prediction intervals for regional sea‐level change via split‐conformal inference and linear regression, and show that Monte Carlo dropout does not yield well‐calibrated uncertainties in this instance. We also demonstrate substantial gains in computational efficiency using both the physics‐based emulator and neural networks in comparison to the numerical model for the complete regional sea‐level solution. Overall, we find the physics‐based emulator modestly outperforms the machine learning emulators for this problem.

58 GEOSCIENCES↗

Evaluation of Physical Security Risk for Potential Implementation of FLEX using Dynamic Simulation Methods

The requirements for United States nuclear power plants to maintain a large onsite physical security force contribute to their large operational costs. The cost of maintaining the current physical security posture is approximately 10% of the overall operation and maintenance budget for commercial nuclear power plants. The goal of the Light Water Reactor Sustainability Program Physical Security Pathway is to develop tools, methods, and technologies and provide the technical basis for an optimized physical security posture. This pathway will analyze and minimize the conservatisms built into current security postures in order to reduce security costs while still ensuring adequate security and operational safety. The research performed at Idaho National Laboratory within this pathway has successfully developed a dynamic force-on-force (FOF) modeling framework using various computer simulation tools and integrated them with the dynamic assessment Event Modeling Risk Assessment using Linked Diagrams (EMRALD) tool. This document provides an overview of lessons learned in applying a dynamic computational framework that links results from a commercially available FOF simulation tool, a commercially available thermal-hydraulic tool, and EMRALD to an operating commercial nuclear power plant. This process of including plant procedures and multiple analysis results is being called Modeling and Analysis for Safety Security using Dynamic EMRALD Framework. Previous reports described how a user could integrate their plant-specific FOF models with the dynamic simulation tool EMRALD, model operator actions, integrate with probabilistic risk assessment tools, such as Computer Aided Fault Tree Analysis System or Systems Analysis Programs for Hands-on Integrated Reliability Evaluations, and with thermal-hydraulic tools, such as RELAP-5. Previous reports applied various combinations of available simulations codes with EMRALD using generic plant models to demonstrate how to perform the analysis. This report documents the results of applying the dynamic computational framework to an actual nuclear facility using their security scenarios and timelines. The purpose of this study was to verify that results achieved using generic models are similar to actual plant results and to refine our guidance of the use of the framework. Such an assessment enables further analysis, such as what-if scenarios and staff-reduction evaluation, thereby optimizing physical security at plants. NOTE: The work performed in this report is based on a generic EMRALD model with actual plant data used for the analysis. However, only the generic model and general results of the analysis are in the report. No plant’s sensitive information is discussed in this report. The discussion shows examples of insights that can be obtained from the MASS-DEF methodology.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

On the Stochastic Stability of Deep Markov Models

Deep Markov models (DMM) are generative models which are scalable and expressive generalization of Markov models for representation, learning, and inference problems. DMMs using deep neural networks to parametrize the transition of Markov probability distributions have recently been shown to provide more expressiveness in modeling sequential data and dynamical system responses. However, the fundamental stochastic stability guarantees of such models have not been thoroughly investigated. In this paper, we present a rigorous analytical method to prove the necessary and sufficient conditions of DMM's stochastic stability. This task is achieved by spectral analysis of the efficiently computed Jacobians of probabilistic maps modeled by deep neural networks. We make theoretical connections between the eigenvalues of neural network's weights and the different activation function types used on the stability and overall dynamic behavior of DMMs with Gaussian distributions. We empirically substantiate our theoretical results on stochastic stability and eigenvalue spectra via several numerical experiments. Formal stability guarantees of DMMs can substantially improve their robustness and trustworthiness, necessary for reliable use in safety-critical real-world applications.

Drgona, Jan↗

Benchmarking the performance of uncertainty quantification methods for neural network-based interatomic potentials

Machine-learned interatomic potentials (ML-IAPs) continue to gain popularity as accurate, computationally efficient replacements for traditional, physics-based interatomic potentials and expensive ab initio methods. Uncertainty quantification (UQ) of ML-IAPs is a growing area of research as UQ is critical in many applications of IAPs, such as developing curated datasets, active learning-based data augmentation, self-improving models, and estimating the uncertainty of molecular dynamics simulations. In this paper, we construct and benchmark a series of different neural network potentials (NNPs) with varying network architectures to determine the performance of these models with respect to both the mean and uncertainty calibration error. Each NNP method is specifically designed to predict either epistemic or aleatoric uncertainty with particular focus on the differences in behavior between the epistemic and aleatoric uncertainty estimates. We benchmark these methods using multiple datasets common in the ML-IAP literature. The results show that the aleatoric uncertainty from single-shot model architectures is a competitive alternative to ensemble-based epistemic uncertainty predictions in regions of sufficient data-density. However, in regions where the representative data is sparse, aleatoric uncertainty models tend to overpredict and epistemic methods tend to underpredict the actual model error. We conclude that the type of UQ is crucial when discussing performance of probabilistic model results as different methods have different performance characteristics depending on the regime in which they are evaluated. Therefore, the type of UQ method should be carefully evaluated against both the data characteristics and requirements for the intended application.

97 MATHEMATICS AND COMPUTING↗