SEARCH · Search NASA
Results for “Distributed training”
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.
Physics informed deep neural network embedded in a chemical transport model for the Amazon rainforest
Secondary organic aerosols (SOA) are fine particles in the atmosphere, which interact with clouds, radiation and affect the Earth’s energy budget. SOA formation involves chemistry in gas phase, aqueous aerosols, and clouds. Simulating these chemical processes involve solving a stiff set of differential equations, which are computationally expensive steps for three-dimensional chemical transport models. Deep neural networks (DNNs) are universal function approximators that could be used to represent the complex nonlinear changes in aerosol physical and chemical processes; however, key challenges such as generalizability to extended time periods, preservation of mass balance, simulating sparse model outputs, and maintaining physical constraints have limited their use in atmospheric chemistry. Here, we develop an approach of using a physics-informed DNN that overcomes previous such challenges and demonstrates its applicability for the chemical formation processes of isoprene epoxydiol SOA (IEPOX-SOA) over the Amazon rainforest. The DNN is trained with data generated by simulating IEPOX-SOA over the entire atmospheric column, using the Weather Research and Forecasting Model coupled with Chemistry (WRF-Chem). The trained DNN is then embedded within WRF-Chem to replace the computationally expensive default solver of IEPOX-SOA formation. The trained DNN predictions generalizes well with the default model simulation of the IEPOX-SOA mass concentrations and its size distribution (20 size bins) over several days of simulations in both dry and wet seasons. The embedded DNN reduces the computational expense of WRF-Chem by a factor of 2. Our approach shows promise in terms of application to other computationally expensive chemistry solvers in climate models.
A physics-constrained deep learning treatment of runaway electron dynamics
An adjoint formulation leveraging a physics-informed neural network (PINN) is employed to advance the density moment of a runaway electron (RE) distribution forward in time. A distinguishing feature of this approach is that once the adjoint problem is solved, its solution can be used to project the RE density forward in time for an arbitrary initial momentum space distribution of REs. Furthermore, by employing a PINN, a parametric solution to the adjoint problem can be learned. Thus, once trained, this adjoint-deep learning framework is able to efficiently project the RE density forward in time across various plasma conditions while still including a fully kinetic description of RE dynamics. As an example application, the temporal evolution of the density of primary electrons is studied, with particular emphasis on evaluating the decay of a RE population when below threshold. Predictions from the adjoint-deep learning framework are found to be in good agreement with a traditional relativistic electron Fokker–Planck solver, for several distinct initial conditions, and across an array of physics parameters. Once trained, the PINN thus provides a means of generating RE density time histories with exceptionally low online execution time.
A Review of Current Research Trends in Power-Electronic Innovations in Cyber–Physical Systems
Here, in this article, a broad overview of the current research trends in power-electronic innovations in cyber–physical systems (CPSs) is presented. The recent advances in semiconductor device technologies, control architectures, and communication methodologies have enabled researchers to develop integrated smart CPSs that can cater to the emerging requirements of smart grids, renewable energy, electric vehicles, trains, ships, the Internet of Things (IoT), and so on. The topics presented in this article include novel power-distribution architectures, protection techniques considering large renewable integration in smart grids, wireless charging in electric vehicles, simultaneous power and information transmission, multihop network-based coordination, power technologies for renewable energy and smart transformer, CPS reliability, transactive smart railway grid, and real-time simulation of shipboard power systems. It is anticipated that the research trends presented in this article will provide a timely and useful overview to the power-electronics researchers with broad applications in CPSs.
Reducing Urban Traffic Congestion Using Deep Learning and Model Predictive Control
This article proposes a deep learning (DL)-based control algorithm—DL velocity-based model predictive control (VMPC)—for reducing traffic congestion with slowly time-varying traffic signal controls. This control algorithm consists of system identification using DL and traffic signal control using VMPC. For the training process of DL, we established a modeling error entropy loss as the criteria inspired by the theory of stochastic distribution control (SDC) originated by the fourth author. Herein, simulation results show that the proposed algorithm can reduce traffic congestion with a slowly varying traffic signal control input. Results of an ablation study demonstrate that this algorithm compares favorably to other model-based controllers in terms of prediction error, signal varying speed, and control effectiveness.
Generative Ensemble Regression: Learning Particle Dynamics from Observations of Ensembles with Physics-informed Deep Generative Models
Here, we propose a new method for inferring the governing stochastic ordinary differential equations (SODEs) by observing particle ensembles at discrete and sparse time instants, i.e., multiple “snapshots.” Particle coordinates at a single time instant, possibly noisy or truncated, are recorded in each snapshot but are unpaired across the snapshots. By training a physics-informed generative model that generates “fake” sample paths, we aim to fit the observed particle ensemble distributions with a curve in the probability measure space, which is induced from the inferred particle dynamics. We employ different metrics to quantify the differences between distributions, e.g., the sliced Wasserstein distances and the adversarial losses in generative adversarial networks. We refer to this method as generative “ensemble-regression” (GER), in analogy to the classic “point-regression,” where we infer the dynamics by performing regression in the Euclidean space. We illustrate the GER by learning the drift and diffusion terms of particle ensembles governed by SODEs with Brownian motions and Lévy processes up to 100 dimensions. We also discuss how to treat cases with noisy or truncated observations. Apart from systems consisting of independent particles, we also tackle nonlocal interacting particle systems with unknown interaction potential parameters by constructing a physics-informed loss function. Finally, we investigate scenarios of paired observations and discuss how to reduce the dimensionality in such cases by proving a convergence theorem that provides theoretical support.
DSS-SimPy-RL (Open-DSS and SimPy based Cyber-Physical RL environment) [SWR-23-29]
Recently, numerous data-driven approaches to control an electric grid using machine learning techniques have been investigated. With the advancement of reinforcement learning (RL) based techniques, gradually the conventional optimization based solvers are being replaced with RL approach where there is uncertainty in the environment such as renewable generation or cyber system emulation. However, to train an agent efficiently, it requires numerous interactions with an environment to learn the best policies. There are numerous RL environments for the power systems based on some well-known simulators, similarly there are environment for communication domains. While majority of the cyber emulators are based in an UNIX environment, the power simulators are based in the Windows-based operating system, the generation of cyber-physical mixed domain RL environment has been challenging. Existing co-simulation methods are efficient but resource and time intensive to generate large scale data set for training RL agents. Hence, this software focuses on development and validation of a mixed domain RL environment using Open DSS for the physical side and leverages a discrete event simulator python package, SimPy, for cyber-side emulation which is Operating Systems agnostic. Further utilizing this software co-simulation and training RL agents for re-routing based resilient control for network reconfiguration and volt-var control in power distribution feeder are performed.
Linking topographic variation in belowground C processes with hydrological processes to improve Earth system models
The effects of topography on belowground processes were investigated at the Susquehanna-Shale Hills Critical Zone Observatory in central Pennsylvania. Hill slope had stronger effects on aboveground biomass allocation than belowground allocation. Hills slope position affected the respiratory response to temperature with a 10-degree increase having more influence in slope positions characterized by higher moisture. A spatial distributed forest model that includes nitrogen transport was developed from a 1-dimensional biogeochemical model and a spatially distributed land surface hydrological model. This model can simulate the impact of topography-driven spatially-distributed soil moisture, soil temperature, solar radiation, and soil mineral N on watershed carbon processes. This project helped in the training of four PhD students.
Machine learning analysis reveals relationship between pomacentrid calls and environmental cues
Sound production rates of fishes can be used as an indicator for coral reef health, providing an opportunity to utilize long-term acoustic recordings to assess environmental change. As acoustic datasets become more common, computational techniques need to be developed to facilitate analysis of the massive data files produced by long-term monitoring. Machine learning techniques demonstrate an advantage in the identification of fish sounds over manual sampling approaches. Here we evaluated the ability of convolutional neural networks to identify and monitor call patterns for pomacentrids (damselfishes) in a tropical reef region of the western Pacific. A stationary hydrophone was deployed for 39 mo (2014-2018) in the National Park of American Samoa to continuously record the local marine acoustic environment. A neural network was trained—achieving 94% identification accuracy of pomacentrids—to demonstrate the applicability of machine learning in fish acoustics and ecology. The distribution of sound production was found to vary on diel and interannual timescales. Additionally, the distribution of sound production was correlated with wind speed, water temperature, tidal amplitude, and sound pressure level. This research has broad implications for state-of-the-art acoustic analysis and promises to be an efficient, scalable asset for ecological research, environmental monitoring, and conservation planning.
Simulation techniques for estimating error in the classification of normal patterns
Methods of efficiently generating and classifying samples with specified multivariate normal distributions were discussed. Conservative confidence tables for sample sizes are given for selective sampling. Simulation results are compared with classified training data. Techniques for comparing error and separability measure for two normal patterns are investigated and used to display the relationship between the error and the Chernoff bound.
Software for Autonomous Spacecraft Maneuvers
The AutoCon computer programs facilitate and accelerate the planning and execution of orbital control maneuvers of spacecraft while analyzing and resolving mission constraints. AutoCon-F is executed aboard spacecraft, enabling the spacecraft to plan and execute maneuvers autonomously; AutoCon-G is designed for use on the ground. The AutoCon programs utilize advanced techniques of artificial intelligence, including those of fuzzy logic and natural-language scripting, to resolve multiple conflicting constraints and automatically plan maneuvers. These programs can be used to satisfy requirements for missions that involve orbits around the Earth, the Moon, or any planet, and are especially useful for missions in which there are requirements for frequent maneuvers and for resolution of complex conflicting constraints. During operations, the software targets new trajectories, places and sizes maneuvers, and controls spacecraft burns. AutoCon-G provides a userfriendly graphical interface, and can be used effectively by an analyst with minimal training. AutoCon-F reduces latency and supports multiple-spacecraft and formation-flying missions. The AutoCon architecture supports distributive processing, which can be critical for formation- control missions. AutoCon is completely object-oriented and can easily be enhanced by adding new objects and events. AutoCon-F was flight demonstrated onboard GSFC's EO-1 spacecraft flying in formation with Landsat-7.
A Human Factors Approach to Bridging Systems and Introducing New Technologies
The application of human factors in aviation has grown to cover a wide range of disciplines and methods capable of assessing human-systems integration at many levels. For example, at the individual level, pilot workload may be studied while at the team level, coordinated workload distribution may be the focal point. At the organizational level, the way in which individuals and teams are supported by training and standards, policies and procedures may introduce additional, relevant topics. A consideration of human factors at each level contributes to our understanding of successes and failures in pilot performance, but this system focused on the flight deck alone -- is only one part of the airspace system. In the FAA's NextGen plan to overhaul the National Airspace System (NAS), new capabilities will enhance flightdeck systems (pilots), flight operations centers (dispatchers) and air traffic control systems (controllers and air traffic managers). At a minimum, the current roles and responsibilities of these three systems are likely to change. Since increased automation will be central to many of the enhancements, the role of automation is also likely to change. Using NextGen examples, a human factors approach for bridging complex airspace systems will be the main focus of this presentation. It is still crucial to consider the human factors within each system, but the successful implementation of new technologies in the NAS requires an understanding of the collaborations that occur when these systems intersect. This human factors approach to studying collaborative systems begins with detailed task descriptions within each system to establish a baseline of the current operations. The collaborative content and context are delineated through the review of regulatory and advisory materials, letters of agreement, policies, procedures and documented practices. Field observations and interviews also help to fill out the picture. Key collaborative functions across systems are identified and placed on a phase-of-flight timeline including information requirements, decision authority and use of automation, as well as level of frequency and criticality.
A First Look at the Upcoming SISO Space Reference FOM
Spaceflight is difficult, dangerous and expensive; human spaceflight even more so. In order to mitigate some of the danger and expense, professionals in the space domain have relied, and continue to rely, on computer simulation. Simulation is used at every level including concept, design, analysis, construction, testing, training and ultimately flight. As space systems have grown more complex, new simulation technologies have been developed, adopted and applied. Distributed simulation is one those technologies. Distributed simulation provides a base technology for segmenting these complex space systems into smaller, and usually simpler, component systems or subsystems. This segmentation also supports the separation of responsibilities between participating organizations. This segmentation is particularly useful for complex space systems like the International Space Station (ISS), which is composed of many elements from many nations along with visiting vehicles from many nations. This is likely to be the case for future human space exploration activities. Over the years, a number of distributed simulations have been built within the space domain. While many use the High Level Architecture (HLA) to provide the infrastructure for interoperability, HLA without a Federation Object Model (FOM) is insufficient by itself to insure interoperability. As a result, the Simulation Interoperability Standards Organization (SISO) is developing a Space Reference FOM. The Space Reference FOM Product Development Group is composed of members from several countries. They contribute experiences from projects within NASA, ESA and other organizations and represent government, academia and industry. The initial version of the Space Reference FOM is focusing on time and space and will provide the following: (i) a flexible positioning system using reference frames for arbitrary bodies in space, (ii) a naming conventions for well-known reference frames, (iii) definitions of common time scales, (iv) federation agreements for common types of time management with focus on time stepped simulation, and (v) support for physical entities, such as space vehicles and astronauts. The Space Reference FOM is expected to make collaboration politically, contractually and technically easier. It is also expected to make collaboration easier to manage and extend.
Tidal Disruption and the Appearance of Periodic Comet Shoemaker-Levy 9
A unified model is presented that quantitatively interprets the observed characteristics of the nuclear train, the two dust trails, and the tail region of P/Shoemaker-Levy 9 in terms of a collisionally modified rotation velocity distribution of the comet's debris.
Collaborative Supervised Learning for Sensor Networks
Collaboration methods for distributed machine-learning algorithms involve the specification of communication protocols for the learners, which can query other learners and/or broadcast their findings preemptively. Each learner incorporates information from its neighbors into its own training set, and they are thereby able to bootstrap each other to higher performance. Each learner resides at a different node in the sensor network and makes observations (collects data) independently of the other learners. After being seeded with an initial labeled training set, each learner proceeds to learn in an iterative fashion. New data is collected and classified. The learner can then either broadcast its most confident classifications for use by other learners, or can query neighbors for their classifications of its least confident items. As such, collaborative learning combines elements of both passive (broadcast) and active (query) learning. It also uses ideas from ensemble learning to combine the multiple responses to a given query into a single useful label. This approach has been evaluated against current non-collaborative alternatives, including training a single classifier and deploying it at all nodes with no further learning possible, and permitting learners to learn from their own most confident judgments, absent interaction with their neighbors. On several data sets, it has been consistently found that active collaboration is the best strategy for a distributed learner network. The main advantages include the ability for learning to take place autonomously by collaboration rather than by requiring intervention from an oracle (usually human), and also the ability to learn in a distributed environment, permitting decisions to be made in situ and to yield faster response time.
Predicting CO 2 Plume Migration in Heterogeneous Formations Using Conditional Deep Convolutional Generative Adversarial Network
Numerical simulation of flow and transport in heterogeneous formations has long been studied, especially for uncertainty quantification and risk assessment. The high computational cost associated with running large-scale numerical simulations in a Monte Carlo sense has motivated the development of surrogate models, which aim to capture the important input-output relations of physics-based models but require only a fraction of the cost of full model runs. In this work, we formulate a conditional deep convolutional generative adversarial network (cDC-GAN) surrogate model to learn the dynamic functional mappings in multiphase models. The cDC-GAN belongs to a class of semisupervised learning methods that can be used to learn the data generation processes. Like the original GAN, a main strength of the cDC-GAN is that it includes a self-training scheme for improving the quality of generative modeling in a game theoretic framework, without requiring extensive statistical knowledge and assumptions on input data distributions. In particular, our cDC-GAN model is designed to learn cross-domain mappings between high-dimensional input (e.g., permeability) and output (e.g., phase saturations) pairs, with the ability to incorporate conditioning information (e.g., prediction time). As a use case, we demonstrate the performance of cDC-GAN for predicting the migration of carbon dioxide (CO 2 ) plume in heterogeneous carbon storage reservoirs, which has both numerical and practical significance because of the safe storage requirements now mandated in many countries. Results show that cDC-GAN achieves high accuracy in predicting the spatial and temporal evolution patterns of the injected CO 2 plume, as compared to the original results obtained using a compositional reservoir simulator. The performance of cDC-GAN models, trained using the same number of training samples, stays relatively robust when the level of spatial heterogeneity is increased. Our cDC-GAN is pattern based and is not limited by the underlying physics. Thus, it provides a general framework for developing surrogate models, and for conducting uncertainty analyses for a wide range of physics-based models used in both groundwater and subsurface energy exploration applications.
FM4NPP: A Scaling Foundation Model for Nuclear and Particle Physics
Large language models have revolutionized artificial intelligence by enabling large, generalizable models trained through self-supervision. This paradigm has inspired the development of scientific foundation models (FMs). However, applying this capability to experimental particle physics is challenging due to the sparse, spatially distributed nature of detector data, which differs dramatically from natural language. This work addresses if an FM for particle physics can scale and generalize across diverse tasks. We introduce a new dataset with more than 11 million particle collision events and a suite of downstream tasks and labeled data for evaluation. We propose a novel self-supervised training method for detector data and demonstrate its neural scalability with models that feature up to 188 million parameters. With frozen weights and task-specific adapters, this FM consistently outperforms baseline models across all downstream tasks. The performance also exhibits robust data-efficient adaptation. Further analysis reveals that the representations extracted by the FM are task-agnostic but can be specialized via a single linear mapping for different downstream tasks.
Station set residual : event classification using historical distribution of observing stations.
Analysts working at the International Data Centre in support of treaty monitoring through the Comprehensive Nuclear-Test-Ban Treaty Organization spend a significant amount of time reviewing hypothesized seismic events produced by an automatic processing system. When reviewing these events to determine their legitimacy, analysts take a variety of approaches that rely heavily on training and past experience. One method used by analysts to gauge the validity of an event involves examining the set of stations involved in the detection of an event. In particular, leveraging past experience, an analyst can say that an event located in a certain part of the world is expected to be detected by Stations A, B, and C. Implicit in this statement is that such an event would usually not be detected by Stations X, Y, or Z. For some well understood parts of the world, the absence of one or more 'expected' stations - or the presence of one or more 'unexpected' stations - is correlated with a hypothesized event's legitimacy and to its survival to the event bulletin. The primary objective of this research is to formalize and quantify the difference between the observed set of stations detecting some hypothesized event, versus the expected set of stations historically associated with detecting similar nearby events close in magnitude. This Station Set Residual can be quantified in many ways, some of which are correlated with the analysts determination of whether or not the event is valid. We propose that this Station Set Residual score can be used to screen out certain classes of 'false' events produced by automatic processing with a high degree of confidence, reducing the analyst burden. Moreover, we propose that the visualization of the historically expected distribution of detecting stations can be immediately useful as an analyst aid during their review process.