SEARCH · Search NASA
Results for “Generative modeling”
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.
Battery Degradation Modeling in Hybrid Power Plants: An Island System Unit Commitment Study: Preprint
As hybrid power plants (HPPs), such as photovoltaic (PV) and battery combinations, become increasingly important in power systems with high renewable energy penetration to address PV variability and ensure grid stability. This paper focuses on the urgent need to model the coordination between PV and battery systems in HPPs while accounting for battery degradation. We present a generation scheduling model that explicitly incorporates PV-battery hybridization in the unit commitment problem. Moreover, the cost function of the HPP scheduling problem endogenously considers battery degradation with adjustable weights to strike a balance between minimizing production costs and prolonging battery life, particularly when providing energy arbitrage and ancillary services. Using a realistic island system simulation, we demonstrate that accounting for battery degradation in the scheduling problem can significantly extend battery life with only minor additional production costs.
Behind-the-Meter Solar Generation Disaggregation using Consumer Mixture Models
To facilitate deep penetration of solar energy in smart grids, we need high observability of solar generation at the edges of the grid. Current advanced metering infrastructures (AMI) only monitor the aggregated measurements from net-metered households, but disaggregated consumption and solar generation components are required for grid optimizations. We propose an unsupervised disaggregation model for disaggregating solar generation from AMI measurements without the need of training data. The model requires only AMI measurements from consumers in a region and the solar irradiance as input, and models the consumption of consumers by neighboring households without rooftop photovoltaics (PV) to perform the disaggregation. We evaluate our results on a real life dataset from Austin, Texas. We show that our model is able to disaggregate consumption and solar generation measurements with 42.24% and 31.67% less mean squared error, respectively, in comparison to a baseline technique that uses supervised learning. This shows that our model is capable of disaggregating historical data even if the dataset has no training data and only contains minimal exogenous data.
Comparing different generations of idealized solar geoengineering simulations in the Geoengineering Model Intercomparison Project (GeoMIP)
Abstract. Solar geoengineering has been receiving increased attention in recent years as a potential temporary solution to offset global warming. One method of approximating global-scale solar geoengineering in climate models is via solar reduction experiments. Two generations of models in the Geoengineering Model Intercomparison Project (GeoMIP) have now simulated offsetting a quadrupling of the CO2 concentration with solar reduction. This simulation is idealized and designed to elicit large responses in the models. Here, we show that energetics, temperature, and hydrological cycle changes in this experiment are statistically indistinguishable between the two ensembles. Of the variables analyzed here, the only major differences involve highly parameterized and uncertain processes, such as cloud forcing or terrestrial net primary productivity. We conclude that despite numerous structural differences and uncertainties in models over the past two generations of models, including an increase in climate sensitivity in the latest generation of models, the models are consistent in their aggregate climate response to global solar dimming.
Low responsiveness of machine learning models to critical or deteriorating health conditions
Machine learning (ML) based mortality prediction models can be immensely useful in intensive care units. Such a model should generate warnings to alert physicians when a patient’s condition rapidly deteriorates, or their vitals are in highly abnormal ranges. Before clinical deployment, it is important to comprehensively assess a model’s ability to recognize critical patient conditions. We develop multiple medical ML testing approaches, including a gradient ascent method and neural activation map. We systematically assess these machine learning models’ ability to respond to serious medical conditions using additional test cases, some of which are time series. Guided by medical doctors, our evaluation involves multiple machine learning models, resampling techniques, and four datasets for two clinical prediction tasks. We identify serious deficiencies in the models’ responsiveness, with the models being unable to recognize severely impaired medical conditions or rapidly deteriorating health. For in-hospital mortality prediction, the models tested using our synthesized cases fail to recognize 66% of the injuries. In some instances, the models fail to generate adequate mortality risk scores for all test cases. Our study identifies similar kinds of deficiencies in the responsiveness of 5-year breast and lung cancer prediction models. Using generated test cases, we find that statistical machine-learning models trained solely from patient data are grossly insufficient and have many dangerous blind spots. Most of the ML models tested fail to respond adequately to critically ill patients. How to incorporate medical knowledge into clinical machine learning models is an important future research direction.
Co-optimized machine-learned manifold models for large eddy simulation of turbulent combustion
Many modeling approaches in large eddy simulation (LES) of turbulent combustion employ a projection of the thermochemical state onto a low-dimensional manifold within state space to reduce the number of transported variables and hence computational cost. Flamelet-generated manifolds (FGM) is an example of a well-established, physics-based approach, but increasingly, principal component analysis (PCA) is being used as a data-driven method for generating manifold models. For both approaches, the nonlinear relationship between the location on the predefined manifold and the outputs of interest, such as reaction rates, can be tabulated or encoded in a neural network. This work proposes a new approach for manifold modeling that extends these existing approaches. A modified neural network structure simultaneously encodes the definition of the manifold variables, the nonlinear mapping, and the subfilter closure for LES. This allows all three of these aspects of the model to be co-optimized, generating a model from any source of combustion thermochemical state data. The manifold parameterizing variables are constrained to be linear combinations of species, as in FGM and PCA-based models, to aid in interpretability and implementation. For LES, subfilter variances of the manifold variables are also included as inputs. Two types of a priori analysis are performed to evaluate the new approach. In the first, the model is trained on data from one-dimensional premixed flames. In this case, the approach recovers the behavior of flamelet-based manifold approaches, and in fact slightly improves performance by identifying an optimized progress variable. The approach is also applied to data from direct numerical simulations of spherical ignition kernels in isotropic turbulence. For any specified manifold dimensionality, the new approach provides substantially lower prediction errors than a PCA-based model developed from the same data set. Additionally, the LES formulation of the new approach can provide accurate predictions for filtered reaction rates across a variety of filter widths.
CosmoGAN: creating high-fidelity weak lensing convergence maps using Generative Adversarial Networks
Abstract Inferring model parameters from experimental data is a grand challenge in many sciences, including cosmology. This often relies critically on high fidelity numerical simulations, which are prohibitively computationally expensive. The application of deep learning techniques to generative modeling is renewing interest in using high dimensional density estimators as computationally inexpensive emulators of fully-fledged simulations. These generative models have the potential to make a dramatic shift in the field of scientific simulations, but for that shift to happen we need to study the performance of such generators in the precision regime needed for science applications. To this end, in this work we apply Generative Adversarial Networks to the problem of generating weak lensing convergence maps. We show that our generator network produces maps that are described by, with high statistical confidence, the same summary statistics as the fully simulated maps.
Model-Based Approaches to Generate Knowledge from Data in a Plant Reliability Context
One challenge that nuclear power plant system engineers are facing is that the amount of equipment reliability (ER) data being continuously generated are extremely large. These data elements come in different forms: textual (e.g., condition reports) and numeric (e.g., generated by monitoring systems) and they provide system engineers with valuable insights and information regarding the discovery of anomalous behaviors or degradation trends, the identification of the possible causes behind such behaviors and trends, and the prediction of their direct consequences. This paper directly targets the generation of knowledge from ER data by putting “data into context”. Here, we employ model-based system engineering (MBSE) models of systems and assets to represent and capture their architecture and functional (i.e., cause-effect) relations. ER data elements are processed by identifying first which elements of the developed MBSE elements they are referring to. This task is much harder for textual data since the information contained in issue or maintenance reports needs to “be understood” by a computational tool. Here we called this process “knowledge extraction” where our methods to extract knowledge from textual data. Lastly, once numeric and textual ER data elements have been processed and “understood”, we discover possible cause-effect relations among them. This is performed by observing if a logical connection through the MBSE models exists, and if there is a temporal relation among them. The logic and temporal are the two main ingredients to perform “machine reasoning” from ER data.
US Renewable Futures in the GCAM
This report examines renewable energy deployment in the United States using a version of the Global Change Assessment Model (GCAM) with a detailed representation of renewables; the GCAM-RE. Electricity generation was modeled in four generation segments and 12-subregions. This level of regional and sector detail allows a more explicit representation of renewable energy generation. Wind, solar thermal power, and central solar PV plants are implemented in explicit resource classes with new intermittency parameterizations appropriate for each technology. A scenario analysis examines a range of assumptions for technology characteristics, climate policy, and long distance transmission.
Particle-in-cell modeling of electron beam generated plasma
Plasmas generated using energetic electron beams are well known for their low electron temperature (T e ) and plasma potential, which makes them attractive for atomic-precision plasma processing applications such as atomic layer etch and deposition. A 2-dimensional particle-in-cell model for an electron beam-generated plasma in argon confined by a constant applied magnetic field is described here in this article. Plasma production primarily occurs in the path of the beam electrons in the center of the chamber. The resulting plasma spreads out in the chamber through non-ambipolar diffusion with a short-circuiting effect allowing unequal electron and ion fluxes to different regions of the bounding conductive chamber walls. The cross-field transport of the electrons (and thus the steady-state characteristics of the plasma) are strongly impacted by the magnetic field. T e is anisotropic in the electron beam region, but low and isotropic away from the plasma production zone. The plasma density increases and the plasma becomes more confined near the region of production when the magnetic field strengthens. The magnetic field reduces both electron physical and energy transport perpendicular to the magnetic field. T e is uniform along the magnetic field lines and slowly decreases perpendicular to it. Electrons are less energetic in the sheath regions where the sheath electric field repels and confines the low-energy electrons from the bulk plasma. Even though electron and ion densities are similar in the bulk plasma due to quasi-neutrality, electron and ion fluxes on the grounded chamber walls are unequal at most locations. Electron confinement by the magnetic field weakens with increasing pressure, and the plasma spread out farther from the electron beam region.
Skeletal Kinetics Reduction for Astrophysical Reaction Networks
A novel methodology is developed to extract accurate skeletal reaction models for nuclear combustion. Local sensitivities of isotope mass fractions with respect to reaction rates are modeled based on the forced optimally time-dependent (f-OTD) scheme. These sensitivities are then analyzed temporally to generate skeletal models. The methodology is demonstrated by conducting skeletal reduction of constant density and temperature burning of carbon and oxygen relevant to Type Ia supernovae (SNe Ia). The 495-isotopes Torch model is chosen as the detailed reaction network. A map of maximum production of 56 Ni in SNe Ia is produced for different temperatures, densities, and proton-to-neutron ratios. The f-OTD simulations and the sensitivity analyses are then performed with initial conditions from this map. A series of skeletal models are derived and their performances are assessed by comparison against currently existing skeletal models. Previous models have been constructed intuitively by assuming the dominance of α-chain reactions. The comparison of the newly generated skeletal models against previous models is based on the predicted energy release and 44 Ti and 56 Ni abundances by each model. The consequences of ye ≠ 0.5 in the initial composition are also explored where ye is the electron fraction. The simulated results show that 56 Ni production decreases by decreasing ye as expected, and that the 43 Sc is a key isotope in proton and neutron channels toward 56 Ni production. It is shown that an f-OTD skeletal model with 150 isotopes can accurately predict the 56Ni abundance in SNe Ia for ye ≲ 0.5 initial conditions.
Impacts of climate change on subannual hydropower generation: a multi-model assessment of the United States federal hydropower plant
Abstract Hydropower is a low-carbon emission renewable energy source that provides competitive and flexible electricity generation and is essential to the evolving power grid in the context of decarbonization. Assessing hydropower availability in a changing climate is technically challenging because there is a lack of consensus in the modeling representation of key dynamics across scales and processes. Focusing on 132 US federal hydropower plants, in this study we evaluate the compounded impact of climate and reservoir-hydropower models’ structural uncertainties on monthly hydropower projections. In particular, instead of relying on one single regression-based hydropower model, we introduce another conceptual reservoir operations-hydropower model in the assessment framework. This multi-model assessment approach allows us to partition uncertainties associated with both climate and hydropower models for better clarity. Results suggest that while at least 70% of the uncertainties at the annual scale and 50% at the seasonal scale can be attributed to the choice of climate models, up to 50% of seasonal variability can be attributed to the choice of hydropower models, particularly in regions over the western US where the reservoir storage is substantial. The analysis identifies regions where multi-model assessments are needed and presents a novel approach to partition uncertainties in hydropower projections. Another outcome includes an updated evaluation of Coupled Model Intercomparison Project Phase 5 (CMIP5)-based federal hydropower projection, at the monthly scale and with a larger ensemble, which can provide a baseline for understanding future assessments based on CMIP6 and beyond.
Integrated Reflection Seismic Monitoring and Reservoir Modeling for Geologic CO 2 Sequestration
The US DOE/NETL CCS MVA program funded a project with Fusion Petroleum Technologies Inc. (now SIGMA) to model the proof of concept of using sparse seismic data in the monitoring of CO 2 injected into saline aquifers. The goal of the project was to develop and demonstrate an active source reflection seismic imaging strategy based on deployment of spatially sparse surface seismic arrays. The primary objective was to test the feasibility of sparse seismic array systems to monitor the CO 2 plume migration injected into deep saline aquifers. The USDOE/RMOTC Teapot Dome (Wyoming) 3D seismic and reservoir data targeting the Crow Mountain formation was used as a realistic proxy to evaluate the feasibility of the proposed methodology. Though the RMOTC field has been well studied, the Crow Mountain as a saline aquifer has not been studied previously as a CO 2 sequestration (storage) candidate reservoir. A full reprocessing of the seismic data from field tapes that included prestack time migration (PSTM) followed by prestack depth migration (PSDM) was performed. A baseline reservoir model was generated from the new imaging results that characterized the faults and horizon surfaces of the Crow Mountain reservoir. The 3D interpretation was integrated with the petrophysical data from available wells and incorporated into a geocellular model. The reservoir structure used in the geocellular model was developed using advanced inversion technologies including Fusion's ThinMAN™ broadband spectral inversion. Seal failure risk was assessed using Fusion's proprietary GEOPRESS™ pore pressure and fracture pressure prediction technology. CO 2 injection was simulated into the Crow Mountain with a commercial reservoir simulator. Approximately 1.2MM tons of CO 2 was simulated to be injected into the Crow Mountain reservoir over 30 years and subsequently let 'soak' in the reservoir for 970 years. The relatively small plume developed from this injection was observed migrating due to gravity to the apexes of the double anticline in the Crow Mountain reservoir of the Teapot dome. Four models were generated from the reservoir simulation task of the project which included three saturation models representing snapshots at different times during and after simulated CO 2 injection and a fully saturated CO 2 fluid substitution model. The saturation models were used along with a Gassmann fluid substitution model for CO 2 to perform fluid volumetric substitution in the Crow Mountain formation. The fluid substitution resulted in a velocity and density model for the 3D volume at each saturation condition that was used to generate a synthetic seismic survey. FPTI's (Fusion Petroleum Technologies Inc.) proprietary SeisModelPRO™ full acoustic wave equation software was used to simulate acquisition of a 3D seismic survey on the four models over a subset of the field area. The simulated acquisition area included the injection wells and the majority of the simulated plume area.
PLUSWIND Derived Data
This dataset consists of annual CSV files containing multiple sources of modeled, hourly wind speeds and generation. For complete information about this dataset, including validation of modeled generation versus recorded generation, please see the Scientific Data article: Millstein, D., Jeong, S., Ancell, A., & Wiser, R. (2023). A database of hourly wind speed and modeled generation for US wind plants based on three meteorological models. Scientific Data, 10(1), 883. https://doi.org/10.1038/s41597-023-02804-w
Generative adversarial networks (GAN) based efficient sampling of chemical composition space for inverse design of inorganic materials
A major challenge in materials design is how to efficiently search the vast chemical design space to find the materials with desired properties. One effective strategy is to develop sampling algorithms that can exploit both explicit chemical knowledge and implicit composition rules embodied in the large materials database. Here, we propose a generative machine learning model (MatGAN) based on a generative adversarial network (GAN) for efficient generation of new hypothetical inorganic materials. Trained with materials from the ICSD database, our GAN model can generate hypothetical materials not existing in the training dataset, reaching a novelty of 92.53% when generating 2 million samples. The percentage of chemically valid (charge-neutral and electronegativity-balanced) samples out of all generated ones reaches 84.5% when generated by our GAN trained with such samples screened from ICSD, even though no such chemical rules are explicitly enforced in our GAN model, indicating its capability to learn implicit chemical composition rules to form compounds. Our algorithm is expected to be used to greatly expand the range of the design space for inverse design and large-scale computational screening of inorganic materials.
Enhancing power grid resilience to winter storms via generator winterization with equity considerations
Here we develop two-stage stochastic programming models for generator winterization that enhance power grid resilience while incorporating social equity. The first stage in our models captures the investment decisions for generator winterization, and the second stage captures the operation of a degraded power grid, with the objective of minimizing load shed and social inequity. To incorporate equity into our models, we propose a concept called adverse effect probability that captures the disproportionate effects of power outages on communities with varying vulnerability levels. Grid operations are modeled using DC power flow, and equity is captured through mean or maximum adverse effects experienced by communities. We apply our models to a synthetic Texas power grid, using winter storm scenarios created from the generator outage data from the 2021 Texas winter storm. Our extensive numerical experiments show that more equitable outcomes, in the sense of reducing adverse effects experienced by vulnerable communities during power outages, are achievable with no impact on total load shed through investing in winterization of generators in different locations and capacities.
Predicting grid-size-dependent fracture strains of DP980 with a microstructure-based post-necking model
Ductile fracture is a local phenomenon, and it is well established that fracture strain levels depend on both stress triaxiality and the resolution (grid size) of strain measurements. Two-dimensional plane strain post-necking models with different representative volume element (RVE) sizes are used to predict the size-dependent fracture strain of a commercial dual-phase steel, DP980. The models are generated from the actual microstructures, and the individual phase flow properties and literature-based individual phase damage parameters for the Johnson-Cook model are used for ferrite and martensite. A monotonic relationship is predicted: the smaller the model size, the higher the fracture strain. Thus, a general framework is developed to quantify the size-dependent fracture strains for multiphase materials. In addition to the RVE sizes, the influences of intrinsic microstructure features, i.e., the flow curve and fracture strains of the two constituent phases, on the predicted fracture strains also are examined. Application of the derived fracture strain versus RVE size relationship is demonstrated with large clearance trimming simulations with different element sizes.