Search NASA⌕ Search

SEARCH · Search NASA

Results for “leverage”

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 523 records · Page 29
All sourcesNASA NTRSDOE OSTI
DOE OSTI2022

Predictive modeling of a subcritical pulverized-coal power plant for optimization: Parameter estimation, validation, and application

As renewable power generation deployment increases, fossil fuel plants are increasingly required to operate more flexibly. Many coal-fired power plants were originally designed to operate at base load and do not operate optimally at partial load. Predictive first-principles plant-wide models can be employed to identify opportunities for flexibility improvements and diagnose low-load operating issues. This paper describes the application of the Institute for the Design of Advanced Energy Systems Integrated Platform (IDAES) to model and optimize flexible power plant operations. The key benefits of using IDAES are that it provides an open-source, fully equation-oriented modeling framework for efficient modular model construction, reuse, and customization, together with a mathematical optimization framework leveraging powerful, state-of-the-art solvers. The process systems engineering workflow from predictive process simulation to parameter estimation, model validation, and plant optimization is applicable to a variety of existing and next-generation energy systems as well as other chemical and environmental processes. Here, to demonstrate this capability, a physics-based, steady-state model was developed to improve full- and part-load performance of the Escalante Generating Station, a 245 MWe (net) subcritical pulverized coal-fired power plant owned and operated by Tri-State Generation and Transmission Association. Specifically, sixty-nine model parameters were simultaneously estimated from several months of operating data enabling prediction of flow rates, temperatures, pressures, and steam quality throughout the plant. The validated model was leveraged by Escalante to reduce the minimum operating load from 90 MW to 50 MW by diagnosing a low-load water-hammer issue, enabling coal usage and emissions reductions during periods of low power demand. Additionally, opportunities for heat rate reduction (i.e., efficiency improvement) through a steeper sliding-pressure approach to load-following and optimization of other boiler operating variables were also identified and quantified. For example, a potential efficiency improvement of 0.7 percentage points was observed at half-load operation.

01 COAL, LIGNITE, AND PEAT↗
DOE OSTI2020

Unified architecture for data-driven metadata tagging of building automation systems

This article presents a Unified Architecture (UA) for automated point tagging of Building Automation System (BAS) data, based on a combination of data-driven approaches. Advanced energy analytics applications—including fault detection and diagnostics and supervisory control—have emerged as a significant opportunity for improving the performance of our built environment. Effective application of these analytics depends on harnessing structured data from the various building control and monitoring systems, but typical BAS implementations do not employ any standardized metadata schema. While standards such as Project Haystack and Brick Schema have been developed to address this issue, the process of structuring the data, i.e., tagging the points to apply a standard metadata schema, has, to date, been a manual process. This process is typically costly, labor-intensive, and error-prone. In this work we address this gap by proposing a UA that automates the process of point tagging by leveraging the data accessible through connection to the BAS, including time-series data and the raw point names. The UA intertwines supervised classification and unsupervised clustering techniques from machine learning and leverages both their deterministic and probabilistic outputs to inform the point tagging process. Furthermore, we extend the UA to embed additional input and output data-processing modules that are designed to address the challenges associated with the real-time deployment of this automation solution. We test the UA on two datasets for real-life buildings: (i) commercial retail buildings and (ii) office buildings from the National Renewable Energy Laboratory (NREL) campus. We report the proposed methodology correctly applied 85–90% and 70–75% of the tags in each of these test scenarios, respectively for two significantly different building types used for testing UA's fully-functional prototype. The proposed UA, therefore, offers promising approach for automatically tagging BAS data as it reaches close to 90% accuracy. Further building upon this framework to algorithmically identify the equipment type and their relationships is an apt future research direction to pursue.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
DOE OSTI2023

GMT: A deep learning approach to generalized multivariate translation for scientific data analysis and visualization

In scientific visualization, despite the significant advances of deep learning for data generation, researchers have not thoroughly investigated the issue of data translation. We present a new deep learning approach called generalized multivariate translation (GMT) for multivariate time-varying data analysis and visualization. Like V2V, GMT assumes a preprocessing step that selects suitable variables for translation. However, unlike V2V, which only handles one-to-one variable translation during training and inference, GMT enables one-to-many and many-to-many variable translation in the same framework. We leverage the recent StarGAN design from multi-domain image-to-image translation to achieve this generalization capability. We experiment with different loss functions and injection strategies to explore the best choices and leverage pre-training for performance improvement. We compare GMT with other state-of-the-art methods (i.e., Pix2Pix, V2V, StarGAN). Furthermore, the results demonstrate the overall advantage of GMT in translation quality and generalization ability.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2020

Multilevel and multifidelity uncertainty quantification for cardiovascular hemodynamics

Standard approaches for uncertainty quantification in cardiovascular modeling pose challenges due to the large number of uncertain inputs and the significant computational cost of realistic three-dimensional simulations. Here, we propose an efficient uncertainty quantification framework utilizing a multilevel multifidelity Monte Carlo (MLMF) estimator to improve the accuracy of hemodynamic quantities of interest while maintaining reasonable computational cost. This is achieved by leveraging three cardiovascular model fidelities, each with varying spatial resolution to rigorously quantify the variability in hemodynamic outputs. We employ two low-fidelity models (zero- and one-dimensional) to construct several different estimators. Our goal is to investigate and compare the efficiency of estimators built from combinations of these two low-fidelity model alternatives and our high-fidelity three-dimensional models. We demonstrate this framework on healthy and diseased models of aortic and coronary anatomy, including uncertainties in material property and boundary condition parameters. Our goal is to demonstrate that for this application it is possible to accelerate the convergence of the estimators by utilizing a MLMF paradigm. Therefore, we compare our approach to single fidelity Monte Carlo estimators and to a multilevel Monte Carlo approach based only on three-dimensional simulations, but leveraging multiple spatial resolutions. We demonstrate significant, on the order of 10 to 100 times, reduction in total computational cost with the MLMF estimators. We also examine the differing properties of the MLMF estimators in healthy versus diseased models, as well as global versus local quantities of interest. As expected, global quantities such as outlet pressure and flow show larger reductions than local quantities, such as those relating to wall shear stress, as the latter rely more heavily on the highest fidelity model evaluations. Similarly, healthy models show larger reductions than diseased models. In all cases, our workflow coupling Dakota’s MLMF estimators with the SimVascular cardiovascular modeling framework makes uncertainty quantification feasible for constrained computational budgets.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2024

Finite elements for Matérn-type random fields: Uncertainty in computational mechanics and design optimization

This work highlights an approach for incorporating realistic uncertainties into scientific computing workflows based on finite elements, focusing on prevalent applications in computational mechanics and design optimization. We leverage Matérn-type Gaussian random fields (GRFs) generated using the SPDE method to model aleatoric uncertainties, including environmental influences, variating material properties, and geometric ambiguities. Our focus lies on delivering practical GRF realizations that accurately capture imperfections and variations and understanding how they impact the predictions of computational models as well as the shape and topology of optimized designs. Here we describe a numerical algorithm based on solving a generalized SPDE to sample GRFs on arbitrary meshed domains. The algorithm leverages established techniques and integrates seamlessly with the open-source finite element library MFEM and associated scientific computing workflows, like those found in industrial and national laboratory settings. Our solver scales efficiently for large-scale problems and supports various domain types, including surfaces and embedded manifolds. We showcase its versatility through biomechanics and topology optimization applications, emphasizing the potential to influence these domains. The flexibility and efficiency of SPDE-based GRF generation empowers us to run large-scale optimization problems on 2D and 3D domains, including finding optimized designs on embedded surfaces, and to generate design features and topologies beyond the reach of conventional techniques. Moreover, these capabilities allow us to model and quantify geometric uncertainties on reconstructed submanifolds, such as the interpolated surfaces of cerebral aneurysms provided by postprocessing CT scans. In addition to offering benefits in these specific domains, the proposed techniques transcend specific applications and generalize to arbitrary forward and backward problems in uncertainty quantification involving finite elements.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

Improving microstructures segmentation via pretraining with synthetic data

Image analysis of material microstructures through microscopy is an integral capability in the field of materials science. The topological and chemical information obtained through microscopy allow us to draw vital connections between material microstructures, properties, and processing. While scanning electron microscopy (SEM) is able to yield a considerable wealth of information interpretable by the intuition of experts, there has been considerable interest in using machine learning, convolutional neural networks (CNNs) in particular, for such image analysis task. Training CNNs for an image analysis task requires a large annotated dataset. However, in many materials science applications, obtaining a large annotated dataset is cost and labor intensive. In this work, we study the use of synthetic data to enlarge the available annotated experimental data of uranium oxide. We utilize a modified Potts model to simulate uranium oxide particles with morphologies similar to those observed experimentally. We then leverage an image-to-image translation model to synthesize the simulated particles as if they are acquired with SEM. Through this process, we obtain pairs of particle images and their corresponding SEM representations, which corresponds to pairs of annotations and images. Unlike previous works, we leverage synthetic data for pretraining a CNN model prior, and finetune that model further with experimental data. We experimentally demonstrate that using synthetic data as incremental learning process benefits the overall performance compared to training a model on combined synthetic and experimental data.

36 MATERIALS SCIENCE↗
DOE OSTI2025

Estimating soybean yields from high-temporal-resolution multi-source data using deep learning

Accurate and timely crop yield prediction is crucial for ensuring food security and maintaining stable agricultural markets. In recent years, there has been a surge in interest in leveraging high-temporal-resolution, multi-source data for effective crop growth monitoring and yield estimation. A notable challenge arises from the difficulty in capturing the intricate interactions between variables across different time steps within these high-temporal-resolution time series datasets. This complexity hinders the reliable extraction of yield information from voluminous and often noisy datasets, especially during periods of extreme weather events. Here, in this study, we propose an Attention and Graph Isomorphism Network-enhanced Bi-directional Long Short-Term Memory network (AGB-LSTM) for estimating county-level soybean yield in the United States. This model integrates a diverse set of remote sensing data, including Near-Infrared Reflectance of Vegetation (NIRv), Sun-Induced chlorophyll Fluorescence (SIF), and Gross Primary Productivity (GPP), along with environmental covariates. The AGB-LSTM effectively leverages information related to crop yield from high-temporal-resolution time series data (5-days), achieving an accuracy of R²= 0.67 and rRMSE = 14.46%. This approach significantly outperforms traditional machine learning methods such as Random Forest (RF) (R²= 0.52, rRMSE = 17.36%) and Bi-LSTM (R²= 0.58, rRMSE = 16.17%). Sensitivity experiments with different time steps and ranges demonstrated that our model could accurately and stably predict yields 1 to 2 months before harvest. Moreover, data with a finer temporal resolution consistently improved prediction performance, resulting in an approximately 20% increase in and an approximately 20% decrease in rRMSE compared to using monthly composites. We also evaluated the robustness of the model under extreme climate events and observed strong performance (R²= 0.50, rRMSE = 21.32%). Finally, yield mapping for major soybean-producing regions in North America in 2023 revealed spatial patterns that closely matched USDA yield reports. Our findings suggest that the AGB-LSTM model is a promising and effective method for estimating yield and has notable potential for global crop yield forecasting.

Deep learning↗
DOE OSTI2023

Investigation of 3D printed lightweight hybrid composites via theoretical modeling and machine learning

Hybrid composites combine two or more different fillers to achieve multifunctional or advanced material properties, such as lightweight and enhanced mechanical properties. The properties of the composites significantly depend on their microstructures, which can be tailored via advanced 3D printing processes. Understanding the process-structure-property relationships is critical to enable the design and engineering of novel hybrid composites for applications in aerospace, automotive, and protective coatings. Here, for this work, we develop 3D printable and lightweight hybrid composites and leverage the conventional design of experiments, a theoretical hybrid model, and an image-driven machine learning (ML) method to investigate their mechanical behaviors. The hybrid composites are formulated with elastomer matrix, microfillers, and thin-shell particles, enabling a significant degree of design freedom of microstructures with densities and mechanical properties varying up to 70% and 91%, respectively. Our statistical analysis indicates that the 3D printing path direction and the microfibers fraction are dominating process parameters with contribution percentages of 45.3% and 57.7% on the specific stiffness and strength, respectively. A hybrid mechanics model is developed based on a simple Weibull distribution function and classical single-filler models to effectively capture the variations in mechanical properties, however, it overestimates the values due to its statistical constraints and idealization of experimental uncertainty. The image-driven ML model leverages the microscale images directly without losing the structural details, shows more accurate predictions with experimental data, and has 48.6% lower root mean square error than the theoretical model.

3D printing↗
DOE OSTI2023

Matching diverse feedstocks to conversion processes for the future bioeconomy

A wide variety of wasted or underutilized organic feedstocks can be leveraged to build a sustainable bioeconomy, ranging from crop residues to food processor residues and municipal wastes. Leveraging these feedstocks is both high-risk and high-reward. Converting mixed, variable, and/or highly contaminated feedstocks can pose engineering and economic challenges. However, converting these materials to fuels and chemicals can divert waste from landfills, reduce fugitive methane emissions, and enable more responsible forest management to reduce the frequency and severity of wildfires. Historically, low-value components, including ash and lignin, are poised to become valuable coproducts capable of supplementing cement and valuable chemicals. Here, we evaluate the challenges and opportunities associated with converting a range of feedstocks to renewable fuels and chemicals.

09 BIOMASS FUELS↗
DOE OSTI2021

A Flexible and Generic Functional Mock-up Unit Based Threat Injection Framework for Grid-interactive Efficient Buildings: A Case Study in Modelica

Grid-interactive efficient buildings (GEBs) have been considered as an important asset to support the power grid reliability by utilizing the demand flexibility offered by GEBs. GEBs are enabled by advances in sensors and controls, and the communication between building equipment, whole buildings, and the grid. The integration of different building technologies and network-based communication system makes GEBs vulnerable to passive threats such as equipment failure and active threats such as cyber-attacks. Modeling and simulation is an effective way to evaluate the impact of threats on the system performance. This paper proposes a generic and flexible threat injection framework for commonly-used building energy simulators such as EnergyPlus and Modelica to support threat modeling and evaluation. This framework leverages functional mock-up unit (FMU) to develop a general modeling interface for threat injection and simulation. A numerical case study using Modelica as a building energy simulator is conducted to demonstrate the capability of the framework for supporting single/multiple-order threat modeling and simulation of a GEB. Four threats and their combinations are injected on a Modelica-based threat-free building energy and control system, including operating supply fan at its full speed, remotely cycling the chiller on and off, blocking the chiller from receiving the chilled water supply temperature setpoints, and hijacking the global zone air temperature setpoint. Simulation results show that the cyber-attack that leads to short-term signal blocking has small effects on the system operation due to the "self-healing" feature of the heating, ventilation, and air-conditioning (HVAC) interactive control system. The threat that takes control of resetting the global zone air temperature setpoints has the most adverse impact on the system energy use, peak power demand, thermal comfort and the provision of demand flexibility. The combination of four threats have aggregative effects on the system but the effects are less than the additive effects of the individual threat.

Fu, Yanyang↗
DOE OSTI2021

Evaluation of Phase Change Plaster/Paste Composites for Building Envelopes

Thermal energy storage (TES) materials increase a building’s thermal mass and provide the flexibility to shave and time-shift energy use. These materials are needed to enable grid-interactive efficient buildings, allowing building occupants to lower costs by leveraging free ambient heating/cooling as well as shifting consumption to cheaper electricity periods. Phase change materials (PCMs) that store heat through a solid-liquid/solid-solid transformation are of particular interest for TES because of their ability to store large amounts of energy across a constant temperature or very small temperature range. Furthermore, previous studies on PCM building envelopes evaluated products that may only be incorporated as part of new construction and are not easily applicable to existing, energy-intensive buildings. In this work, we leverage laboratory PCM characterization and whole-building energy simulation to study the energy and cost savings of PCM plaster/paste coatings, which can be applied post construction to both new builds and retrofits of existing buildings (about 40% of residential buildings in the United States fall into the retrofit category). We evaluated the energy impact of PCM plaster/paste composites for a single-family residential building in the United States, including layer design (location and thickness), composite properties (phase change temperature, energy density, and shape of the enthalpy curve), and U.S. climate zone. The enthalpy curve shape of PCM shows that phase change temperature and energy density are not the only two parameters that must be considered for PCM selection. Maximizing the energy density amid the effective TES zone is key to boosting energy savings. We also studied PCM price targets with the aim of achieving reasonable payback periods based on both block and time-of-use electricity rates. Moreover, we developed an empirical equation to provide a rough estimate of the energy savings potential of incorporating PCM without the need to perform detailed simulations. These results will help guide the selection and design of PCM plaster/paste composite elements for residential building envelopes.

ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATION↗
DOE OSTI2025

Designing robust energy policy packages under deep uncertainty: A multi-metric decision support framework

The complexity of transitioning to sustainable energy systems requires policy frameworks capable of balancing multiple objectives while addressing deep uncertainty. However, existing approaches often lack systematic methods to identify combinations of policy levers that remain effective across a wide range of uncertain futures. This paper presents a novel decision support framework that guides the selection of robust policy packages based on their performance across multiple objectives under uncertainty. Our method leverages a large ensemble of scenarios and applies scenario discovery techniques to identify influential policy levers. Here, we introduce new indicators to assess the robustness of policies by evaluating their ability to mitigate adverse outcomes across metrics. These indicators support an iterative process to build a robust policy package. Finally, we map the technological and energy pathways associated with the robust policy package by leveraging an energy system optimization model. We illustrate the application of this framework to the Spanish energy system, providing insights into how specific combinations of policy levers shape decarbonization pathways under uncertainty.

Decision-support method↗
DOE OSTI2023

Range-wide population assessments for subalpine fir indicate widespread disturbance-driven decline

Subalpine forests in western North America are threatened by rapid climate change, increased activity by endemic and exotic insects and diseases, and changing wildfire regimes. The interactive effects of these stressors have resulted in pronounced population declines in many subalpine tree species; however, a systematic assessment of the status and trends of subalpine forests is lacking. Subalpine fir (Abies lasiocarpa) is a widespread species across the western United States, with documented population declines in many parts of its distribution. Here we use subalpine fir as an initial leverage point to build a more complete understanding of subalpine forest baseline conditions and responses to environmental change. Specifically, we leverage the USDA Forest Service Forest Inventory and Analysis (FIA) database to (1) ask how subalpine fir populations are changing across the species’ distribution in the western US, (2) assess the drivers of recent subalpine fir population trends, and (3) explore whether those changes imply generalized species-wide and/or system-wide decline. We found that subalpine fir abundance and basal area are declining concurrently across ~ 62% of the species’ distribution, and increasing across ~ 19%. Range-wide, we estimated 25.02 ± 2.74 % subalpine fir mortality between 2000 and 2009 and 2010–2019 FIA inventory periods, with higher mortality concentrated in the eastern Oregon Cascades, central Idaho, and parts of southern Colorado. High regeneration density did not predict positive population trajectories, which were instead associated with higher rates of adult recruitment. While the importance of different mortality agents varied substantially between ecoregions, 83.4% of total range-wide mortality was related to fire or biological disturbance. Declining subalpine fir basal area coincided with declines in the basal area of other co-occurring tree species in 39% of subalpine forest area, and with increases in conspecific basal area in 22% of forest area. Fire disturbance was the single largest cause of subalpine fir mortality; however, even where subalpine fir fire mortality was high, mortality among other species was primarily caused by insects. In conclusion, our results suggest that subalpine fir declines across large portions of the western United States are driven by forest disturbance, and that declines in subalpine fir populations may be indicative of negative change in subalpine forest systems broadly.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2025

Co-optimization of fuel properties, combustion system geometry, and injection strategy for conventional diesel fuel

Here, studies have shown that fuel properties can impact an engine’s operation in several ways, including ignition delay, sooting tendency, mixture formation, and combustion temperature. In mixing-controlled compression ignition (MCCI) engines, the fuel system design and piston bowl geometry significantly affect combustion performance and emissions. Based on current information, it is difficult to draw conclusions about fuel property effects and sensitivities. The central fuel hypothesis approach used in the US Department of Energy Co-Optima program has worked well for spark ignition fuels: identifying critical fuel property ranges is sufficient to screen fuel blends that are expected to maximize efficiency and reduce pollutant emissions. However, for MCCI-relevant fuels, the information gained from past studies is not sufficient to build such a merit function or to allow for performing a similar screening of fuel blends. It is hypothesized that a co-optimization of a fuel’s physical and chemical properties, combustion system geometry, and injection strategy could leverage synergies between the effects of the fuel properties and geometries, resulting in improved performance over state-of-the-art. A machine learning–assisted unconstrained global optimization algorithm was used to explore a design space comprising 23 independent variables. The results show that physical property effects were minimal even for large variations in fuel properties, and the only interaction effect that was observed was the effect of varied fuel density parameters on fuel/air mixture formation. Nevertheless, these interactions were not sufficient in magnitude to significantly affect optimization results. Therefore, analysis of the results suggests that fuel physical properties cannot be leveraged in a co-optimization context to increase engine efficiency.

33 ADVANCED PROPULSION SYSTEMS↗
DOE OSTI2024

Efficient distributed continual learning for steering experiments in real-time

Deep learning has emerged as a powerful method for extracting valuable information from large volumes of data. However, when new training data arrives continuously (i.e., is not fully available from the beginning), incremental training suffers from catastrophic forgetting (i.e., new patterns are reinforced at the expense of previously acquired knowledge). Training from scratch each time new training data becomes available would result in extremely long training times and massive data accumulation. Rehearsal-based continual learning has shown promise for addressing the catastrophic forgetting challenge, but research to date has not addressed performance and scalability. To fill this gap, we propose an approach based on a distributed rehearsal buffer that efficiently complements data-parallel training on multiple GPUs to achieve high accuracy, short runtime, and scalability. It leverages a set of buffers (local to each GPU) and uses several asynchronous techniques for updating these local buffers in an embarrassingly parallel fashion, all while handling the communication overheads necessary to augment input minibatches using unbiased, global sampling. We further propose a generalization of rehearsal buffers to support both classification and generative learning tasks, as well as more advanced rehearsal strategies (notably Dark Experience Replay, leveraging knowledge distillation). We illustrate this approach with a real-life HPC streaming application from the domain of ptychographic image reconstruction. Furthermore, we run extensive experiments on up to 128 GPUs of the ThetaGPU supercomputer to compare our approach with baselines representative of training-from-scratch (the upper bound in terms of accuracy) and incremental training (the lower bound). Results show that rehearsal-based continual learning achieves a top-5 validation accuracy close to the upper bound, while simultaneously exhibiting a runtime close to the lower bound.

Asynchronous data management↗
DOE OSTI2024

Analyzing inference workloads for spatiotemporal modeling

Ensuring power grid resiliency, forecasting climate conditions, and optimization of transportation infrastructure are some of the many application areas where data is collected in both space and time. Spatiotemporal modeling is about modeling those patterns for forecasting future trends and carrying out critical decision-making by leveraging machine learning/deep learning. Once trained offline, field deployment of trained models for near real-time inference could be challenging because performance can vary significantly depending on the environment, available compute resources and tolerance to ambiguity in results. Users deploying spatiotemporal models for solving complex problems can benefit from analytical studies considering a plethora of system adaptations to understand the associated performance-quality trade-offs. To facilitate the co-design of next-generation hardware architectures for field deployment of trained models, it is critical to characterize the workloads of these deep learning (DL) applications during inference and assess their computational patterns at different levels of the execution stack. In this paper, we develop several variants of deep learning applications that use spatiotemporal data from dynamical systems. We study the associated computational patterns for inference workloads at different levels, considering relevant models (Long short-term Memory, Convolutional Neural Network and Spatio-Temporal Graph Convolution Network), DL frameworks (Tensorflow and PyTorch), precision (FP16, FP32, AMP, INT16 and INT8), inference runtime (ONNX and AI Template), post-training quantization (TensorRT) and platforms (Nvidia DGX A100 and Sambanova SN10 RDU). Overall, our findings indicate that although there is potential in mixed-precision models and post-training quantization for spatiotemporal modeling, extracting efficiency from contemporary GPU systems might be challenging. Instead, co-designing custom accelerators by leveraging optimized High Level Synthesis frameworks (such as SODA High-Level Synthesizer for customized FPGA/ASIC targets) can make workload-specific adjustments to enhance the efficiency.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

Self-supervised and multi-fidelity learning for extended predictive soil spectroscopy

Infrared spectroscopy is a cost-effective, non-destructive, and environmentally benign technology that is increasingly recognized as an important solution for meeting the global demand for soil data. While both near-infrared (NIR) and mid-infrared (MIR) diffuse reflectance spectroscopy enable rapid estimation of soil properties, they present a significant trade-off: NIR offers superior scalability and lower operational costs, whereas MIR provides higher analytical fidelity by capturing fundamental molecular vibrations. In this study, we propose a self-supervised, multi-fidelity learning framework designed to bridge this gap. Our approach leverages large-scale MIR spectral libraries to learn a compact, transferable latent representation, into which NIR spectra are subsequently aligned for downstream prediction. The workflow consists of pretraining a latent model on a large MIR library, adapting the representation using a smaller paired NIR–MIR dataset, and evaluating generalization on an independent external test set. Across a range of chemical and physical soil properties, we found that MIR-derived embeddings improved prediction accuracy relative to baseline models that used raw MIR inputs. Predictions derived from the spectrum conversion (NIR to MIR) task did not match the performance of the original MIR spectra but were similar or superior to predictive performance of NIR-only models, suggesting the unified spectral latent space can effectively leverage the larger and more diverse MIR dataset for prediction of soil properties not well represented in current NIR libraries.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2021

Deep Learning Explicit Differentiable Predictive Control Laws for Buildings

We present a differentiable predictive control (DPC) methodology for learning constrained control laws for unknown nonlinear systems. DPC poses an approximate solution to multiparametric programming problems emerging from explicit nonlinear model predictive control (MPC). Contrary to approximate MPC, DPC does not require supervision by an expert controller. Instead, a system dynamics model is learned from a small dataset of recorded observations of the perturbed system's dynamics and the control law is optimized offline by interaction with the learned system model. The DPC method is based on two sequential steps, i) system identification using a constrained neural state-space model, and ii) optimization of an explicit control law parametrized by another neural network in closed-loop simulation with the identified neural state-space model. The combination of a differentiable closed-loop system and penalty methods for constraint handling of system outputs and inputs allows us to optimize the control law's parameters directly by backpropagating economic MPC loss through the learned system model. By incorporating domain knowledge and leveraging established techniques from optimal control, our method leverages deep neural networks as nonlinear function approximators for system identification and control while avoiding concomitant costs of intractably large datasets, and computationally expensive over-parametrized models. The scalability, data efficiency, and constrained optimal control capability of the proposed DPC method are demonstrated in simulation using a multi-zone building emulator.

Drgona, Jan↗