Search NASA⌕ Search

SEARCH · Search NASA

Results for “network optimization”

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 289 records · Page 16
All sourcesNASA NTRSDOE OSTI
DOE OSTI2018

Deep Learning Algorithm for Auto-Delineation of High-Risk Oropharyngeal Clinical Target Volumes With Built-In Dice Similarity Coefficient Parameter Optimization Function

Automating and standardizing the contouring of clinical target volumes (CTVs) can reduce interphysician variability, which is one of the largest sources of uncertainty in head and neck radiation therapy. In addition to using uniform margin expansions to auto-delineate high-risk CTVs, very little work has been performed to provide patient- and disease-specific high-risk CTVs. The aim of the present study was to develop a deep neural network for the auto-delineation of high-risk CTVs.

62 RADIOLOGY AND NUCLEAR MEDICINE↗
DOE OSTI2025

Understanding and Estimating Error Propagation in Neural Networks for Scientific Data Analysis

Neural networks are increasingly integrated into scientific discovery, where input data reduction and model quantization play a key role in accelerating inference. However, understanding and mitigating the impact of these techniques on output error is critical for ensuring reliable results, particularly in tasks demanding high numerical precision. This paper introduces a comprehensive framework for optimizing neural network inference in scientific computing by combining data reduction and weight quantization while maintaining error-controlled outcomes. We develop theoretical analyses to bound error propagation under these reductions and propose a framework that balances computational performance with error constraints. Evaluation on real-world learning-based combustion simulations and satellite image classification demonstrates that our derived error bounds accurately predict observed errors while enabling significant computational speedup under our framework. This work highlights the potential for further leveraging advancements in modern lossy compression algorithms and hardware accelerators that support lower-precision formats.

He, Weiming [New Jersey Institute of Technology]↗
DOE OSTI2020

Learning to Optimize Variational Quantum Circuits to Solve Combinatorial Problems

Quantum computing is a computational paradigm with the potential to outperform classical methods for a variety of problems. Proposed recently, the Quantum Approximate Optimization Algorithm (QAOA) is considered as one of the leading candidates for demonstrating quantum advantage in the near term. QAOA is a variational hybrid quantum-classical algorithm for approximately solving combinatorial optimization problems. The quality of the solution obtained by QAOA for a given problem instance depends on the performance of the classical optimizer used to optimize the variational parameters. In this paper, we formulate the problem of finding optimal QAOA parameters as a learning task in which the knowledge gained from solving training instances can be leveraged to find high-quality solutions for unseen test instances. To this end, we develop two machine-learning-based approaches. Our first approach adopts a reinforcement learning (RL) framework to learn a policy network to optimize QAOA circuits. Our second approach adopts a kernel density estimation (KDE) technique to learn a generative model of optimal QAOA parameters. In both approaches, the training procedure is performed on small-sized problem instances that can be simulated on a classical computer; yet the learned RL policy and the generative model can be used to efficiently solve larger problems. Furthermore, extensive simulations using the IBM Qiskit Aer quantum circuit simulator demonstrate that our proposed RL- and KDE-based approaches reduce the optimality gap by factors up to 30.15 when compared with other commonly used off-the-shelf optimizers.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2019

LP-BNN: Ultra-low-Latency BNN Inference with Layer Parallelism

High inference latency seriously limits the deployment of DNNs in real-time domains such as autonomous driving, robotic control, and many others. To address this emerging challenge, researchers have proposed approximate DNNs with reduced precision, e.g., Binarized Neural Networks (BNNs). While BNNs can be built to have little loss in accuracy, latency reduction still has much room for improvement. In this paper, we propose a single-FPGA-based BNN accelerator that achieves microsecond-level ultra-low-latency inference of ImageNet, LP- BNN. We obtain this performance via several design optimizations. First, we optimize the network structure by removing Batch Normalization (BN) functions which leads to significant latency in BNNs without any loss on accuracy. Second, we propose a parameterized architecture which is based on layer parallelism and supports nearly perfect load balancing for various types of BNNs. Third, we fuse all the convolution layers and the first fully connected layer. We process them in parallel through fine- grained inter-layer pipelining. With our proposed accelerator, the inference of binarized AlexNet, VGGNet, and ResNet are completed within 21.5µs, 355µs, and 67.8µs respectively, with no loss in accuracy as compared with other BNN implementations.

Geng, Tong↗
DOE OSTI2025

Energy-efficient multimodal mobility networks in transportation digital twins: Strategies and optimization

The study proposes a comprehensive Transportation Mobility (TransitMo) framework covering conceptual design, model formulation, optimization, simulation, and impact analysis of the transportation mobility system. TransitMo is composed of a transportation digital twin developed in Simulation of Urban MObility (SUMO) and an Intelligent Traffic Management and Control Center (ITMCC) that identifies the best ways to improve the movement of people within urban areas using various modes of transportation. This study encompasses advanced modeling techniques, algorithms, and strategic testing to optimize energy efficiency and mobility in a multimodal shared mobility network. TransitMo’s practical applications are exemplified through a city-scaled simulation network in Chattanooga, TN, employing demographic data to analyze historical traffic patterns and forecast future demands. Central to this methodology are three models: the User Preference Model (UP), the Energy Consumption Model (EC), and the System Optimization Model (SO). These models work in concert to iteratively devise the optimal travel incentives and minimize the total system cost in a real-time manner. In conclusion, test results verified that the proposed adaptive incentive program and optimized bus scheduling can improve network performance by increasing public transit ridership.

42 ENGINEERING↗
DOE OSTI2022

Wabash CarbonSAFE Roadmap for Network and Storage Deployment: Subtask 10.3 (Technical Report)

Wabash Valley Resources is incorporating carbon capture and storage (CCS) technologies into its hydrogen production plant in Terre Haute, Indiana, at an expected annual capture rate that will make Wabash Valley Resources one of the largest carbon capture facilities in the world. In addition to storing its own captured CO 2 , the geographic location and subsurface geology at Wabash Valley Resources suggest it may be an advantageous site for a carbon storage hub. This report analyzes the potential of nearby sources to transport their captured CO 2 to Wabash Valley Resources through an optimal pipeline network.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗
DOE OSTI2025

Assessing the limitations of commercial sensors and models for supporting marine carbon dioxide removal monitoring: a case study

Several unknowns remain surrounding marine Carbon Dioxide Removal (mCDR) monitoring, reporting, and verification (MRV) practices and capabilities. Current in-situ sensor technology is limited (primarily pH and pCO 2 ), requiring calculations and assumptions to estimate changes in carbonate chemistry parameters, including total alkalinity (TA). Considering that cost, energy consumption, and accuracy of commercial sensors can vary by orders of magnitude, understanding how well existing sensors perform in an mCDR context is important for this emerging community. Likewise, documenting sensor limitations and how relatively simple models can optimize sensor deployments will improve MRV efforts and support protocol development. Here we (1) compare performance a variety of commercially available sensors in a blind mesocosm experiment simulating ocean alkalinity enhancement (OAE), and how sensor performance impacted carbonate chemistry estimates; (2) evaluate if sensors can distinguish the OAE signal from natural variability during a small scale OAE field test in Sequim Bay, WA, USA, and (3) use an idealized ocean biogeochemistry model to explore optimal sensor network design based on (1) and (2). Our mesocosm results indicate that correctly constraining pH uncertainty will be critical for accurate TA estimates with current sensor technology compared to the less impactful variation caused by uncertainty in pCO 2 (pH data that are presented throughout are reported on the total scale (pH T ) unless otherwise noted). Our pilot field test demonstrated that sensors were capable of distinguishing mCDR signatures from natural variability under optimal real-world conditions. Idealized modeling simulations of the field test showed that a range of sparse and dense (3 to 100) sensors sampling areas of detectable increases will underestimate the net change in surface pH by at least 35–55%, at both realistic and highly elevated alkalinity input levels. We also highlight the limitations of current sensing technology for MRV, and the importance of ocean biogeochemistry models as critical tools for predicting when and where mCDR signals will be detectable using available sensors. Overall, our findings suggest that commercially available pCO 2 sensors and some pH sensors will form an important backbone for mCDR MRV tasks, though complete MRV characterization will require these data to be used in combination with other tools.

OAE↗
DOE OSTI2024

Optimizing Carbon Capture, Transport, and Storage: Overcoming Challenges with Machine Learning and Cost-Benefit Analysis

Carbon Capture, Utilization, and Storage (CCUS) is a critical strategy for reducing CO₂ emissions and mitigating climate change. However, its widespread deployment faces numerous challenges across the capture, transport, and storage phases. These challenges include the technical complexity of predicting subsurface behaviors during CO₂ injection, ensuring long-term storage integrity, optimizing transportation networks, and balancing the economic and environmental trade-offs. Addressing these issues requires an integrated approach combining advanced subsurface modeling with system-level analyses to assess costs, risks, and benefits. This presentation provides an overview of studies conducted by the National Energy Technology Laboratory (NETL) to tackle these challenges. NETL’s efforts encompass cutting-edge research in subsurface fluid behavior machine learning predictions, alongside the development of innovative tools for system optimization and economic evaluation. By bridging technical expertise and strategic analysis, NETL aims to advance the deployment of CCUS technologies to support global decarbonization efforts. Presented at the Carnegie Mellon University CEE IESS Student Seminar October 4, 2024.

Shih, Chung Yan↗
DOE OSTI2020

hp -VPINNs: Variational physics-informed neural networks with domain decomposition

We formulate a general framework for hp-variational physics-informed neural networks (hp-VPINNs) based on the nonlinear approximation of shallow and deep neural networks and hp-refinement via domain decomposition and projection onto the space of high-order polynomials. The trial space is the space of neural network, which is defined globally over the entire computational domain, while the test space contains piecewise polynomials. Specifically in this study, the hp-refinement corresponds to a global approximation with a local learning algorithm that can efficiently localize the network parameter optimization. Here, we demonstrate the advantages of hp-VPINNs in both accuracy and training cost for several numerical examples of function approximation and in solving differential equations.

42 ENGINEERING↗
DOE OSTI2022

Anisotropic Thermally Conductive Composites Enabled by Acoustophoresis and Stereolithography

Opportunities to improve thermal management in electronic devices are currently hindered by processing constraints that limit thermal conductivity in polymer-matrix composites. Active patterning of filler particles is a promising route to improve conductivity while retaining processability by improving particle contact density and directing heat along optimized pathways. Here, we employ acoustic patterning to align and compact filler particles into stripes during stereolithographic 3D printing. This approach produces polymer-based composite materials with highly efficient embedded heat transport pathways which reach 95 vol% particle utilization (relative to the parallel conduction upper limit). These composites exhibit anisotropic thermal conductivity up to 300% higher than unpatterned composites, with in-plane anisotropy ratios of up to 350%. Combining this high conductivity with 3D printing enables materials with engineered heat networks that optimize transport from hot spots to heatsinks while maintaining low viscosity for fast particle patterning and for infiltration around electronic components. Finally, numerical simulations of acoustic assembly of particles with varied geometry, when compared to experimentally characterized particle packing, illuminate pathways for further improving conductivity by optimizing particle geometry for alignment and stacking of particles with maximum contact surface area.

36 MATERIALS SCIENCE↗
DOE OSTI2020

A Topology Optimization Framework to Facilitate Adoption of Advanced District Thermal Energy Systems

Advanced district thermal energy systems, which circulate water at temperatures near ambient conditions, and facilitate the utilization of waste heat and renewable thermal sources, can lower the carbon-intensity of urban districts, advancing the U.N. Sustainable Development Goals. Optimization of the network topology — the selection of the best subset of buildings and the best network to connect them, to minimize life cycle cost — can increase adoption of these system in appropriate applications. The potential "solution space" of the topology optimization problem grows factorially with the number of buildings in the district, motivating the consideration of a design heuristic. In this study, a heuristic for the network selection was evaluated with an exhaustive search, for a prototypical four-building district. For the prototypical district considered, the heuristic was effective in selecting an optimal network topology. Additionally, it was found that, in this case, the selection of the subset of buildings was more influential on the life cycle cost than the selection of the network topology. This work is part of a larger effort to develop a topology optimization framework for district thermal energy systems, which is anticipated to address barriers to adoption of ambient-temperature systems.

28 EE - Advanced Manufacturing Office (EE-5A)↗
DOE OSTI2023

NSGA-PINN: A Multi-Objective Optimization Method for Physics-Informed Neural Network Training

This paper presents NSGA-PINN, a multi-objective optimization framework for the effective training of physics-informed neural networks (PINNs). The proposed framework uses the non-dominated sorting genetic algorithm (NSGA-II) to enable traditional stochastic gradient optimization algorithms (e.g., ADAM) to escape local minima effectively. Additionally, the NSGA-II algorithm enables satisfying the initial and boundary conditions encoded into the loss function during physics-informed training precisely. We demonstrate the effectiveness of our framework by applying NSGA-PINN to several ordinary and partial differential equation problems. In particular, we show that the proposed framework can handle challenging inverse problems with noisy data.

Lu, Binghang (ORCID:0009000160016632)↗
DOE OSTI2021

High-Performance All-Polymer Solar Cells and Photodetectors Enabled by a High-Mobility n-Type Polymer and Optimized Bulk-Heterojunction Morphology

All-polymer solar cells (all-PSCs) and all-polymer photodetectors (all-PPDs) are promising for application in flexible electronics, but their performance is greatly limited by the lack of polymer acceptors and the difficulties in morphology control. Herein, we report the use of a high-mobility n-type polymer semiconductor, PNDI-DTBT, for fabricating high-performance all-PSCs and all-PPDs. By pairing PNDI-DTBT with a matched polymer donor, a prominent power conversion efficiency of 8.5% was obtained in all-PSCs, demonstrating the great potential of high-mobility n-type polymers with a conventional donor–acceptor skeleton for photovoltaic application. Moreover, all-PPDs with a low dark current density of 1.32 × 10 –8 A cm –2 at –0.1 V bias and a high specific detectivity of 4.77 × 10 12 Jones were achieved, which belong to the best results of organic photodetectors. Morphology investigations revealed that the formation of a bicontinuous interpenetrating network with optimal phase separation scale, high domain purity, and preferential vertical composition distribution in the bulk-heterojunction active layer contribute to the remarkable device performance. Furthermore, these results suggest that combining high-mobility n-type polymers and morphology optimization is fruitful to achieving high-performance all-PSCs and all-PPDs.

36 MATERIALS SCIENCE↗
DOE OSTI2024

Optimal control of the electron temperature profile in DIII-D using machine learning surrogate models

The viability of the tokamak as a potential fusion reactor depends on the ability to keep the plasma in a stable regime while achieving temperatures, densities, and confinement times that are as high as possible. Tokamak scenario development attempts to find plasma regimes that achieve all of these conditions and are accessible with a given set of hardware constraints. This requires the ability to control plasma properties such as the normalized beta, the internal inductance, safety factor, rotation, etc. One property that has received less attention than some of the others, but is no less critical to achieving high performance, is the electron temperature (T e ) profile. In this work, Linear Quadratic Integral (LQI) control is used to develop a controller for the electron temperature profile in DIII-D. The controller is based on a linearized model derived from the transport equation that describes the evolution of the electron temperature, and includes contributions from the neural network surrogate models NubeamNet and MMMnet. Furthermore, the controller is tested in simulation using COTSIM, and is proven capable of tracking a target T e profile.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2025

Advanced Transmission Technologies – GETs and HPCs Session 2: Advanced Power Flow Control and Transmission Topology Optimization

The INL TADA GETs Cohort Session 2, held on November 7, 2025, conducted in collaboration with ScottMadden, focused on two core Advanced Transmission Technologies (ATTs): Advanced Power Flow Control (APFC) and Transmission Topology Optimization (TTO). These technologies are pivotal in enhancing grid flexibility, reliability, and cybersecurity resilience. APFC, particularly through modular FACTS devices like Modular Static Synchronous Series Compensators (M-SSSCs), enables dynamic voltage injection to reroute power flows. The session highlighted the deployment benefits of APFC, such as rapid installation, minimal civil works, and re-deployability. Regulatory drivers like FERC Order 2023 mandate the inclusion of Grid-Enhancing Technologies (GETs) in interconnection studies. Case studies from Central Hudson, CAISO, and National Grid (UK) demonstrated APFC’s effectiveness in congestion relief and cost savings. The session also addressed cybersecurity concerns, including firmware vulnerabilities, SCADA integration risks, and supply chain dependencies. Participants engaged in interactive exercises to rank cybersecurity and supply chain risks, emphasizing the need for robust digital assurance strategies. TTO involves software-based reconfiguration of transmission networks to optimize power flow without new infrastructure. The session showcased its operational value, with examples from SPP, PJM, and MISO showing significant congestion cost reductions. Cybersecurity vulnerabilities were discussed, particularly in API security and software supply chains, referencing incidents like SolarWinds and attacks on Danish utilities. Digital assurance exercises explored worst-case scenarios, attack paths, and mitigation responsibilities between vendors and utilities. Reliability challenges such as algorithm stability, vendor dependency, and operator trust were also examined. Cross-cutting themes emphasized the importance of digital assurance tools, including Software Bills of Materials (SBOMs) and hardware-in-loop testing. Human performance, training, and operational confidence were identified as critical enablers of technology adoption. The session concluded with a preview of Session 3, which will focus on High Performance Conductors (HPCs) and risk-based cybersecurity tools. Session 2 of 3.

24 - POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2022

Deep learning to estimate permeability using geophysical data

Time-lapse electrical resistivity tomography (ERT) is a popular geophysical method to estimate three-dimensional (3D) permeability fields from electrical potential difference measurements. Traditional inversion and data assimilation methods are used to ingest this ERT data into hydrogeophysical models to estimate permeability. Due to ill-posedness and the curse of dimensionality, existing inversion strategies provide poor estimates and low resolution of the 3D permeability field. Recent advances in deep learning provide us with powerful algorithms to overcome this challenge. This paper presents a deep learning (DL) framework to estimate the 3D subsurface permeability from time-lapse ERT data. To test the feasibility of the proposed framework, we train DL-enabled inverse models on simulation data. Each measurement in both synthetic and field data is standardized by removing the mean and scaling the time-series to unit variance. This pre-processing step is necessary to bring simulation data closer to field observations. Subsurface process models based on hydrogeophysics are used to generate this synthetic data. Training performed on limited simulation data resulted in the DL model over-fitting. An advanced data augmentation based on mixup is implemented to generate additional training samples to overcome this issue. This mixup technique creates weakly labeled (low-fidelity) samples from strongly labeled (high-fidelity) data. The weakly labeled training data is then used to develop DL-enabled inverse models and reduce over-fitting. As both time-lapse ERT (1133048 features/realization) and 3D permeability (585453 features/realization) data samples are from a high-dimensional space, principal component analysis (PCA) is employed to reduce dimensionality. Encoded ERT and encoded permeability are generated using the trained PCA estimators. A deep neural network is then trained to map the encoded ERT to encoded permeability. This mixup training and unsupervised learning allowed us to build a fast and reasonably accurate DL-based inverse model under limited simulation data. Results show that proposed weak supervised learning can capture salient spatial features in the 3D permeability field. Quantitatively, the average mean squared error (in terms of the natural log) on the strongly labeled training, validation, and test datasets is less than 0.5. The R 2 -score (global metric) is greater than 0.75, and the percent error in each cell (local metric) is less than 10%. Finally, an added benefit in terms of computational cost is that the proposed DL-based inverse model is at least O(10 4 ) times faster than running a forward model once it is trained. Data generation, DL model training, and hyperparameter tuning to identify optimal neural network architectures utilized high-performance computing resources while the DL inference is performed on a standard laptop. Approximately, O(10 5 ) processor hours are used for generating data and DL tuning and training. We acknowledge that the data generation and DL model development are expensive. But once a DL model is trained, it can be re-used for inversion rapidly for the given system, with set physics and domain. Note that traditional inversion may require multiple forward model simulations (e.g., in the order of 10 to 1000), which are very expensive. This computational savings ≈ O(10 5 ) – O(10 7 )) makes the proposed DL-based inverse model attractive for subsurface imaging and real-time ERT monitoring applications due to fast and yet reasonably accurate estimations of permeability field.

58 GEOSCIENCES↗
DOE OSTI2024

Learning Constrained Parametric Differentiable Predictive Control Policies With Guarantees

We present differentiable predictive control (DPC), a method for offline learning of constrained neural control policies for nonlinear dynamical systems with performance guarantees. We show that the sensitivities of the parametric optimal control problem can be used to obtain direct policy gradients. Specifically, we employ automatic differentiation (AD) to efficiently compute the sensitivities of the model predictive control (MPC) objective function and constraints penalties. To guarantee safety upon deployment, we derive probabilistic guarantees on closed-loop stability and constraint satisfaction based on indicator functions and Hoeffding’s inequality. We empirically demonstrate that the proposed method can learn neural control policies for various parametric optimal control tasks. In particular, we show that the proposed DPC method can stabilize systems with unstable dynamics, track time-varying references, and satisfy nonlinear state and input constraints. Our DPC method has practical time savings compared to alternative approaches for fast and memory-efficient controller design. Specifically, DPC does not depend on a supervisory controller as opposed to approximate MPC based on imitation learning. We demonstrate that, without losing performance, DPC is scalable with greatly reduced demands on memory and computation compared to implicit and explicit MPC while being more sample efficient than model-free reinforcement learning (RL) algorithms.

97 MATHEMATICS AND COMPUTING↗