Search NASA⌕ Search

SEARCH · Search NASA

Results for “Distributed 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 73 records · Page 4

DaYu: Optimizing Distributed Scientific Workflows by Decoding Dataflow Semantics and Dynamics

The combination of ever-growing scientific datasets and distributed workflow complexity creates I/O performance bottlenecks due to data volume, velocity, and variety. Although the increasing use of descriptive data formats (e.g., HDF5, netCDF) helps organize these datasets, it also creates obscure bottlenecks due to the need to translate high level operations into file addresses and then into low-level I/O operations. To address this challenge, we introduce DaYu, a method and toolset for analyzing (a) semantic relationships between logical datasets and file addresses, (b) how dataset operations translate into I/O, and (c) the combination across entire workflows. DaYu's analysis and visualization enables identification of critical bottlenecks and reasoning about remediation. We describe our methodology and propose optimization guidelines. Evaluation on scientific workflows demonstrates up to 3.7x performance improvements in I/O time for obscure bottlenecks. The time and storage overhead for DaYu's time-ordered data is typically under 0.2% of runtime and 0.25% of data volume, respectively.

Tang, Meng↗

Multigrid solution of a distributed optimal control problem constrained by the Stokes equations

We construct multigrid preconditioners to accelerate the solution process of a linear-quadratic optimal control problem constrained by the Stokes system. The first order optimality conditions of the control problem form a linear system (the KKT system) connecting the state, adjoint, and control variables. Our approach is to eliminate the state and adjoint variables by essentially solving two Stokes systems, and to construct efficient multigrid preconditioners for the Schur-complement of the block associated with the state and adjoint variables. These multigrid preconditioners are shown to be of optimal order with respect to the convergence properties of the discrete methods used to solve the Stokes system. In particular, the number of conjugate gradient iterations is shown to decrease as the resolution increases, a feature shared by similar multigrid preconditioners for elliptic constrained optimal control problems.

97 MATHEMATICS AND COMPUTING↗

Spectral Analytics of Solar Photovoltaic Power Output for Optimal Distributed Energy Resource Utilization

This paper examines the spectral analytics of solar photovoltaic (PV) power in order to understand its frequency content. This information is then used to illustrate that the different frequency components of PV generation can be consumed locally by controlling local distributed energy resources in residential/commercial buildings. The solar PV generation signal is divided into three components: high frequency (second-level), medium frequency (minute-level), and low-frequency that correlate with the solar activity. One year of solar PV power data is analyzed with 1-second resolution to find the ideal bounds for the different frequency bands. Results show that by employing intelligent control of Heating, Ventilation and Air-Conditioning (HVAC) systems, HVAC loads are able to accommodate the low and medium frequency components of the PV generation while local energy storage systems can be used to offset the high frequency components. This demonstrates the ability to spatially-local consumption of PV generation using controllable loads so as to minimize impact on the grid, reduce size of storage devices, and increase solar PV penetration levels.

Olama, Mohammed M.↗

Least-cost Optimal Distribution Grid Expansion (LODGE) v1

LODGE finds the least-cost portfolio of traditional distribution system upgrades to integrate new loads and behind-the-meter DERs in combination with advanced non-wire alternative (NWA) planning solutions, such as utility-owned storage and distributed PV. Working with a set of least-cost solutions per feeder/substation allows us to benchmark and compare techno-economic performance of different traditional and NWA planning solutions. An important aspect of the LODGE model is that it includes power flow operation constraints. This means that the least-cost planning solutions delivered by LODGE are technically feasible. Available projects in LODGE portfolio include: 1) feeder reconductoring; 2) transformers upgrades; 3) non-wire alternatives, such as strategic siting and sizing of storage and distributed PV; 4) voltage regulators.

Heleno, Miguel↗

Optimizing Distributed Energy Storage Sizing in Puerto Rico: Leveraging Increased Distributed Generation for Enhanced Resilience

Distributed solar photovoltaic (PV) and battery energy storage systems (BESS) adoption has increased rapidly in Puerto Rico since Hurricane Maria in 2017 as customers seek local resilient energy solutions. To investigate how to best leverage these many distributed PV and BESS systems, in this paper, we introduce a modeling framework that can be used by entities planning resilience investments to provide insights about the sizing and operational needs to achieve desired resilience benefits.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Optimizing Distributed Training on Frontier for Large Language Models

Large language models (LLMs) have demonstrated remarkable success as foundational models, benefiting various downstream applications through fine-tuning. Loss scaling studies have demonstrated the superior performance of larger LLMs compared to their smaller counterparts. Nevertheless, training LLMs with billions of parameters poses significant challenges and requires considerable computational resources. For example, training a one trillion parameter GPT-style model on 20 trillion tokens requires a staggering 120 million exaflops. This research explores efficient distributed training strategies to extract this computation from Frontier, the world's first exascale supercomputer. We enable and investigate various model and data parallel training techniques, such as tensor parallelism, pipeline parallelism, and sharded data parallelism, to facilitate training a trillion-parameter model on Frontier. We empirically assess these techniques and their associated parameters to determine their impact on memory footprint, communication latency, and GPU's computational efficiency. We analyze the complex interplay among these techniques and find a strategy to combine them to achieve high throughput through hyperparameter tuning. We have identified efficient strategies for training large LLMs of varying sizes through empirical analysis and hyperparameter tuning. For 22 Billion, 175 Billion, and 1 Trillion parameters, we achieved GPU throughputs of 38.38%, 36.14%, and 31.96%, respectively. For the training of the 175 Billion parameter model and the 1 Trillion parameter model, we achieved 100% weak scaling efficiency on 1024 and 3072 Mi250X GPUs, respectively. We also achieved strong scaling efficiencies of 89% and 87% for these two models. We trained these models only tens of iterations instead of training till completion.

Yin, Junqi↗

Spatial and temporal temperature distribution optimization for a geostationary antenna

The Geostationary Microwave Precipitation Radiometer antenna is considered and a thermal design analysis is performed to determine a design that would minimize on-orbit antenna temporal and spatial temperature gradients. The final design is based on an optically opaque radome which covered the antenna. The average orbital antenna temperature is found to be 9 C with maximum temporal and spatial variations of 34 C and 1 C, respectively. An independent thermal distortion analysis showed that this temporal variation would give an antenna figure error of 14 microns.

Tsuyuki, G.↗

HyperSpace: Distributed Bayesian Hyperparameter Optimization

As machine learning models continue to increase in complexity, so does the potential number of free model parameters commonly known as hyperparameters. While there has been considerable progress toward finding optimal configurations of these hyperparameters, many optimization procedures are treated as black boxes. We believe optimization methods should not only return a set of optimized hyperparameters, but also give insight into the effects of model hyperparameter settings. To this end, we present HyperSpace, a parallel implementation of Bayesian sequential model-based optimization. HyperSpace leverages high performance computing (HPC) resources to better understand unknown, potentially non-convex hyperparameter search spaces. We show that it is possible to learn the dependencies between model hyperparameters through the optimization process. By partitioning large search spaces and running many optimization procedures in parallel, we also show that it is possible to discover families of good hyperparameter settings over a variety of models including unsupervised clustering, regression, and classification tasks.

Computer Science↗

Optimal PV Inverter Control in Distribution Systems via Data-Driven Distributionally Robust Optimization

Distribution systems with high penetration of uncertain solar generation call for advanced control strategies of photovoltaics (PVs) inverters. This paper proposes a data-driven distributionally robust optimization (DDDRO) approach to optimally controlling the PV inverters to improve the system operation performance under solar power uncertainties. In the proposed DDDRO approach, a Wasserstein ball-based method is proposed to construct the distributional ambiguity set to model the uncertainties of PV generation through partial observations of historical data without knowing exact probability distributions. We further reformulate the computationally intractable DDDRO model to a mixed integer second order cone programming (MISOCP) problem. The effectiveness and out-of-sample performance of the proposed approach have been demonstrated on a modified IEEE 33-node system. We conduct a comparative study to compare the proposed method with traditional chance constrained programming (CCP). It shows that the proposed DDDRO approach can provide a less conservative yet robust solution to minimize the worse-case expectation of the total network loss while maintaining nodal voltages in a secure range.

Xue, Yaosuo↗