Search NASA⌕ Search

SEARCH · Search NASA

Results for “distributed algorithm”

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 487 records · Page 27

ScaWL: Scaling k-WL (Weisfeiler-Lehman) Algorithms in Memory and Performance on Shared and Distributed-Memory Systems

The k-dimensional Weisfeiler-Lehman (k-WL) algorithm—developed as an efficient heuristic for testing if two graphs are isomorphic—is a fundamental kernel for node embedding in the emerging field of graph neural networks. Unfortunately, the k-WL algorithm has exponential storage requirements, limiting the size of graphs that can be handled. This work presents a novel k-WL scheme with a storage requirement orders of magnitude lower while maintaining the same accuracy as the original k-WL algorithm. Due to the reduced storage requirement, our scheme allows for processing much bigger graphs than previously possible on a single compute node. For even bigger graphs, we provide the first distributed-memory implementation. Our k-WL scheme also has significantly reduced communication volume and offers high scalability. Our experimental results demonstrate that our approach is significantly faster and has superior scalability compared to five other implementations employing state-of-the-art techniques.

algorithims↗

Empirical comparison of heuristic load distribution in point-to-point multicomputer networks

The study compared several load placement algorithms using instrumented programs and synthetic program models. Salient characteristics of these program traces (total computation time, total number of messages sent, and average message time) span two orders of magnitude. Load distribution algorithms determine the initial placement for processes, a precursor to the more general problem of load redistribution. It is found that desirable workload distribution strategies will place new processes globally, rather than locally, to spread processes rapidly, but that local information should be used to refine global placement.

Grunwald, Dirk C.↗

Tunable thermal conductivity via domain structure engineering in ferroelectric thin films: A phase-field simulation

Effective thermal conductivity as a function of domain structure is studied by solving the heat conduction equation using a spectral iterative perturbation algorithm in materials with inhomogeneous thermal conductivity distribution. Using this proposed algorithm, the experimentally measured effective thermal conductivities of domain-engineered {001} p- BiFeO 3 thin films are quantitatively reproduced. In conjunction with two other testing examples, this proposed algorithm is proven to be an efficient tool for interpreting the relationship between the effective thermal conductivity and micro-/domain-structures. By combining this algorithm with the phase-field model of ferroelectric thin films, the effective thermal conductivity for PbZr 1-x Ti x O 3 films under different composition, thickness, strain, and working conditions is predicted. It is shown that the chemical composition, misfit strain, film thickness, film orientation, and a Piezoresponse Force Microscopy tip can be used to engineer the domain structures and tune the effective thermal conductivity. Furthermore, we expect our findings will stimulate future theoretical, experimental and engineering efforts on developing devices based on the tunable effective thermal conductivity in ferroelectric nanostructures.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

Neural chaos: A spectral stochastic neural operator

Building surrogate models for operators with uncertainty quantification capabilities is essential for many engineering applications where randomness–such as variability in material properties, boundary conditions, and initial conditions–is unavoidable. Polynomial Chaos Expansion (PCE) is widely recognized as a go-to method for constructing stochastic surrogates in both intrusive and non-intrusive ways, and it has recently been used in the context of operator learning. However, its application becomes challenging for complex or high-dimensional processes, as achieving accuracy requires higher-order polynomials, which can increase computational demand and/or the risk of overfitting. Furthermore, PCE requires specialized treatments to manage random variables that are not independent, and these treatments may be problem-dependent or may fail with increasing complexity. Here, in this work, we adopt the same formalism as the spectral expansion used in PCE; however, we replace the classical polynomial basis functions with neural network (NN) basis functions to leverage their expressivity. To achieve this, we propose an algorithm that identifies NN-parameterized basis functions in a purely data-driven manner, without any prior assumptions about the joint distribution of the random variables involved, whether independent or dependent, or about their marginal distributions. The proposed algorithm identifies each NN-parameterized basis function sequentially, ensuring they are orthogonal with respect to the data distribution. The basis functions are constructed directly on the joint stochastic variables without requiring a tensor product structure or assuming independence of the random variables. This approach may offer greater flexibility for complex stochastic models, while simplifying implementation compared to the tensor product structures typically used in PCE to handle random vectors. This is particularly advantageous given the current state of open-source packages, where building and training neural networks can be done with just a few lines of code and extensive community support. We demonstrate the effectiveness of the proposed scheme through several numerical examples of varying complexity and provide comparisons with classical PCE.

Polynomial chaos expansion↗

Networked Microgrids for Self-healing Power Systems

This paper proposes a transformative architecture for the normal operation and self-healing of networked microgrids (MGs). MGs can support and interchange electricity with each other in the proposed infrastructure. The networked MGs are connected by a physical common bus and a designed two-layer cyber communication network. The lower layer is within each MG where the energy management system (EMS) schedules the MG operation; the upper layer links a number of EMSs for global optimization and communication. In the normal operation mode, the objective is to schedule dispatchable distributed generators (DGs), energy storage systems (ESs) and controllable loads to minimize the operation costs and maximize the supply adequacy of each MG. When a generation deficiency or fault happens in a MG, the model switches to the self-healing mode and the local generation capacities of other MGs can be used to support the on-emergency portion of the system. A consensus algorithm is used to distribute portions of the desired power support to each individual MG in a decentralized way. The allocated portion corresponds to each MG’s local power exchange target which is used by its EMS to perform the optimal schedule. The resultant aggregated power output of networked MGs will be used to provide the requested power support. Test cases demonstrate the effectiveness of the proposed methodology.

Consensus Algorithm↗

Decentralized Microgrid Protection Through Relative Fault Direction Classification: Preprint

Protection in inverter-based resources (IBRs) dominated microgrids generally face significant challenges due to the low fault current and inconsistent fault behaviors from IBRs. Recently, machine learning-based approaches have attracted considerable attention to address these challenges. This paper introduces a novel decentralized protection strategy for microgrids. The proposed method decomposes the protection challenge into several distributed learning tasks, enabling individual relays to autonomously determine the direction of faults using a binary classification framework based on support vector machine (SVM) algorithms. Following the distributed fault direction estimation, classifier outcomes are shared among neighboring relays, facilitating a local decision-making process to ascertain the presence of faults within the neighborhood. Finally, a tripping signal is generated based on the classifier results of each relay to operate the circuit breaker. To test and validate this approach, a 100% renewable microgrid model is simulated in MATLAB/Simulink. In the numerical analysis, the application of SVM classifiers in our approach yields impressive results: an average relay classification accuracy of 98%, and a 96% accuracy in circuit breaker control. These findings highlight the potential of machine-learning-based approaches in enhancing the efficiency and reliability of microgrid protection systems.

decentralized algorithm↗

Maximal clique enumeration with data-parallel primitives

The enumeration of all maximal cliques in an undirected graph is a fundamental problem arising in several research areas. We consider maximal clique enumeration on shared-memory, multi-core architectures and introduce an approach consisting entirely of data-parallel operations, in an effort to achieve efficient and portable performance across different architectures. We study the performance of the algorithm via experiments varying over benchmark graphs and architectures. Overall, we observe that our algorithm achieves up to a 33-time speedup and 9-time speedup over state-of-the-art distributed and serial algorithms, respectively, for graphs with higher ratios of maximal cliques to total cliques. Further, we attain additional speedups on a GPU architecture, demonstrating the portable performance of our data-parallel design.

97 MATHEMATICS AND COMPUTING↗

Pre-Launch GOES-R Risk Reduction Activities for the Geostationary Lightning Mapper

The GOES-R Geostationary Lightning Mapper (GLM) is a new instrument planned for GOES-R that will greatly improve storm hazard nowcasting and increase warning lead time day and night. Daytime detection of lightning is a particularly significant technological advance given the fact that the solar illuminated cloud-top signal can exceed the intensity of the lightning signal by a factor of one hundred. Our approach is detailed across three broad themes which include: Data Processing Algorithm Readiness, Forecast Applications, and Radiance Data Mining. These themes address how the data will be processed and distributed, and the algorithms and models for developing, producing, and using the data products. These pre-launch risk reduction activities will accelerate the operational and research use of the GLM data once GOES-R begins on-orbit operations. The GLM will provide unprecedented capabilities for tracking thunderstorms and earlier warning of impending severe and hazardous weather threats. By providing direct information on lightning initiation, propagation, extent, and rate, the GLM will also capture the updraft dynamics and life cycle of convective storms, as well as internal ice precipitation processes. The GLM provides information directly from the heart of the thunderstorm as opposed to cloud-top only. Nowcasting applications enabled by the GLM data will expedite the warning and response time of emergency management systems, improve the dispatch of electric power utility repair crews, and improve airline routing around thunderstorms thereby improving safety and efficiency, saving fuel and reducing delays. The use of GLM data will assist the Bureau of Land Management (BLM) and the Forest Service in quickly detecting lightning ground strikes that have a high probability of causing fires. Finally, GLM data will help assess the role of thunderstorms and deep convection in global climate, and will improve regional air quality and global chemistry/climate modeling. The GLM has a robust design that benefits and improves upon its strong heritage of NASA-developed LEO predecessors, the Optical Transient Detector (OTD) and the Lightning Imaging Sensor (LIS). GLM will have a substantially larger number of pixels within the focal plane, two lens systems, and multiple Real-Time Event Processors REPS for on-board event detection and data compression to provide continuous observations of the Americas and adjacent oceans.

Goodman, S. J.↗

An Efficient, Multi-Layered Crown Delineation Algorithm for Mapping Individual Tree Structure Across Multiple Ecosystems

Deriving individual tree information from discrete return, small footprint LiDAR data may improve forest above ground biomass estimates, and provide tree-level information that is important in many ecological studies. Several crown delineation algorithms have been developed to extract individual tree information from LiDAR point clouds or rasterized canopy height models (CHM), but many of these algorithms have difficulty discriminating between overlapping crowns, and also may fail to detect understory trees. Our approach uses a watershed based delineation of a CHM, which is subsequently refined using the LiDAR point cloud. Individual tree detection was validated with stem mapped field data from the Smithsonian Environmental Research Center (SERC), Maryland, and on a plot and stand level through comparisons of stem density and basal area to delineated metrics at both SERC and a study area in the Sierra Nevada, California. For individual tree detection, the algorithm correctly identified 70% of dominant trees, 58% of co-dominant trees, 35% of intermediate trees and 21% of suppressed trees at SERC. The algorithm had difficulty distinguishing between crowns of small, dense understory trees of approximately the same height. Delineated crown volume alone explained 53% and 84% of the variability in basal area at the SERC and Sierra Nevada sites, respectively. The algorithm produced crown area distributions comparable to diameter at breast height (DBH) size class distributions observed in the field in both study sites. The algorithm detected understory crowns better in the conifer-dominated Sierra Nevada site than in the closed-canopy deciduous site in Maryland. The ability for the algorithm to reproduce both accurate tree size distributions and individual crown geometries in two dissimilar and complex forests suggests great promise for applicability to a wide range of forest systems.

LiDAR↗

Learning a Probability Distribution Efficiently and Reliably

A new algorithm, called the CDF-Inversion Algorithm, is described. Using it, one can efficiently learn a probability distribution over a finite set to a specified accuracy and confidence. The algorithm can be extended to learn joint distributions over a vector space. Some implementation results are described.

Laird, Philip↗

CO component estimation based on the independent component analysis

Fast Independent Component Analysis (FastICA) is a component separation algorithm based on the levels of non-Gaussianity. Here we apply FastICA to the component separation problem of the microwave background, including carbon monoxide (CO) line emissions that are found to contaminate the PLANCK High Frequency Instrument (HFI) data. Specifically, we prepare 100 GHz, 143 GHz, and 217 GHz mock microwave sky maps, which include galactic thermal dust, NANTEN CO line, and the cosmic microwave background (CMB) emissions, and then estimate the independent components based on the kurtosis. We find that FastICA can successfully estimate the CO component as the first independent component in our deflection algorithm because its distribution has the largest degree of non-Gaussianity among the components. Thus, FastICA can be a promising technique to extract CO-like components without prior assumptions about their distributions and frequency dependences.

79 ASTRONOMY AND ASTROPHYSICS↗

Scaling Deep Learning on GPU and Knights Landing clusters

The speed of deep neural networks training has become a big bottleneck of deep learning research and development. For example, training GoogleNet by ImageNet dataset on one Nvidia K20 GPU needs 21 days. To speed up the training process, the current deep learning systems heavily rely on the hardware accelerators. However, these accelerators have limited on-chip memory compared with CPUs. To handle large datasets, they need to fetch data from either CPU memory or remote processors. We use both self-hosted Intel Knights Landing (KNL) clusters and multi-GPU clusters as our target platforms. From an algorithm aspect, current distributed machine learning systems are mainly designed for cloud systems. These methods are asynchronous because of the slow network and high fault-tolerance requirement on cloud systems. We focus on Elastic Averaging SGD (EASGD) to design algorithms for HPC clusters. Original EASGD used round-robin method for communication and updating. The communication is ordered by the machine rank ID, which is inefficient on HPC clusters. First, we redesign four efficient algorithms for HPC systems to improve EASGD's poor scaling on clusters. Async EASGD, Async MEASGD, and Hogwild EASGD are faster \textcolor{black}{than} their existing counterparts (Async SGD, Async MSGD, and Hogwild SGD, resp.) in all the comparisons. Finally, we design Sync EASGD, which ties for the best performance among all the methods while being deterministic. In addition to the algorithmic improvements, we use some system-algorithm codesign techniques to scale up the algorithms. By reducing the percentage of communication from 87% to 14%, our Sync EASGD achieves 5.3x speedup over original EASGD on the same platform. We get 91.5% weak scaling efficiency on 4253 KNL cores, which is higher than the state-of-the-art implementation.

60 APPLIED LIFE SCIENCES↗

Optimization and Control Theory for Smart (Power) Grids [Slides]

The national interest in a broad class of problems loosely characterized as the smart grid is rapidly growing. At its core, the smart grid is the modernization of our existing electric power systems through improvements in efficiency, increased robustness and resiliency, integration of time-variable renewable generation, and reduction of environmental impacts. To date, the focus of most commercial entities has been on development of hardware such as smart meters and green technologies. However, with the exception of a handful of academic researchers, there has been limited research on optimization, control and more broadly, the information science required to effectively utilize this hardware. Our proposal targets development of the basic information science that will effectively bridge this gap. Using our expertise in information theory, network analysis, infrastructure analysis and power engineering, we will develop new algorithms for optimal design of the nation's next generation power transmission and distribution system that address the realities and challenges of the future grid. We focus on three core information science challenges in Grid Design, Grid Control and Grid Stability by considering a set of challenging problems. First, we study efficient and robust integration of geographically widespread renewable generation by solving large, cutting-edge optimization problems that incorporate difficult constraints imposed by the transmission of electrical power. Second, we consider two load balancing problems in the low-voltage distribution grid: development of robust and efficient distributed control techniques and algorithms based on queuing theory. These tools will enable high penetration of small-scale distributed generation and plug-in hybrid vehicles with limited or no communications. We will also develop efficient control algorithms for maintaining an operational distribution grid via switching of redundant lines. Third, we will develop a preventive toolbox capable of detecting instabilities and failures prior to their occurrence. Damaging failures and cascading outages, which can result in billions of dollars worth of damage, have substantial economic consequences. We will focus on estimating the probability of these outages, identifying their signatures and precursors, and developing detection algorithms for alerting the transmission grid operator. The new challenges in IS&T for smart grids require coherent advances in analysis and control, stability and reliability metrics, state estimation, data aggregation and assimilation, and communication middleware for the grid as well as understanding consumer response.

24 POWER TRANSMISSION AND DISTRIBUTION↗

New Results on Communication- and Memory-Aware Load Balancing Model and Algorithms

While load balancing in distributed-memory computing has been well-studied, we present an innovative approach to this problem: a unified, reduced-order model that combines three key components to describe “work” in a distributed system: computation, communication, and memory. Our model enables an optimizer to explore complex tradeoffs in task placement, such as augmented parallelism, at the expense of data replication increasing memory usage. We propose a fully distributed, heuristic-based load balancing optimization algorithm, and demonstrate that it quickly finds close-to-optimal solutions. We formalize the complex optimization problem as a mixed-integer linear program, and compare it to our strategy. Finally, we show that when applied to an electromagnetics code, our approach obtains up to 2.3x speedups for the imbalanced execution.

97 MATHEMATICS AND COMPUTING↗

Topology Reconfiguration and distributed Volt Var Control (TR-dVVC) v1

Topology Reconfiguation is a simple algorithm to choose a network configuration with the least losses in a system with high penetration of renewable energy. It uses a linearized optimal power flow algorithm to identify the optimal setpoints of PV in a system, then performs an exhaustive search of switch configurations to find the loss minimizing solution, it repeats this process until convergence. At the end, the operator has a switch schedule for a 24 hour period that should minimize losses. VVC is a distributed algorithm in which PV units drive voltage at points within the network to 1 per unit through reactive power control. The VVC algorithm relies on extremum seeking control to provide the control. The software package is a simulation of VVC on a standard IEEE 34 bus three phase network.

MacDonald, Jason↗

Π4U: A high performance computing framework for Bayesian uncertainty quantification of complex models

We present Π4U,{sup 1} an extensible framework, for non-intrusive Bayesian Uncertainty Quantification and Propagation (UQ+P) of complex and computationally demanding physical models, that can exploit massively parallel computer architectures. The framework incorporates Laplace asymptotic approximations as well as stochastic algorithms, along with distributed numerical differentiation and task-based parallelism for heterogeneous clusters. Sampling is based on the Transitional Markov Chain Monte Carlo (TMCMC) algorithm and its variants. The optimization tasks associated with the asymptotic approximations are treated via the Covariance Matrix Adaptation Evolution Strategy (CMA-ES). A modified subset simulation method is used for posterior reliability measurements of rare events. The framework accommodates scheduling of multiple physical model evaluations based on an adaptive load balancing library and shows excellent scalability. In addition to the software framework, we also provide guidelines as to the applicability and efficiency of Bayesian tools when applied to computationally demanding physical models. Theoretical and computational developments are demonstrated with applications drawn from molecular dynamics, structural dynamics and granular flow.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗