Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Real World”

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 415 records · Page 23
All sourcesNASA NTRSDOE OSTI
DOE OSTI2021

Early Battery Performance Prediction for Mixed Use Charging Profiles Using Hierarchal Machine Learning

A key step limiting how fast batteries can be deployed is the time necessary to provide evaluation and validation of performance. Using data analysis approaches, such as machine learning, the validation process can be accelerated. However, questions on the validity of projecting models trained on limited data or simple cycling profiles, such as constant current cycling, to real-world scenarios with complex loads remains. Here, we present the ability to predict performance with less than 1.2% mean absolute percent error when trained on cells aged using complex electric vehicle discharge profiles, and either AC Level 2 charge or DC Fast charge profiles, using only the first 45 cycles, namely 5% of the total testing time. While error is low across the projections, this study also highlights that battery lifetime analysis using only cycling data may not extrapolate safely to certain real-world conditions due to the impact of calendar degradation.

25 ENERGY STORAGE↗
DOE OSTI2021

Data-Driven Forgetting and Discount Factors for Vehicle Speed Forecasting in Ecological Adaptive Cruise Control

Abstract This paper investigates temporal correlations in human driving behavior using real-world driving to improve speed forecasting accuracy. These correlations can point to a measurement weighting function with two parameters: a forgetting factor for past speed measurements that the vehicle itself drove with, and a discount factor for the speeds of vehicles ahead based on information from vehicle-to-vehicle communication. The developed weighting approach is applied to a vehicle speed predictor using polynomial regression, a prediction method well-known in the literature. The performance of the developed approach is then assessed in both real-world and simulated traffic scenarios for accuracy and robustness. The new weighting method is applied to an ecological adaptive cruise control system, and its influence is analyzed on the prediction accuracy and the performance of the ecological adaptive cruise control in an electric vehicle powertrain model. The results show that the new prediction method improves energy saving from the eco-driving by up to 4.7% compared to a baseline least-square-based polynomial regression. This is a 10% improvement over the constant speed/acceleration model, a conventional speed predictor.

Automation & Control Systems↗
DOE OSTI2022

A Privacy-Preserving Strategy for the Trust Layer of the Energy Grid of Things Distributed Energy Resource Management System

Emergent from the shadows of the traditional grid flaws, the Smart Grid (SG) idea was born and led by government mandates toward cleaner energy production. The SG represents the next generation of electricity distribution systems that subsume recent technological innovations. It uses digital communication between its components and entities to attain more automation, self-sufficiency, and reliability. Unfortunately, this relatively new concept is not flawless; the intrinsic reliance on increased digital communication spreads open attack paths for adversaries. Therefore, finding solutions that address information exchange vulnerabilities has become imperative. The Energy Grid of Things (EGoT) is Portland State University’s (PSU’s) implementation of a Distributed Energy Resource Management System (DERMS). The EGoT DERMS requires access to customers’ information to achieve operational objectives. The system’s access to customers’ information needs to be restricted such that it does not violate customers’ privacy. Applying privacy protection models such as K-anonymity to EGoT DERMS sub-components safeguards that privacy. This thesis work proposes a strategy to ensure communication in the EGoT DERMS is privacy-preserving and secure. Specifically, it provides an approach to applying the Mondrian Algorithm to ensure data within the system excludes Personally Identifiable Information (PII) and provides means for securing the communication according to industry standards (IEEE 2030.5). Results suggest that the generalization hierarchy derived for the EGoT DERMS exhibits an Identical Generalization Hierarchy structure. Guarantees of sameness manifested in the test feeder topology would not hold in real-world scenarios. This thesis work proposes a strategy to ensure communication in the EGoT DERMS is privacy-preserving and secure. Specifically, it provides an approach to applying the Mondrian Algorithm to ensure data within the system excludes Personally Identifiable Information (PII) and provides means for securing the communication according to industry standards (IEEE 2030.5). Results suggest that the generalization hierarchy derived for the EGoT DERMS exhibits an Identical Generalization Hierarchy structure. Guarantees of sameness manifested in the test feeder topology would not hold in real-world scenarios.

Alsiad, Mohammed↗
DOE OSTI2025

AI Benchmark Democratization and Carpentry

Benchmarks are a cornerstone of modern machine learning, enabling reproducibility, comparison, and scientific progress. However, AI benchmarks are increasingly complex, requiring dynamic, AI-focused workflows. Rapid evolution in model architectures, scale, datasets, and deployment contexts makes evaluation a moving target. Large language models often memorize static benchmarks, causing a gap between benchmark results and real-world performance. Beyond traditional static benchmarks, continuous adaptive benchmarking frameworks are needed to align scientific assessment with deployment risks. This calls for skills and education in AI Benchmark Carpentry. From our experience with MLCommons, educational initiatives, and programs like the DOE's Trillion Parameter Consortium, key barriers include high resource demands, limited access to specialized hardware, lack of benchmark design expertise, and uncertainty in relating results to application domains. Current benchmarks often emphasize peak performance on top-tier hardware, offering limited guidance for diverse, real-world scenarios. Benchmarking must become dynamic, incorporating evolving models, updated data, and heterogeneous platforms while maintaining transparency, reproducibility, and interpretability. Democratization requires both technical innovation and systematic education across levels, building sustained expertise in benchmark design and use. Benchmarks should support application-relevant comparisons, enabling informed, context-sensitive decisions. Dynamic, inclusive benchmarking will ensure evaluation keeps pace with AI evolution and supports responsible, reproducible, and accessible AI deployment. Community efforts can provide a foundation for AI Benchmark Carpentry.

von Laszewski, Gregor [Virginia U.]↗
DOE OSTI2022

Port of New York and New Jersey Drayage Electrification Analysis

The National Renewable Energy Laboratory (NREL) evaluated the potential for drayage electrification in the Port of New York and New Jersey (PoNYNJ), with a focus on operators: Harbor Freight Transport (HF), Safeway Trucking (SWT), and International Motor Freight Inc (IMF). This report summarizes the data collection and electrification evaluation of all three drayage operators, includes detailed operational data, and identifies the performance requirements for battery electric tractors (BETs) and corresponding infrastructure operated within the context of PoNYNJ drayage operation. This report also details a methodology to evaluate opportunities, strategies, and challenges associated with future expansions of BETs in meeting PANYNJ emissions goals. The Port Authority has established a goal of achieving Net Zero carbon emissions by 2050 across all facilities, including from tenant and stakeholder sources such as drayage trucks. NREL used real-world performance data collected on the three PoNYNJ drayage operations, along with modeling and analysis tools to compare BET to diesel trucks. From March to July 2021, NREL collected 1Hz vehicle and engine data from 46 drayage trucks at the three operators totaling nearly 121,000 miles of operation, providing enough information to assess vehicle operations for electrification potential. A Future Automotive Systems Technology Simulator (FASTSim) electric truck powertrain model was validated using PoNYNJ data and scenarios were run to evaluate drayage truck electrification requirements over the real-world cycles. The first scenario examined BET viability with minimal changes to existing operations. This assumes the trucks charge when stopped for two hours or longer, have a functional battery size of 375 kWh, and can charge at 270 kilowatts (kW) average which are the specification of the commercially available Freightliner eCascadia. The second scenario looked at what operational, charging infrastructure, and BET technology changes would be needed to fully electrify. Finally, detailed analysis was run on charging rate structure to understand operational costs to the fleets. The studied drayage trucks averaged 5.1 MPG, spent roughly 9% of their energy at idle, and drove an average of 140 miles per day with a maximum daily distance of 573 miles. The FASTSim model results indicate a comparable BET would use 417 kWh of energy per day on average accounting for cargo weight, which is close to the full usable capacity of the eCascadia currently available on the market. Based on the daily average operating data, partial fleet electrification is possible with current technology. However, some specific days of operation would require over 1,600 kWh of energy due to longer distances traveled by the trucks and more intense operation. Trucks used for long distance and intense operation cannot be readily electrified with current technology without operational changes. Full adoption of BETs could reduce CO 2 emissions from these fleets by roughly 75% today, eliminating 76 metric tons of CO 2 (MTCO 2 ) per vehicle each year, which equates to 24,100 MTCO 2 per year for all three operators. Commercially available direct current fast chargers (DCFC) have charge rates up to 350 kW. Based on the average daily modeled energy use for each operator, current industrial rate structures, and the assumption of 350 kW peak charging, full drayage electrification would increase electricity consumption. In addition, peak demand usage would increase with unmanaged charging along with cost of electricity having a direct impact on cost per mile for electric vehicles. The resulting cost per mile for BETs along with comparable cost per mile for conventional diesel trucks are also examined at $\$$4.00 per gallon of diesel. It will be important for PANYNJ and the drayage operators within the PoNYNJ to consider these load impacts to their existing electrical infrastructure and devise operational strategies that avoid coincident charging of vehicles to mitigate demand charges. Despite these electricity cost increases, savings from reductions in diesel consumption will help offset the costs of this increased electricity consumption. However, prices of both electricity and diesel are subject to change based on various factors meaning the realized savings will vary over time. This shows BETs could be cost-competitive on an energy cost per mile basis for all scenarios while diesel is above $\$$3.00/gal. Further, if diesel prices dropped to the 15-year low of $2.33/gal, it would still be cost competitive to operate the EVs with electricity costs of 16.3 ¢/kWh or less.

33 ADVANCED PROPULSION SYSTEMS↗
DOE OSTI2024

Rapid Detection of Anomalies in Battery Energy Storage System Data

Data analytics is pivotal in assessing the technical characteristics and performance of Battery Energy Storage Systems (BESS), underpinning BESS modeling, optimization, and control. However, raw datasets frequently harbor anomalies from measurement errors and equipment malfunctions, impacting BESS reliability and analysis accuracy To address the challenge, this paper presents a novel methodology for the rapid detection of anomalous charge or discharge cycles within BESS operational data, expediting the cleaning process while ensuring data integrity. We’ve collected diverse and comprehensive real-world BESS operational datasets in collaboration with the Electric Power Research Institute and multiple Washington State utilities. These datasets serve dual roles: enabling comprehensive data exploration and analysis for understanding underlying challenges and method development, while also acting as a vital validation resource, demonstrating practical effectiveness. The proposed method detects anomalies and aids in their resolution, improving system performance characterization precision. It also reveals recurring data anomaly sources, offering insights for data collection and handling enhancement. Practitioners can gain valuable insights from the identified anomalous cycles in the real-world datasets along with the investigative process for root cause analyses and essential data cleaning steps.

Crawford, Aladsair J.↗
DOE OSTI2025

Dual aging pathways of Cu-SSZ-13 SCR catalysts: Hydrothermal vs. sulfur-induced deactivation

Hydrothermal aging (HTA) and chemical poisoning are two primary factors contributing to the real-world degradation of Cu-SSZ-13 SCR catalysts. Investigating field-returned samples offers valuable insights into performance degradation caused by these mechanisms. However, the simultaneous presence of both deactivation pathways complicates the isolation of their individual effects in post-mortem analyses. In this study, we separately prepared model Cu-SSZ-13 SCR catalysts subjected to hydrothermal-aging and sulfur-induced chemical poisoning. Using various characterization techniques, we elucidated the specific role of each aging process in catalyst deactivation and compared the results to real-world field-aged catalysts. Our findings show that hydrothermal aging at 650 °C for 100 h caused dealumination of the zeolite framework but no significant CuO x cluster formation. In contrast, sulfur aging (via sulfur exposure, calcination at 550 °C, and desulfation up to 750 °C) led to CuO x formation without any observable dealumination. On model catalysts, sulfur poisoning was found to reduce Cu mobility and the amount of active Cu sites, thus degrading catalyst activity. Although some activity was recovered upon desulfation, a portion of the initial catalyst activity remained irreversibly lost due to CuO x formation. We demonstrate that this occurs because sulfated species impede the ability of multi-nuclear Cu species (e.g., Cu dimers) to split back into their isolated form, leading to CuSO 4 -clusters that oxidatively desulfate to CuO x species. This degradation pathway explains the significant reduction in activity of field-aged samples, where substantial CuSO 4 -cluster accumulation leads to reduced active Cu and subsequent conversion to CuO x . Furthermore, the conclusions from model catalysts were extended directly to field-aged commercial samples, elucidating the decline in activity and chemical properties during field deployment.

Catalyst Deactivation↗
DOE OSTI2025

Decentralized Distributed Proximal Policy Optimization (DD-PPO) for High Performance Computing Scheduling on Multi-User Systems

Resource allocation in High Performance Computing (HPC) environments presents a complex and multifaceted challenge for job scheduling algorithms. Beyond the efficient allocation of system resources, schedulers must account for and optimize multiple performance metrics, including job wait time and system throughput. Traditional heuristic-based scheduling algorithms increasingly struggle and lack the efficiency needed to meet the demands and address the complexity and scale of modern HPC systems. Consequently, recent research efforts have focused on leveraging advancements in Artificial Intelligence (AI) and Deep Learning (DL), particularly Reinforcement Learning (RL), to develop more adaptable and intelligent scheduling strategies. Previous RL-based scheduling approaches have explored a range of algorithms, from Deep Q-Networks (DQN) to Proximal Policy Optimization (PPO), and more recently, hybrid methods that integrate Graph Neural Networks (GNNs) with RL techniques. However, a common limitation across these methods is their reliance on relatively small datasets, with few methods being evaluated using large-scale, multi-million-job trace datasets representative of real-world HPC workloads. Moreover, existing RL schedulers face scalability issues due to centralized policy updates, which hinder training efficiency and performance when applied to large datasets. This study introduces a novel RL-based scheduler utilizing Decentralized Distributed Proximal Policy Optimization (DD-PPO) algorithm, which supports large-scale distributed training across multiple workers without requiring parameter synchronization at every step. By eliminating reliance on centralized updates to a shared policy, the DD-PPO scheduler enhances scalability, training efficiency, and sample utilization. Experimental validation using a large real-world dataset containing over 11.5 million job traces collected from petascale HPC systems over six years assesses the influence of dataset scale on training effectiveness and compares DD-PPO performance to traditional and advanced scheduling approaches. The experimental results demonstrate improved scheduling performance in comparison to both heuristic-based schedulers and existing RL-based scheduling algorithms.

AI↗
DOE OSTI2025

A Scalable Multi-Modal Framework for High-Fidelity Distributed Human Mobility Simulations

The development of data-driven models for human mobility in urban settings requires access to substantial and diverse real-world data. However, existing historical data often presents challenges such as limited volume, variety, and veracity, as well as missing data and privacy preservation concerns. Also, urban mobility modeling is inherently time-variant, complex, and multi-modal, encompassing everything from individual walking and running to private road travel and large-scale public transportation. These challenges call for innovative solutions to overcome data limitations and compute needs to model mobility behaviors accurately. To address these challenges, we propose a distributed, co-simulation-based architecture DURMOSim that integrates real-world data with scalable, high-fidelity simulations, demonstrating distributed co-simulation feasibility with existing mobility models. DURMOSim underpins a modular integration that would enable using any available mobility simulators for greater extensibility and scalability in performing various urban scenarios. In this paper, we present the design, implementation, and performance evaluation of DURMOSim, highlighting its capability to model population-scale mobility patterns. Our initial results show its ability to dynamically synchronize multiple simulation models at runtime with negligible computational overhead. We believe DURMOSim could be a robust tool for advancing urban mobility research and intelligent transportation systems.

Yoginath, Srikanth [ORNL] (ORCID:0000000184236050)↗
DOE OSTI2021

Safeguards-Informed Hybrid Imagery Dataset [Poster]

Deep Learning computer vision models require many thousands of properly labelled images for training, which is especially challenging for safeguards and nonproliferation, given that safeguards-relevant images are typically rare due to the sensitivity and limited availability of the technologies. Creating relevant images through real-world staging is costly and limiting in scope. Expert-labeling is expensive, time consuming, and error prone. We aim to develop a data set of both realworld and synthetic images that are relevant to the nuclear safeguards domain that can be used to support multiple data science research questions. In the process of developing this data, we aim to develop a novel workflow to validate synthetic images using machine learning explainability methods, testing among multiple computer vision algorithms, and iterative synthetic data rendering. We will deliver one million images – both real-world and synthetically rendered – of two types uranium storage and transportation containers with labelled ground truth and associated adversarial examples.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

V1G Frequency Regulation: Algorithm Development, Validation & Analysis at Scale

Researchers at Argonne National Laboratory developed and validated a high-fidelity digital twin of a smart charging (V1G) ecosystem to model the participation of up to 1,000 unique electric vehicles (EVs) in the PJM frequency regulation market. Utilizing a discrete-event framework, the simulation models complex interactions, from dynamic grid signals (updated every 2 seconds) to individual EV charging dynamics. The simulation incorporates multiple EV models created from real-world lab test data. Researchers tested multiple control algorithms to balance the dual objectives of maximizing aggregator’s revenue and driver charging needs. Results demonstrate that aggregated EVs function as a controllable, highly effective grid resource, achieving high PJM Performance Scores (80–90%). Additionally, an optimized, market-aware bidding strategy was identified as key to profitability. The platform was shown to provide drivers with an average charging discount of nearly 50%. The algorithm was further validated in the lab using production EVs and charging stations to compare simulation results with real-world performance.

Manne, Nithin↗
NASA NTRS1990

Analyzing spacecraft configurations through specialization and default reasoning

For an intelligent system to describe a real-world situation using as few statements as possible, it is necessary to make inferences based on observed data and to incorporate general knowledge of the reasoning domain into the description. These reasoning processes must reduce several levels of specific descriptions into only those few that most precisely describe the situation. Moreover, the system must be able to generate descriptions in the absence of data, as instructed by certain rules of inference. The deductions applied by the system, then, generate a high-level description from the low-level evidence provided by the real and default data sources. An implementation of these ideas in a real-world situation is described. The application concerns evaluation of Space Shuttle electromechanical system configurations by console operators in the Mission Control Center. A production system provides the reasoning mechanism through which the default assignments and specializations occur. Examples are provided within this domain for each type of inference, and the suitability is discussed of each toward achieving the goal of describing a situation in the fewest statements possible. Finally, several enhancements are suggested that will further increase the intelligence of similar spacecraft monitoring applications.

Barry, Matthew R.↗
DOE OSTI2025

MLCommons Science Benchmarks

Benchmarks are a cornerstone of modern machine learning practice, providing standardized eval- uations that enable reproducibility, comparison, and scientific progress. Yet, as AI systems particularly deep learning models become increasingly dynamic, traditional static benchmarking approaches are losing their relevance. Models rapidly evolve in architecture, scale, and capability; datasets shift; and deployment contexts continuously change, creating a moving target for evaluation. Without adaptive benchmarking frame- works, both scientific assessment and real-world de- ployment risk becoming misaligned with actual system behavior. Drawing on our experience from MLCommons, educa- tional initiatives, and government programs such as the DOE s Million Parameter Consortium, we identify key barriers that hinder the broader adoption and utility of benchmarking in AI. These include substantial resource demands, limited access to specialized hardware, lack of expertise in benchmark design, and uncertainty among practitioners about how to relate benchmark results to their own application domains. Moreover, current benchmarks often emphasize peak performance on leadership-class hardware, offering limited guidance for more diverse, real-world deployment scenarios. We argue that benchmarking itself must become dy- namic in order to incorporate evolving models, updated data, and heterogeneous computational platforms while maintaining transparency, reproducibility, and inter- pretability. Democratizing this process requires not only technical innovation, but also systematic educational efforts spanning undergraduate to professional levels to develop sustained expertise in benchmark design and use. Finally, benchmarks should be framed and com- municated to support application-relevant comparisons, enabling both developers and users to make informed, context-sensitive decisions. Advancing dynamic and inclusive benchmarking practices will be essential to ensure that evaluation keeps pace with the evolving AI landscape and supports responsible, reproducible, and accessible AI deployment.

Hawks, Benjamin G. [Fermilab]↗
NASA NTRSDate not supplied

High-Dimensional Similarity Search with Quantum-Assisted Variational Autoencoder

Recent progress in quantum algorithms and hardware indicates the potential importance of quantum computing in the near future. However, finding suitable application areas remains an active area of research. Quantum machine learning is touted as a potential approach to demonstrate quantum advantage within both the gate-model and the adiabatic schemes. For instance, the QVAE has been proposed as a quantum enhancement to the discrete VAE. We extend on previous work and study the real-world applicability of a QVAE by presenting a proof-of-concept for similarity search in large-scale high-dimensional datasets. While exact and fast similarity search algorithms are available for low dimensional datasets, scaling to high-dimensional data is non-trivial. We show how to construct a space-efficient search index based on the latent space representation of a QVAE. Our experiments show a correlation between the Hamming distance in the embedded space and the Euclidean distance in the original space on the MODIS dataset. Further, we find real-world speedups compared to linear search and demonstrate memory-efficient scaling to half a billion data points.

Data mining, similarity search, quantum machine le↗
DOE OSTI2020

Dissertation Proposal: Innovations in Pulsed Power and Plasma Science Theory Simulation and Experiment

Pulsed power and plasma physics are topics of great study at both Sandia National Laboratories (SNL or Sandia) and the University of New Mexico (UNM). The goal of this research is to further knowledge and understanding of these fields using the resources of both SNL and UNM in three ways. The first way is through the comprehension, application, and testing of theory. Reading and analytically deriving theoretical solutions of problems both real-world and simplified will allow for a fresh perspective and the furthering of the theory. One such theory is Ottinger's generalized theory for voltage measurement in magnetically insulated transmission lines (MITLs). By working through the math, a deeper understanding of the theory is gained from which one may add more physically accurate and/or more detailed physics into the theory. Additionally, understanding the theory lays a good foundation from which one can analyze, test, and compare results to the theory in the following two ways that will advance the fields of pulsed power and plasma physics. The second way is through the modeling and simulation of real-world and simplified problems that utilize and test the afore mentioned theories. Theory can be applied to a simulation domain by using the unstructured time-domain electromagnetic (UTDEM) codes EMPHASIS and EMPIRE as well as the physical modeling software CUBIT, all of which were developed at SNL. Problems such as the modeling and design of the extended MITL on HERMES III, the understanding of space-charge-limited emission from vacuum cathodes, and the interaction between a relativistic electron beam and an ideal gas can all be modeled, simulated, and analyzed with this set of codes. Here the advantage is three-fold. Firstly, theory that describes our understanding of these problems can be put to the test and advanced through iterative simulation and analysis. Secondly, the understanding of these problems will have a positive impact on national security through the advancement of the technological capability of the United States of America. Thirdly, and not unrelated to the prior advantage, is the validation and verification of EMPIRE and EMPHASIS. This segues into the third way, which is through experiment and the comparison of experiment to simulated and theoretical results. Performing experimental comparisons completes the scientific method and grounds all of the work in reality. Being able to physically test theory and simulation is necessary for any real conclusions to be drawn. Another advantage for carrying out experimental work is to advance the physical testing capabilities of SNL. Several systems will be developed and tested through the course of this work that positively impact technological advancement of Sandia National Labs. Lastly, all of the above work will converge to yield a well-rounded perspective that ties the three categories of research together.

42 ENGINEERING↗
DOE OSTI2020

MimicGAN: Robust Projection onto Image Manifolds with Corruption Mimicking

In the past few years, Generative Adversarial Networks (GANs) have dramatically advanced our ability to represent and parameterize high-dimensional, non-linear image manifolds. As a result, they have been widely adopted across a variety of applications, ranging from challenging inverse problems like image completion, to problems such as anomaly detection and adversarial defense. A recurring theme in many of these applications is the notion of projecting an image observation onto the manifold that is inferred by the generator. In this context, Projected Gradient Descent (PGD) has been the most popular approach, which essentially optimizes for a latent vector that minimizes the discrepancy between a generated image and the given observation. However, PGD is a brittle optimization technique that fails to identify the right projection (or latent vector) when the observation is corrupted, or perturbed even by a small amount. Such corruptions are common in the real world, for example images in the wild come with unknown crops, rotations, missing pixels, or other kinds of non-linear distributional shifts which break current encoding methods, rendering downstream applications unusable. To address this, we propose corruption mimicking—a new robust projection technique, that utilizes a surrogate network to approximate the unknown corruption directly at test time, without the need for additional supervision or data augmentation. The proposed method is significantly more robust than PGD and other competing methods under a wide variety of corruptions, thereby enabling a more effective use of GANs in real-world applications. Finally, more importantly, we show that our approach produces state-of-the-art performance in several GAN-based applications—anomaly detection, domain adaptation, and adversarial defense, that benefit from an accurate projection.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Reinforcement learning building control approach harnessing imitation learning

Reinforcement learning (RL) has shown significant success in sequential decision making in fields like autonomous vehicles, robotics, marketing and gaming industries. This success has attracted the attention to the RL control approach for building energy systems which are becoming complicated due to the need to optimize for multiple, potentially conflicting, goals like occupant comfort, energy use and grid interactivity. However, for real world applications, RL has several drawbacks like requiring large training data and time, and unstable control behavior during the early exploration process making it infeasible for an application directly to building control tasks. To address these issues, an imitation learning approach is utilized herein where the RL agents starts with a policy transferred from accepted rule based policies and heuristic policies. This approach is successful in reducing the training time, preventing the unstable early exploration behavior and improving upon an accepted rule-based policy - all of these make RL a more practical control approach for real world applications in the domain of building controls.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
DOE OSTI2021

Class imbalance in out-of-distribution datasets: Improving the robustness of the TextCNN for the classification of rare cancer types

In the last decade, the widespread adoption of electronic health record documentation has created huge opportunities for information mining. Natural language processing (NLP) techniques using machine and deep learning are becoming increasingly widespread for information extraction tasks from unstructured clinical notes. Disparities in performance when deploying machine learning models in the real world have recently received considerable attention. In the clinical NLP domain, the robustness of convolutional neural networks (CNNs) for classifying cancer pathology reports under natural distribution shifts remains understudied. In this research, we aim to quantify and improve the performance of the CNN for text classification on out-of-distribution (OOD) datasets resulting from the natural evolution of clinical text in pathology reports. We identified class imbalance due to different prevalence of cancer types as one of the sources of performance drop and analyzed the impact of previous methods for addressing class imbalance when deploying models in real-world domains. Our results show that our novel class-specialized ensemble technique outperforms other methods for the classification of rare cancer types in terms of macro F1 scores. We also found that traditional ensemble methods perform better in top classes, leading to higher micro F1 scores. Based on our findings, we formulate a series of recommendations for other ML practitioners on how to build robust models with extremely imbalanced datasets in biomedical NLP applications.

60 APPLIED LIFE SCIENCES↗