Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “learned priors”

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 217 records · Page 12
All sourcesNASA NTRSDOE OSTI
DOE OSTI2022

General-Purpose Bayesian Tensor Learning With Automatic Rank Determination and Uncertainty Quantification

A major challenge in many machine learning tasks is that the model expressive power depends on model size. Low-rank tensor methods are an efficient tool for handling the curse of dimensionality in many large-scale machine learning models. The major challenges in training a tensor learning model include how to process the high-volume data, how to determine the tensor rank automatically, and how to estimate the uncertainty of the results. While existing tensor learning focuses on a specific task, this paper proposes a generic Bayesian framework that can be employed to solve a broad class of tensor learning problems such as tensor completion, tensor regression, and tensorized neural networks. We develop a low-rank tensor prior for automatic rank determination in nonlinear problems. Our method is implemented with both stochastic gradient Hamiltonian Monte Carlo (SGHMC) and Stein Variational Gradient Descent (SVGD). We compare the automatic rank determination and uncertainty quantification of these two solvers. We demonstrate that our proposed method can determine the tensor rank automatically and can quantify the uncertainty of the obtained results. We validate our framework on tensor completion tasks and tensorized neural network training tasks.

Bayesian inference↗
DOE OSTI2021

Connectivity‐based parcellation of normal and anatomically distorted human cerebral cortex

Abstract For over a century, neuroscientists have been working toward parcellating the human cortex into distinct neurobiological regions. Modern technologies offer many parcellation methods for healthy cortices acquired through magnetic resonance imaging. However, these methods are suboptimal for personalized neurosurgical application given that pathology and resection distort the cerebrum. We sought to overcome this problem by developing a novel connectivity‐based parcellation approach that can be applied at the single‐subject level. Utilizing normative diffusion data, we first developed a machine‐learning (ML) classifier to learn the typical structural connectivity patterns of healthy subjects. Specifically, the Glasser HCP atlas was utilized as a prior to calculate the streamline connectivity between each voxel and each parcel of the atlas. Using the resultant feature vector, we determined the parcel identity of each voxel in neurosurgical patients ( n = 40) and thereby iteratively adjusted the prior. This approach enabled us to create patient‐specific maps independent of brain shape and pathological distortion. The supervised ML classifier re‐parcellated an average of 2.65% of cortical voxels across a healthy dataset ( n = 178) and an average of 5.5% in neurosurgical patients. Our patient dataset consisted of subjects with supratentorial infiltrating gliomas operated on by the senior author who then assessed the validity and practical utility of the re‐parcellated diffusion data. We demonstrate a rapid and effective ML parcellation approach to parcellation of the human cortex during anatomical distortion. Our approach overcomes limitations of indiscriminately applying atlas‐based registration from healthy subjects by employing a voxel‐wise connectivity approach based on individual data.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2021

Limited-angle computed tomography with deep image and physics priors

Computed tomography is a well-established x-ray imaging technique to reconstruct the three-dimensional structure of objects. It has been used extensively in a variety of fields, from diagnostic imaging to materials and biological sciences. One major challenge in some applications, such as in electron or x-ray tomography systems, is that the projections cannot be gathered over all the angles due to the sample holder setup or shape of the sample. This results in an ill-posed problem called the limited angle reconstruction problem. Typical image reconstruction in this setup leads to distortion and artifacts, thereby hindering a quantitative evaluation of the results. To address this challenge, we use a generative model to effectively constrain the solution of a physics-based approach. Our approach is self-training that can iteratively learn the nonlinear mapping from partial projections to the scanned object. Because our approach combines the data likelihood and image prior terms into a single deep network, it is computationally tractable and improves performance through an end-to-end training. We also complement our approach with total-variation regularization to handle high-frequency noise in reconstructions and implement a solver based on alternating direction method of multipliers. We present numerical results for various degrees of missing angle range and noise levels, which demonstrate the effectiveness of the proposed approach.

47 OTHER INSTRUMENTATION↗
DOE OSTI2023

TransformerG2G: Adaptive time-stepping for learning temporal graph embeddings using transformers

Dynamic graph embedding has emerged as a very effective technique for addressing diverse temporal graph analytic tasks (i.e., link prediction, node classification, recommender systems, anomaly detection, and graph generation) in various applications. Such temporal graphs exhibit heterogeneous transient dynamics, varying time intervals, and highly evolving node features throughout their evolution. Hence, incorporating long-range dependencies from the historical graph context plays a crucial role in accurately learning their temporal dynamics. In this paper, we develop a graph embedding model with uncertainty quantification, TransformerG2G, by exploiting the advanced transformer encoder to first learn intermediate node representations from its current state (t) and previous context (over timestamps [t–1,t–l], l is the length of context). Moreover, we employ two projection layers to generate lower-dimensional multivariate Gaussian distributions as each node's latent embedding at timestamp t. We consider diverse benchmarks with varying levels of "novelty" as measured by the TEA (Temporal Edge Appearance) plots. Here, our experiments demonstrate that the proposed TransformerG2G model outperforms conventional multi-step methods and our prior work (DynG2G) in terms of both link prediction accuracy and computational efficiency, especially for high degree of novelty. Furthermore, the learned time-dependent attention weights across multiple graph snapshots reveal the development of an automatic adaptive time stepping enabled by the transformer. Importantly, by examining the attention weights, we can uncover temporal dependencies, identify influential elements, and gain insights into the complex interactions within the graph structure. For example, we identified a strong correlation between attention weights and node degree at the various stages of the graph topology evolution.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Statistical learning for predicting density–matrix-based electron dynamics

In this study, we consider the problem of learning density-dependent molecular Hamiltonian matrices from time series of electron density matrices, all in the context of Hartree–Fock theory. Prior work developed a solution to this problem for small molecular systems with density and Hamiltonian matrices of size at most 6 × 6. Here, using a battery of techniques, we scale prior methods to larger molecular systems with, for example, 29 × 29 matrices. This includes systems that either have more electrons or are expressed in large basis sets such as 6-311++G**. Scaling the method to larger systems enhances its relevance for realistic applications in chemistry and physics. To achieve this scaling, we apply dimensionality reduction, ridge regression and analytic computation of Hessians. Through the combination of these techniques, we are able to learn Hamiltonians by minimizing an objective function that encodes local propagation error. Importantly, these learned Hamiltonians can then be used to predict electron dynamics for thousands of steps: When we use our learned Hamiltonians to numerically solve the time-dependent Hartree–Fock equation, we obtain predicted dynamics that are in close quantitative agreement with ground truth dynamics. This includes field-off trajectories similar to the training data and field-on trajectories outside of the training data.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2020

Normal or abnormal? Machine learning for the leakage detection in carbon sequestration projects using pressure field data

The international commitments for atmospheric carbon reduction will require a rapid increase in carbon capture and storage (CCS) projects. The key to any successful CCS project lies in the long term storage and prevention of leakage of stored carbon dioxide (CO 2 ). In addition to being a greenhouse gas, CO 2 leaks reaching the surface can accumulate in low-lying areas resulting in a serious health risk. Among several alternatives, some of the more promising CCS storage formations are depleted oil and gas reservoirs, where the reservoirs had good geological seals prior to hydrocarbon extraction. With more CCS wells coming online, it is imperative to implement permanent, automated monitoring tools. We apply machine learning models to automate the leakage detection process in carbon storage reservoirs using rates of (CO 2 ) injection and pressure data measured by simple harmonic pulse testing (HPT). To validate the feasibility of this machine learning based workflow, we use data from HPT experiments carried out in the Cranfield oil field, Mississippi, USA. The data consist of a series of pulse tests conducted with baseline parameters and with an artificially introduced leak. Here, in this study, we pose the leakage detection task as an anomaly detection problem where deviation from the predicted behavior indicates leaks in the reservoir. Results show that different machine learning architectures such as multi-layer feed forward network, Long Short-Term Memory, and convolutional neural network are able to identify leakages and can provide early warning. These warnings can then be used to take remedial measures.

58 GEOSCIENCES↗
DOE OSTI2022

Bayesian operator inference for data-driven reduced-order modeling

This work proposes a Bayesian inference method for the reduced-order modeling of time-dependent systems. Informed by the structure of the governing equations, the task of learning a reduced-order model from data is posed as a Bayesian inverse problem with Gaussian prior and likelihood. The resulting posterior distribution characterizes the operators defining the reduced-order model, hence the predictions subsequently issued by the reduced-order model are endowed with uncertainty. The statistical moments of these predictions are estimated via a Monte Carlo sampling of the posterior distribution. Since the reduced models are fast to solve, this sampling is computationally efficient. Furthermore, the proposed Bayesian framework provides a statistical interpretation of the regularization term that is present in the deterministic operator inference problem, and the empirical Bayes approach of maximum marginal likelihood suggests a selection algorithm for the regularization hyperparameters. The proposed method is demonstrated on two examples: the compressible Euler equations with noise-corrupted observations, and a single-injector combustion process.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2020

Retrospective on Recent DOE-Funded Studies Concerning the Extraction of Rare Earth Elements & Lithium from Geothermal Brines (Final Report)

Rare earth elements (REE) and lithium are non-toxic metals that are considered critical materials due to their use in electronics, magnets, batteries, and a wide variety of industrial processes important for the economy and military preparedness. Demand for REE and lithium is increasing and these critical materials are imported, so identifying and exploiting domestic sources of REE and lithium is a national priority. The U.S. Department of Energy (DOE) Geothermal Technologies Office (GTO) has been in the forefront of sponsoring research investigating the potential recovery of REE, lithium, and other critical minerals from geothermal brines. It has been proposed that the future of geothermal energy should include “hybrid systems” that combine electricity generation with other revenue-generating activities, such as recovery of valuable and critical minerals, including REE and lithium. Two recent GTO funding opportunities have focused on the recovery of REE and other valuable minerals from geothermal brines. The research supported by the GTO’s mineral recovery program is focused on three areas: resource characterization, technology for the extraction of REE, and technology for the extraction of lithium (Tables 1 and 2). This report is a retrospective study examining the outcome of GTO’s two recent mineral recovery programs (DE-FOA-0001016 in FY 2014 and DE-FOA-0001376 in FY 2016). In this report, the knowledge, technology, and techniques that were developed by researchers funded by GTO are summarized and discussed. Four projects were funded to assess the concentrations and amounts of REE found in geothermal brines and oil field produced waters. The GTO-funded studies compiled publically available data on REE concentrations from brines and produced water from all over the USA. In addition, new samples were collected and characterized from major geothermal and hydrocarbon basins in the Western USA. The studies examined the relationship between lithology and REE concentrations and developed models examining the influence of geology on REE concentrations in produced brines. It was determined that REE are frequently found at higher concentrations in oil field produced water than geothermal brines, but that some geothermal areas had significant REE resources. Significant reservoirs of REE were identified in the Western USA. In some cases, concentrations of REE were more than 1000 times the concentrations found in seawater. Collectively, these studies represent a comprehensive picture of REE resources associated with geothermal and hydrocarbon systems in the USA. The studies did not examine lithium resources, but in some cases, lithium concentration data was collected. Data from these studies are housed in the Geothermal Data Repository (GDR) and represent a significant information resource and it is recommended that these data be further analyzed in a future study. Eight projects were funded to develop new technology for REE extraction from geothermal fluids. These projects investigated sorption as an approach for removal and recovery of REE from geothermal brines. The projects investigated cutting-edge technology for selective sorption of ions from complex solutions, including the application of metal-organic frameworks and biosorbent proteins. The REE sorption studies tested different combinations of metal-binding ligands and solid supports. The most promising metal-binding ligands for REE included phosphonic acid, thiol, and carboxylic acid functional groups. Ligands were attached or incorporated into a wide variety of solid supports. In most cases, attachment was via covalent bonding to organic resins, polymers, or silica-based supports. Most of the REE projects were conducted at a low technology readiness level (TRL) and showed promise, but direct comparison between technologies was not possible based on the available information. It is recommended that testing and reporting be standardized to the extent possible to facilitate comparisons between technologies. Two projects were directed at novel lithium extraction technology. Both projects investigated the use of inorganic sorbents, including manganese oxides. One study also examined the use of metal- ion imprinted polymers as selective ion-exchange resins for the separation of lithium and manganese from brines. Both approaches showed promise for the selective extraction of lithium from brines, including potentially geothermal brines. Results from these GTO studies indicated that selective REE and lithium extraction is possible, but interference from co-occurring solutes, such as calcium, magnesium, or heavy metals, will interfere with process efficiency and negatively impact process economics. Techno-economic analysis conducted as part of the resource and technology studies suggest extraction of REE from geothermal brines is unlikely to be economically viable, especially since non-geothermal produced waters frequently have higher REE concentrations. It is recommended that benchmarks for techno-economic analysis be established to the extent possible for future studies, to facilitate direct comparison of various technologies. Based on the collective results of this program, it appears that hybrid geothermal power would benefit more from recovery of lithium and other metals, rather than REE. It is recommended that future studies be conducted at a higher- TRL and that sorbents be tested against actual geothermal fluid samples. Prior higher-TRL efforts to extract metals from geothermal brines should be further evaluated for lessons learned.

36 MATERIALS SCIENCE↗
DOE OSTI2024

Recruiting Military Talent through Solar Ready Vets Training Partnerships Webinar

SkillBridge training programs provide opportunities for service members to gain valuable civilian work experience through career training prior to separation. Employers are invited to join this webinar hosted by Solar Energy Industries Association to learn more about these programs, featuring success stories from across the industry.

Cohen Hall, E'Lon↗
DOE OSTI2021

Evaluating Probability of Containment Effectiveness at a GCS Sites using integrated assessment modeling approach with Bayesian decision Networks

Improved scientific and engineering understanding of the behavior of geologic CO2 storage together with established regulatory framework and incentive structures raise the prospects for accelerated, large-scale deployment of this greenhouse gas emissions reduction approach. Incentive structures call for the establishment of appropriate verification and accounting approaches to support claims of the integrity of a geologic storage complex and to justify taking credit for long-term storage. In this study, we present a framework for assessing the probability of containment effectiveness over the lifetime of a geologic carbon storage site (e.g., after 70 years of injection and post-injection site performance) using forward stochastic model realizations based on site characterization data and using a monitoring-informed Bayesian network based on hypothetical detectability from surface seismic surveys over the site injection and post-injection phases. The National Risk Assessment Partnership’s open-source Integrated Assessment Model (NRAP-Open-IAM) was utilized to develop an ensemble of 10,000 a priori stochastic forecasts of CO2 containment. Those simulations were used to train the Bayesian network model to estimate the prior probabilities of the CO2 leakage mass into overlying, monitorable aquifers considering the uncertainties in the reservoir properties, permeability of potentially leaky wells and the overlying aquifers. The conditional probabilities in the Bayesian network were either learned from the NRAP-Open-IAM simulations or derived from the predefined detection thresholds for the monitoring method. Observations obtained from monitoring, over time during the site operation phases were then used to generate updated posterior probabilities of containment (and any loss from containment) in the Bayesian network by propagating the prior probabilities through the conditional probabilities. We demonstrate how to construct and use the Bayesian network for verifying the long-term storage complex effectiveness informed by monitoring based on the NRAP-Open-IAM simulations previously developed for the FutureGen 2.0 site. This approach may have relevance for stake holders to demonstrate secure geologic storage, provide a defensible, probabilistic approach to claim credit for geologic storage, and to estimate the likelihood that any fraction of the claimed credit may need to be refunded to the creditor based on available monitoring information.

Bayesian network, Risk assessment, Monitoring, car↗
DOE OSTI2022

Generalizability analysis of tool condition monitoring ensemble machine learning models

Tool condition monitoring (TCM) is an essential research area for the optimization and automation of metal machining processes, and could help manufacturers reduce costs, production time, machine downtime, energy use, and part scrappage. However, TCM systems developed in prior studies have struggled to reach the high level of generalizability which is necessary for industrial applications. This study addresses TCM system generalizability to new machining conditions, how variations in machining and environmental conditions may be used to improve model generalizability, and ensemble machine learning techniques for TCM. Further, milling tool life experiments were conducted using various machining conditions, and the processes' sound, spindle power, and axial load signals were collected. Different machine learning models were evaluated for the prediction of tool wear levels, including four individual models and five ensemble models. Changes in cutting speed were found to display a large effect on model performance, while the chip load showed some effect, and the feed rate had little effect. A simulated noise data augmentation technique for model improvement is applied within TCM for the first time, and resulted in increased model generalizability and reduced overfitting. Across several performance metrics the extremely randomized trees ensemble machine learning model generally performed the best for this application, achieving a leave-one-group-out cross validation accuracy score of 92.4 %, a 10-fold cross validation score of 98.9 %, and an averaged accuracy across 11 generalizability tests of 87.3 %.

42 ENGINEERING↗
DOE OSTI2020

Physics-constrained Deep Recurrent Neural Models of Building Thermal Dynamics

We develop physics-constrained and control-oriented predictive deep learning models for the thermal dynamics of a real-world commercial office building. The proposed method is based on the systematic encoding of physics-based prior knowledge into a structured recurrent neural architecture. Specifically, our model mimics the structure of the building thermal dynamics model and leverages penalty methods to model inequality constraints. Additionally, we use constrained matrix parameterization based on the Perron-Frobenius theorem to bound the eigenvalues of the learned network weights. We interpret the stable eigenvalues as dissipativeness of the learned building thermal model. We demonstrate the effectiveness of the proposed approach on a dataset obtained from an office building with $20$ thermal zones.

Building Energy, structured neural networks↗
DOE OSTI2022

Sequential Decision Making (SDM) for Mesh Refinement and Model Selection in Multiscale, Multi-Physics Applications

Intelligent automation and decision support are needed to enhance computational efficiency and robustness in multiscale and multi-physics problems, including materials science, manufacturing, and climate and weather modeling. Current scientific computing approaches for enabling decisions by scientists fail to explore the role of learning, reasoning, and probabilistic planning. Often these decisions are not performed in real-time during the computation but are made prior to the start of the computation, which must be interrupted in order to make changes to the prior choices. Such interruptions at different stages of the computation increase the total computing time and the need for a human expert to frequently monitor the results. State of art scientific computing methods consist of rule-based algorithms that cannot automatically adapt to a dynamically changing computing environment. The development of a Sequential Decision Making (SDM) framework will automate scientific computing by optimizing the policies for mesh refinement, time-stepping, model and algorithm selection, resource allocation, and pre and post-processing. Our agent SDM framework for scientific computing will consist of data-driven learning (Classifier), automated reasoning (contextual knowledge), and probabilistic planning (Reinforcement Learning). In this project, we focused on three problems to demonstrate our SDM framework on a set of ordinary and partial differential equations. Classification of Lorenz system regions using Feed-Forward Neural Networks examined learning in the SDM framework. On the other hand, reasoning and planning in the SDM framework were used in two problems: adaptive time-stepping for nonlinear ODEs using on-policy RL algorithms, and adaptive mesh refinement for 2-D PDEs using off-policy RL algorithms.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Grid-connected heat pump water heater benefits for low-income households in the Southeastern United States

Energy efficient heat pump water heaters (HPWHs) reduce customer energy costs; grid-connected controls further increase HPWH value by enabling energy storage and shifting demand to cheaper, off-peak periods. However, a HPWH’s first cost premium can put it out of reach for low-income customers. This paper will present results of an ongoing study exploring the benefits of load shifting HPWHs for 24 low-income households, aiming to support increased product deployment in this demographic. The study uses EcoPort modules to shift HPWH load for a study sample consisting primarily of low-income adults and seniors in North Carolina. Building upon prior studies, the study’s HPWH operating schedules are designed to maximize shifted energy and minimize participant electricity costs based on local time-of-use rates. The study documents load profiles of grid-connected HPWHs in a less-studied demographic group and explores how certain groups may have particularly flexible loads due to their unique usage profiles. The results are relevant for HPWH product performance in the Southeast. The study will quantify HPWH load shifting performance during North Carolina’s hot summer, temperate shoulder, and occasionally sub-freezing winter seasons, and the study includes installations in conditioned, semi-conditioned, and unconditioned spaces. Finally, study results will also explore how controlled HPWHs can provide value to the regional grid by reducing seasonal peaks via demand response. Lessons learned include HPWH acceptance for low-income and senior users, best practices for maintaining HPWH connectivity among users without prior product experience, and strategies to leverage remote monitoring to ensure optimal operation and enhance HPWH reliability.

Urigwe, Daniela↗
DOE OSTI2021

A physics-informed and hierarchically regularized data-driven model for predicting fluid flow through porous media

This paper presents a new deep learning data-driven model for predicting structure dependent pore-fluid velocity fields in rock. The model is based on a Convolutional Auto-Encoder (CAE) artificial neural network capable of learning from image data generated by direct numerical simulations of fluid flow through pore-structures, such as by Lattice Boltzmann or molecular dynamics methods. The main novelty of the model in comparison to previous CAE-based data-driven approaches consists of three parts. The first is a methodology for decomposing the full-domain of the porous media into sub-regions, or “sub-domains”, in order to reduce the overall size of the CAE, batch process the sub-domains in parallel, and enable the CAE to learn local and generalizable nonlinear mappings of pore-fluid velocities. The second consists of embedding the finite difference solutions of the incompressible Navier-Stokes and continuity equations into convolutional layers prior to the CAE in order to provide the CAE with knowledge of fluid dynamics physics (PhyFlow). The third main novelty is that the training of the CAE is regularized with a hierarchical loss function that encourages the learning of fluid flow patterns (in a way similar to ranked modes in principal component analysis), ranking from most to least important. This is shown to increase the stability in learning, reduce over-fitting, and promote interpretability of the CAE neural network layers (HierCAE). The comprehensive new data-driven model, which we call the PhyFlow-HierCAE model, is shown to exhibit improved accuracy and generalizability of flow field predictions over conventional CAE models, attributable to the embedded physical knowledge and the hierarchical regularization, as well as realize orders of magnitude speed-ups in computation times as a surrogate for the direct numerical simulations. Examples of training and forward predictions on unseen pore-structures are provided and evaluated for data from Lattice Boltzmann and molecular dynamics simulations of pore-fluid flow. The model is shown to be a fast and accurate emulator (or “surrogate”) for predicting effective permeability of unseen pore-structures based on learning from relatively small direct numerical simulation datasets.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2026

MOSAIC-CONUS: A Multimodal, Multi-Temporally Paired Dataset for Earth Sciences

Earth embeddings—vector representations of geographic locations indexed in space and time—are emerging as a unifying interface for geospatial AI. However, their quality depends not only on model design, but on how multimodal Earth observation (EO) data are spatially indexed, temporally aligned, and cross-modally associated during pretraining. We introduce MOSAIC-CONUS (Multimodal Observations with Spatially Aligned Imagery, Urban Points of Interest, In-Situ Measurements and Text Captions), a large-scale EO dataset over the contiguous United States, organized around 250,000 stratified point indices that serve as stable spatial keys across seven modalities: active radar, passive optical imagery, lidar-derived elevation, land cover, functional context, hydrometeorological measurements, and textual summaries. Unlike existing EO datasets, MOSAIC-CONUS introduces four contributions not jointly addressed in prior work: 1. an open-source, large-scale multimodal EO corpus structured around point-indexed data designed to support Earth embedding learning; 2. explicit radar-optical pairing tables spanning twelve temporal alignment regimes, formalizing cross-sensor alignment as a controllable variable for analyzing how temporal mismatch across modalities influences learned embeddings quality; 3. a benchmark suite spanning cross-modal retrieval, annual nightlights regression, and basin-held-out streamflow prediction, positioning MOSAIC-CONUS as a benchmark-ready resource for multimodal AI systems; and 4. a language-based embedding layer through co-registered textual summaries, enabling Earth embeddings to function as a queryable interface for agentic AI systems. The dataset and pairing protocols are publicly released.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2026

A transfer learning approach to energy-efficient control of small and medium-sized commercial buildings

Model-free reinforcement learning (RL) provides a data-driven and adaptive approach to optimize building energy use while satisfying occupant comfort. This powerful tool does not need any prior knowledge about the environment and system it is optimizing and can adapt its policy based on the changes in captures. Like any other data-driven tool, it faces high training costs due to the extensive agent-environment interactions required to capture long-term building dynamics and user comfort. Transfer learning, particularly policy distillation, offers a promising way to accelerate training by leveraging pretrained RL agents in different building and system types. Here, this study investigates online student distillation, in which the student model updates its neural network weights using outputs from teacher models. The work introduces a student distillation strategy designed for efficient knowledge transfer, along with a teacher selection method that ensures high-quality guidance. The approach is validated using a highly calibrated whole building energy model for a small/medium commercial building test facility. Results show substantial reductions in training time and data requirements while surpassing the performance of ASHRAE Guideline 36, an advanced rule-based control strategy. The distilled RL model required 45% less data and achieved 20% higher cumulative rewards than a state-of-the-art RL model, with faster convergence and lower energy consumption. These outcomes demonstrate that effective transfer learning enables a scalable and data-efficient energy management solution for commercial buildings.

ASHRAE guideline 36↗
DOE OSTI2023

Internal calibration of transient kinetic data via machine learning

The temporal analysis of products (TAP) reactor provides a vast amount of transient kinetic information that may be used to describe a variety of chemical features including residence time distributions, kinetic coefficients, number of active sites, reaction mechanism, etc. However, as with any measurement device, the TAP reactor signal is convoluted with noise and drift is common. In order to reduce the uncertainty of the kinetic measurement and any derived parameters or mechanisms, proper preprocessing must be performed prior to any advanced type of analysis. This preprocessing includes baseline correction, i.e., a shift in the voltage response, and calibration, i.e., a scaling of the flux response based on prior experiments. The traditional methodology of preprocessing requires significant user discretion and reliance on separate calibration experiments that may drift over time. Herein we use machine learning techniques combined with physical constraints to understand the noise and drift that is being generated within and between experiments for enhancement of the chemical kinetic signal. As such, the proposed methodology demonstrates clear benefits over the traditional preprocessing approach by eliminating the need for separate calibration experiments or heuristic input from the user.

36 MATERIALS SCIENCE↗