Search NASA⌕ Search

SEARCH · Search NASA

Results for “learning based systems”

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 991 records · Page 55

Cluster-Graph Fingerprinting: A Framework for Quantitative Analysis of Machine-Learned Interatomic Model Training and Simulation Data

Machine-learned interatomic models represent a significant advancement in simulation methods, extending the predictive ability of first-principles methods to previously inaccessible length and time scales. However, the data-driven nature of these models can lead to difficult-to-detect errors that can compromise prediction accuracy. To address this challenge, we introduce a novel fingerprinting approach based on the Chebyshev Interaction Model for Efficient Simulation (ChIMES) ML-IAM graph-based descriptor. Our strategy enables efficient and statistically rigorous analysis of system configurations used in ML-IAM training and those generated by their application, e.g., in molecular dynamics simulations. We demonstrate that these fingerprints can effectively assess novelty of a configuration relative to an existing data set and determine dissimilarity among individual configurations, which are two key tasks in workflows for active learning-based ML-IAM training, data set curation, and on-the-fly uncertainty quantification.

36 MATERIALS SCIENCE↗

Generative learning for slow manifolds and bifurcation diagrams

In dynamical systems characterized by separation of time scales, the approximation of so called “slow manifolds”, on which the long term dynamics lie, is a useful step for model reduction. Initializing on such slow manifolds is a useful step in modeling, since it circumvents fast transients, and is crucial in multiscale algorithms (like the equation-free approach) alternating between fine scale (fast) and coarser scale (slow) simulations. In a similar spirit, when one studies the infinite time dynamics of systems depending on parameters, the system attractors (e.g., its steady states) lie on bifurcation diagrams (curves for one-parameter continuation, and more generally, on manifolds in state parameter space. Sampling these manifolds gives us representative attractors (here, steady states of ODEs or PDEs) at different parameter values. Algorithms for the systematic construction of these manifolds (slow manifolds, bifurcation diagrams) are required parts of the “traditional” numerical nonlinear dynamics toolkit. In more recent years, as the field of Machine Learning develops, conditional score-based generative models (cSGMs) have been demonstrated to exhibit remarkable capabilities in generating plausible data from target distributions that are conditioned on some given label. It is tempting to exploit such generative models to produce samples of data distributions (points on a slow manifold, steady states on a bifurcation surface) conditioned on (consistent with) some quantity of interest (QoI, observable). In this work, we present a framework for using cSGMs to quickly (a) initialize on a low-dimensional (reduced-order) slow manifold of a multi-time-scale system consistent with desired value(s) of a QoI (a “label”) on the manifold, and (b) approximate steady states in a bifurcation diagram consistent with a (new, out-of-sample) parameter value. This conditional sampling can help uncover the geometry of the reduced slow-manifold and/or approximately “fill in” missing segments of steady states in a bifurcation diagram. Finally, the quantity of interest, which determines how the sampling is conditioned, is either known a priori or identified using manifold learning-based dimensionality reduction techniques applied to the training data.

Dynamical systems↗

Securing Grid-interactive Efficient Buildings (GEB) through Cyber Defense and Resilient System (CYDRES)

The DOE CYDRES project is driven by the urgent need to address critical research gaps in the domain of cyber-physical security of smart buildings, including Grid-interactive Efficient Buildings (GEBs). CYDRES, a real-time advanced building resilient platform, aims to enhance the cyber-attack-immune capabilities of buildings through multi-layered prevention, detection, and adaptation mechanisms. CYDRES consists of five key modules: a multi-layer network analyzer, an Automatic Fault Detection, Diagnosis, and Prognosis (AFDDP) framework, an intelligent mode selector, a cyber-resilient control framework, and a situation awareness platform. The Network Analyzer employs a data-driven framework that includes a protocol state learning tool and a CRF (Conditional Random Field) command validator. In Hardware-In-the-Loop (HIL) testbeds, it achieved 100% detection accuracy with a false alarm rate of 3%, validating its efficacy in identifying selected cyber-attacks. The AFDDP framework leverages pattern matching, PCA (Principal Component Analysis)-based strategies, and a DBN (Dynamic Bayesian Network)-based fault diagnosis approach to pinpoint the causes of physical system abnormalities using Building Automation System (BAS) data. In HIL experiments, the AFDDP module attained a detection accuracy of over 95% with a false alarm rate below 7%. Additionally, the fault detector utilized machine learning (Random Forest) and deep learning (Multi-Layer Perceptron) methods with acoustic sensor data to achieve a 100% fault detection accuracy in Heating, Ventilation, and Air-Conditioning (HVAC) equipment. The Mode Selector offered real-time impact analysis, allowing immediate actions to protect BASs in the face of emerging threats. The cyber-resilient control framework included an adaptive Model Predictive Control (MPC) and a measurement compensator, reducing temperature violations by up to 94% and improving the total demand flexibility by up to 70% in HIL experiments. Such HIL experiments covered a cyber-attack case and a physical fault case, showcasing CYDRES’ efficiency in maintaining operational continuity during threats. The situation awareness platform in Grafana enhanced real-time threat detection and response visualization, augmenting the operational awareness for building operators. CYDRES demonstrated high technical effectiveness in various test scenarios, particularly in HIL environments. The project's phased development approach ensured efficient use of resources, highlighting its practical feasibility and readiness for commercialization. By enhancing the security and resilience of building operations, CYDRES represents a significant advance in mitigating risks associated with cyber-physical systems, thereby enhancing public confidence in the safety of modern building infrastructure. Future directions for the project include expanding testing protocols, refining AFDDP methodologies, exploring more comprehensive resilient control strategies, and testing in real commercial buildings.

42 ENGINEERING↗

2022 American Conference on Neutron Scattering (ACNS 2022)

The 11th American Conference on Neutron Scattering (ACNS 2022) will be held on June 5-9, 2022, in Boulder, CO. The Conference will provide essential information on the breadth and depth of current neutron-related research worldwide. Hosted by the Neutron Scattering Society of America, this year’s Conference will feature a combination of invited and contributed talks, poster sessions, and tutorials. Topics of the conference are: Advances in Neutron Facilities, Instrumentation and Software: Developments in sources, instrumentation, sample environments and control software. Hard Condensed Matter: Magnetism, correlated metals, quantum/topological materials, superconductors, ferroelectrics, multiferroics, glasses, and disorder phenomena. Submissions outlining examples of neutron scattering in industrial and engineering applications involving hard condensed matter systems are also encouraged. Soft Matter: Neutron studies of soft materials and related fields including in situ and in operando studies. Polymers, surfactants, emulsions, gels, nanoparticles, colloidal suspensions and more. Submissions of computational studies or applications of machine learning beneficial to neutron scattering experiments, as well as examples of neutron scattering in industrial and engineering applications are strongly encouraged. Biology, Biophysics and Biotechnology: Neutron studies of biological and biologically relevant systems. Proteins, bio membranes, biological assemblies, natural materials, nucleic acids, drug-delivery platforms and biomedical systems. Submissions of computational studies or applications of machine learning beneficial to biological neutron scattering experiments, as well as examples of neutron scattering in applied research involving biological systems, are strongly encouraged. Materials Chemistry and Energy: Neutron-based studies of functional materials and materials for energy applications. Examples include porous materials such as metal organic frameworks (MOFs), zeolites; phosphors; novel pigments; electrolytes; catalysts; ionic conductors/cathode materials; photovoltaic materials (hybrid perovskites); thermoelectrics; magnetocalorics/electrocalorics. Structural Materials and Engineering: Neutron scattering studies of materials and engineering processes including structural materials, concrete and metals, as well as engineering processes including combustion, corrosion, additive manufacturing, and others. Neutron Physics: Fundamental physical studies of the neutron and related areas. Emerging Applications in Neutron Scattering: Machine Learning and Data Science: Advances in computing power have contributed to rapidly evolving machine learning and data science fields that can be leveraged to the benefit of the neutron scattering community. The purpose of this session is to highlight recent advances in machine learning and data science and to serve as the foundation of a parallel data and computation track highlighting computation advances and applications in neutron scattering throughout the conference.

36 MATERIALS SCIENCE↗

Advances in genetic tools for metabolic engineering of non-conventional yeasts

Non-conventional yeasts are emerging as powerful alternatives to Saccharomyces cerevisiae for metabolic engineering, owing to their innate stress tolerance, broad substrate utilization, and distinctive metabolic capabilities. These attributes position them as promising chassis for producing biofuels, pharmaceuticals, and specialty chemicals. This review synthesizes recent advances in genetic toolkits for four such species—Pichia kudriavzevii (Issatchenkia orientalis), Starmerella bombicola, Debaryomyces hansenii, and Pachysolen tannophilus—highlighting progress across plasmid architectures (episomal and integrative), identification of autonomously replicating sequences and centromeric elements, and the development of safe-harbor genomic loci. We summarize promoter and terminator libraries enabling tunable expression, the expansion of auxotrophic and antifungal selection markers with recycling strategies, and the rapid adaptation of CRISPR-based systems (Cas9 and Cas12a) with optimized guide RNA expression, multiplex editing, and approaches that enhance homologous recombination (e.g., KU70/80 disruption). We also review landing-pad platforms for modular, repeated integrations and transposon-based tools (e.g., piggyBac) that facilitate multigene pathway assembly. Collectively, these innovations are accelerating design-build-test-learn cycles and enabling precise, scalable engineering of non-conventional yeasts. Remaining challenges—including limited species-specific episomal systems, variable transformation efficiencies, genome-stability concerns, and alternative codon usage—define clear priorities for future toolkit development. Together, these advances and open needs chart a path toward robust, sustainable biomanufacturing using diverse non-conventional yeast chassis.

59 BASIC BIOLOGICAL SCIENCES↗

A Barrier-Certificated Reinforcement Learning Approach for Enhancing Power System Transient Stability

Increasing integration of renewable resources brings more flexibility and poses new challenges to modern power systems, leading to highly nonlinear and complex dynamics. Here, this paper aims to provide a general solution framework to traditional control problems, such as frequency control and voltage control, which attempt to maintain the stability of either synchronous generators-governed or inverter-governed systems when subjected to a disturbance and simultaneously guarantee operational constraints, providing a complete complement to existing works on control design. Building on reinforcement learning (RL) and control barrier functions, the framework includes two subsystems, i.e., a model-free controller and a barrier-certification system, which discover RL-based control actions and sequentially filter them using a barrier certificate to satisfy operational constraints. Calculating a barrier function is generally challenging for a complex power system. This is addressed by representing the barrier function using neural networks (NNs) and data-based approaches. An adaptive method is introduced to certify the neural barrier function that perseveres barrier conditions, which is more compatible with online implementation. The proposed framework synthesizes a stabilizing controller that satisfies predefined safety regions. The effectiveness of the proposed framework is demonstrated via several comparative case studies.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Transient Data Library of Solar Grid Integrated Distributed System

This submission contains an open-source library of transient events in distributed system with high solar PV. The library includes the collected data, related documents and scripts for loading the data. The data library is built for transient event detection and machine learning based analysis algorithm development. The data was collected via both field test and software simulation. The units for the data are included in the data file headers for each data series. A text editor or spreadsheet software, such as Excel, and Matlab is required to view the data.

algorithms↗

Survival analysis of localized prostate cancer with deep learning

In recent years, data-driven, deep-learning-based models have shown great promise in medical risk prediction. By utilizing the large-scale Electronic Health Record data found in the U.S. Department of Veterans Affairs, the largest integrated healthcare system in the United States, we have developed an automated, personalized risk prediction model to support the clinical decision-making process for localized prostate cancer patients. This method combines the representative power of deep learning and the analytical interpretability of parametric regression models and can implement both time-dependent and static input data. To collect a comprehensive evaluation of model performances, we calculate time-dependent C-statistics C td over 2-, 5-, and 10-year time horizons using either a composite outcome or prostate cancer mortality as the target event. The composite outcome combines the Prostate-Specific Antigen (PSA) test, metastasis, and prostate cancer mortality. Our longitudinal model Recurrent Deep Survival Machine (RDSM) achieved C td 0.85 (0.83), 0.80 (0.83), and 0.76 (0.81), while the cross-sectional model Deep Survival Machine (DSM) attained C td 0.85 (0.82), 0.80 (0.82), and 0.76 (0.79) for the 2-, 5-, and 10-year composite (mortality) outcomes, respectively. In addition to estimating the survival probability, our method can quantify the uncertainty associated with the prediction. The uncertainty scores show a consistent correlation with the prediction accuracy. We find PSA and prostate cancer stage information are the most important indicators in risk prediction. Our work demonstrates the utility of the data-driven machine learning model in prostate cancer risk prediction, which can play a critical role in the clinical decision system.

60 APPLIED LIFE SCIENCES↗

Adaptive Data-Driven Model Predictive Control for Heat Pipe Microreactors

To establish a technical basis for self-regulating microreactors, a model predictive control (MPC) system is investigated to proactively respond to anomalies and disturbances in anticipation of potential deviations from operating setpoints. Due to the difficulty of developing a physics-based surrogate model that can accurately match plant data in various operating conditions, machine learning algorithms are used in MPC, which allow for learning from both simulation and operation data, thus efficiently describing the targeted transient with arbitrary accuracy. However, one of the biggest concerns in applying ML algorithms like artificial neural networks (ANNs) is that the predictive capabilities of ANN are limited by training data. If there are gaps between the training and target domain, the accuracy of an ANN can degrade significantly when it is used to predict unseen data. To improve the predictive capability of ANN and enable a confident use of data-driven MPCs outside the training data, this study proposes an adaptive data-driven MPC framework. The system will monitor the discrepancy between plant responses and surrogate predictions, fine-tune the ANN-based surrogate when a large discrepancy is detected, and continue MPC operation with updated surrogates. The framework is demonstrated on a point kinetic model for microreactors. The hyperparameters of the update strategy, including layers to update, error thresholds, learning rate discount, and number of data points used for fine-tuning, are optimized so the simulated microreactor is able to follow changes in setpoint with the smallest of deviations.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Adaptable Data Driven Model Predictive Control for Heat Pipe Microreactors

To establish a technical basis for self-regulating microreactors, a model predictive control (MPC) system is investigated to proactively respond to anomalies and disturbances in anticipation of potential deviations from operating setpoints. Due to the difficulty of developing a physics-based surrogate model that can accurately match plant data in various operating conditions, machine learning algorithms are used in MPC, which allow for learning from both simulation and operation data, thus efficiently describing the targeted transient with arbitrary accuracy. However, one of the biggest concerns in applying ML algorithms like artificial neural networks (ANNs) is that the predictive capabilities of ANN are limited by training data. If there are gaps between the training and target domain, the accuracy of an ANN can degrade significantly when it is used to predict unseen data. To improve the predictive capability of ANN and enable a confident use of data-driven MPCs outside the training data, this study proposes an adaptive data-driven MPC framework. The system will monitor the discrepancy between plant responses and surrogate predictions, fine-tune the ANN-based surrogate when a large discrepancy is detected, and continue MPC operation with updated surrogates. The framework is demonstrated on a point kinetic model for microreactors. The hyperparameters of the update strategy, including layers to update, error thresholds, learning rate discount, and number of data points used for fine-tuning, are optimized so the simulated microreactor is able to follow changes in setpoint with the smallest of deviations.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Short-Term Probabilistic Solar Forecasting via Reinforcement Learning over ECMWF

In this paper, we present an innovative reinforcement learning approach for short-term solar forecasting, leveraging data from the European Centre for Medium-Range Weather Forecasts (ECMWF). The methodology begins with the application of the System Advisor Model (SAM) to transform various ECMWF numerical weather prediction members into predictive photovoltaic power generation. To enhance the precision of deterministic forecasting, we introduce a dynamic model selection algorithm based on Q-learning. This algorithm dynamically identifies and utilizes the most accurate ensemble member for forecasting purposes. Furthermore, we employ a support vector regression surrogate model with a Gaussian distribution to generate probabilistic forecasts, providing a holistic view of solar energy generation uncertainty. To expedite the training process and make it more practical for real-world applications, we integrate a rolling update workflow. This innovative workflow reduces the training period from months to a mere 19 days, making our method highly efficient. Numerical results of the case study show that in comparison to benchmark models, the proposed method improves the deterministic and probabilistic solar forecasting accuracy by up to 40.84% and 48.42%, respectively.

ensemble forecasting↗

Tree-Based Ensemble Learning Models for Wall Temperature Predictions in Post-Critical Heat Flux Flow Regimes at Subcooled and Low-Quality Conditions

Accurately predicting post-critical heat flux (CHF) heat transfer is an important but challenging task in water-cooled reactor design and safety analysis. Although numerous heat transfer correlations have been developed to predict post-CHF heat transfer, these correlations are only applicable to relatively narrow ranges of flow conditions due to the complex physical nature of the post-CHF heat transfer regimes. In this paper, a large quantity of experimental data is collected and summarized from the literature for steady-state subcooled and low-quality film boiling regimes with water as the working fluid in vertical tubular test sections. In addition, a low-quality water film boiling (LWFB) database is consolidated with a total of 22,813 experimental data points, which cover a wide flow range of the system pressure from 0.1 to 9.0 MPa, mass flux from 25 to 2750 kg/m 2 s, and inlet subcooling from 1 to 70 °C. Two machine learning (ML) models, based on random forest (RF) and gradient boosted decision tree (GBDT), are trained and validated to predict wall temperatures in post-CHF flow regimes. The trained ML models demonstrate significantly improved accuracies compared to conventional empirical correlations. To further evaluate the performance of these two ML models from a statistical perspective, three criteria are investigated and three metrics are calculated to quantitatively assess the accuracy of these two ML models. For the full LWFB database, the root-mean-square errors between the measured and predicted wall temperatures by the GBDT and RF models are 5.7% and 6.2%, respectively, confirming the accuracy of the two ML models.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Machine-Learning-Based High-Resolution Earthquake Catalog Reveals How Complex Fault Structures Were Activated during the 2016–2017 Central Italy Sequence

The 2016–2017 central Italy seismic sequence occurred on an 80 km long normal-fault system. The sequence initiated with the Mw 6.0 Amatrice event on 24 August 2016, followed by the Mw 5.9 Visso event on 26 October and the Mw 6.5 Norcia event on 30 October. We analyze continuous data from a dense network of 139 seismic stations to build a high-precision catalog of ~900,000 earthquakes spanning a 1 yr period, based on arrival times derived using a deep-neural-network-based picker. Our catalog contains an order of magnitude more events than the catalog routinely produced by the local earthquake monitoring agency. Aftershock activity reveals the geometry of complex fault structures activated during the earthquake sequence and provides additional insights into the potential factors controlling the development of the largest events. Activated fault structures in the northern and southern regions appear complementary to faults activated during the 1997 Colfiorito and 2009 L’Aquila sequences, suggesting that earthquake triggering primarily occurs on critically stressed faults. Delineated major fault zones are relatively thick compared to estimated earthquake location uncertainties, and a large number of kilometer-long faults and diffuse seismicity were activated during the sequence. These properties might be related to fault age, roughness, and the complexity of inherited structures. The rich details resolvable in this catalog will facilitate continued investigation of this energetic and well-recorded earthquake sequence.

58 GEOSCIENCES↗

HPC Analytics of Fused Thermal Plants Data to Optimize Operating Envelope

In this project, ORNL extensively reviewed the ORAP RAM data, and it guided us to develop machine learning models that can predict time to next failures and forecast failure trends, which will be useful for optimizing power plant operation strategies. More specifically, we trained multiple random forest models and evaluated the model accuracy to validate with 10+ years of historical data. In addition, we implemented a web-based graphical user interface system for the models to show how our models can be used in more intuitive ways. This proof of concept allowed exploration of model use with power plant operators in mind. Developed machine learning models will be helpful for managing risks, planning maintenance and operation, ultimately reducing the down time and increasing the service hours. For future work, there are several interesting research topics including but not limited to model enhancement, creating synergy with traditional failure modeling approaches, and data-driven actionable recommendation and suggestions.

20 FOSSIL-FUELED POWER PLANTS↗

An Edge-Cloud Integrated Solution for Buildings Demand Response Using Reinforcement Learning

Buildings, as major energy consumers, can provide great untapped demand response (DR) resources for grid services. However, their participation remains low in real-life. One major impediment for popularizing DR in buildings is the lack of cost-effective automation systems that can be widely adopted. Existing optimization-based smart building control algorithms suffer from high costs on both building-specific modeling and on-demand computing resources. To tackle these issues, this paper proposes a cost-effective edge-cloud integrated solution using reinforcement learning (RL). Beside RL’s ability to solve sequential optimal decision-making problems, its adaptability to easy-to-obtain building models and the off-line learning feature are likely to reduce the controller’s implementation cost. Using a surrogate building model learned automatically from building operation data, an RL agent learns an optimal control policy on cloud infrastructure, and the policy is then distributed to edge devices for execution. Simulation results demonstrate the control efficacy and the learning efficiency in buildings of different sizes. A preliminary cost analysis on a 4-zone commercial building shows the annual cost for optimal policy training is only 2.25% of the DR incentive received. Results of this study show a possible approach with higher return on investment for buildings to participate in DR programs.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Environmentally Assisted Fatigue in Light Water Reactor Environment

This report summarizes the Environmentally Assisted Fatigue (EAF) research conducted at ANL under the US DOE Light Water Reactor Sustainability (LWRS) program. Starting from a rich background in theoretical and experimental EAF, ANL previously developed an approach to evaluate fatigue performance of reactor materials in light water reactor environments with the correction factor F en . The approach was based on a large body of experimental work performed at ANL and elsewhere, and was consistent with American Society of Mechanical Engineers (ASME)’s methodology governing the design and construction of reactor components. In recent years, the program was focused on component fatigue prediction and made several major and fundamental contributions in this area. These accomplishments help meet the needs identified by the industry concerning component level fatigue predictions in complex, transient conditions. The main contribution of the ANL program involved the development of a system-level model for estimating residual strain and life of nuclear reactor coolant system components under connected-system-thermal-mechanical boundary conditions. The goal was to predict the stress hotspots, strain residuals, strain amplitudes and the resulting fatigue lives. Thermal-mechanical stress analysis was performed considering thermal stratification and a design-basis reactor loading cycle. Based on the finite element (FE) model results, the strain residuals, strain amplitudes and resulting fatigue lives of reactor coolant system (RCS) components were predicted. The results show that some of the RCS components can have significantly different strain amplitudes, residual strain, and fatigue lives, despite having similar geometry and material. In addition, the simulated component-level strain profile can guide the selection of appropriate test inputs for conducting laboratory-scale EAF tests. Building upon the system-level model, ANL developed a digital twin (DT) framework to predict the structural states and associated fatigue life of components in real-time. This framework is a comprehensive system designed to predict the structural states and fatigue lives of reactor components. It includes multiple models and integrates artificial intelligence (AI), machine learning (ML), and FE based modeling tools to evaluate the structural states and fatigue lives.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Machine Learning Based Online Performance Prediction for Runtime Parallelization and Task Scheduling

With the emerging many-core paradigm, parallel programming must extend beyond its traditional realm of scientific applications. Converting existing sequential applications as well as developing next-generation software requires assistance from hardware, compilers and runtime systems to exploit parallelism transparently within applications. These systems must decompose applications into tasks that can be executed in parallel and then schedule those tasks to minimize load imbalance. However, many systems lack a priori knowledge about the execution time of all tasks to perform effective load balancing with low scheduling overhead. In this paper, we approach this fundamental problem using machine learning techniques first to generate performance models for all tasks and then applying those models to perform automatic performance prediction across program executions. We also extend an existing scheduling algorithm to use generated task cost estimates for online task partitioning and scheduling. We implement the above techniques in the pR framework, which transparently parallelizes scripts in the popular R language, and evaluate their performance and overhead with both a real-world application and a large number of synthetic representative test scripts. Our experimental results show that our proposed approach significantly improves task partitioning and scheduling, with maximum improvements of 21.8%, 40.3% and 22.1% and average improvements of 15.9%, 16.9% and 4.2% for LMM (a real R application) and synthetic test cases with independent and dependent tasks, respectively.

99 GENERAL AND MISCELLANEOUS↗

DRAS: Deep Reinforcement Learning for Cluster Scheduling in High Performance Computing

Cluster schedulers are crucial in high-performance computing (HPC). They determine when and which user jobs should be allocated to available system resources. Existing cluster scheduling heuristics are developed by human experts based on their experience with specific HPC systems and workloads. However, the increasing complexity of computing systems and the highly dynamic nature of application workloads have placed tremendous burden on manually designed and tuned scheduling heuristics. More aggressive optimization and automation are needed for cluster scheduling in HPC. In this work, we present an automated HPC scheduling agent named DRAS (Deep Reinforcement Agent for Scheduling) by leveraging deep reinforcement learning. DRAS is built on a hierarchical neural network incorporating special HPC scheduling features such as resource reservation and backfilling. An efficient training strategy is presented to enable DRAS to rapidly learn the target environment. Once being provided a specific scheduling objective given by the system manager, DRAS automatically learns to improve its policy through interaction with the scheduling environment and dynamically adjusts its policy as workload changes. We implement DRAS into a HPC scheduling platform called CQGym. CQGym provides a common platform allowing users to flexibly evaluate DRAS and other scheduling methods such as heuristic and optimization methods. Furthermore, the experiments using CQGym with different production workloads demonstrate that DRAS outperforms the existing heuristic and optimization approaches by up to 50%.

97 MATHEMATICS AND COMPUTING↗