Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “distributed machine learning”

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 1,405 records · Page 78
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

Embedding Neural Thermal Scattering (NeTS) Modules in SERPENT for Higher Fidelity Advanced Reactor Analysis

When a neutron born in fission thermalizes to the order of $k$ $B$ $T$, it’s de-Broglie wavelength and energy approach the order of inter-atomic spacing and elementary lattice oscillations, respectively. $S$($a,β,t$) or the scattering law, uuantify these temperature-dependent crystallographic contributions to total cross section (or reaction rate). In a Monte Carlo analysis, cumulative distribution functions (CDFs) of $S$($a,β,t$) are loaded to memory from “A Compact ENDF” (ACE) files for stochastically selecting thermal scattered neutron trajectories. In this work, novel neural thermal scattering (NeTS) modules for $S$($a,β,t$) CDFs are designed, trained, serialized and embedded within SERPENT using Python’s limited C-API for on-the-fly deployment of crystalline graphite $S$($a,β,t$) sampling. Torchscript tracing and Numba just-in-time (JIT) compilation streamline neural inference on NVIDIA GPUs with CUDA libraries. Demonstrations of bare sphere thermalization of fast and thermal sources show excellent agreement between embedded NeTS in SERPENT and MCNP. With an explicit model of the reactor, NeTS can predict on-the-fly changes in TREAT neutron spectra as a function of local temperature, which can serve to improve transient and accident predictions in a multiphysics analysis framework. This framework can be further extended to account on-the-fly for changes in local graphitic microstructure to scattering cross sections, and outlines a novel coupling of modern machine learning with state-of-the-art reactor physics methods.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS2020

California & Oregon Ecological Forecasting: Detecting and Forecasting Fog Occurrence, Frequency, and Change to Support Coast Redwood (Sequoia sempervirens) Habitat Assessments

Fog and low clouds play an important role in providing moisture to coastal ecosystems. Coast redwood (Sequoia sempervirens) forests are currently distributed along a narrow strip of coastline in California and Oregon and rely on the presence of marine fog for moisture availability during the dry season (June-October). Recent time series analyses presented an uncertain future of fog frequency; however, a decline in fog presence may have adverse effects on the coast redwood habitat. To support Save the Redwoods League, a non-profit organization dedicated to coast redwood forest management, the team analyzed hourly fog data from the Geospatial Operational Environmental Satellite 17 (GOES-17) Advanced Baseline Imager (ABI) and daily cloud cover data from the Moderate Resolution Imaging Spectroradiometer (MODIS) aboard the Terra satellite. To explore present day fog longevity, GOES-17 was utilized to map the number of fog hours per day for the 2019 and 2020 dry seasons. The MODIS cloud flag was used to map the presence or absence of daily fog, which was summarized to create a monthly fog frequency dataset and identify trends in fog presence between 2000-2020. Both datasets were used as inputs into the random forest machine learning algorithm to identify climatic drivers of fog presence and longevity over the landscape. The present-day models suggested that daily temperature difference is a driving force behind fog presence and longevity. Trends in fog presence from 2000-2020 indicated great interannual variability. Finally, fog presence was modeled under a 2080 climate projection to shed light on the future of fog presence under a projected warmer climate. Model results projected an overall decline in fog presence during the dry season in the 2080s. Decreased fog presence as a result of increased temperature difference under a warmer climate remains to be a topic of investigation as to the impact on future redwood habitat suitability.

DEVELOP Project Summary↗
NASA NTRS1991

From biological neural networks to thinking machines: Transitioning biological organizational principles to computer technology

The three-dimensional organization of the vestibular macula is under study by computer assisted reconstruction and simulation methods as a model for more complex neural systems. One goal of this research is to transition knowledge of biological neural network architecture and functioning to computer technology, to contribute to the development of thinking computers. Maculas are organized as weighted neural networks for parallel distributed processing of information. The network is characterized by non-linearity of its terminal/receptive fields. Wiring appears to develop through constrained randomness. A further property is the presence of two main circuits, highly channeled and distributed modifying, that are connected through feedforward-feedback collaterals and biasing subcircuit. Computer simulations demonstrate that differences in geometry of the feedback (afferent) collaterals affects the timing and the magnitude of voltage changes delivered to the spike initiation zone. Feedforward (efferent) collaterals act as voltage followers and likely inhibit neurons of the distributed modifying circuit. These results illustrate the importance of feedforward-feedback loops, of timing, and of inhibition in refining neural network output. They also suggest that it is the distributed modifying network that is most involved in adaptation, memory, and learning. Tests of macular adaptation, through hyper- and microgravitational studies, support this hypothesis since synapses in the distributed modifying circuit, but not the channeled circuit, are altered. Transitioning knowledge of biological systems to computer technology, however, remains problematical.

Ross, Muriel D.↗
DOE OSTI2022

Assessments of epistemic uncertainty using Gaussian stochastic weight averaging for fluid-flow regression

Here, we use Gaussian stochastic weight averaging (SWAG) to assess the epistemic uncertainty associated with neural-network-based function approximation relevant to fluid flows. SWAG approximates a posterior Gaussian distribution of each weight, given training data, and a constant learning rate. Having access to this distribution, it is able to create multiple models with various combinations of sampled weights, which can be used to obtain ensemble predictions. The average of such an ensemble can be regarded as the 'mean estimation', whereas its standard deviation can be used to construct 'confidence intervals', which enable us to perform uncertainty quantification (UQ) with regard to the training process of neural networks. We utilize representative neural-network-based function approximation tasks for the following cases: (i) a two-dimensional circular-cylinder wake; (ii) the DayMET dataset (maximum daily temperature in North America); (iii) a three-dimensional square-cylinder wake; and (iv) urban flow, to assess the generalizability of the present idea for a wide range of complex datasets. SWAG-based UQ can be applied regardless of the network architecture, and therefore, we demonstrate the applicability of the method for two types of neural networks: (i) global field reconstruction from sparse sensors by combining convolutional neural network (CNN) and multi-layer perceptron (MLP); and (ii) far-field state estimation from sectional data with two-dimensional CNN. We find that SWAG can obtain physically-interpretable confidence-interval estimates from the perspective of epistemic uncertainty. This capability supports its use for a wide range of problems in science and engineering.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2022

Data-driven based coordinated smart inverter control for distributed energy resources

Smart inverters (SI) for distributed energy resources (DER) are becoming popular since they have the ability to stabilize as well as restore the voltage and frequency of power systems. Aiming at establishing the mathematical models combined with SI control methods, multiple optimization methods are developed. However, the computational complexity of solving such a mathematical model with various uncertainties limits the real-time application of the SI control. To conquer this challenge, a data-driven-based SI control approach is developed to achieve coordinated control in the high penetration DER system. First, an optimization problem for maximizing the active power generation and minimizing the power loss is designed using the Volt/VAR control. To reduce the time consumption, the recurrent neural network (RNN) is proposed to model the relationship between the uncertainties and control actions during the offline site. The RNN with different sub-structures such as the long short-term memory cell and gated recurrent unit cell are included to enrich the diversity of features. In the last stage, different experiment comparisons, including multiple uncertainties maps and stateof- art machine learning methods, are conducted to verify the effectiveness of the proposed method based on the IEEE 123 bus power system. The results demonstrate that the proposed method can effectively achieve a rapid and coordinated control with a lower error rate.

Qiu, Wei↗
NASA NTRS2000

Autonomous Performance Monitoring System: Monitoring and Self-Tuning (MAST)

Maintaining the long-term performance of software onboard a spacecraft can be a major factor in the cost of operations. In particular, the task of controlling and maintaining a future mission of distributed spacecraft will undoubtedly pose a great challenge, since the complexity of multiple spacecraft flying in formation grows rapidly as the number of spacecraft in the formation increases. Eventually, new approaches will be required in developing viable control systems that can handle the complexity of the data and that are flexible, reliable and efficient. In this paper we propose a methodology that aims to maintain the accuracy of flight software, while reducing the computational complexity of software tuning tasks. The proposed Monitoring and Self-Tuning (MAST) method consists of two parts: a flight software monitoring algorithm and a tuning algorithm. The dependency on the software being monitored is mostly contained in the monitoring process, while the tuning process is a generic algorithm independent of the detailed knowledge on the software. This architecture will enable MAST to be applicable to different onboard software controlling various dynamics of the spacecraft, such as attitude self-calibration, and formation control. An advantage of MAST over conventional techniques such as filter or batch least square is that the tuning algorithm uses machine learning approach to handle uncertainty in the problem domain, resulting in reducing over all computational complexity. The underlying concept of this technique is a reinforcement learning scheme based on cumulative probability generated by the historical performance of the system. The success of MAST will depend heavily on the reinforcement scheme used in the tuning algorithm, which guarantees the tuning solutions exist.

Peterson, Chariya↗
DOE OSTI2023

The entropy of galaxy spectra: how much information is encoded?

Abstract The inverse problem of extracting the stellar population content of galaxy spectra is analysed here from a basic standpoint based on information theory. By interpreting spectra as probability distribution functions, we find that galaxy spectra have high entropy, thus leading to a rather low effective information content. The highest variation in entropy is unsurprisingly found in regions that have been well studied for decades with the conventional approach. We target a set of six spectral regions that show the highest variation in entropy – the 4000 Å break being the most informative one. As a test case with real data, we measure the entropy of a set of high-quality spectra from the Sloan Digital Sky Survey, and contrast entropy-based results with the traditional method based on line strengths. The data are classified into star-forming (SF), quiescent (Q), and active galactic nucleus (AGN) galaxies, and show – independently of any physical model – that AGN spectra can be interpreted as a transition between SF and Q galaxies, with SF galaxies featuring a more diverse variation in entropy. The high level of entanglement complicates the determination of population parameters in a robust, unbiased way, and affects traditional methods that compare models with observations, as well as machine learning (especially deep learning) algorithms that rely on the statistical properties of the data to assess the variations among spectra. Entropy provides a new avenue to improve population synthesis models so that they give a more faithful representation of real galaxy spectra.

Ferreras, Ignacio (ORCID:0000000345843127)↗
DOE OSTI2026

Inverse Mapping of the Collision Kernel and Wall Flux Scaling in a Tall Convection‐Cloud Chamber Using Local Sensors and Knowledge‐Informed Deep Learning

Droplet collision–coalescence is a crucial process in cloud physics, but accurately representing this process under different dynamical conditions remains challenging. A proposed future convective‐cloud chamber aims to investigate this key process, but the method for observing it remains unclear, even though it is theoretically established that collision‐coalescence will occur. This study serves as a proof‐of‐concept demonstration of how knowledge‐informed deep learning, combined with measurement data from local sensors in the chamber, can be used to estimate the collision kernels, which determine how the droplet size distribution evolves during collision‐coalescence. In addition to estimating the collision kernel, we also address wall fluxes, another uncertain but important process that acts as a source of heat and moisture in the chamber. Ensemble runs of large‐eddy simulations are conducted by scaling the wall fluxes and the collision kernel, while the measured flow and cloud properties are used as inputs for a neural network. Results indicate that this approach successfully maps the scaling of wall fluxes and the collision kernel with biases of approximately 1% or less relative to the range of the target data. This proof‐of‐concept lays the groundwork for future applications; when the real measurements are available, real sensor data combined with the trained model presented in this work will enable estimation of the actual wall fluxes and collision kernel.

cloud chamber↗
DOE OSTI2022

A Robust Event Diagnostics Platform: Integrating Tensor Analytics and Machine Learning into Real-time Grid Monitoring

The objective of this project is to develop a robust event diagnostics (RED) platform by integrating state-of-the-art tensor analytics and machine learning into real-time grid monitoring. The proposed platform can effectively analyze and discover the information hiding within the provided PMU data for effective real-time grid monitoring. The proposed RED platform provides a set of robust diagnostics tools for grid operation and management, including 1) data quality assessment, 2) data completion, 3) event detection, and 4) robust event classification. All the functionalities of the RED platform can help the operator to make informed decisions and respond in a timely manner. The developed RED platform will serve as an innovative advisory tool to reliably identify key events and discover new insights about the events and grid characteristics in the PMU data, and contribute to the efficient, safe, reliable operation and design of the nation’s electric system.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2025

A Multi-Sensor Approach for Measuring Bird and Bat Collisions with Offshore Wind Turbines (Final Technical Report)

Collision of birds and bats with wind turbines is a conservation concern for both land-based and offshore wind projects. The fatality rates of birds and bats at land-based turbines are well documented. The measurement strategies on land focus on finding carcasses following collision, estimating the number of carcasses missed through searcher efficiency, carcass persistence trials and carcass fall distributions, and modeling statistically robust fatality rates. Few technologies have been developed to monitor offshore bird and bat collisions, and many that have been developed focused on detecting collisions with large birds. The few studies that have attempted to document collisions at offshore turbines do not account for smaller bodied animals or for collisions that might be missed, which prevents the calculation of statistically robust fatality rates. The overall goal of this report, A Multi-Sensor Approach for Measuring Bird and Bat Collisions with Offshore Wind Turbines (Project), was to develop an effective multi-sensor system for quantifying bird and bat collision rates, specifically for offshore wind facilities. The Project goal and resulting automated collision detection system was achieved through two major technological advancements: 1) refining The Netherlands Organisation for Applied Scientific Research’s (TNO’s) existing WT-Bird® vibration sensing system, that had successfully detected large bird collisions during daytime, to allow for improved detection of smaller birds and bats during both daytime and nighttime hours and 2) improving image processing systems and developing and integrating machine learning algorithms to automatically detect and classify small and large bird and bat collisions with offshore turbines. This final technical report (FTR) summarizes Methods , Results , Conclusions , and Lessons Learned during each of the five Tasks identified for this research and development effort. This FTR includes summaries of the following: Task 1. Initial Engineering Tests to Improve WT-Bird® Task 2. Installation of WT‐Bird® on a Utility-scale Turbine at the National Wind Technology Center – National Renewable Energy Laboratory Task 3. Field Tests and Refinement of the Object Detection System Task 4. Validation of WT-Bird® on a Land-based Turbine Task 5. Preparation for the Implementation of WT-Bird® on an Offshore Turbine. This research and development effort documented successful improvement of the WT Bird® collision detection system to detect small birds and bats, and WT-Bird® is the first collision detection system to validate results compared to land-based post-construction monitoring. The collision trials provide estimates of missed targets that can be used to estimate fatality rates, a significant improvement relative to other offshore collision monitoring systems. Advances were made in developing an edge-processing solution to reduce data storage requirements, which is important if the system is deployed for long periods of time at offshore turbines. The improved WT-Bird® system also provides an important option for wind operators on land or offshore who need to document specific details about when collisions occur, particularly efforts to further research on bat impact minimization, or when standard fatality searches are impractical (e.g. offshore) or inadequate (e.g. challenging locations on land).

17 WIND ENERGY↗
NASA NTRS2020

California & Oregon Ecological Forecasting: Detecting and Forecasting Fog Occurrence, Frequency, and Change to Inform Coast Redwood (Sequoia sempervirens) Habitat Assessments

Fog and low clouds play an important role in providing moisture to coastal ecosystems. Coast redwood (Sequoia sempervirens) forests are currently distributed along a narrow strip of coastline in California and Oregon and rely on the presence of marine fog for moisture availability during the dry season (June-October). Recent time series analyses presented an uncertain future of fog frequency; however, a decline in fog presence may have adverse effects on the coast redwood habitat. To complement ongoing work by Save the Redwoods League, a non-profit organization dedicated to coast redwood forest management, the team analyzed hourly fog data from the Geospatial Operational Environmental Satellite 17 (GOES-17) Advanced Baseline Imager (ABI) and daily cloud cover data from the Moderate Resolution Imaging Spectroradiometer (MODIS) aboard the Terra satellite. To explore present day fog longevity, GOES-17 was utilized to map the number of fog hours per day for the 2019 and 2020 dry seasons. The MODIS cloud flag was used to map the presence or absence of daily fog, which was summarized to create a monthly fog frequency dataset and identify trends in fog presence between 2000-2020. Both datasets were used as inputs into the random forest machine learning algorithm to identify climatic drivers of fog presence and longevity over the landscape. The present-day models suggested that daily temperature difference is a driving force behind fog presence and longevity. Trends in fog presence from 2000-2020 indicated great interannual variability. Finally, fog presence was modeled under a 2080 climate projection to shed light on the future of fog presence under a projected warmer climate. Model results projected an overall decline in fog presence during the dry season in the 2080s. Decreased fog presence as a result of increased temperature difference under a warmer climate remains to be a topic of investigation as to the impact on future redwood habitat suitability.

DEVELOP Technical Paper↗
DOE OSTI2020

On the Solution of ℓ 0 -Constrained Sparse Inverse Covariance Estimation Problems

The sparse inverse covariance matrix is used to model conditional dependencies between variables in a graphical model to fit a multivariate Gaussian distribution. Estimating the matrix from data are well known to be computationally expensive for large-scale problems. Sparsity is employed to handle noise in the data and to promote interpretability of a learning model. Although the use of a convex ℓ 1 regularizer to encourage sparsity is common practice, the combinatorial ℓ 0 penalty often has more favorable statistical properties. In this paper, we directly constrain sparsity by specifying a maximally allowable number of nonzeros, in other words, by imposing an ℓ 0 constraint. Here, we introduce an efficient approximate Newton algorithm using warm starts for solving the nonconvex ℓ 0 -constrained inverse covariance learning problem. Numerical experiments on standard data sets show that the performance of the proposed algorithm is competitive with state-of-the-art methods.

$\ell_0$-Constrained↗
DOE OSTI2022

Control of a Three-Phase Grid-Connected Voltage-Sourced Converter Using Long Short-Term Memory Networks

With the rise of inverter-based resources (IBRs) within the power system, the control of grid-connected converters (GCCs) has become pertinent due to the fact they interface IBRs to the grid. The conventional method of control for a GCC such as the voltage-sourced converter (VSC) is through a decoupled control loop in the synchronous reference frame. However, this model-based control method is sensitive to parameter changes causing deterioration in controller performance. Data-driven approaches such as machine learning can be utilized to design controllers that are capable of operating GCCs in various system conditions. This work explores a deep learning-based control method for a three-phase grid-connected VSC, specifically utilizing a long short-term memory (LSTM) network for robust control. Simulations of a conventional controlled VSC are conducted using Simulink to collect data for training the LSTM-based controller. The LSTM model is built and trained using the Keras and TensorFlow libraries in Python and tested in Simulink. The performance of the LSTM-based controller is evaluated under different case studies and compared to the conventional method of control. Simulation results demonstrate the effectiveness of this approach by outperforming the conventional controller and maintaining stability under different system parameter changes.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2020

Quantifying the Dynamics of Protein Self-Organization Using Deep Learning Analysis of Atomic Force Microscopy Data

The dynamics of protein self-assembly on the inorganic surface and the resultant geometric patterns are visualized using high-speed atomic force microscopy. The time dynamics of the classical macroscopic descriptors such as 2D fast Fourier transforms, correlation, and pair distribution functions are explored using the unsupervised linear unmixing, demonstrating the presence of static ordered and dynamic disordered phases and establishing their time dynamics. Here, the deep learning (DL)-based workflow is developed to analyze detailed particle dynamics and explore the evolution of local geometries. Finally, we use a combination of DL feature extraction and mixture modeling to define particle neighborhoods free of physics constraints, allowing for a separation of possible classes of particle behavior and identification of the associated transitions. Overall, this work establishes the workflow for the analysis of the self-organization processes in complex systems from observational data and provides insight into the fundamental mechanisms.

36 MATERIALS SCIENCE↗
DOE OSTI2022

Scalable and Energy-Efficient Methods for Interactive Exploration of Scientific Data

The main scientific contributions of this project are the following novel concepts for multidimensional arrays: shape-based similarity join (SIGMOD 2016), incremental view maintenance (SIGMOD 2017), user-defined stencil functions (HPDC 2017), and distributed caching for in-situ processing (SSDBM 2018). Building on our collaboration with the astrophysics group at LBNL, we applied these techniques to the data generated in the Palomar Transient Factory (PTF) astronomical survey. They played a pivotal role in the first-ever observation of a neutron star merger, which produces gravitational waves and turns out to be the origin of heavy elements, including gold. This has lead to a Science magazine article that has received extensive media coverage on ACM TechNews, Slashdot, FiveThirtyEight, and Quanta Magazine, among others. Additionally, two other articles detailing related aspects of the same discovery have been published in the Astrophysical Journal Letters journal. These publications have more than 3,000 citations according to Google Scholar (as of February 2022). This cross-disciplinary collaboration provided very good opportunities to apply database techniques to real-life scientific problems. The fact that they facilitated major discoveries in astrophysics proves the importance of our research. In addition to the work on multidimensional array databases, this project has also developed stochastic gradient descent (SGD) optimization algorithms for training large scale machine learning models, methods for querying in-situ data, and a database query optimizer based on sketch synopses.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2024

Smart Pixels: In-pixel AI for on-sensor data filtering

We present a smart pixel prototype readout integrated circuit (ROIC) designed in CMOS 28 nm bulk process, with in-pixel implementation of an artificial intelligence (AI) / machine learning (ML) based data filtering algorithm designed as proof-of-principle for a Phase III upgrade at the Large Hadron Collider (LHC) pixel detector. The first version of the ROIC consists of two matrices of 256 smart pixels, each 25$\times$25 $\mu$m$^2$ in size. Each pixel consists of a charge-sensitive preamplifier with leakage current compensation and three auto-zero comparators for a 2-bit flash-type ADC. The frontend is capable of synchronously digitizing the sensor charge within 25 ns. Measurement results show an equivalent noise charge (ENC) of $\sim$30e$^-$ and a total dispersion of $\sim$100e$^-$ The second version of the ROIC uses a fully connected two-layer neural network (NN) to process information from a cluster of 256 pixels to determine if the pattern corresponds to highly desirable high-momentum particle tracks for selection and readout. The digital NN is embedded in-between analog signal processing regions of the 256 pixels without increasing the pixel size and is implemented as fully combinatorial digital logic to minimize power consumption and eliminate clock distribution, and is active only in the presence of an input signal. The total power consumption of the neural network is $\sim$ 300 $\mu$W. The NN performs momentum classification based on the generated cluster patterns and even with a modest momentum threshold, it is capable of 54.4% - 75.4% total data rejection, opening the possibility of using the pixel information at 40MHz for the trigger. The total power consumption of analog and digital functions per pixel is $\sim$ 6 $\mu$W per pixel, which corresponds to $\sim$ 1 W/cm$^2$ staying within the experimental constraints.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2022

Cross-Layered Distributed Data-Driven Framework for Enhanced Smart Grid Cyber-Physical Security

Smart Grid (SG) research and development has drawn much attention from academia, industry and government due to the great impact it will have on society, economics and the environment. Securing the SG is a considerably significant challenge due the increased dependency on communication networks to assist in physical process control, exposing them to various cyber-threats. In addition to attacks that change measurement values using False Data Injection (FDI) techniques, attacks on the communication network may disrupt the power system's real-time operation by intercepting messages, or by flooding the communication channels with unnecessary data. Addressing these attacks requires a cross-layer approach. In this paper a cross-layered strategy is presented, called Cross-Layer Ensemble CorrDet with Adaptive Statistics(CECD-AS), which integrates the detection of faulty SG measurement data as well as inconsistent network inter-arrival times and transmission delays for more reliable and accurate anomaly detection and attack interpretation. Numerical results show that CECD-AS can detect multiple False Data Injections, Denial of Service (DoS) and Man In The Middle (MITM) attacks with a high F1-score compared to current approaches that only use SG measurement data for detection such as the traditional physics-based State Estimation, Ensemble CorrDet with Adaptive Statistics strategy and other machine learning classification-based detection schemes.

cyber-physical security↗
DOE OSTI2025

Stable Simulation of the Community Atmosphere Model Using Machine‐Learning Physical Parameterization Trained With Experience Replay

In recent years, machine learning (ML) models have been used to improve physical parameterizations of general circulation models (GCMs). A significant challenge of integrating ML models into GCMs is the online instability when they are coupled for long‐term simulation. We present a new strategy that demonstrates robust online stability when the physical parameterization package of an atmospheric GCM is replaced by a deep ML model. The method uses experience replay with a multistep training scheme of the ML model in which the model's own output at the previous time step is used in the training. Predicted physics tendencies in the replay buffer with the most recent errors in the training iterations are reused, making the ML model learn from its own errors. The training method reduces the gap between the offline and online environments of the ML model. The method is used to train the ML model as the physical parameterization of the Community Atmosphere Model (CAM5) with training data from the Multi‐scale Modeling Framework high resolution simulations. Three 6‐year online simulations of the CAM5 are carried out by using the ML physics package. The simulated spatial distributions of precipitation, surface temperature and zonally averaged atmospheric fields demonstrate overall better accuracy than that of the standard CAM5 and benchmark model even without the use of additional physical constraints or tuning. This work is the first to demonstrate a solution to address the online instability problem in climate modeling with ML physics by using experience replay.

54 ENVIRONMENTAL SCIENCES↗