SEARCH · Search NASA
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.
Hedrick, E., K. Hedrick, D. Bhattacharyya, S.E. Zitney, and B.P. Omell, "Integrating Reinforcement Learning with Model Predictive Control for Adaptive Control of Energy Systems," Clearwater Clean Energy Conference, Clearwater, FL, August 1-4, 2022.
The economic feasibility of nuclear microreactors will depend on minimizing operating costs through advancements in autonomous control, especially when these microreactors are operating alongside other types of energy systems (e.g., renewable energy). This study explores the application of deep reinforcement learning (RL) for real-time drum control in microreactors, exploring performance in regard to load-following scenarios. By leveraging a point kinetics model with thermal and xenon feedback, we first establish a baseline using a single-output RL agent, then compare it against a traditional proportional–integral–derivative (PID) controller. This study demonstrates that RL controllers, including both single- and multi-agent RL (MARL) frameworks, can achieve similar or even superior load-following performance as traditional PID control across a range of load-following scenarios. In short transients, the RL agent was able to reduce the tracking error rate in comparison to PID by one half to one third. Over extended 300-minute load-following scenarios in which xenon feedback becomes a dominant factor, PID maintained better accuracy, but RL still remained within a 1% error margin despite being trained only on short-duration scenarios. This highlights RL’s strong ability to generalize and extrapolate to longer, more complex transients, affording substantial reductions in training costs and reduced overfitting. Furthermore, when control was extended to multiple drums, MARL enabled independent drum control as well as maintained reactor symmetry constraints without sacrificing performance---an objective that standard single-agent RL could not learn. We also found that, as increasing levels of Gaussian noise were added to the power measurements, the RL controllers were able to maintain lower error rates than PID, and to do so with at least 10% and upwards of 150% less control effort. These findings illustrate RL's potential for autonomous nuclear reactor control, laying the groundwork for future integration into high-fidelity simulations and experimental validation efforts.
Federated machine learning (FL) is gaining significant popularity to develop cybersecurity solutions in power grids because of its advanced capability to support decentralized data handing at local devices, its privacy preservation, and its low-bandwidth requirement. However, the evolving adversarial machine learning (AML) threats raise significant concerns for the cybersecurity of FL architectures. The FL-based split neural network (SplitNN) achieves high performance through the decentralized training of local neural network models while preserving data privacy across multiple entities. In this paper, we propose a methodology for evaluating the performance of a vertical FLbased anomaly detector against different types of AML attacks, including denial-of-service attacks, adversarial data injection attacks, and replay attacks on the trained local models deployed in the grid network. For a case study, we consider the modified IEEE 13-bus system, and we develop SplitNN-based binary and multiclass classification models to detect, locate, and identify different types of data integrity attacks on the volt-watt control with two pooling layers: maximum pooling and AvgPool. Our experimental results, computed through performance metrics, reveal that the severity of these AML attacks varies with the integrated pooling mechanism, the type of classification model, and the nature of the cyberattack. Further, the AML attacks negatively impacted the prediction time per sample for the pretrained SplitNN during the online testing.
This paper develops a reinforcement learning (RL) approach to solve a cooperative, multi-agent Volt-Var Control (VVC) problem for high solar penetration distribution systems. The ingenuity of our RL method lies in a novel two-stage progressive training strategy that can effectively improve training speed and convergence of the machine learning algorithm. In Stage 1 (individual training), while holding all the other agents inactive, we separately train each agent to obtain its own optimal VVC actions in the action space: fconsume, generate, do-nothingg. In Stage 2 (cooperative training), all agents are trained again coordinatively to share VVC responsibility. Rewards and costs in our RL scheme include (i) a system-level reward (for taking an action), (ii) an agent-level reward (for doing-nothing), and (iii) an agent-level action cost function. This new framework allows rewards to be dynamically allocated to each agent based on their contribution while accounting for the trade-off between control effectiveness and action cost. The proposed methodology is tested and validated in a modified IEEE 123-bus system using realistic PV and load profiles. Simulation results confirm that the proposed approach is robust and computationally efficient; and it achieves desirable volt-var control performance under a wide range of operation conditions.
Emergency frequency control is one of the most critical approaches to maintain power system stability after major disturbances. With the increasing number of grid-connected renewable energy sources, existing model-based methods of frequency control are facing up with challenges of computational speed and scalability for large-scale systems. In this paper, the emergency frequency control problem is formulated as a Markov Decision Process (MDP) and solved through a novel Distributional Deep Reinforcement Learning (DDRL) method, namely the distributional soft actor critic (DSAC) method. Compared with other RL methods that only estimate the mean value, the proposed DSAC model estimates the distribution of value function over returns. This advancement can lead to more insights and knowledge for the agent, with the benefit of a much faster and more stable learning process, and the improved frequency control performance. Here, the simulation results on IEEE 39-bus and IEEE 118-bus systems demonstrate the effectiveness and robustness of proposed models, as well as the advantage compared to other state-of-the-art DRL algorithms.
Autonomous control systems provide recommendations to help operators in decision-making during plant operations ranging from normal operation to accident management. An important step of autonomous control is prognosis. In nuclear engineering domain, prognosis is the process of predicting future conditions of a system or equipment based on present signs and symptoms of a fault. The prognosis model allows predicting future reactor states for possible candidate control strategies so that the outcomes can be evaluated to determine the best control strategy. The prognosis model requires representing direct relationships between the symptoms and the predictions. In nuclear engineering, computational simulations are approximate representations of the operation of the real system. However, prognosis with computational simulations requires high computation power and time due to possible large number of scenarios. Necessary computation resources can be reduced with machine learning (ML) approach for fast predictions by building a surrogate function using the simulation data. A critical issue is, ML models are ignorant of physical knowledge, and these models approximate statistical relationships between the system variables. This ignorance can produce results that are inconsistent with physical laws, even if an optimal result is achieved from a mathematical point of view. Physics-guided machine learning (PGML) is an approach to tackle this issue. Here, this work formulates and illustrates a framework to guide development and assessment of the ML-based prognosis model for autonomous control systems. The development of the prognosis model considers the training of a ML model which consists of optimizing many aspects of the ML approach. The assessment of the prognosis model considers training data limitations and uncertainties of the ML approach. Prognosis models with standalone ML and PGML are developed and assessed on the loss-of-flow scenario of Experimental Breeder Reactor II. The results indicate that PGML based prognosis model has the best performance compared to other prognosis models.
This codebase contains a multi-agent, actor-critic reinforcement learning implementation for cyber-resilient grid voltage control. It uses a 123-bus OpenDSS system as the environment, with three-phase power flow translating nodal power injections into solved nodal voltages. The reward function penalizes deviations from nominal voltage as well as reactive power dispatch, while encouraging agents to take actions that result in fast convergence to nominal conditions. The codebase models false data injection attacks and includes functionality for training, testing, hyper-parameter tuning, and visualization.
Due to the occurrence of over-fitting at the learning phase, the modeling of chemical processes via artificial neural networks (ANN) by using corrupted data (i.e., noisy data) is an ongoing challenge. Therefore, this work investigates the effect of both Gaussian and non-Gaussian noise on the performance of process-structure based recurrent neural networks (RNN) models, which take the form of partially-connected RNN models in this work, that are used to approximate a class of multi-input-multi-outputs nonlinear systems. Furthermore, two different techniques, specifically Monte Carlo dropout and co-teaching, are utilized in the development of partially-connected RNN models. Here, these two techniques are employed to reduce the over-fitting in ANNs when noisy data is used in the training process and, hence, to improve the open-loop accuracy as well as the closed-loop performance under a Lyapunov-based model predictive controller (MPC). Aspen Plus Dynamics, a well-known high-fidelity process simulator, is used to simulate a large-scale chemical process application in order to demonstrate the anticipated improvements in both open-loop approximation and closed-loop controller performance in the presence of Gaussian and non-Gaussian noise in the data set using physics-informed RNNs.
Deep learning has demonstrated an exceptional ability to solve complex tasks (an engineering success); however, it has done so at the expense of the ability to generate new knowledge (a scientific failure). We propose an alternative framework—entitled Deep Symbolic Regression (DSR)—in which artificial neural networks (NNs) rapidly generate hypotheses about physical relationships among inputs. This framework bypasses the need to interpret an NN altogether, while still leveraging the representational power of deep learning. The resulting models are tractable mathematical expressions, which are inherently and readily human interpretable and can provide insights into underlying physical phenomena. Further, we fold this methodology into the scientific process by allowing the scientist to directly integrate a priori knowledge and beliefs to accelerate learning. We demonstrate this methodology on symbolic regression—the problem of rediscovering underlying expressions describing a dataset—and achieve state-of-the-art performance across a wide variety of symbolic regression problems. Further, we generalize our DSR framework to apply to the more general class of symbolic optimization problems, in which one seeks to optimize a sequence of symbols or “tokens” under a black-box reward function. Examples of other symbolic optimization problems include neural architecture search and computational antibody design. Our generalized tool, Deep Symbolic Optimization (DSO), has been demonstrated on the task of learning symbolic control policies for reinforcement learning environments, and has been adopted as an enabling capability for computational antibody design.
We present a learning-based predictive control methodology using the differentiable programming framework with sufficient Lyapunov-based guarantees. The neural Lyapunov differentiable predictive control (NLDPC) learns the policy by constructing a computational graph encompassing the system dynamics, state, and input constraints, and the necessary Lyapunov certification constraints, and thereafter using the automatic differentiation to update the neural policy parameters. In conjunction, our approach jointly learns a Lyapunov function that certifies the regions of state-space with stable dynamics. We also provide an initial condition sampling-based statistical guarantee for the training of NLDPC. Our offline training approach provides a computationally efficient and scalable alternative to classical model predictive control solutions, which now directly comes with Lyapunov based guarantees as shown in this paper. We substantiate the advantages of the proposed approach with simulations to stabilize the double integrator model and on an example of controlling an aircraft model.
Topological variations in power system is a common phenomenon and can impose significant challenges to traditional controllers of power system. Recent study revealed the strength of deep reinforcement learning (DRL) based approaches in power system preventive and corrective control. But topological variations are difficult to capture using classical fully connected neural network (FCN) model and has not been explicitly modeled in previous work. Hence, we develop a Graph Convolutional Network (GCN) based DRL framework to tackle topology changes in control design of power system. The GCN model exploits the graph structure of the power network and helps the DRL agent to embed the topology information during learning process. Our GCN based approach is evaluated using the IEEE-39 bus system and it outperforms the FCN-based DRL scheme in terms of training convergence and control performance considering grid topology changes.
Quality control of Laser Powder Bed Fusion (PBF-LB) additively manufactured parts is an important hurdle inhibiting the technology’s use structural applications. Acoustic monitoring of the laser powder bed fusion process can detect defects in-situ that are known to degrade mechanical properties. However, processing-structure-property (PSP) relationships are required to extrapolate from detected defects to part performance. Here, this study explores how acoustics may be a suitable signature linking processing conditions to properties, thus effectively substituting for structure in the PSP relationship. Establishing such a relationship would enable a part’s mechanical performance to be directly predicted from its acoustic signature, reducing the need for destructive testing or microstructural analysis to ensure a part will meet performance requirements. One hundred CoCrFeMnNi high entropy alloy tensile bars were printed across 13 process conditions in a series of 6 prints. The acoustic signatures of these tensile bars were used to train machine learning models to predict each part’s mechanical properties. By using both process information and acoustic information to predict mechanical properties, yield strength was predicted 18% more accurately and ductility to failure was predicted 10% more accurately than is achieved when using duplicate parts to predict part performance. Finally, individual acoustic frequencies were investigated to determine why acoustic signatures improve mechanical property predictions and the potential physical origins of these signatures. This work demonstrates how blending acoustics, process information, and machine learning can provide in-situ diagnostics of mechanical properties and improve the reliability of the PBF-LB process.
To reduce the modeling burden for control of spark-ignition engines, reinforcement learning (RL) has been applied to solve the dilute combustion limit problem. Q-learning was used to identify an optimal control policy to adjust the fuel injection quantity in each combustion cycle. A physics-based model was used to determine the relevant states of the system used for training the control policy in a data-efficient manner. The cost function was chosen such that high cycle-to-cycle variability (CCV) at the dilute limit was minimized while maintaining stoichiometric combustion as much as possible. Experimental results demonstrated a reduction of CCV after the training period with slightly lean combustion, contributing to a net increase in fuel conversion efficiency of 1.33%. To ensure stoichiometric combustion for three-way catalyst compatibility, a second feedback loop based on an exhaust oxygen sensor was incorporated into the fuel quantity controller using a slow proportional-integral (PI) controller. The closed-loop experiments showed that both feedback loops can cooperate effectively, maintaining stoichiometric combustion while reducing combustion CCV and increasing fuel conversion efficiency by 1.09%. Finally, a modified cost function was proposed to ensure stoichiometric combustion with a single controller. In addition, the learning period was shortened by half to evaluate the RL algorithm performance on limited training time. Experimental results showed that the modified cost function could achieve the desired CCV targets, however, the learning time was reduced by half and the fuel conversion efficiency increased only by 0.30%.
A first of its kind fully data-driven system has been developed and implemented into the DIII-D plasma control system to detect and control Alfvén eigenmodes (AE) in real-time. Susceptibility to fast ion-induced AE is a challenge in fully non-inductive tokamak operation, which significantly reduces fast-particle confinement and results in degraded fusion gain. Controlling AEs in real-time to improve fast-ion confinement is, hence, important for future advanced tokamak fusion reactors. The models were implemented and tested in experiments which showed that neural networks (NN) are highly effective in detecting 5 types of AE (BAE, EAE, LFM, RSAE, TAE) using high resolution ECE. To estimate the neutron deficit, a NN has been trained that outputs the classical neutron rate using similar inputs to NUBEAM. Also a preliminary ML-based proportional control has been designed and gone through initial testing in experiment to use feedback-control on the neutral beam power to achieve desired amplitude of AE modes and neutron deficits. The effect of AEs on fast-ion confinement is measured by analysing the gap in classical neutron rate from the proposed NN-based NUBEAM and the measured neutron rate.
Solid-state batteries are projected to exhibit improved energy densities and safety compared to liquid-electrolyte-based systems. Consequently, the development of solid electrolytes that can conduct cations at rates comparable to that of liquid electrolytes is an active area of research. To facilitate this search it is desirable to understand the chemical and structural features that control ion transport through a crystalline lattice. Here, machine learning (ML) is used to identify features that influence ion mobility in solids, quantify the relative importance of those features, and demonstrate how the variation of individual features alters ion mobility. Adopting the anti-perovskite lattice as a model system, ML algorithms were trained to predict migration barriers from a data set containing hundreds of barriers calculated using density functional theory. In this study, more than 10 6 feature sets of non-redundant descriptors were evaluated for their ability to predict migration barriers. The present analysis correctly reiterates the importance of several features that are known to influence mobility – for example, anion polarizability – but also identifies others whose importance is not widely recognized. Lattice properties such as hopping distance and channel width exhibit the greatest influence on cation mobility. Individual conditional expectation analyses performed on these features shows that barriers decrease as hopping distance decreases and as channel width increases. The defect formation energy – which is commonly associated with the concentration of defects, but not with their mobility – is identified as an important feature for interstitial migration. In sum, this analysis aids in the design of optimal solid electrolytes by simplifying the multi-dimensional design space to a sub-set of properties that are the most important.
This paper provides a model-free framework for real-time control of wind farms to accurately track a power reference signal. This problem requires tractable dynamical models for capturing the aerodynamic interaction between wind turbines and controllers that can make decisions in realtime given varying atmospheric conditions. In this paper, we propose a deep reinforcement learning framework to provide real-time yaw control of a wind farm. Modifications have been made to FLOw Redirection and Induction in Steady State (FLORIS), a modeling tool that incorporates transient wake behavior. The control problem is formulated to track a synthetic power reference signal based on historical atmospheric (wind speed and direction) information, price signals, and regulation deployment data from U.S. regional transmission operators. Results indicate that a wind farm, with this control paradigm, can achieve good tracking performance when tested with real atmospheric data.
The PIP-II superconducting linac at Fermilab is designed to deliver multi-megawatt proton beams for neutrino physics and other high-intensity applications. To expedite commissioning and enhance operational reliability, we have developed an EPICS-based data flow framework that seamlessly integrates digital twins (DT) with physical twins (PT). These digital twins comprise high-fidelity beam dynamics models or data-driven surrogate models connected to their physical counterparts through real-time diagnostics and advanced machine-learning algorithms.Central to this framework is Linac_Gen, an accelerated simulation tool that incorporates convolutional neural networks, random forests, and genetic algorithms to provide up to a tenfold speedup in optimizing the accelerator geometry model. An EPICS translator layer ensures interoperability by efficiently mapping lattice parameters across diverse simulation platforms.Our EPICS-based framework supports multiple operational modes—monitoring, passive learning, closed-loop control, and online learning—covering the entire machine lifecycle. By leveraging HPC resources and multi-objective optimization techniques, the digital twin enables adaptive trajectory correction, real-time fault detection, and predictive modeling of beam stability. This comprehensive approach paves the way for robust, high-intensity operation and data-driven accelerator R&D at Fermilab.
Three gaps impede the development of cost-effective and accurate data-driven building energy modeling/models (DBEM) for energy forecasting and predictive control strategies. Gap 1: data bias is common in building operation data, but this topic is hardly studied in DBEM; Gap 2: high data dimensionality is common in DBEM, but a systematic and scalable methodology is lacking to solve the problem; Gap 3: the interactions between data bias and high dimensionality have not been systematically studied for DBEM and predictive control in buildings. In this work, to address the three gaps mentioned above, we develop a framework that integrates active learning and feature selection for DBEM used for whole building data predictive control (or DPC, which is a branch of model predictive control). The framework provides a systematic methodology and automatic workflow that starts with raw data from building automation systems to the establishment of data-driven energy models for DPC controllers. The developed strategies and framework are evaluated in a virtual testbed based on EnergyPlus and BCVTB. Improved performance and reduced computational complexity are observed from the DBEM built with the developed framework, as well as the DPC controller based on that DBEM, indicating the effectiveness of the developed framework.