SEARCH · Search NASA
Results for “Computational optimization”
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.
Toward real-time optimization through model reduction and model discrepancy sensitivities
Optimization problems arise in a range of scenarios, from optimal control to model parameter estimation. In many applications, such as the development of digital twins, it is essential to solve these optimization problems within wall-clock-time limitations. However, this is often unattainable for complex systems, such as those modeled by nonlinear partial differential equations. One strategy for mitigating this issue is to construct a reduced-order model (ROM) that enables more rapid optimization. In particular, the use of nonintrusive ROMs—those that do not require access to the full-order model at evaluation time—is popular because they facilitate the computation of optimization solutions within the wall-clock time requirements. However, the optimization solution will be unreliable if the iterates move outside the ROM training data. This article proposes the use of hyper-differential sensitivity analysis with respect to model discrepancy (HDSA-MD) as a computationally efficient tool to augment ROM-constrained optimization and improve its reliability. The proposed approach consists of two phases: (i) an offline phase where several full-order model evaluations are computed to train the ROM, and (ii) an online phase where a ROM-constrained optimization problem is solved, a limited number of full-order model evaluations are computed, and HDSA-MD is used to enhance the optimization solution. Numerical results are demonstrated for two examples, atmospheric contaminant control and wildfire ignition location estimation, in which a ROM is trained offline using inaccurate atmospheric data. In conclusion, the HDSA-MD update yields a significant improvement in the ROM-constrained optimization solution using only one full-order model evaluation online with corrected atmospheric data.
Stochastic Optimization of Complex Systems
This project focused on methodologies for the solution of stochastic optimization problems based on relaxation and penalty methods, Monte Carlo simulation, parallel processing, and inverse optimization. The main results of the project were the development of a convergent method for the solution of models that include expectation constraints as in equilibrium models, improvement of Monte Carlo convergence through the use of a new method of sample batch optimization, the development of new parallel processing methods for stochastic unit commitment models, and the development of improved methods in combination with parallel processing for incorporating automatic differentiation methods into optimization.
Approximate Boltzmann distributions in quantum approximate optimization
Approaches to compute or estimate the output probability distributions from the quantum approximate optimization algorithm (QAOA) are needed to assess the likelihood it will obtain a quantum computational advantage. We analyze output from QAOA circuits solving 7200 random MaxCut instances, with $n$ = 14–23 qubits and depth parameter $p$ ≤ 12 and find that the average basis state probabilities follow approximate Boltzmann distributions: The average probabilities scale exponentially with their energy (cut value), with a peak at the optimal solution. Furthermore, we describe the rate of exponential scaling or effective temperature in terms of a series with a leading-order term $T$ ~ $C$ min /$n$ $\sqrt{p}$, with $C$ min the optimal solution energy. Using this scaling, we generate approximate output distributions with up to 38 qubits and find these give accurate accounts of important performance metrics in cases we can simulate exactly.
Optimal Power Management for Large-Scale Battery Energy Storage Systems via Bayesian Inference
Large-scale battery energy storage systems (BESS) have found ever-increasing use across industry and society to accelerate clean energy transition and improve energy supply reliability and resilience. However, their optimal power management poses significant challenges: the underlying high-dimensional nonlinear nonconvex optimization lacks computational tractability in real-world implementation, and the uncertainty of the exogenous power demand makes exact optimization difficult. This paper presents a new solution framework to address these bottlenecks. The solution pivots on introducing power-sharing ratios to specify each cell’s power quota from the output power demand. To find the optimal power-sharing ratios, we formulate a nonlinear model predictive control (NMPC) problem to achieve power-loss-minimizing BESS operation while complying with safety, cell balancing, and power supply-demand constraints. We then propose a parameterized control policy for the power-sharing ratios, which utilizes only three parameters, to reduce the computational demand in solving the NMPC problem. This policy parameterization allows us to translate the NMPC problem into a Bayesian inference problem for the sake of 1) computational tractability, and 2) overcoming the nonconvexity of the optimization problem. We leverage the ensemble Kalman inversion technique to solve the parameter estimation problem. Concurrently, a low-level control loop is developed to seamlessly integrate our proposed approach with the BESS to ensure practical implementation. This low-level controller receives the optimal power-sharing ratios, generates output power references for the cells, and maintains a balance between power supply and demand despite uncertainty in output power. We conduct extensive simulations and experiments on a 20-cell prototype to validate the proposed approach.
Annual report for DOE VTO
Carbon fiber (CF)/polymer composites are a transformative class of high-performance, lightweight material, where high aspect-ratio CFs reinforce a polymer matrix and exceed the strength of steel alloys at a fraction of the density. Despite the advantages of such a class of material, the broader implementation of CF composites in a range of automotive, aerospace, and energy applications is hindered by limitations of current manufacturing methods. These current techniques (e.g., hand lay-up, wet filament winding) are costly and impose severe limitations on fiber placement, orientation, and angle, and thus a composite’s ultimate properties. Today’s CF composites are expensive to manufacture, limited in form factor, and utilize costly and sub-optimal continuous filament CF. Advanced additive manufacturing (AM) processes, combined with computational design optimization and new approaches to resin development, offer alternative design and manufacturing paradigms that have the realistic potential to lift these constraints. Such integrated AM approaches could thus help to realize the full potential of CF composite materials. One relevant application of CF composite materials where manufacturing constraints limit the cost-benefit ratio is in the manufacture of high-performance composite pressure vessels for onboard compressed natural gas (CNG) storage. Current CNG storage vessels (Types 3–5) are made from load-bearing filament-wound carbon-fiber composite and are ~3.5 times as expensive as an all-metallic Type-1 vessel. This cost is invariably tied to the complex and labor-intensive nature of conventional filament winding processes and the large volumes of expensive high tensile-strength CF tow feedstock required in manufacture. Our proposed approach to CNG storage vessel manufacture is based on a combination of AM technologies for CF composite printing and design optimization tools that were pioneered at Lawrence Livermore National Laboratory (LLNL), with advances in resin/composite formulation enabled by chemical and nano-material modification. Through the successful development of this technology, LLNL seeks to demonstrate the capability for advanced CNG storage vessel manufacture at reduced cost with no reduction in performance versus the most advanced, extant Type-5 designs.
Language model-accelerated deep symbolic optimization
Symbolic optimization methods have been used to solve varied challenging and relevant problems such as symbolic regression and neural architecture search. However, the current state of the art typically learns each problem from scratch and is unable to leverage pre-existing knowledge and datasets that are available for many applications. Here, inspired by the similarity between sequence representations learned in natural language processing and the formulation of symbolic optimization as a discrete sequence optimization problem, we propose language model-accelerated deep symbolic optimization (LA-DSO), a method that leverages language models to learn symbolic optimization solutions more efficiently. We demonstrate LA-DSO in two tasks: symbolic regression, which allows us to perform extensive experimentation due to its low computation requirements, and computational antibody optimization, which shows that our proposal accelerates learning in challenging real-world problems.
Load Forecasting Based Distribution System Network Reconfiguration -- A Distributed Data-Driven Approach
In this paper, a short-term load forecasting approach based network reconfiguration is proposed in a parallel manner. Specifically, a support vector regression (SVR) based short-term load forecasting approach is designed to provide an accurate load prediction and benefit the network reconfiguration. Because of the nonconvexity of the three-phase balanced optimal power flow, a second-order cone program (SOCP) based approach is used to relax the optimal power flow problem. Then, the alternating direction method of multipliers (ADMM) is used to compute the optimal power flow in distributed manner. Considering the limited number of the switches and the increasing computation capability, the proposed network reconfiguration is solved in a parallel way. The numerical results demonstrate the feasible and effectiveness of the proposed approach.
Linear optimal power flow system and method
An electric power system or power grid is optimized using a computer-implemented tool the represents in computer memory the optimization function and at least one constraint, which the processor operates upon using a linear programming solver algorithm. The constraints are represented in memory as data structures that include both real and reactive power terms, corresponding to at least one of a power flow model and a transmission line model. The transmission line model is represented using a piecewise linear representation. The power flow model may also include for each node in the power system a real power loss term representing transmission line loss allocated to that node.
Data-driven optimized control of the COVID-19 epidemics
Optimizing the impact on the economy of control strategies aiming at containing the spread of COVID-19 is a critical challenge. We use daily new case counts of COVID-19 patients reported by local health administrations from different Metropolitan Statistical Areas (MSAs) within the US to parametrize a model that well describes the propagation of the disease in each area. We then introduce a time-varying control input that represents the level of social distancing imposed on the population of a given area and solve an optimal control problem with the goal of minimizing the impact of social distancing on the economy in the presence of relevant constraints, such as a desired level of suppression for the epidemics at a terminal time. We find that with the exception of the initial time and of the final time, the optimal control input is well approximated by a constant, specific to each area, which contrasts with the implemented system of reopening ‘in phases’. For all the areas considered, this optimal level corresponds to stricter social distancing than the level estimated from data. Proper selection of the time period for application of the control action optimally is important: depending on the particular MSA this period should be either short or long or intermediate. We also consider the case that the transmissibility increases in time (due e.g. to increasingly colder weather), for which we find that the optimal control solution yields progressively stricter measures of social distancing. We finally compute the optimal control solution for a model modified to incorporate the effects of vaccinations on the population and we see that depending on a number of factors, social distancing measures could be optimally reduced during the period over which vaccines are administered to the population.
Acceleration of the matrix multiplication of Radiance three phase daylighting simulations with parallel computing on heterogeneous hardware of personal computer
Building designers are increasingly relying on complex fenestration systems to reduce energy consumed for lighting and HVAC in low energy buildings. Radiance, a lighting simulation program, has been used to conduct daylighting simulations for complex fenestration systems. Depending on the configurations, the simulation can take hours or even days using a personal computer. This paper describes how to accelerate the matrix multiplication portion of a Radiance three-phase daylight simulation by conducting parallel computing on heterogeneous hardware of a personal computer. The algorithm was optimized and the computational part was implemented in parallel using OpenCL. The speed of new approach was evaluated using various daylighting simulation cases on a multicore central processing unit and a graphics processing unit. Based on the measurements and analysis of the time usage for the Radiance daylighting simulation, further speedups can be achieved by using fast I/O devices and storing the data in a binary format.
A Combined Day-ahead and Real-time approach for Scheduling Real and Reactive power dispatch of Battery Energy Storage System
A combined day-ahead (look-ahead) and real-time optimization approach is proposed to schedule BESS real and reactive power dispatch. The look ahead scheduling agent maximizes profit and essentially performs the function of energy arbitrage by looking at forecasted prices, network constraints and BESS operating conditions to generate real power dispatch schedule. While this schedule is followed, the real time optimization agent computes an optimum reactive power set-point (for a particular objective function) using optimal power flow framework. The concepts are implemented by co-simulating legacy tools and demonstrated on a large real world distribution system. In addition to the profits from energy arbitrage, optimum reactive power set point computation results in achieving desirable network attributes like flat voltage profiles and minimizing voltage deviations from desired value.
Data‐Driven Engineering of Thermostable Collagen‐Mimetic Peptoid Triple Helices
Collagen-mimetic peptides (CMPs) are engineered molecules designed to replicate the triple-helical structure of natural collagen. A repeating x–y-Gly sequence is the defining motif of CMPs and is critical to their triple-helical structure and stability. Substitutions to the residues occupying the x and y positions present a means to modulate the CMP structure and properties. Peptoid residues—N-substituted glycine derivatives—present an attractive potential substitution due to their thermal stability, proteolytic resistance, biocompatibility, and diverse palette of non-natural side chains, but also tend to introduce a high degree of backbone flexibility that can diminish the stability of the triple helix. In this work, we report a computational active learning cycle comprising molecular dynamics simulation, Gaussian process regression, and Bayesian optimization to computationally identify a number of promising peptoid substitutions predicted to stabilize the desired quaternary structure through side chain interactions and produce stable peptoid-based collagen-like triple helices. To experimentally test the computational predictions, a top candidate identified by the screen was synthesized and imaged using scanning electron microscopy to resolve fibril-like bundles consistent with collagen-like triple helices. This work predicts a number of CMP peptoid substitutions capable of forming stable triple-helical structures, presents a generalizable design strategy for engineering desired peptoid structures, and opens new avenues for the design of peptoid-based biomimetic materials.
DLIO: A DATA-CENTRIC BENCHMARK FOR DEEP LEARNING APPLICATIONS
SF-22-136 Deep learning has been shown as a successful method for various tasks, and its popularity results in numerous open-source deep learning software tools. Deep learning has been applied to a broad spectrum of scientific domains such as cosmology, particle physics, computer vision, fusion, and astrophysics. Scientists have performed a great deal of work to optimize the computational performance of deep learning frameworks. However, the same cannot be said for I/O performance. As deep learning algorithms rely on big-data volume and variety to effectively train neural networks accurately, I/O is a significant bottleneck on large-scale distributed deep learning training. DLIO, is a novel representative benchmark suite built based on the I/O profiling of the selected workloads. DLIO can be utilized to accurately emulate the I/O behavior of modern deep learning applications. Using DLIO, application developers and system software solution architects can identify potential I/O bottlenecks in their applications and guide optimizations to boost the I/O performance leading to lower training times. The storage vendor can also use DLIO as a guide for designing and optimize the storage and filesystem targeting at deep learning application.
A Machine Learning-Genetic Algorithm (ML-GA) Approach for Rapid Optimization Using High-Performance Computing
A Machine Learning-Genetic Algorithm (ML-GA) approach was developed to virtually discover optimum designs using training data generated from multi-dimensional simulations. Machine learning (ML) presents a pathway to transform complex physical processes that occur in a combustion engine into compact informational processes. In the present work, a total of over 2000 sector-mesh computational fluid dynamics (CFD) simulations of a heavy-duty engine were performed. These were run concurrently on a supercomputer to reduce overall turnaround time. The engine being optimized was run on a low-octane (RON70) gasoline fuel under partially premixed compression ignition (PPCI) mode. A total of nine input parameters were varied, and the CFD simulation cases were generated by randomly sampling points from this nine-dimensional input space. These input parameters included fuel injection strategy, injector design, and various in-cylinder flow and thermodynamic conditions at intake valve closure (IVC). The outputs (targets) of interest from these simulations included five metrics related to engine performance and emissions. Over 2000 samples generated from CFD were then used to train an ML model that could predict these five targets based on the nine input features. A robust super learner approach was employed to build the ML model, where results from a collection of different ML algorithms were pooled together. Thereafter, a stochastic global optimization genetic algorithm (GA) was used, with the ML model as the objective function, to optimize the input parameters based on a merit function so as to minimize fuel consumption while satisfying CO and NOx emissions constraints. The optimized configuration from ML-GA was found to be very close to that obtained from a sequentially performed CFD-GA approach, where a CFD simulation served as the objective function. In addition, the overall turnaround time was (at least) 75% lower with the ML-GA approach, as the training data was generated from concurrent CFD simulations and employing the ML model as the objective function significantly accelerated the GA optimization. Finally, this study demonstrates the potential of ML-GA and high-performance computing (HPC) to reduce the number of CFD simulations to be performed for optimization problems without loss in accuracy, thereby providing significant cost savings compared to traditional approaches.
Spectral method for a kinetic swarming model
Here we present the first numerical method for a kinetic description of the Vicsek swarming model. The kinetic model poses a unique challenge, as there is a distribution dependent collision invariant to satisfy when computing the interaction term. We use a spectral representation linked with a discrete constrained optimization to compute these interactions. To test the numerical scheme we investigate the kinetic model at different scales and compare the solution with the microscopic and macroscopic descriptions of the Vicsek model. Lastly, we observe that the kinetic model captures key features such as vortex formation and traveling waves.
Redirector: Designing Cell Factories by Reconstructing the Metabolic Objective
Advances in computational metabolic optimization are required to realize the full potential of new in vivo metabolic engineering technologies by bridging the gap between computational design and strain development. We present Redirector, a new Flux Balance Analysis-based framework for identifying engineering targets to optimize metabolite production in complex pathways. Previous optimization frameworks have modeled metabolic alterations as directly controlling fluxes by setting particular flux bounds. Redirector develops a more biologically relevant approach, modeling metabolic alterations as changes in the balance of metabolic objectives in the system. This framework iteratively selects enzyme targets, adds the associated reaction fluxes to the metabolic objective, thereby incentivizing flux towards the production of a metabolite of interest. These adjustments to the objective act in competition with cellular growth and represent up-regulation and down-regulation of enzyme mediated reactions. Using the iAF1260 E. coli metabolic network model for optimization of fatty acid production as a test case, Redirector generates designs with as many as 39 simultaneous and 111 unique engineering targets. These designs discover proven in vivo targets, novel supporting pathways and relevant interdependencies, many of which cannot be predicted by other methods. Redirector is available as open and free software, scalable to computational resources, and powerful enough to find all known enzyme targets for fatty acid production.
Inference-Optimized AI and High Performance Computing for Gravitational Wave Detection at Scale
We introduce an ensemble of artificial intelligence models for gravitational wave detection that we trained in the Summit supercomputer using 32 nodes, equivalent to 192 NVIDIA V100 GPUs, within 2 h. Once fully trained, we optimized these models for accelerated inference using NVIDIA TensorRT. We deployed our inference-optimized AI ensemble in the ThetaGPU supercomputer at Argonne Leadership Computer Facility to conduct distributed inference. Using the entire ThetaGPU supercomputer, consisting of 20 nodes each of which has 8 NVIDIA A100 Tensor Core GPUs and 2 AMD Rome CPUs, our NVIDIA TensorRT-optimized AI ensemble processed an entire month of advanced LIGO data (including Hanford and Livingston data streams) within 50 s. Our inference-optimized AI ensemble retains the same sensitivity of traditional AI models, namely, it identifies all known binary black hole mergers previously identified in this advanced LIGO dataset and reports no misclassifications, while also providing a 3X inference speedup compared to traditional artificial intelligence models. We used time slides to quantify the performance of our AI ensemble to process up to 5 years worth of advanced LIGO data. In this synthetically enhanced dataset, our AI ensemble reports an average of one misclassification for every month of searched advanced LIGO data. We also present the receiver operating characteristic curve of our AI ensemble using this 5 year long advanced LIGO dataset. This approach provides the required tools to conduct accelerated, AI-driven gravitational wave detection at scale.