Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “robust regression”

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 181 records · Page 10

An overview of uncertainty quantification techniques with application to oceanic and oil‐spill simulations

Abstract We give an overview of four different ensemble‐based techniques for uncertainty quantification and illustrate their application in the context of oil plume simulations. These techniques share the common paradigm of constructing a model proxy that efficiently captures the functional dependence of the model output on uncertain model inputs. This proxy is then used to explore the space of uncertain inputs using a large number of samples, so that reliable estimates of the model's output statistics can be calculated. Three of these techniques use polynomial chaos (PC) expansions to construct the model proxy, but they differ in their approach to determining the expansions' coefficients; the fourth technique uses Gaussian Process Regression (GPR). An integral plume model for simulating the Deepwater Horizon oil‐gas blowout provides examples for illustrating the different techniques. A Monte Carlo ensemble of 50,000 model simulations is used for gauging the performance of the different proxies. The examples illustrate how regression‐based techniques can outperform projection‐based techniques when the model output is noisy. They also demonstrate that robust uncertainty analysis can be performed at a fraction of the cost of the Monte Carlo calculation.

Iskandarani, Mohamed↗

Assessing robustness of factor ranking for supersaturated designs

Supersaturated designs can potentially be a beneficial tool for efficiently exploring a large number of factors with a moderately sized design. However, because more factors are being considered than there are runs, the stability of the identified factors depends heavily on effect sparsity and the lack of highly influential observations. A helpful tool for the analysis of supersaturated designs is least absolute shrinkage and selection operation (LASSO), which is useful when the effects of many explanatory variables are sparse in a high-dimensional dataset. To understand the impact of individual observations on the selected factors, the LASSO influence plot was created. This work describes an application of this plot and its variants that can be used to identify influential points, increase understanding of the impact of individual observations on model parameters, and the robustness of results in analyses with supersaturated designs. These graphical methods can serve as a complement to other regression diagnostics techniques in the LASSO regression setting.

3-D LASSO influence plots↗

Gap-filling eddy covariance methane fluxes: Comparison of machine learning model predictions and uncertainties at FLUXNET-CH4 wetlands

Time series of methane fluxes measured by eddy-covariance require gap-filling to estimate annual emissions. Gap-filling methane fluxes is challenging because of high variability and complex responses to multiple drivers. To date, there is no widely established gap-filling standard for methane, with regards both to the best model algorithms and predictors. In this study, we address the need for standardization by synthesizing results of gap-filling methods applied at 17 wetland sites spanning boreal to tropical regions including all major wetlands classes and two rice paddies. We introduce new procedures for: 1) creating realistic artificial gap scenarios, 2) training and evaluating gap-filling models without overstating performance, and 3) predicting half-hourly methane fluxes and annual emissions with robust uncertainty estimates. We tested a conventional method (marginal distribution sampling) and four machine learning algorithms - penalized linear regression, artificial neural networks, random forests, and boosted decision trees - and four predictor sets, including temporal, meteorological, ecosystem carbon and energy flux, and soil predictors. We find that the conventional method can achieve similar median performance to the machine learning models but is worse than the best machine learning models and relatively insensitive to predictor choices. Of the machine learning models, decision tree algorithms performed the best in cross-validation experiments, even with a baseline predictor set, and artificial neural networks showed comparable performance when using all predictors. Soil temperature was frequently the most important predictor whilst water table depth was important at sites with substantial water table fluctuations, highlighting the value of data on soil conditions. Raw gap-filling uncertainties from the machine learning models were underestimated and we propose a method to calibrate uncertainties to observations. Finally, we gap-fill and provide summary evaluation metrics for all 81 sites in the FLUXNET-CH4 community dataset and publicly release the python code for model development, evaluation, and uncertainty estimation.

42 ENGINEERING↗

Low-Cost Spectral Sensor Development Description

Solar spectral data for all parts of the US is limited due in part to the high cost of commercial spectrometers. Solar spectral information is necessary for accurate photovoltaic (PV) performance forecasting, especially for large utility-scale PV installations. A low-cost solar spectral sensor would address the obstacles and needs. In this report, a novel low-cost, discrete-band sensor device, comprised of five narrow-band sensors, is described. The hardware is comprised of commercial-off-the-shelf components to keep the cost low. Data processing algorithms were developed and are being refined for robustness. PV module short-circuit current ($I_{sc}$) prediction methods were developed based on interaction-terms regression methodology and spectrum reconstruction methodology for computing $I_{sc}$. The results suggest the computed spectrum using the reconstruction method agreed well with the measured spectrum from the wide-band spectrometer (RMS error of 38.2 W/m 2 -nm). Further analysis of computed $I_{sc}$ found a close correspondence of 0.05 A RMS error. The goal is for ubiquitous adoption of the low-cost spectral sensor in solar PV and other applications such as weather forecasting.

Armijo, Kenneth Miguel↗

Robust PCA-Deep Belief Network Surrogate Model for Distribution System Topology Identification with DERs

With the expansion of distribution networks and increased penetration of distributed energy resources (DERs), it is becoming increasingly important to obtain accurate distribution network topology in real-time. In this paper, a robust principal component analysis coupled deep belief network (PCA-DBN) surrogate model is proposed for distribution system topology identification. It integrates the benefits of robust feature extraction from PCA to deal with data quality issues and filter out noise, and the strength of DBN in capturing the nonlinear relationship between voltage amplitudes and the binary states of switchable connections. This also significantly reduces the DBN training complexity without loss of accuracy. It is shown that the widely used standard deviation of voltage drop and the voltage covariance matrix features yield less accuracy as compared to that of the voltage amplitudes in presence of high penetration of DERs and ZIP loads. Comparison results with other alternatives, such as the random forest (RF), multi-output regression (MOR) and the traditional DBN methods demonstrate that the proposed method can achieve a much higher topology identification accuracy while maintaining robustness to missing data and measurement noise under various penetration levels of DERs.

deep belief network↗

A robust gene expression-based prognostic risk score predicts overall survival of lung adenocarcinoma patients

Identification of reliable predictive biomarkers and new therapeutic targets is a critical step for significant improvement in patient outcomes. Here, we developed a multi-step bioinformatics analytic strategy to mine large omics and clinical data to build a prognostic scoring system for predicting the overall survival (OS) of lung adenocarcinoma (LuADC) patients. In latter we first identified 1327 significantly and robustly deregulated genes, 600 of which were significantly associated with the OS of LuADC patients. Gene co-expression network analysis revealed the biological functions of these 600 genes in normal lung and LuADCs, which were found to be enriched for cell cycle-related processes, blood vessel development, cell-matrix adhesion and metabolic processes. Finally, we implemented a multiple resampling method combined with Cox regression analysis to identify a 27-gene signature associated with OS, and then created a prognostic scoring system based on this signature. This scoring system robustly predicted OS of LuADC patients in 100 sampling test sets and was further validated in four independent LuADC cohorts. In addition, in comparison to other existing prognostic gene signatures published in the literature, our signature was significantly superior in predicting OS of LuADC patients. In summary, our multiomics and clinical data integration study created a 27-gene prognostic risk score that can predict OS of LuADC patients independent of age, gender and clinical stage. This score could guide therapeutic selection and allow stratification in clinical trials.

59 BASIC BIOLOGICAL SCIENCES↗

Advanced stationary and nonstationary kernel designs for domain-aware Gaussian processes

Gaussian process regression is a widely-applied method for function approximation and uncertainty quantification. The technique has gained popularity recently in the machine learning community due to its robustness and interpretability. The mathematical methods we discuss in this paper are an extension of the Gaussian-process framework. We are proposing advanced kernel designs that only allow for functions with certain desirable characteristics to be elements of the reproducing kernel Hilbert space (RKHS) that underlies all kernel methods and serves as the sample space for Gaussian process regression. These desirable characteristics reflect the underlying physics; two obvious examples are symmetry and periodicity constraints. In addition, non-stationary kernel designs can be defined in the same framework to yield flexible multi-task Gaussian processes. We will show the impact of advanced kernel designs on Gaussian processes using several synthetic and two scientific data sets. The results of our research show that including domain knowledge, communicated through advanced kernel designs, has a significant impact on the accuracy and relevance of the function approximation.

97 MATHEMATICS AND COMPUTING↗

Computationally efficient and error aware surrogate construction for numerical solutions of subsurface flow through porous media

Limiting the injection rate to restrict the pressure below a threshold at a critical location can be an important goal of simulations that model the subsurface pressure between injection and extraction wells. The pressure is approximated by the solution of Darcy’s partial differential equation for a given permeability field. The subsurface permeability is modeled as a random field since it is known only up to statistical properties. This induces uncertainty in the computed pressure. Solving the partial differential equation for an ensemble of random permeability simulations enables estimating a probability distribution for the pressure at the critical location. These simulations are computationally expensive, and practitioners often need rapid online guidance for real-time pressure management. An ensemble of numerical partial differential equation solutions is used to construct a Gaussian process regression model that can quickly predict the pressure at the critical location as a function of the extraction rate and permeability realization. The Gaussian process surrogate analyzes the ensemble of numerical pressure solutions at the critical location as noisy observations of the true pressure solution, enabling robust inference using the conditional Gaussian process distribution. Our first novel contribution is to identify a sampling methodology for the random environment and matching kernel technology for which fitting the Gaussian process regression model scales as O ( n log n ) instead of the typical O ( n 3 ) rate in the number of samples n used to fit the surrogate. The surrogate model allows almost instantaneous predictions for the pressure at the critical location as a function of the extraction rate and permeability realization. Our second contribution is a novel algorithm to calibrate the uncertainty in the surrogate model to the discrepancy between the true pressure solution of Darcy’s equation and the numerical solution. Finally, although our method is derived for building a surrogate for the solution of Darcy’s equation with a random permeability field, the framework broadly applies to solutions of other partial differential equations with random coefficients.

54 ENVIRONMENTAL SCIENCES↗

Evaluation of Data-Driven Models for Predicting Solar Photovoltaics Power Output

This research was undertaken to evaluate different inverse models for predicting power output of solar photovoltaic (PV) systems under different practical scenarios. In particular, we have investigated whether PV power output prediction accuracy can be improved if module/cell temperature was measured in addition to climatic variables, and also the extent to which prediction accuracy degrades if solar irradiation is not measured on the plane of array but only on a horizontal surface. We have also investigated the significance of different independent or regressor variables, such as wind velocity and incident angle modifier in predicting PV power output and cell temperature. The inverse regression model forms have been evaluated both in terms of their goodness-of-fit, and their accuracy and robustness in terms of their predictive performance. Given the accuracy of the measurements, expected CV-RMSE of hourly power output prediction over the year varies between 3.2% and 8.6% when only climatic data are used. Depending on what type of measured climatic and PV performance data is available, different scenarios have been identified and the corresponding appropriate modeling pathways have been proposed. The corresponding models are to be implemented on a controller platform for optimum operational planning of microgrids and integrated energy systems.

14 SOLAR ENERGY↗

Poster — Thur Eve — 44: Linearization of Compartmental Models for More Robust Estimates of Regional Hemodynamic, Metabolic and Functional Parameters using DCE-CT/PET Imaging

Quantitative analysis of dynamic positron emission tomography (PET) data usually involves minimizing a cost function with nonlinear regression, wherein the choice of starting parameter values and the presence of local minima affect the bias and variability of the estimated kinetic parameters. These nonlinear methods can also require lengthy computation time, making them unsuitable for use in clinical settings. Kinetic modeling of PET aims to estimate the rate parameter k{sub 3}, which is the binding affinity of the tracer to a biological process of interest and is highly susceptible to noise inherent in PET image acquisition. We have developed linearized kinetic models for kinetic analysis of dynamic contrast enhanced computed tomography (DCE-CT)/PET imaging, including a 2-compartment model for DCE-CT and a 3-compartment model for PET. Use of kinetic parameters estimated from DCE-CT can stabilize the kinetic analysis of dynamic PET data, allowing for more robust estimation of k{sub 3}. Furthermore, these linearized models are solved with a non-negative least squares algorithm and together they provide other advantages including: 1) only one possible solution and they do not require a choice of starting parameter values, 2) parameter estimates are comparable in accuracy to those from nonlinear models, 3) significantly reduced computational time. Our simulated data show that when blood volume and permeability are estimated with DCE-CT, the bias of k{sub 3} estimation with our linearized model is 1.97 ± 38.5% for 1,000 runs with a signal-to-noise ratio of 10. In summary, we have developed a computationally efficient technique for accurate estimation of k{sub 3} from noisy dynamic PET data.

60 APPLIED LIFE SCIENCES↗

Equation Discovery Using Fast Function Extraction: a Deterministic Symbolic Regression Approach

Advances in machine learning (ML) coupled with increased computational power have enabled identification of patterns in data extracted from complex systems. ML algorithms are actively being sought in recovering physical models or mathematical equations from data. This is a highly valuable technique where models cannot be built using physical reasoning alone. In this paper, we investigate the application of fast function extraction (FFX), a fast, scalable, deterministic symbolic regression algorithm to recover partial differential equations (PDEs). FFX identifies active bases among a huge set of candidate basis functions and their corresponding coefficients from recorded snapshot data. This approach uses a sparsity-promoting technique from compressive sensing and sparse optimization called pathwise regularized learning to perform feature selection and parameter estimation. Furthermore, it recovers several models of varying complexity (number of basis terms). FFX finally filters out many identified models using non-dominated sorting and forms a Pareto front consisting of optimal models with respect to minimizing complexity and test accuracy. Numerical experiments are carried out to recover several ubiquitous PDEs such as wave and heat equations among linear PDEs and Burgers, Korteweg–de Vries (KdV), and Kawahara equations among higher-order nonlinear PDEs. Additional simulations are conducted on the same PDEs under noisy conditions to test the robustness of the proposed approach.

42 ENGINEERING↗

End–to–End Metasurface Design for Temperature Imaging via Broadband Planck‐Radiation Regression

A theoretical framework is presented for temperature imaging from long-wavelength infrared (LWIR) thermal radiation (e.g., 8–12 µm) through the end-to-end design of a metasurface-optics frontend and a computational-reconstruction backend. A new nonlinear reconstruction algorithm, “Planck regression”, is introduced to reconstruct the temperature map from a gray scale sensor image, even in the presence of severe chromatic aberration, by exploiting black body and optical physics particular to thermal imaging. This algorithm is combined with an end-to-end approach that optimizes manufacturable, single-layer metasurfaces to yield the most accurate reconstruction. The designs demonstrate high-quality, noise-robust reconstructions of arbitrary temperature maps (including completely random images) in simulations of an ultra-compact thermal-imaging device. Here, it is also shown that Planck regression is much more generalizable to arbitrary images than a straightforward neural-network reconstruction, which requires a large training set of domain-specific images.

36 MATERIALS SCIENCE↗

Quantifying PM2.5-Meteorology Sensitivities in a Global Climate Model

Climate change can influence fine particulate matter concentrations (PM2.5) through changes in air pollution meteorology. Knowledge of the extent to which climate change can exacerbate or alleviate air pollution in the future is needed for robust climate and air pollution policy decision-making. To examine the influence of climate on PM2.5, we use the Geophysical Fluid Dynamics Laboratory Coupled Model version 3 (GFDL CM3), a fully-coupled chemistry-climate model, combined with future emissions and concentrations provided by the four Representative Concentration Pathways (RCPs). For each of the RCPs, we conduct future simulations in which emissions of aerosols and their precursors are held at 2005 levels while other climate forcing agents evolve in time, such that only climate (and thus meteorology) can influence PM2.5 surface concentrations. We find a small increase in global, annual mean PM2.5 of about 0.21 micro-g/cu m3 (5%) for RCP8.5, a scenario with maximum warming. Changes in global mean PM2.5 are at a maximum in the fall and are mainly controlled by sulfate followed by organic aerosol with minimal influence of black carbon. RCP2.6 is the only scenario that projects a decrease in global PM2.5 with future climate changes, albeit only by -0.06 micro-g/cu m (1.5%) by the end of the 21st century. Regional and local changes in PM2.5 are larger, reaching upwards of 2 micro-g/cu m for polluted (eastern China) and dusty (western Africa) locations on an annually averaged basis in RCP8.5. Using multiple linear regression, we find that future PM2.5 concentrations are most sensitive to local temperature, followed by surface wind and precipitation. PM2.5 concentrations are robustly positively associated with temperature, while negatively related with precipitation and wind speed. Present-day (2006-2015) modeled sensitivities of PM2.5 to meteorological variables are evaluated against observations and found to agree reasonably well with observed sensitivities (within 10e50% over the eastern United States for several variables), although the modeled PM2.5 is less sensitive to precipitation than in the observations due to weaker convective scavenging. We conclude that the hypothesized "climate penalty" of future increases in PM2.5 is relatively minor on a global scale compared to the influence of emissions on PM2.5 concentrations.

PM2.5↗

Determinants of efficiency in the provision of municipal street-cleaning and refuse collection services

Effective waste management systems can make critical contributions to public health, environmental sustainability and economic development. The challenge affects every person and institution in society, and measures cannot be undertaken without data collection and a quantitative analysis approach. In this paper, the two-stage double bootstrap procedure of is used to estimate the efficiency determinants of Spanish local entities in the provision of public street-cleaning and refuse collection services. The purpose is to identify factors that influence efficiency. The final sample comprised 1072 municipalities. In the first stage, robust efficiency estimates are obtained with Data Envelopment Analysis (DEA). We apply the second stage, based on a truncated-regression, to estimate the effect of a group of environmental factors on DEA estimates. The results show the existence of a significant relation between efficiency and all the variables analysed (per capita income, urban population density, the comparative index of the importance of tourism and that of the whole economic activity). We have also considered the influence of a dummy categorical variable - the political sign of the governing party - on the efficient provision of the services under study. The results from the methodology proposed show that municipalities governed by progressive parties are more efficient.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

Robust Design Under Uncertainty in Quantum Error Mitigation

Error mitigation techniques are crucial to achieving near-term quantum advantage. Classical postprocessing of quantum computation outcomes is a popular approach for error mitigation, which includes methods, such as zero noise extrapolation, virtual distillation, and learning-based error mitigation. However, these techniques have limitations due to the propagation of uncertainty resulting from the finite shot number of a quantum measurement. In this work, we introduce general and unbiased methods for quantifying the uncertainty and error of error-mitigated observables based on the strategic sampling of error mitigation outcomes. We then extend our approach to demonstrate the optimization of performance and robustness of error mitigation under uncertainty. To illustrate our methods, we apply them to zero noise extrapolation and Clifford date regression in the ground state of the XY model simulated using depolarizing and International Business Machines Corporation (IBM) Toronto noise models, respectively. In particular, we optimize the choice of noise levels and the allocation of shots for zero noise extrapolation and the distribution of the training circuits for Clifford data regression. While our methods are readily applicable to any postprocessing-based error mitigation approach, in practice they must not be prohibitively expensive—even though they perform optimizations of the error mitigation hyperparameters requiring sampling of a statistical distribution of error mitigation outcomes. By leveraging surrogate-based optimization, we show that our methods can efficiently perform optimal design for a zero noise extrapolation implementation. We then further demonstrate the transferability of learned zero noise extrapolation hyperparameters to other similar circuits.

97 MATHEMATICS AND COMPUTING↗

Generation of a Parabolic Trough Collector Efficiency Curve from Separate Measurements of Outdoor Optical Efficiency and Indoor Receiver Heat Loss

The thermal efficiency of a parabolic trough collector is a function of both the fraction of direct normal radiation absorbed by the receiver (the optical efficiency) and the heat lost to the environment when the receiver is at operating temperature. The thermal efficiency can be determined by testing the collector under actual operating conditions or by separately measuring these two components. This paper describes how outdoor measurement of the optical efficiency is combined with laboratory measurements of receiver heat loss to obtain the thermal efficiency curve. This paper describes this approach and also makes the case that there are advantages to plotting collector efficiency versus the difference between the operating temperature and the ambient temperature at which the receiver heat loss was measured divided by radiation to a fractional power (on the order of 1/3 but obtained via data regression) - as opposed to the difference between operating and ambient temperatures divided by the radiation. The results are shown to be robust over wide ranges of ambient temperature, sky temperature, and wind speed.

14 SOLAR ENERGY↗

Integration of LIBS with Machine Learning for Real-Time Monitoring of Feedstock in H 2 Gasification Applications

This project, funded by the U.S. Department of Energy (DOE) – Office of Fossil Energy under Award Number DE-FE0032177, aimed to assess the feasibility of an integrated Laser-Induced Breakdown Spectroscopy (LIBS) system with advanced machine learning (ML) models for real-time characterization and potential control of hydrogen gasifiers running on waste materials as feedstocks. This was a multidisciplinary effort that encompassed the acquisition and standardized analysis of individual and blended feedstocks—comprising biomass, coal waste, and plastic waste, followed by the development of a dynamic LIBS bench system for material sample analysis and development of predictive ML models. Comprehensive laboratory testing enabled the creation of a robust elemental dataset that served as the foundation for ML model training. Techniques such as Random Forest, Gradient Boosting, Support Vector Regression, and Neural Networks were employed to predict key feedstock properties, including higher heating value (HHV), moisture content, thermal conductivity, and ash composition with high accuracy. The results were validated against experimental data and demonstrated strong potential for real-time application in gasifier control systems. The project concluded with a study on the integration of the LIBS+ML approach for gasifier control and a techno-economic analysis of the implementation of the approach into hydrogen (H 2 ) gasification systems. Dissemination of results was carried out at a DOE meeting. This work establishes a scalable framework for automated, in-line feedstock quality assessment, offering significant implications for process optimization and emissions reduction in hydrogen production.

01 COAL, LIGNITE, AND PEAT↗

cTULIP: application of a human-based RNA-seq primary tumor classification tool for cross-species primary tumor classification in canine

The domestic dog, Canis familiaris, is quickly gaining traction as an advantageous model for use in the study of cancer, one of the leading causes of death worldwide. Naturally occurring canine cancers share clinical, histological, and molecular characteristics with the corresponding human diseases. In this study, we take a deep-learning approach to test how similar the gene expression profile of canine glioma and bladder cancer (BLCA) tumors are to the corresponding human tumors. We likewise develop a tool for identifying misclassified or outlier samples in large canine oncological datasets, analogous to that which was developed for human datasets. We test a number of machine learning algorithms and found that a convolutional neural network outperformed logistic regression and random forest approaches. We use a recently developed RNA-seq-based convolutional neural network, TULIP, to test the robustness of a human-data-trained primary tumor classification tool on cross-species primary tumor prediction. Our study ultimately highlights the molecular similarities between canine and human BLCA and glioma tumors, showing that protein-coding one-to-one homologs shared between humans and canines, are sufficient to distinguish between BLCA and gliomas. The results of this study indicate that using protein-coding one-to-one homologs as the features in the input layer of TULIP performs good primary tumor prediction in both humans and canines. Furthermore, our analysis shows that our selected features also contain the majority of features with known clinical relevance in BLCA and gliomas. Our success in using a human-data-trained model for cross-species primary tumor prediction also sheds light on the conservation of oncological pathways in humans and canines, further underscoring the importance of the canine model system in the study of human disease.

60 APPLIED LIFE SCIENCES↗