Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Error Modeling”

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 793 records · Page 44

Convergence of Weak-SINDy Surrogate Models

In this paper, we give an in-depth error analysis for surrogate models generated by a variant of the Sparse Identification of Nonlinear Dynamics (SINDy) method. We start with an overview of a variety of nonlinear system identification techniques, namely SINDy, weak-SINDy, and the occupation kernel method. Under the assumption that the dynamics are a finite linear combination of a set of basis functions, these methods establish a linear system to recover coefficients. We illuminate the structural similarities between these techniques and establish a projection property for the weak-SINDy technique. Following the overview, we analyze the error of surrogate models generated by a simplified version of weak-SINDy. In particular, under the assumption of boundedness of a composition operator given by the solution, we show that (i) the surrogate dynamics converges towards the true dynamics and (ii) the solution of the surrogate model is reasonably close to the true solution. Finally, as an application, we discuss the use of a combination of weak-SINDy surrogate modeling and proper orthogonal decomposition (POD) to build a surrogate model for partial differential equations (PDEs).

97 MATHEMATICS AND COMPUTING↗

PROGRESS TOWARDS NEXT GENERATION, WAVEFORM BASED THREE-DIMENSIONAL MODELS AND METRICS TO IMPROVE NUCLEAR EXPLOSION MONITORING IN THE MIDDLE EAST

Efforts to update current wave speed models of the Middle East require a thoroughly tested database of sources and recordings. Recordings of seismic waves traversing the region from Tibet to the Red Sea will be the principal metric in guiding improvements to the current wave speed model. Precise characterizations of the earthquakes, specifically depths and faulting mechanisms, are essential to avoid mapping source errors into the refined wave speed model. Errors associated with the source are manifested in amplitude and phase changes. Source depths and paths near nodal planes are particularly error prone as small changes may severely affect the resulting wavefield. Once sources are quantified, regions requiring refinement will be highlighted using adjoint tomography methods based on spectral element simulations [Komatitsch and Tromp (1999)]. An initial database of 250 regional Middle Eastern events from 1990-2007, was inverted for depth and focal mechanism using teleseismic arrivals [Kikuchi and Kanamori (1982)] and regional surface and body waves [Zhao and Helmberger (1994)]. From this initial database, we reinterpreted a large, well recorded subset of 201 events through a direct comparison between data and synthetics based upon a centroid moment tensor inversion [Liu et al. (2004)]. Evaluation was done using both a 1D reference model [Dziewonski and Anderson (1981)] at periods greater than 80 seconds and a 3D model [Kustowski et al. (2008)] at periods of 25 seconds and longer. The final source reinterpretations will be within the 3D model, as this is the initial starting point for the adjoint tomography. Transitioning from a 1D to 3D wave speed model shows dramatic improvements when comparisons are done at shorter periods, (25 s). Synthetics from the 1D model were created through mode summations while those from the 3D simulations were created using the spectral element method. To further assess errors in source depth and focal mechanism, comparisons between the three methods were made. These comparisons help to identify problematic stations and sources which may bias the final solution. Estimates of standard errors were generated for each event's source depth and focal mechanism to identify poorly constrained events. A final, well characterized set of sources and stations will be then used to iteratively improve the wave speed model of the Middle East. After a few iterations during the adjoint inversion process, the sources will be reexamined and relocated to further reduce mapping of source errors into structural features. Finally, efforts continue in developing the infrastructure required to 'quickly' generate event kernels at the n-th iteration and invert for a new, (n+1)-th, wave speed model of the Middle East. While development of the infrastructure proceeds, initial tests using a limited number of events shows the 3D model, while showing vast improvement compared to the 1D model, still requires substantial modifications. Employing our new, full source set and iterating the adjoint inversions at successively shorter periods will lead to significant changes and refined wave speed structures of the Middle East.

58 GEOSCIENCES↗

Deployment of Traditional and Hybrid Machine Learning for Critical Heat Flux Prediction in the CTF Thermal-Hydraulics Code

Critical heat flux (CHF) marks the transition from nucleate to film boiling, where heat transfer to the working fluid can rapidly deteriorate. Accurate CHF prediction is essential for efficiency, safety, and preventing equipment damage, particularly in nuclear reactors. Although widely used, empirical correlations frequently exhibit discrepancies when compared to experimental data, limiting their reliability in diverse operational conditions. Traditional machine learning (ML) approaches have demonstrated potential for CHF prediction but often suffer from limited interpretability, data scarcity, and insufficient knowledge of physical principles. Hybrid model approaches, which combine data-driven ML with base models, mitigate these concerns by incorporating prior knowledge of the domain. This study integrates an externally trained purely data-driven ML model and two hybrid models (using the Biasi and Bowring CHF correlations) within the CTF subchannel code via a custom Fortran framework. Performance was evaluated using two validation cases: a subset of the Nuclear Regulatory Commission (NRC) CHF database and the Bennett dryout experiments. In both cases, the hybrid models demonstrated significantly lower error metrics compared to conventional empirical correlations, with the best models often reducing relative error by about 5 percentage points. The pure ML model achieved comparable accuracy, outperforming the hybrid Biasi model in the NRC test case (3.3% versus 5.5% relative error) but exhibiting slightly higher error against the hybrid Bowring model in the Bennett test case (7.7% versus 6.1%). Trend analysis of error parity indicated that ML-based models reduced the tendency for CHF overprediction, improving overall accuracy. These results demonstrate that ML-based CHF models can be effectively integrated into subchannel codes and could potentially increase performance compared to conventional methods.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Training models using forces computed by stochastic electronic structure methods

Abstract Quantum Monte Carlo (QMC) can play a very important role in generating accurate data needed for constructing potential energy surfaces. We argue that QMC has advantages in terms of a smaller systematic bias and an ability to cover phase space more completely. The stochastic noise can ease the training of the machine learning model. We discuss how stochastic errors affect the generation of effective models by analyzing the errors within a linear least squares procedure, finding that there is an advantage to having many relatively imprecise data points for constructing models. We then analyze the effect of noise on a model of many-body silicon finding that noise in some situations improves the resulting model. We then study the effect of QMC noise on two machine learning models of dense hydrogen used in a recent study of its phase diagram. The noise enables us to estimate the errors in the model. We conclude with a discussion of future research problems.

Ceperley, David M. (ORCID:0000000150826271)↗

A first principles framework to predict the transient performance of latent heat thermal energy storage

Thermal energy storage (TES) is increasingly recognized as an essential component of efficient Combined Heat and Power (CHP), Concentrated Solar Power (CSP), Heating Ventilation and Air Conditioning (HVAC), and refrigeration as it reduces peak demand while helping to manage intermittent availability of energy (e.g., from solar or wind). Latent Heat Thermal Energy Storage (LHTES) is a viable option because of its high energy storage density. Parametric analysis of LHTES in terms of dimensionless numbers is highly desired as a tool to model LHTES systems. One approach is to develop a model equation so as to minimize the error between the model and data obtained from experiments or simulations. While this approach can produce an accurate correlation applicable within the range of data used for its creation, it does not provide physical understanding of the rate-limiting process controlling the transient behavior of the device. In this paper we present an alternative approach whereby the potential rate-limiting processes are identified from first principles and then the key process is determined as a function of time as a LHTES device is charged. For example, in a simple geometry, the melt-fraction can be expected to vary linearly in time if the heat transfer rate is limited by natural convection of the phase changing material and we show it scales with the PCM Grashof number as $Gr^1_p$ and PCM Prandtl number as $Pr_p^{(1/3)}$. On the other hand, if surface area of solid PCM limits the heat transfer rate, the melt fraction increases asymptotically to reach full melting. The existence of these linear and asymptotic regions and the $Gr^1_pP r^{1/3}_p$ shape of the melt fraction curve is verified using our database of 64 simulations. Of practical importance in designing LHTES devices is the melt fraction at which the heat transfer rate ceases to be limited by convection, after which the heat storage rate deteriorates. For our geometry, this is found to be about 90%. This test case of our methodology shows the value of our approach, that predicting heat storage rate based on the rate-limiting physical phenomenon as a function of time is an effective approach to modeling LHTES devices.

25 ENERGY STORAGE↗

Understanding reliability and some limitations of the images and spectra reconstructed from a multi-monochromatic x-ray imager

Temperature and density asymmetry diagnosis is critical to advance inertial confinement fusion (ICF) science. A multi-monochromatic x-ray imager (MMI) is an attractive diagnostic for this purpose. The MMI records the spectral signature from an ICF implosion core with time resolution, 2-D space resolution, and spectral resolution. While narrow-band images and 2-D space-resolved spectra from the MMI data constrain temperature and density spatial structure of the core, the accuracy of the images and spectra depends not only on the quality of the MMI data but also on the reliability of the post-processing tools. In this paper, we synthetically quantify the accuracy of images and spectra reconstructed from MMI data. Errors in the reconstructed images are less than a few percent when the space-resolution effect is applied to the modeled images. The errors in the reconstructed 2-D space-resolved spectra are also less than a few percent except those for the peripheral regions. Spectra reconstructed for the peripheral regions have slightly but systematically lower intensities by ~6% due to the instrumental spatial-resolution effects. However, this does not alter the relative line ratios and widths and thus does not affect the temperature and density diagnostics. We also investigate the impact of the pinhole size variation on the extracted images and spectra. A 10% pinhole size variation could introduce spatial bias to the images and spectra of ~10%. A correction algorithm is developed, and it successfully reduces the errors to a few percent. Finally, it is desirable to perform similar synthetic investigations to fully understand the reliability and limitations of each MMI application.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Validation Analysis of the Shoal Groundwater Flow and Transport Model

Environmental restoration at the Shoal underground nuclear test is following a process prescribed by a Federal Facility Agreement and Consent Order (FFACO) between the U.S. Department of Energy, the U.S. Department of Defense, and the State of Nevada. Characterization of the site included two stages of well drilling and testing in 1996 and 1999, and development and revision of numerical models of groundwater flow and radionuclide transport. Agreement on a contaminant boundary for the site and a corrective action plan was reached in 2006. Later that same year, three wells were installed for the purposes of model validation and site monitoring. The FFACO prescribes a five-year proof-of-concept period for demonstrating that the site groundwater model is capable of producing meaningful results with an acceptable level of uncertainty. The corrective action plan specifies a rigorous seven step validation process. The accepted groundwater model is evaluated using that process in light of the newly acquired data. The conceptual model of ground water flow for the Project Shoal Area considers groundwater flow through the fractured granite aquifer comprising the Sand Springs Range. Water enters the system by the infiltration of precipitation directly on the surface of the mountain range. Groundwater leaves the granite aquifer by flowing into alluvial deposits in the adjacent basins of Fourmile Flat and Fairview Valley. A groundwater divide is interpreted as coinciding with the western portion of the Sand Springs Range, west of the underground nuclear test, preventing flow from the test into Fourmile Flat. A very low conductivity shear zone east of the nuclear test roughly parallels the divide. The presence of these lateral boundaries, coupled with a regional discharge area to the northeast, is interpreted in the model as causing groundwater from the site to flow in a northeastward direction into Fairview Valley. Steady-state flow conditions are assumed given the absence of groundwater withdrawal activities in the area. The conceptual and numerical models were developed based upon regional hydrogeologic investigations conducted in the 1960s, site characterization investigations (including ten wells and various geophysical and geologic studies) at Shoal itself prior to and immediately after the test, and two site characterization campaigns in the 1990s for environmental restoration purposes (including eight wells and a year-long tracer test). The new wells are denoted MV-1, MV-2, and MV-3, and are located to the northnortheast of the nuclear test. The groundwater model was generally lacking data in the north-northeastern area; only HC-1 and the abandoned PM-2 wells existed in this area. The wells provide data on fracture orientation and frequency, water levels, hydraulic conductivity, and water chemistry for comparison with the groundwater model. A total of 12 real-number validation targets were available for the validation analysis, including five values of hydraulic head, three hydraulic conductivity measurements, three hydraulic gradient values, and one angle value for the lateral gradient in radians. In addition, the fracture dip and orientation data provide comparisons to the distributions used in the model and radiochemistry is available for comparison to model output. Goodness-of-fit analysis indicates that some of the model realizations correspond well with the newly acquired conductivity, head, and gradient data, while others do not. Other tests indicated that additional model realizations may be needed to test if the model input distributions need refinement to improve model performance. This approach (generating additional realizations) was not followed because it was realized that there was a temporal component to the data disconnect: the new head measurements are on the high side of the model distributions, but the heads at the original calibration locations themselves have also increased over time. This indicates that the steady-state assumption of the groundwater model is in error. To test the robustness of the model despite the transient nature of the heads, the newly acquired MV hydraulic head values were trended back to their likely values in 1999, the date of the calibration measurements. Additional statistical tests are performed using both the backward-projected MV heads and the observed heads to identify acceptable model realizations. A jackknife approach identified two possible threshold values to consider. For the analysis using the backward-trended heads, either 458 or 818 realizations (out of 1,000) are found acceptable, depending on the threshold chosen. The analysis using the observed heads found either 284 or 709 realizations acceptable. The impact of the refined set of realizations on the contaminant boundary was explored using an assumed starting mass of a single radionuclide and the acceptable realizations from the backward-trended analysis.

54 ENVIRONMENTAL SCIENCES↗

Investigation of a breathing surrogate prediction algorithm for prospective pulmonary gating

Purpose: A major challenge of four dimensional computed tomography (4DCT) in treatment planning and delivery has been the lack of respiration amplitude and phase reproducibility during image acquisition. The implementation of a prospective gating algorithm would ensure that images would be acquired only during user-specified breathing phases. This study describes the development and testing of an autoregressive moving average (ARMA) model for human respiratory phase prediction under quiet respiration conditions. Methods: A total of 47 4DCT patient datasets and synchronized respiration records was utilized in this study. Three datasets were used in model development and were removed from further evaluation of the ARMA model. The remaining 44 patient datasets were evaluated with the ARMA model for prediction time steps from 50 to 1000 ms in increments of 50 and 100 ms. Thirty-five of these datasets were further used to provide a comparison between the proposed ARMA model and a commercial algorithm with a prediction time step of 240 ms. Results: The optimal number of parameters for the ARMA model was based on three datasets reserved for model development. Prediction error was found to increase as the prediction time step increased. The minimum prediction time step required for prospective gating was selected to be half of the gantry rotation period. The maximum prediction time step with a conservative 95% confidence criterion was found to be 0.3 s. The ARMA model predicted peak inhalation and peak exhalation phases significantly better than the commercial algorithm. Furthermore, the commercial algorithm had numerous instances of missed breath cycles and falsely predicted breath cycles, while the proposed model did not have these errors. Conclusions: An ARMA model has been successfully applied to predict human respiratory phase occurrence. For a typical CT scanner gantry rotation period of 0.4 s (0.2 s prediction time step), the absolute error was relatively small, 0.06{+-}0.02 s at peak inhalation and 0.05{+-}0.04 s at peak exhalation. The application of the ARMA model for prospective pulmonary gating has been demonstrated.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Improved Prediction of Cold-Air Pools in the Weather Research and Forecasting Model Using a Truly Horizontal Diffusion Scheme for Potential Temperature

The terrain-following vertical coordinate system used by many atmospheric models, including the Weather Research and Forecasting (WRF) Model, is prone to errors in regions of complex terrain. These errors stem, in part, from the calculation of horizontal gradients within the diffusion term of the momentum or scalar evolution equations. In WRF, such gradients can be calculated along coordinate surfaces, or using metric terms that help account for grid skewness. However, neither of these options ensures a truly horizontal gradient calculation, especially if a grid cell is skewed enough that the heights of the neighboring grid points used in the calculation fall outside the vertical range of the cell. In this work, an improved scheme that uses Taylor series approximations to vertically interpolate variables to the level necessary for a truly horizontal gradient calculation is implemented in WRF for the diffusion of potential temperature. The scheme is validated using an atmosphere-at-rest configuration, in which spurious flows develop only as a result of numerical errors and can thus be used as a proxy for model performance. Following validation, the method is applied to the simulation of cold-air pools (CAPs), which occur in regions of complex terrain and are characterized by strong near-surface temperature gradients. Using the truly horizontal scheme, idealized simulations demonstrate reduced numerical mixing in a quiescent CAP, and a realistic case study in the Columbia River basin shows a reduction in positive wind speed bias by up to roughly 20% compared to observations from the Second Wind Forecast Improvement Project.

54 ENVIRONMENTAL SCIENCES↗

Modelling the impact of quasar redshift errors on the full-shape analysis of correlations in the Lyman-α forest.

In preparation for the first cosmological measurements from the full shape of the Lyman-α (Lyα) forest from DESI, we must carefully model all relevant systematics that might bias our analysis. It was shown in Youles et al. (2022) that random quasar redshift errors produce a smoothing effect on the mean quasar continuum in the Lyα forest region. This, in turn, gives rise to spurious features in the Lyα autocorrelation and its cross-correlation with quasars. Using synthetic data sets based on the DESI survey, we confirm that the impact on BAO measurements is small, but that a bias is introduced to parameters which depend on the full shape of our correlations. We combine a model of this contamination in the cross-correlation (Youles et al. 2022) with a new model we introduce here for the auto-correlation. These are parametrised by 3 parameters, which, when included in a joint fit to both correlation functions, successfully eliminate any impact of redshift errors on our full-shape constraints. We also present a strategy for removing this contamination from real data, by removing ∼0.3% of correlating pairs.

cosmology↗

Predicting the mineral composition of dust aerosols – Part 2: Model evaluation and identification of key processes with observations

A global compilation of nearly sixty measurement studies is used to evaluate two methods of simulating the mineral composition of dust aerosols in an Earth system model. Both methods are based upon a Mean Mineralogical Table (MMT) that relates the soil mineral fractions to a global atlas of arid soil type. The Soil Mineral Fraction (SMF) method assumes that the aerosol mineral fractions match the fractions of the soil. The MMT is based upon soil measurements after wet sieving, a process that destroys aggregates of soil particles that would have been emitted from the original, undisturbed soil. The second method approximately reconstructs the emitted aggregates. This model is referred to as the Aerosol Mineral Fraction (AMF) method because the mineral fractions of the aerosols differ from those of the wet-sieved parent soil, partly due to reaggregation. The AMF method remedies some of the deficiencies of the SMF method in comparison to observations. Only the AMF method exhibits phyllosilicate mass at silt sizes, where they are abundant according to observations. In addition, the AMF quartz fraction of silt particles is in better agreement with measured values, in contrast to the overestimated SMF fraction. Measurements at distinct clay and silt particle sizes are shown to be more useful for evaluation of the models, in contrast to the sum over all particles sizes that is susceptible to compensating errors, as illustrated by the SMF experiment. Model errors suggest that allocation of the emitted silt fraction of each mineral into the corresponding transported size categories is an important remaining source of uncertainty. Evaluation of both models and the MMT is hindered by the limited number of size-resolved measurements of mineral content that sparsely sample aerosols from the major dust sources. In conclusion, the importance of climate processes dependent upon aerosol mineral composition shows the need for global and routine mineral measurements.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Machine learning assisted hybrid models can improve streamflow simulation in diverse catchments across the conterminous US

Incomplete representations of physical processes often lead to structural errors in process-based (PB) hydrologic models. Machine learning (ML) algorithms can reduce streamflow modeling errors but do not enforce physical consistency. As a result, ML algorithms may be unreliable if used to provide future hydroclimate projections where climates and land use patterns are outside the range of training data. Here we test hybrid models built by integrating PB model outputs with a ML algorithm known as Long Short-Term Memory (LSTM) network on their ability to simulate streamflow in 531 catchments representing diverse conditions across the Conterminous United States. Model performance of hybrid models as measured by Nash-Sutcliffe efficiency (NSE) improved relative to standalone PB and LSTM models. More importantly, hybrid models provide highest improvement in catchments where PB models fail completely (i.e., NSE < 0). However, all models performed poorly in catchments with extended low flow periods, suggesting need for additional research.

54 ENVIRONMENTAL SCIENCES↗

QuadTune version 1: a regional tuner for global atmospheric models

When a new, better-formulated physical parameterization is introduced into a global atmospheric model, aspects of the global model solutions are sometimes degraded. Then, in order to use the new global model to address science questions, there is an incentive to restore its accuracy. Oftentimes this restoration is achieved by tuning of model parameter values. Unfortunately, the retuning process is expensive because characterizing the parameter dependence requires numerous time-consuming global simulations. To reduce the cost of tuning, this manuscript introduces a “poor man's” model tuner, “QuadTune”. QuadTune carves the globe into regions and approximates the model parameter dependence through the use of an uncorrelated quadratic emulator (i.e., response surface). The simplicity of the emulator reduces the required number of global model simulations and aids explainability of tuner behavior. Tuning removes parametric error but leaves behind model structural error. Structural error manifests itself as regional residual biases, such as stubborn biases and tuning trade-offs. To visualize these residual biases, QuadTune's software includes a set of diagnostic plots. This paper illustrates the use of the plots for characterizing residual biases with an example tuning problem.

Environmental sciences↗

Probabilistic locked mode predictor in the presence of a resistive wall and finite island saturation in tokamaks

We present a framework for estimating the probability of locking to an error field in a rotating tokamak plasma. This leverages machine learning methods trained on data from a mode-locking model, including an error field, resistive magnetohydrodynamics modeling of the plasma, a resistive wall, and an external vacuum region, leading to a fifth-order ordinary differential equation (ODE) system. It is an extension of the model without a resistive wall introduced by Akçay et al. [Phys. Plasmas 28, 082106 (2021)]. Tearing mode saturation by a finite island width is also modeled. We vary three pairs of control parameters in our studies: the momentum source plus either the error field, the tearing stability index, or the island saturation term. The order parameters are the time-asymptotic values of the five ODE variables. Normalization of them reduces the system to 2D and facilitates the classification into locked (L) or unlocked (U) states, as illustrated by Akçay et al., [Phys. Plasmas 28, 082106 (2021)]. This classification splits the control space into three regions: L̂, with only L states; Û, with only U states; and a hysteresis (hysteretic) region Ĥ, with both L and U states. In regions L̂ and Û, the cubic equation of torque balance yields one real root. Region Ĥ has three roots, allowing bifurcations between the L and U states. The classification of the ODE solutions into L/U is used to estimate the locking probability, conditional on the pair of the control parameters, using a neural network. We also explore estimating the locking probability for a sparse dataset, using a transfer learning method based on a dense model dataset.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Method of rotation angle measurement in machine vision based on calibration pattern with spot array

We propose a method of rotation angle measurement with high precision in machine vision. An area scan CCD camera, imaging lens, and calibration pattern with a spot array make up the measurement device for measuring the rotation angle. The calibration pattern with a spot array is installed at the rotation part, and the CCD camera is set at a certain distance from the rotation components. The coordinates of the spots on the calibration pattern is acquired through the vision image of the calibration pattern captured by the CCD camera. At the initial position of the calibration pattern, the camera is calibrated with the spot array; the mathematical model of distortion error of the CCD camera is built. With the equation of coordinate rotation measurement, the rotation angle of the spot array is detected. In the theoretic simulation, noise of different levels is added to the coordinates of the spot array. The experiment results show that the measurement device can measure the rotation angle precisely with a noncontact method. The standard deviation of rotation angle measurement is smaller than 3 arc sec. The measurement device can measure both microangles and large angles.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Temporal Error Correlations in a Terrestrial Carbon Cycle Model Derived by Comparison to Carbon Dioxide Eddy Covariance Flux Tower Measurements

Abstract Atmospheric CO 2 flux inversions require as input an estimate of spatial and temporal correlations of errors in their estimate of the prior mean. Some previous studies have used the differences in CO 2 daily average flux estimates produced by terrestrial carbon cycle models and eddy covariance measurements to constrain the flux error correlations. Since inversions are starting to resolve the daily cycle, we set out to examine the correlations at sub‐daily time scales, as well as the correlations across years. To this end, we examine the autocorrelations in the difference between net ecosystem‐atmosphere exchange measurements from 75 AmeriFlux towers and temporally downscaled high‐spatial‐resolution flux estimates from the Carnegie‐Ames‐Stanford Approach (CASA) terrestrial carbon cycle model. We find that the daily cycle is prominent in these hourly autocorrelations and that these autocorrelations persist across years. We propose a family of functions to model these temporal correlations in atmospheric inversions, and use cross validation to determine which of the correlation functions best fits autocorrelation data from towers not in the training set. Correlation functions with a component that attempts to model the daily cycle in the differences match correlations from other towers better than those without. Those models that reproduce the same correlation structures at 1‐year intervals while modulating the amplitudes of the correlations between those intervals improve the fit still further.

54 ENVIRONMENTAL SCIENCES↗

Mitigating Errors in External Respiratory Surrogate-Based Models of Tumor Position

Purpose: To investigate the effect of tumor site, measurement precision, tumor-surrogate correlation, training data selection, model design, and interpatient and interfraction variations on the accuracy of external marker-based models of tumor position. Methods and Materials: Cyberknife Synchrony system log files comprising synchronously acquired positions of external markers and the tumor from 167 treatment fractions were analyzed. The accuracy of Synchrony, ordinary-least-squares regression, and partial-least-squares regression models for predicting the tumor position from the external markers was evaluated. The quantity and timing of the data used to build the predictive model were varied. The effects of tumor-surrogate correlation and the precision in both the tumor and the external surrogate position measurements were explored by adding noise to the data. Results: The tumor position prediction errors increased during the duration of a fraction. Increasing the training data quantities did not always lead to more accurate models. Adding uncorrelated noise to the external marker-based inputs degraded the tumor-surrogate correlation models by 16% for partial-least-squares and 57% for ordinary-least-squares. External marker and tumor position measurement errors led to tumor position prediction changes 0.3-3.6 times the magnitude of the measurement errors, varying widely with model algorithm. The tumor position prediction errors were significantly associated with the patient index but not with the fraction index or tumor site. Partial-least-squares was as accurate as Synchrony and more accurate than ordinary-least-squares. Conclusions: The accuracy of surrogate-based inferential models of tumor position was affected by all the investigated factors, except for the tumor site and fraction index.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Machine Learning–Adjusted WRF Forecasts to Support Wind Energy Needs in Black Start Operations

Abstract The push for increased capacity of renewable sources of electricity has led to the growth of wind-power generation, with a need for accurate forecasts of winds at hub height. Forecasts for these levels were uncommon until recently, and that, combined with the nocturnal collapse of the well-mixed boundary layer and daytime growth of the boundary layer through the levels important for energy generation, has contributed to errors in numerical modeling of wind generation resources. The present study explores several machine learning algorithms to both forecast and correct standard WRF Model forecasts of winds and temperature at hub height within wind turbine plants over several different time periods that are critical for the anticipation of potential blackouts and aiding in black start operations on the power grid. It was found that mean square error for day-2 wind forecasts from the WRF Model can be improved by over 90% with the use of a multioutput neural network, and that 60-min forecasts of WRF error, which can then be used to adjust forecasts, can be made with an LSTM with great accuracy. Nowcasting of temperature and wind speed over a 10-min period using an LSTM produced very low error and especially skillful forecasts of maximum and minimum values over the turbine plant area.

17 WIND ENERGY↗