Search NASA⌕ Search

SEARCH · Search NASA

Results for “Integrated mean squared error”

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 145 records · Page 8

Improved evaluation of safeguards parameters from spent fuel measurements with the Differential Die-Away (DDA) instrument

The Differential Die-Away (DDA) technique is a highly sensitive non-destructive assay method for characterizing and detecting the presence of fissile material within an item of interest. DDA utilizes a series of pulses from a neutron generator (NG) to actively interrogate an item of interest. The die-away time of the neutron population induced by this active interrogation and the integral of the total differential die-away signal can be used to characterize items such as nuclear waste drums and spent nuclear fuel assemblies. In this work, Los Alamos National Laboratory (LANL) conceptualized, designed, and fabricated a DDA instrument that was deployed for field test measurements at the Central Interim Storage Facility for Spent Nuclear Fuel (Clab) in Oskarshamn, Sweden. The instrument performed multiple static measurements at fixed locations and dynamic axial scans of 15 pressurized water reactor (PWR) and 10 boiling water reactor (BWR) spent fuel assemblies, collecting both passive and active measurement data. The static assays of the assemblies measured the differential die-away signal, die-away time, and total passive neutron emission rate to create calibration curves for the evaluation of assembly multiplication, burnup, initial enrichment, effective fissile mass, and total elemental plutonium mass. Each calibration curve was optimized by minimizing the relative root mean square error (RRMSE) of assembly assay results compared to declared assembly parameters. The same quantities were also measured with the axial scans, and the resulting data were applied in two ways: (1) in the creation of calibration curves to improve evaluation of the same safeguards parameters as static assays, and (2) for comparison to simulation. In most cases, across both PWR and BWR assemblies, axial scan data improved the estimation of the above parameters, quantified by decreasing the calibration curve RRMSE. These axial scan results demonstrate the ability of the DDA instrument and analysis method to characterize spent PWR and BWR fuel as well as, or better than, a static assay of the same assembly. Furthermore, the DDA instrument’s unique ability to obtain both active and passive data in a single, axial scan of an entire spent fuel assembly represents a more efficient and accurate way of assaying spent fuel for verification purposes. These results represent a significant advancement for characterizing spent nuclear fuel compared to current technologies.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Bayesian Optimized Deep Ensemble for Uncertainty Quantification of Deep Neural Networks: a System Safety Case Study on Sodium Fast Reactor Thermal Stratification Modeling

Deep neural networks (DNNs) are increasingly important to scientific computing and engineering system simulations. Accurate uncertainty quantification (UQ) for DNNs is critical in safety-sensitive engineering domains. Traditional Deep Ensemble (DE) methods, while easy to implement, frequently suffer from poorly calibrated uncertainty estimates and limited predictive accuracy due to reliance on fixed architectures with varied weight initializations. To address these issues, we introduce a workflow that combines Bayesian Optimization (BO) and DE. The workflow is modular, scalable, and integrates parallel BO initialized with Sobol sequences to individually optimize the hyperparameters of each ensemble member. This method enhances ensemble diversity, improves predictive accuracy, and provides reliable uncertainty estimates. We evaluate the proposed BODE approach in a sodium fast reactor thermal stratification modeling case study, where we used a densely connected convolutional neural network to predict turbulent viscosity during the reactor transient with consideration of data noise. We benchmark its performance against several optimization approaches, including baseline deep ensemble, evolutionary algorithm-optimized ensemble, ensemble formed via random search combined with greedy selection, and a BO ensemble using random initialization. Here, our results demonstrate superior performance of the developed BODE approach. In noise-free scenarios, BODE notably reduces incorrect aleatoric uncertainty and significantly enhances predictive accuracy. Under conditions of 5% and 10% Gaussian noise, BODE adaptively quantifies uncertainty proportional to data noise, achieving up to an 80% reduction in root mean square error compared to baseline methods and producing well-calibrated prediction intervals.

Bayesian optimization↗

Estimation of stochastic volatility with long memory for index prices of FTSE Bursa Malaysia KLCI

In recent years, modeling in long memory properties or fractionally integrated processes in stochastic volatility has been applied in the financial time series. A time series with structural breaks can generate a strong persistence in the autocorrelation function, which is an observed behaviour of a long memory process. This paper considers the structural break of data in order to determine true long memory time series data. Unlike usual short memory models for log volatility, the fractional Ornstein-Uhlenbeck process is neither a Markovian process nor can it be easily transformed into a Markovian process. This makes the likelihood evaluation and parameter estimation for the long memory stochastic volatility (LMSV) model challenging tasks. The drift and volatility parameters of the fractional Ornstein-Unlenbeck model are estimated separately using the least square estimator (lse) and quadratic generalized variations (qgv) method respectively. Finally, the empirical distribution of unobserved volatility is estimated using the particle filtering with sequential important sampling-resampling (SIR) method. The mean square error (MSE) between the estimated and empirical volatility indicates that the performance of the model towards the index prices of FTSE Bursa Malaysia KLCI is fairly well.

97 MATHEMATICS AND COMPUTING↗

Optimisation of the Kaplan hydropower system via PID 2 and digital twin

Here, this paper proposes a proportional–integral-double–derivative (PID 2 ) optimisation method for the Kaplan hydropower system by building a digital twin. The study first uses one multilayer perceptron (MLP) to model the hydroturbine dynamic and then adopts three connected MLPs to model the generator dynamic, both in an open-loop fashion. Inspired by stochastic distribution control (SDC) theory, we regard the training of the turbine's neural network model as a process control problem, and we propose minimising entropy loss to update the network parameters. The next step is to build the digital twin by connecting the neural network models with a PID 2 controller and a lead-lag exciter and run the whole model in a closed-loop fashion. After that, a binary search approach is applied to optimise the PID 2 parameters based on the obtained digital twin model. The simulation results show that the proposed method can reduce the mean square tracking error by more than 90%. Furthermore, the method is extended to jointly optimise the PID 2 controller and excitation system gains through multiobjective optimisation, leveraging Pareto frontier analysis to balance active power and voltage tracking performance. Simulation results confirm the effectiveness of the proposed method, achieving a 83.46% reduction in relative mean square error of active power, a 47.13% reduction in terminal voltage tracking error, and an 82.78% improvement in the overall scalarized objective.

Hydropower system↗

Short-Term Global Horizontal Irradiance Forecasting Based on Sky Imaging and Pattern Recognition

Accurate short-term forecasting is crucial for solar integration in the power grid. In this paper, a classification forecasting framework based on pattern recognition is developed for 1-hour-ahead global horizontal irradiance (GHI) forecasting. Three sets of models in the forecasting framework are trained by the data partitioned from the preprocessing analysis. The first two sets of models forecast GHI for the first four daylight hours of each day. Then the GHI values in the remaining hours are forecasted by an optimal machine learning model determined based on a weather pattern classification model in the third model set. The weather pattern is determined by a support vector machine (SVM) classifier. The developed framework is validated by the GHI and sky imaging data from the National Renewable Energy Laboratory (NREL). Results show that the developed short-term forecasting framework outperforms the persistence benchmark by 16% in terms of the normalized mean absolute error and 25% in terms of the normalized root mean square error.

14 SOLAR ENERGY↗

Self-Evolving Machine: A Continuously Improving Model for Molecular Thermochemistry

Because collecting precise and accurate chemistry data is often challenging, chemistry data sets usually only span a small region of chemical space, which limits the performance and the scope of applicability of data-driven models. To address this issue, we integrated an active learning machine with automatic ab initio calculations to form a self-evolving model that can continuously adapt to new species appointed by the users. In the present work, we demonstrate the self-evolving concept by modeling the formation enthalpies of stable closed-shell polycyclic species calculated at the B3LYP/6-31G(2df,p) level of theory. By combining a molecular graph convolutional neural network with a dropout training strategy, the model we developed can predict density functional theory (DFT) enthalpies for a broad range of polycyclic species and assess the quality of each predicted value. For the species which the current model is uncertain about, the automatic ab initio calculations provide additional training data to improve the performance of the model. For a test set composed of 2858 cyclic and polycyclic hydrocarbons and oxygenates, the enthalpies predicted by the model agree with the reference DFT values with a root-mean-square error of 2.62 kcal/mol. Finally, we found that a model originally trained on hydrocarbons and oxygenates can broaden its prediction coverage to nitrogen-containing species via an active learning process, suggesting that the continuous learning strategy is not only able to improve the model accuracy but is also capable of expanding the predictive capacity of a model to unseen species domains.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Di-CNN: Domain-Knowledge-Informed Convolutional Neural Network for Manufacturing Quality Prediction

In manufacturing, convolutional neural networks (CNNs) are widely used on image sensor data for data-driven process monitoring and quality prediction. However, as purely data-driven models, CNNs do not integrate physical measures or practical considerations into the model structure or training procedure. Consequently, CNNs’ prediction accuracy can be limited, and model outputs may be hard to interpret practically. This study aims to leverage manufacturing domain knowledge to improve the accuracy and interpretability of CNNs in quality prediction. A novel CNN model, named Di-CNN, was developed that learns from both design-stage information (such as working condition and operational mode) and real-time sensor data, and adaptively weighs these data sources during model training. It exploits domain knowledge to guide model training, thus improving prediction accuracy and model interpretability. A case study on resistance spot welding, a popular lightweight metal-joining process for automotive manufacturing, compared the performance of (1) a Di-CNN with adaptive weights (the proposed model), (2) a Di-CNN without adaptive weights, and (3) a conventional CNN. The quality prediction results were measured with the mean squared error (MSE) over sixfold cross-validation. Model (1) achieved a mean MSE of 6.8866 and a median MSE of 6.1916, Model (2) achieved 13.6171 and 13.1343, and Model (3) achieved 27.2935 and 25.6117, demonstrating the superior performance of the proposed model.

47 OTHER INSTRUMENTATION↗

Deep learning–based digital twins for heat pumps

Heat pumps are effective cooling and heating appliances to save energy in buildings. However, traditional heat pump models are challenging to integrate with building demands in a co-simulation environment because of the nonlinear thermodynamics of refrigerants. Developing digital twin representatives for heat pumps capable of faster calculations with good accuracy is desirable. This study aimed to establish a generic deep learning–based digital twin for heat pumps with a large amount of high-fidelity data. Two refrigerants for two different heat pumps were considered: an air source heat pump with refrigerant R-410A, an air source heat pump with refrigerant CO 2 , a water source heat pump with refrigerant R-410A, and a water source heat pump with refrigerant CO 2 . Furthermore, results showed that the deep learning (long short-term memory) models effectively represented these four heat pumps as a digital twin: (a) accuracy for training and testing showed smaller than 0.02 for heating electricity and heating demands, and (b) the digital twins showed good consistency with original data for heating electricity and heating demands (root mean square errors of less than 0.12 W and 0.19 W, respectively). Therefore, deep learning–based heat pump models can be used in the co-simulation of building mechanical systems.

Air source heat pump↗

Integrating very-high-resolution UAS data and airborne imaging spectroscopy to map the fractional composition of Arctic plant functional types in Western Alaska

Widespread changes in vegetation cover and composition are driving strong impacts on Arctic ecosystem functioning and global climate feedbacks. An accurate characterization of tundra vegetation composition is required to understand how the Arctic will respond to future climate change. However, quantifying tundra vegetation composition over large areas is challenging as commonly-used satellite observations are too coarse, spatially and spectrally, to differentiate low-lying tundra vegetation types. Recent airborne and spaceborne imaging spectroscopy platforms provide better data to characterize vegetation composition. Yet, our ability to characterize vegetation composition with imaging spectroscopy remains largely unexplored in the Arctic, particularly due to a lack of ground observations needed to train and test classification models. To address this problem, we collected very-high-resolution (VHR, ~5 cm) unoccupied aerial system (UAS) imagery at three low-Arctic tundra sites located on the Seward Peninsula, western Alaska. In this paper, we examine the feasibility of integrating imagery from the UAS and the hyperspectral Airborne Visible/Infrared Imaging Spectrometer, Next Generation (AVIRIS-NG) airborne instrument to map the fractional composition of 12 key Arctic plant functional types (PFTs). To this end, we first mapped the 12 PFTs from our VHR UAS imagery using random forest classification. We then used these UAS-derived PFT maps as ground truth to develop partial least squares regression (PLSR) models to predict the fractional cover (FCover) of each PFT from AVIRIS-NG imagery. Further, we evaluated the performance of our PLSR models using reserved UAS samples, as well as by mapping PFT FCover and dominant PFT for large tundra landscapes. Our results show that 1) Arctic PFTs can be effectively mapped using VHR UAS imagery, with overall accuracy between 86% and 92%, 2) when the UAS mapped PFTs were used to inform PLSR scaling models, the FCover of the 12 PFTs could be effectively estimated from AVIRIS-NG imagery with a mean absolute error (MAE) <0.13, and 3) our PLSR models outperformed traditional, fully constrained least-squares (FCLS) linear mixture analysis and produced high-quality, spatially contiguous PFT FCover and PFT maps that captured vegetation spatial patterns with similar accuracy to those developed from UAS imagery. The developed PLSR models have the potential to be broadly applied for quantifying vegetation composition with AVIRIS-NG images to help monitor tundra vegetation dynamics and improve process-based modeling of tundra ecosystems.

54 ENVIRONMENTAL SCIENCES↗

Image quality in thoracic 4D cone-beam CT: A sensitivity analysis of respiratory signal, binning method, reconstruction algorithm, and projection angular spacing

Purpose: Respiratory signal, binning method, and reconstruction algorithm are three major controllable factors affecting image quality in thoracic 4D cone-beam CT (4D-CBCT), which is widely used in image guided radiotherapy (IGRT). Previous studies have investigated each of these factors individually, but no integrated sensitivity analysis has been performed. In addition, projection angular spacing is also a key factor in reconstruction, but how it affects image quality is not obvious. An investigation of the impacts of these four factors on image quality can help determine the most effective strategy in improving 4D-CBCT for IGRT. Methods: Fourteen 4D-CBCT patient projection datasets with various respiratory motion features were reconstructed with the following controllable factors: (i) respiratory signal (real-time position management, projection image intensity analysis, or fiducial marker tracking), (ii) binning method (phase, displacement, or equal-projection-density displacement binning), and (iii) reconstruction algorithm [Feldkamp–Davis–Kress (FDK), McKinnon–Bates (MKB), or adaptive-steepest-descent projection-onto-convex-sets (ASD-POCS)]. The image quality was quantified using signal-to-noise ratio (SNR), contrast-to-noise ratio, and edge-response width in order to assess noise/streaking and blur. The SNR values were also analyzed with respect to the maximum, mean, and root-mean-squared-error (RMSE) projection angular spacing to investigate how projection angular spacing affects image quality. Results: The choice of respiratory signals was found to have no significant impact on image quality. Displacement-based binning was found to be less prone to motion artifacts compared to phase binning in more than half of the cases, but was shown to suffer from large interbin image quality variation and large projection angular gaps. Both MKB and ASD-POCS resulted in noticeably improved image quality almost 100% of the time relative to FDK. In addition, SNR values were found to increase with decreasing RMSE values of projection angular gaps with strong correlations (r ≈ −0.7) regardless of the reconstruction algorithm used. Conclusions: Based on the authors’ results, displacement-based binning methods, better reconstruction algorithms, and the acquisition of even projection angular views are the most important factors to consider for improving thoracic 4D-CBCT image quality. In view of the practical issues with displacement-based binning and the fact that projection angular spacing is not currently directly controllable, development of better reconstruction algorithms represents the most effective strategy for improving image quality in thoracic 4D-CBCT for IGRT applications at the current stage.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Forecasting district-scale energy dynamics through integrating building network and long short-term memory learning algorithm

With the development of data-driven techniques, building energy prediction on a district level has attracted increasing attention in recent years for revealing energy use patterns and reduction potentials. However, data acquisition in large building groups is difficult and adjacent buildings also interact with each other. To reduce data cost and incorporate the inter-building impact with the data-driven building energy model, this study proposes a deep learning predictive approach that fuses the building network model with a long short-term memory learning model for district-scale building energy modeling. The building network was constructed based on correlations between the energy use intensity of buildings, which can significantly reduce the computational complexity of the deep learning models for energy dynamic prediction. Five typical building groups with energy use data from 2015 to 2018 on two institutional campuses were selected to perform the validation experiment with TensorFlow. Based on the prediction error assessments, the results suggest that for total building energy use intensity prediction, the proposed model can achieve a mean absolute percentage error of 6.66% and a root mean square error of 0.36 kWh/m 2 , compared to 12.05% and 0.63 kWh/m 2 of the conventional artificial neural network model and to 11.06% and 0.89 kWh/m 2 for the support vector regression model.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Detection of False Data Injection Attacks in Battery Stacks Using Input Noise-Aware Nonlinear State Estimation and Cumulative Sum Algorithms

Grid-scale battery energy storage systems (BESSs) are vulnerable to false data injection attacks (FDIAs), which could be used to disrupt state of charge (SoC) estimation. Inaccurate SoC estimation has negative impacts on system availability, reliability, safety, and the cost of operation. In this article a combination of a Cumulative Sum (CUSUM) algorithm and an improved input noise-aware extended Kalman filter (INAEKF) is proposed for the detection and identification of FDIAs in the voltage and current sensors of a battery stack. The series-connected stack is represented by equivalent circuit models, the SoC is modeled with a charge reservoir model and the states are estimated using the INAEKF. Further, the root mean squared error of the states’ estimation by the modified INAEKF was found to be superior to the traditional EKF. By employing the INAEKF, this article addresses the research gap that many state estimators make asymmetrical assumptions about the noise corrupting the system. Additionally, the INAEKF estimates the input allowing for the identification of FDIA, which many alternative methods are unable to achieve. The proposed algorithm was able to detect attacks in the voltage and current sensors in 99.16% of test cases, with no false positives. Utilizing the INAEKF compared to the standard EKF allowed for the identification of FDIA in the input of the system in 98.43% of test cases.

25 ENERGY STORAGE↗

Pixel multiplexing technique for real-time three-dimensional-imaging laser detection and ranging system using four linear-mode avalanche photodiodes

The avalanche-photodiode-array (APD-array) laser detection and ranging (LADAR) system has been continually developed owing to its superiority of nonscanning, large field of view, high sensitivity, and high precision. However, how to achieve higher-efficient detection and better integration of the LADAR system for real-time three-dimensional (3D) imaging continues to be a problem. In this study, a novel LADAR system using four linear mode APDs (LmAPDs) is developed for high-efficient detection by adopting a modulation and multiplexing technique. Furthermore, an automatic control system for the array LADAR system is proposed and designed by applying the virtual instrumentation technique. The control system aims to achieve four functions: synchronization of laser emission and rotating platform, multi-channel synchronous data acquisition, real-time Ethernet upper monitoring, and real-time signal processing and 3D visualization. The structure and principle of the complete system are described in the paper. The experimental results demonstrate that the LADAR system is capable of achieving real-time 3D imaging on an omnidirectional rotating platform under the control of the virtual instrumentation system. The automatic imaging LADAR system utilized only 4 LmAPDs to achieve 256-pixel-per-frame detection with by employing 64-bit demodulator. Moreover, the lateral resolution is ∼15 cm and range accuracy is ∼4 cm root-mean-square error at a distance of ∼40 m.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Snow Distribution Patterns Revisited: A Physics-Based and Machine Learning Hybrid Approach to Snow Distribution Mapping in the Sub-Arctic

Snowpack distribution in Arctic and alpine landscapes often occurs in repeating, year-to-year patterns due to local topographic, weather, and vegetation characteristics. Previous studies have suggested that with years of observational data, these snow distribution patterns can be statistically integrated into a snow process modeling workflow. Recent advances in snow hydrology and machine learning (ML) have increased our ability to predict snowpack distribution using in-situ observations, remote sensing data sets, and simple landscape characteristics that can be easily obtained for most environments. Here, we propose a hybrid approach to couple a ML snow distribution pattern (MLSDP) map with a physics-based, snow process model. We trained a random forest ML algorithm on tens of thousands of snow survey observations from a subarctic study area on the Seward Peninsula, Alaska, collected during peak snow water equivalent (SWE). We validated hybrid model outputs using in-situ snow depth and SWE observations, as well as a light detection and ranging data set and a distributed temperature profiling sensor data set. When the hybrid results were compared with the physics-based method, the hybrid method more accurately depicted the spatial patterns of the snowpack, areas of drifting snow, and years when no in-situ observations were used in the random forest ML training data set. The hybrid method also showed improvements in root mean squared error at 61% of locations where time-series estimations of snow depth were observed. These results can be applied to any physics-based model to improve the snow distribution patterning to reflect observed conditions in high latitude and high elevation cold region environments.

54 ENVIRONMENTAL SCIENCES↗

Resampling and data augmentation for short-term PV output prediction based on an imbalanced sky images dataset using convolutional neural networks

Integrating photovoltaics (PV) into electricity grids is challenged by potentially large fluctuations in power generation. In recent years, sky image-based PV output prediction using convolutional neural networks (CNNs) has emerged as a promising approach to forecasting fluctuations. A key challenge is imbalanced sky image datasets: because of the geography of solar PV system installations, sky image datasets are often rich in sunny condition data but deficient in cloudy condition data. This imbalance contrasts with the fact that model errors are dominated by cloudy condition performance. In this study, we attempt to remedy this by exploring the enrichment and augmentation of an imbalanced sky images dataset for two PV output prediction tasks: nowcasting (predicting concurrent PV output) and forecasting (predicting 15-minute-ahead future PV output). We empirically examine the efficacy of using different resampling and data augmentation approaches to create a rebalanced dataset for model development. A three-stage greedy search is used to determine the optimal resampling approach, data augmentation techniques and over-sampling rate. The results show that for the nowcast problem, resampling and data augmentation can effectively enhance the model performance, reducing overall root mean squared error (RMSE) by an average of 4%, or a 15 std. (standard deviation) of improvement compared to the variability of the baseline model. In contrast, the treatment RMSE for the forecast problem nearly always overlaps the baseline performance at the ± 2 std. level. The optimal resampling approach expands on the original dataset by over-sampling the minority cloudy data, with the best results from large over-sampling rate (e.g., 4 ~ 6 times over-sampling of cloudy images).

14 SOLAR ENERGY↗

Integrating Reanalysis and Satellite Cloud Information to Estimate Surface Downward Long-Wave Radiation

The estimation of downward long-wave radiation (DLR) at the surface is very important for the understanding of the Earth’s radiative budget with implications in surface–atmosphere exchanges, climate variability, and global warming. Theoretical radiative transfer and observationally based studies identify the crucial role of clouds in modulating the temporal and spatial variability of DLR. In this study, a new machine learning algorithm that uses multivariate adaptive regression splines (MARS) and the combination of near-surface meteorological data with satellite cloud information is proposed. The new algorithm is compared with the current operational formulation used by the European Organization for the Exploitation of Meteorological Satellites (EUMETSAT) Satellite Application Facility on Land Surface Analysis (LSA-SAF). Both algorithms use near-surface temperature and dewpoint temperature along with total column water vapor from the latest European Centre for Medium-range Weather Forecasts (ECMWF) reanalysis ERA5 and satellite cloud information from the Meteosat Second Generation. The algorithms are trained and validated using both ECMWF-ERA5 and DLR acquired from 23 ground stations as part of the Baseline Surface Radiation Network (BSRN) and the Atmospheric Radiation Measurement (ARM) user facility. Results show that the MARS algorithm generally improves DLR estimation in comparison with other model estimates, particularly when trained with observations. When considering all the validation data, root mean square errors (RMSEs) of 18.76, 23.55, and 22.08 W·m –2 are obtained for MARS, operational LSA-SAF, and ERA5, respectively. The added value of using the satellite cloud information is accessed by comparing with estimates driven by ERA5 total cloud cover, showing an increase of 17% of the RMSE. The consistency of MARS estimate is also tested against an independent dataset of 52 ground stations (from FLUXNET2015), further supporting the good performance of the proposed model.

54 ENVIRONMENTAL SCIENCES↗

Implementation of a New Method for Dynamic Multileaf Collimator Tracking of Prostate Motion in Arc Radiotherapy Using a Single KV Imager

Purpose: To implement a method for real-time prostate motion estimation with a single kV imager during arc radiotherapy and to integrate it with dynamic multileaf collimator (DMLC) target tracking. Methods and Materials: An arc field with a circular aperture and 358 deg. gantry rotation was delivered to a motion phantom with a fiducial marker under continuous kV X-ray imaging at 5 Hz, perpendicular to the treatment beam. A pretreatment gantry rotation of 120 deg. in 20 sec with continuous imaging preceded the treatment. During treatment, each kV image was first used together with all previous images to estimate the three-dimensional (3D) target probability density function and then used together with this probability density function to estimate the 3D target position. The MLC aperture was then adapted to the estimated 3D target position. Tracking was performed with five patient-measured prostate trajectories that represented characteristic prostate motion patterns. Two data sets were recorded during tracking: (1) the estimated 3D target positions, for off-line comparison with the actual phantom motion; and (2) continuous portal images, for independent off-line calculation of the 2D tracking error as the positional difference between the marker and the MLC aperture center in each portal image. All experiments were also made with 1- Hz kV imaging. Results: The mean 3D root-mean-square error of the trajectory estimation was 0.6 mm. The mean root-mean-square tracking error was 0.7 mm, both parallel and perpendicular to the MLC. The accuracy degraded slightly for 1- Hz imaging. Conclusions: Single-imager DMLC prostate tracking that allows arbitrary beam modulation during arc radiotherapy was implemented. It has submillimeter accuracy for most prostate motion types.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

A deep learning-based workflow for fast prediction of 3D state variables in geological carbon storage: A dimension reduction approach

Deep learning (DL) models are extensively used as surrogate models for high-fidelity simulations of multiphase fluid flow in porous media at large scales, enabling fast forecasts of the spatial–temporal evolution of three-dimensional (3D) state variables in geological carbon storage (GCS). However, training these models in high-dimensional space remains computationally demanding and prone to overfitting because of limited training data. This paper presents a novel workflow to address these challenges by integrating dimension reduction (DR) methods. Here, the proposed workflow employed pre-trained DR models to extract the latent variables of geological models and state variables and utilized the multi-layer perceptron (MLP) for constructing mapping functions between the input and output variables in latent spaces. Subsequently, the pre-trained reconstruction models converted the MLP-predicted latent state variables to their original high-dimensional form. Furthermore, we proposed a novel strategy for the DR and reconstruction of 3D saturation fields to account for the unique data characteristics of sparsity, nonuniformity, and discontinuity. The proposed strategy applied PCA and inverse PCA for 2D average saturation fields and developed a DL-based 3D reconstruction model, leveraging three 2D average saturation fields as input to produce a 3D saturation field as output. The pre-training of DR and reconstruction models and training of MLP models were conducted on 84 Gulf of Mexico (GoM) simulations and evaluated on 12 testing simulations. Each simulation contained 720 monthly time steps, with the first 360 months as the injection period and the rest as the post-injection period. The proposed workflow, incorporating DR and DL models, accurately predicts the normalized 3D pressure fields, achieving mean square error (MSE) of 2.92 × 10 -7 compared to the ground truth obtained from a full-physics simulator. Furthermore, the proposed strategy outperformed PCA and convolutional autoencoder (CAE) models on 3D saturation fields, resulting in minor workflow prediction errors with an MSE of 2.93 × 10 -5 . The results suggest the proposed workflow provides sufficient predictive fidelity across temporal and spatial scales, and enables a speedup of 160 times compared to the full-physics simulator, facilitating improved decision-making and risk assessment for large-scale GCS management in real-time scenarios.

3D reconstruction model↗