Tuning Of Model Parameters Research Articles

Modelling and optimization of shallow underground thermal energy storage

Shallow geothermal reservoirs are excellent candidates for low-enthalpy energy storage, and can serve as heat batteries providing constant discharge of base heat, as well as rapid discharge of heat in periods of high demand. Recharging can be done by pumping down hot water, heated using excess heat from e.g. waste incinerators. In addition to having a very low carbon footprint, such systems also require limited surface infrastructure, and can easily be placed near or under the end-user, such as below residential buildings. The geological setting is typically complex, with horizons, faults, and intertwined patterns of natural fractures, and the nearwell region is often hydraulically fractured to enhance inter-well communication. Therefore, in order to fully utilize the potential of shallow geothermal heat storage, numerical simulations are imperative. In this work, we show how to practically model such systems, including generation of computational grids with a large number of wells and fractures, numerical discretizations with discrete fractures, and complex storage strategies with multiple wells working together under common group targets. We also discuss how adjoint-based methods can be used to tune model parameters (e.g. well injectivities, rock properties, and hydraulic fracture conductivities) so that the model fits observed data, and to find well controls (e.g. rates and temperatures) that optimize storage operations. The methodology is demonstrated using one artificial and two real underground thermal energy storage projects currently under development, and we highlight important challenges and our suggested solutions related to each of them. Thematic collection: This article is part of the Earth as a thermal battery: future directions in subsurface thermal energy storage systems collection available at: https://www.lyellcollection.org/topic/collections/thermal-energy

Generation, validation, and benchmarking of a commercial independent Monte Carlo calculation beam model for multi-target SRS

BackgroundDosimetric validation of single isocenter multi-target radiosurgery plans is difficult due to conditions of electronic disequilibrium and the simultaneous irradiation of multiple off-axis lesions dispersed throughout the volume. Here we report the benchmarking of a customizable Monte Carlo secondary dose calculation algorithm specific for multi-target radiosurgery which future users may use to guide their commissioning and clinical implementation. PurposeTo report the generation, validation, and clinical benchmarking of a volumetric Monte Carlo (MC) dose calculation beam model for single isocenter radiosurgery of intracranial multi-focal disease. MethodsThe beam model was prepared within SciMoCa (ScientificRT, Munich Germany), a commercial independent dose calculation software, with the aim of broad availability via the commercial software for use with single isocenter radiosurgery. The process included (1) definition & acquisition of measurement data required for beam modeling, (2) tuning model parameters to match measurements, (3) validation of the beam model via independent measurements and end-to-end testing, and finally, (4) clinical benchmarking and validation of beam model utility in a patient specific QA setting. We utilized a 6X Flattening-Filter-Free photon beam from a TrueBeam STX linear accelerator (Siemens Healthineers, Munich Germany). ResultsIn addition to the measured data required for standard IMRT/VMAT (depth dose, central axis profiles & output factors, leaf gap), beam modeling and validation for single-isocenter SRS required central axis and off axis (5 cm & 9 cm) small field output factors and comparison between measurement and simulation of backscatter with aperture for jaw much greater than MLCs. Validation end-to-end measurements included SRS MapCHECK in StereoPHAN geometry (2%/1 mm Gamma = 99.2% ± 2.2%), and OSL & scintillator measurements in anthropomorphic STEEV phantom (6 targets, volume = 0.1–4.1cc, distance from isocenter = 1.2–7.9 cm) for which mean difference was −1.9% ± 2.2%. For 10 patient cases, MC for individual PTVs was −0.8% ± 1.5%, −1.3% ± 1.7%, and −0.5% ± 1.8% for mean dose, D95%, and D1%, respectively. This corresponded to custom passing rates action limits per AAPM TG-218 guidelines of ±5.2%, ±6.4%, and ±6.3%, respectively. ConclusionsThe beam modeling, validation, and clinical action criteria outlined here serves as a benchmark for future users of the customized beam model within SciMoCa for single isocenter radiosurgery of multi-focal disease.

An oscillatory mechanism for multi-level storage in short-term memory

Oscillatory activity is commonly observed during the maintenance of information in short-term memory, but its role remains unclear. Non-oscillatory models of short-term memory storage are able to encode stimulus identity through their spatial patterns of activity, but are typically limited to either an all-or-none representation of stimulus amplitude or exhibit a biologically implausible exact-tuning condition. Here we demonstrate a simple mechanism by which oscillatory input enables a circuit to generate persistent or sequential activity that encodes information not only in the spatial pattern of activity, but also in the amplitude of activity. This is accomplished through a phase-locking phenomenon that permits many different amplitudes of persistent activity to be stored without requiring exact tuning of model parameters. Altogether, this work proposes a class of models for the storage of information in working memory, a potential role for brain oscillations, and a dynamical mechanism for maintaining multi-stable neural representations.

Adaptive tuning of uncertain parameters in a numerical weather prediction model based upon data assimilation

AbstractIn numerical weather prediction models, near‐surface quantities like 10‐m wind speed (FF10M) or 2‐m temperature (T2M) tend to exhibit significantly larger forecast errors than the related variables in the free troposphere. Besides representativeness issues of surface stations, this is primarily related to parametrization errors and insufficient knowledge of relevant physical properties of the soil and the surface layer. For instance, the vegetation roughness length is usually derived from land‐cover classifications that may contain errors and reflect only part of the natural variability. This article describes a methodology implemented into Deutscher Wetterdienst's operational numerical weather prediction model ICON in order to improve the estimate of such parameters by using information from the data assimilation system. Building upon the condition that FF10M, T2M, and 2‐m relative humidity are assimilated, time‐filtered assimilation increments are calculated for the respective fields at the lowest model level. These are taken as proxies for the related model biases. For T2M, an additional weighted increment field is computed that indicates the model bias in the diurnal temperature amplitude. Based on these increment fields, several physical parameter fields and a few model tuning parameters are varied around their base values. This adaptive parameter adjustment is used operationally in the global and regional forecasting systems of Deutscher Wetterdienst. The ensuing reduction of the FF10M, T2M, and 2‐m relative humidity errors typically lies on the order of 5% on a hemispheric average but has substantial regional and seasonal variability that depends on the original magnitude of the model error. A weaker but still statistically significant positive impact is seen in the radiosonde verification of wind speed, humidity, and temperature in the lower troposphere, giving confidence that the adaptive tuning indeed reduces model errors rather than pushing the model towards unrepresentative station observations.

Amplifying the Effects of Contrast Agents on Magnetic Resonance Images Using a Deep Learning Method Trained on Synthetic Data.

Artificial intelligence (AI) methods can be applied to enhance contrast in diagnostic images beyond that attainable with the standard doses of contrast agents (CAs) normally used in the clinic, thus potentially increasing diagnostic power and sensitivity. Deep learning-based AI relies on training data sets, which should be sufficiently large and diverse to effectively adjust network parameters, avoid biases, and enable generalization of the outcome. However, large sets of diagnostic images acquired at doses of CA outside the standard-of-care are not commonly available. Here, we propose a method to generate synthetic data sets to train an "AI agent" designed to amplify the effects of CAs in magnetic resonance (MR) images. The method was fine-tuned and validated in a preclinical study in a murine model of brain glioma, and extended to a large, retrospective clinical human data set. A physical model was applied to simulate different levels of MR contrast from a gadolinium-based CA. The simulated data were used to train a neural network that predicts image contrast at higher doses. A preclinical MR study at multiple CA doses in a rat model of glioma was performed to tune model parameters and to assess fidelity of the virtual contrast images against ground-truth MR and histological data. Two different scanners (3 T and 7 T, respectively) were used to assess the effects of field strength. The approach was then applied to a retrospective clinical study comprising 1990 examinations in patients affected by a variety of brain diseases, including glioma, multiple sclerosis, and metastatic cancer. Images were evaluated in terms of contrast-to-noise ratio and lesion-to-brain ratio, and qualitative scores. In the preclinical study, virtual double-dose images showed high degrees of similarity to experimental double-dose images for both peak signal-to-noise ratio and structural similarity index (29.49 dB and 0.914 dB at 7 T, respectively, and 31.32 dB and 0.942 dB at 3 T) and significant improvement over standard contrast dose (ie, 0.1 mmol Gd/kg) images at both field strengths. In the clinical study, contrast-to-noise ratio and lesion-to-brain ratio increased by an average 155% and 34% in virtual contrast images compared with standard-dose images. Blind scoring of AI-enhanced images by 2 neuroradiologists showed significantly better sensitivity to small brain lesions compared with standard-dose images (4.46/5 vs 3.51/5). Synthetic data generated by a physical model of contrast enhancement provided effective training for a deep learning model for contrast amplification. Contrast above that attainable at standard doses of gadolinium-based CA can be generated through this approach, with significant advantages in the detection of small low-enhancing brain lesions.

Toward Improved Machine Learning-Based Intrusion Detection for Internet of Things Traffic

The rapid development of Internet of Things (IoT) networks has revealed multiple security issues. On the other hand, machine learning (ML) has proven its efficiency in building intrusion detection systems (IDSs) intended to reinforce the security of IoT networks. In fact, the successful design and implementation of such techniques require the use of effective methods in terms of data and model quality. This paper encloses an empirical impact analysis for the latter in the context of a multi-class classification scenario. A series of experiments were conducted using six ML models, along with four benchmarking datasets, including UNSW-NB15, BOT-IoT, ToN-IoT, and Edge-IIoT. The proposed framework investigates the marginal benefit of employing data pre-processing and model configurations considering IoT limitations. In fact, the empirical findings indicate that the accuracy of ML-based IDS detection rapidly increases when methods that use quality data and models are deployed. Specifically, data cleaning, transformation, normalization, and dimensionality reduction, along with model parameter tuning, exhibit significant potential to minimize computational complexity and yield better performance. In addition, MLP- and clustering-based algorithms outperformed the remaining models, and the obtained accuracy reached up to 99.97%. One should note that the performance of the challenger models was assessed using similar test sets, and this was compared to the results achieved using the relevant pieces of research.

BT-CKBQA: An efficient approach for Chinese knowledge base question answering

Knowledge Base Question Answering (KBQA), as an increasingly essential application, can provide accurate responses to user queries. ensuring that users obtain relevant information and make decisions promptly. The deep learning-based approaches have achieved satisfactory QA results by leveraging the neural network models. However, these approaches require numerous parameters, which increases the workload of tuning model parameters. To address this problem, we propose BT-CKBQA, a practical and highly efficient approach incorporating BM25 and Template-based predicate mapping for CKBQA. Besides, a concept lattice based approach is proposed for summarizing the knowledge base, which can largely improve the execution efficiency of QA with little loss of performance. Concretely, BT-CKBQA leverages the BM25 algorithm and custom dictionary to detect the subject of a question sentence. A template-based predicate generation approach is then proposed to generate candidate predicates. Finally, a ranking approach is provided with the joint consideration of character similarity and semantic similarity for predicate mapping. Extensive experiments are conducted over the NLPCC-ICCPOL 2016 and 2018 KBQA datasets, and the experimental results demonstrate the superiority of the proposed approach over the compared baselines. Particularly, the averaged F1-score result of BT-CKBQA for mention detection is up to 98.25%, which outperforms the best method currently available in the literature. For question answering, the proposed approach achieves superior results than most baselines with the F1-score value of 82.68%. Compared to state-of-the-art baselines, the execution efficiency and performance of QA per unit time can be improved with up to 56.39% and 44.06% gains, respectively. The experimental results for the diversification of questions indicate that the proposed approach performs better for diversified questions than domain-specific questions. The case study over a constructed COVID-19 knowledge base illustrates the effectiveness and practicability of BT-CKBQA.

Modeling and experimental validation of electrodynamic maglev systems

The pursuit of novel, environmentally friendly solutions for mobility and transportation has been a relevant area of technical and scientific progress in recent years. In this regard, the Hyperloop concept is a novel solution for sustainable transportation of people and goods at high speed. The electrodynamic levitation system in Hyperloop represents a key enabling technology. Nevertheless, its unstable nature is crucial and requires proper analysis and understanding. This paper proposes the experimental stabilization of electrodynamic maglev systems by means of passive components. A downscaled test setup is used to validate the numerical models that govern passive maglev systems. Setups for quasi-static and dynamic testing are proposed. The former experiments lead to model parameter tuning. The latter reproduce unstable behavior and assist in testing a stabilization method. Results in the time and frequency domains are used to analyze the system response and validate the proposed approach.

Deep reinforcement learning-based controller for dynamic positioning of an unmanned surface vehicle

Dynamic positioning (DP) system is of great significance for the unmanned surface vehicle (USV) to achieve fully autonomous navigation. Traditional control schemes have problems such as model accuracy, parameter tuning, and complex design. In addition, although the deep reinforcement learning (DRL) is widely used in the field of vessel motion control, the learning efficiency is not high, and insufficient robustness in the face of changing environmental. In order to improve the anti-disturbance ability, robustness and convergence speed of the controller during training, a deep reinforcement learning control method based on priority experience replay (PER) is proposed for dynamic positioning of the USV. The mathematical models are established based on the kinematic and dynamic of the USV. Markov decision process (MDP) models are constructed according to the DP tasks. The simulation results show that compared with other DRL algorithms, the proposed method has higher reward value, faster convergence speed, higher control precision and smoother control output.

Input data to match model and prototype states of WWER NPP

The paper considers some possible input data of NPP power unit model for solving the problem of model state matching with prototype state. In the course of work on matching the state of NPP model with the prototype, a numerical experiment using full-scale modeling complex PROSTOR was conducted. Its essence was to obtain steady states of the model with capacity from 94% to 104% with a step of 0,05%. For modeling we used data of Kalinin NPP unit 3 loading 4 on 248 effective days. When analyzing the results of the numerical experiment, two effects were found. One of them, expressed in the bend of the derivative of the obtained graph, is connected with reaching the upper limit of the possibility to regulate the pressure in the main steam collector. The regulating valves on the turbine generator are the executing mechanism of pressure regulator in main steam collector. At the break of the graph they open at 100%, which causes further inability to regulate pressure. The pressure in the main steam collector starts to increase. This effect corresponds to the real processes occurring at the NPP. In this case, as one of the model tuning parameters, it is proposed to use the adaptive coefficient to the concentration of boric acid in the reactor, for the control of which, it is proposed to use a mathematical boric regulator. It was also shown that it is impossible to accurately control the pressure in the reactor due to the peculiarities of operation of the volume compensator electric heaters automatics. In this regard, it was proposed to use an additional mathematical volume compensator electric heaters regulator to match the state of the model with the prototype. These regulators have been successfully tested as part of the PROSTOR software package to match the states of the model and the prototype.

40 Ar/ 39 Ar eruption ages of Turkana Basin tuffs: millennial-scale resolution constrains palaeoclimate proxy tuning models and hominin fossil ages

The Turkana Basin in NW Kenya and SW Ethiopia hosts remarkable fossil-rich sediments that are central to our understanding of early hominin evolution, with interbedded volcanic tuffs providing critical time markers. However, the resolution of existing Early Pleistocene–Pliocene ages is limited to c. 20–60 kyr, inhibiting the evaluation of climatic and environmental drivers of evolution. We present high-precision, single-feldspar 40 Ar/ 39 Ar age and elemental data for four stratigraphically significant tuffs. These samples exhibit variably dispersed age distributions correlated with feldspar compositional trends, interpreted to indicate the partial retention of inherited 40 Ar related to crustal ‘cold storage’ and rapid melt infiltration preceding eruption. We evaluated various statistical methods and calculated astronomically calibrated Bayesian age estimates of 1879.1 ± 0.6 ka (±2.4 ka including external errors) for the Kay Behrensmeyer Site (KBS)/H2 Tuff, 1837.4 ± 0.9 ka (±2.4 ka) for the Malbe/H4 Tuff, 1357.5 ± 1.8 ka (±2.5 ka) for the Chari/L Tuff and 1315.4 ± 1.9 ka (±2.5 ka) for the Gele Tuff. Our results permit refined age constraints for important early Homo fossils, including the cranium KNM-ER1813 ( Homo habilis ) and various Homo erectus fossils. The KBS Tuff age also provides an important calibration locus for orbital tuning of palaeoclimate proxy records, revealing the complex interplays between palaeoclimate and geological drivers of sedimentation. Supplementary material : The supplementary tables and figures include previously published ages (Table S1), sample and irradiation information (Table S2), electron probe microanalytical results for tuff glasses (Table S3, Figs S2 and S3) and feldspars (Table S4), 40 Ar/ 39 Ar analytical data (Tables S5 and S6, Fig. S4), orbital tuning model parameters (Table S7), a summary of previous and revised hominin ages (Table S8) and a worked Bayesian estimation example (Fig. S1) and are available at https://doi.org/10.6084/m9.figshare.c.6602994

Learning-Based Visual-Strain Fusion for Eye-in-Hand Continuum Robot Pose Estimation and Control

Image processing has significantly extended the practical value of the eye-in-hand camera, enabling and promoting its applications for quantitative measurement. However, fully vision-based pose estimation methods sometimes encounter difficulties in handling cases with deficient features. In this article, we fuse visual information with the sparse strain data collected from a single-core fiber inscribed with fiber Bragg gratings (FBGs) to facilitate continuum robot pose estimation. An improved extreme learning machine algorithm with selective training data updates is implemented to establish and refine the FBG-empowered (F-emp) pose estimator <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">online</i> . The integration of F-emp pose estimation can improve sensing robustness by reducing the number of times that visual tracking is lost given moving visual obstacles and varying lighting. In particular, this integration solves pose estimation failures under full occlusion of the tracked features or complete darkness. Utilizing the fused pose feedback, a hybrid controller incorporating kinematics and data-driven algorithms is proposed to accomplish fast convergence with high accuracy. The online-learning error compensator can improve the target tracking performance with a 52.3%–90.1% error reduction compared with constant-curvature model-based control, without requiring fine model-parameter tuning and prior data acquisition.

NSGA-II parameterization for the optimal pressure sensor location in water distribution networks

ABSTRACT The optimal location of pressure sensors is typicallysolved using heuristic algorithms. Non-dominated Sorting Genetic Algorithm II (NSGA-II) is one of the most used algorithms in the water industry, requiring a preliminary parameter tuning process. The lack of guidelines on how to tune model parameters generally limits the use of these algorithms by researchers or practitioners and, as such, fails to be used in real-life problems. The current paper explores different NSGA-II parameterizations for the optimal location of pressure sensors by using a multi-objective optimization methodology applied to a real distribution network. Results show that (i) both the uniform and simulated binary crossover operators (depending on the internal parameters) produce the best results, being the former recommended since it does not require further parameter tuning; (ii) polynomial mutation with lower probability value should be chosen; and (iii) the distribution indices of polynomial mutation have a minor effect on NSGA-II performance.

Investigating the Electromechanical Properties of Carbon Black-Based Conductive Polymer Composites via Stochastic Modeling.

Conductive polymer composites (CPCs) have shown potential for structural health monitoring applications based on repeated findings of irreversible transducer electromechanical property change due to fatigue. In this research, a high-fidelity stochastic modeling framework is explored for predicting the electromechanical properties of spherical element-based CPC materials at bulk scales. CPC dogbone specimens are manufactured via casting and their electromechanical properties are characterized via uniaxial tensile testing. Model parameter tuning, demonstrated in previous works, is deployed for improved simulation fidelity. Modeled predictions are found in agreement with experimental results and compared to predictions from a popular analytical model in the literature.

Integrated method for intelligent structural design of steel frames based on optimization and machine learning algorithm

Optimization methods using metaheuristic algorithms have been widely used in steel frame design to improve the inefficient traditional design method due to repeated model tuning and massive mechanical analysis. However, the random search feature of them may easily result in poor performances. In this paper, combining metaheuristic algorithms and machine learning methods, a highly integrated method based on an on-line model training, updating and parameter tuning process is proposed to improve the performance of the optimization algorithm with general forms and parameters. It reduces the impact of the iterative mechanism and parameter setting of metaheuristic algorithms on their performance. Such method is introduced to intelligent structural design of steel frames including three steps. The standard optimization process is conducted to search optimal design and simultaneously collect the mechanical analysis data of the structure. Then the data is adopted to generate and update surrogate models of structural responses dynamically while an analysis-based feature engineering and an automatic model tuning technique are employed to improve the model accuracy. Finally, a much more efficient procedure is presented to obtain potential solutions which are used to improve the convergence rate and performance of standard optimization. Four cases are used to study the effectiveness of the integrated method and the influence of different settings is discussed, as well as its generality. As a conclusion, the proposed method can achieve structural safety and economic benefit of steel frames, which exhibits superiorly in terms of robustness, optimal results and computational cost even in large-scale optimization problems of complicated frames.

Evaluating neural network models in site-specific solar PV forecasting using numerical weather prediction data and weather observations

The effective use of solar photovoltaic (PV) installations implies the integration of solar PV output into overall energy consumption planning, optimization, and control. Moreover, day-ahead trading of electricity in Europe makes day-ahead solar PV forecasting utterly important, and thus its accuracy becomes of particular interest. Data-driven PV forecasting models are typically trained using numerical weather prediction (NWP) data, the availability of which represents one of the main obstacles in modeling. In this study, we investigate an alternative scenario, in which an artificial neural network (ANN) is trained on weather observations and then tested on NWP data to simulate the model's use in operational PV forecasting. In the experiments, solar PV output data, historical weather observations, and historical NWP data were collected from three sites in eastern Finland. The results showed that, although training ANN on observational data leads to a slight decrease in its performance compared to ANN trained on NWP data, it still outperforms a physical model. In practice, this alternative scenario means that if historical NWP data are not available for model training, observational data allow effective model selection and parameter tuning, and then generalization error estimates are gradually updated using online NWP data.

Mixing in the upper western equatorial Pacific driven by westerly wind event

Diapycnal mixing in the upper western equatorial Pacific (WEP) plays an important role in the tropical air–sea interactions and in the formation of the global climate system. Yet, the WEP is uniquely rich in water masses originating from the two hemispheres and in multiscale processes of different dynamical nature, thus creating a complex regime of mixing remains to be fully characterized by elaborate observations. Here, on the basis of microstructure measurements in the WEP, we report the observations on a strong deep cycle turbulence extending well into the upper thermocline by westerly wind event, with the turbulent kinetic energy dissipation rate ε~O(10−8–10−7) W kg−1 and diapycnal diffusivity Kρ∼O(10−4) m2 s−1. Below the deep cycle turbulence layer, turbulence and mixing are generally weak with ε~O(10−10–10−9) W kg−1 and Kρ∼O(10−7–10−6) m2 s−1, a prototype of the weak mixing nature of the low-latitude western Pacific. The observed turbulence below the deep cycle turbulence layer can be satisfactorily scaled by either the MacKinnon–Gregg model or the Richardson number–based model with tuned model parameters.

Machine Learning-Based Screening of Risk Factors and Prediction of Deep Vein Thrombosis and Pulmonary Embolism After Hip Arthroplasty.

Prophylactic anticoagulation is a standard strategy for patients undergoing total hip arthroplasty (THA) to prevent deep venous thromboembolism (DVT) and pulmonary embolism (PE). Nevertheless, some patients still experience these complications during their hospital stay. Current risk assessment methods like the Caprini and Geneva scores are not specifically designed for THA and may not accurately predict DVT or PE postoperatively. This study used machine learning techniques to establish models for early diagnosis of DVT and PE in patients undergoing THA. Data were collected from 1481 patients who received perioperative prophylactic anticoagulation. Model establishment and parameter tuning were performed using a training set and evaluated using a test set. Among the models, extreme gradient boosting (XGBoost) performed the best, with an area under the receiver operating characteristic curve (AUC) of 0.982, sensitivity of 0.913, and specificity of 0.998. The main features used in the XGBoost model were direct and indirect bilirubin, partial activation prothrombin time, prealbumin, creatinine, D-dimer, and C-reactive protein. Shapley Additive Explanations analysis was conducted to further analyze these features. This study presents a model for early diagnosis DVT or PE after THA and demonstrates bilirubin could be a potential predictor in the assessment of DVT or PE. Compared to traditional risk assessment, XGBoost has a high sensitivity and specificity to predict DVT and PE in the clinical setting. Furthermore, the results of this study were converted into a web calculator that can be used in clinical practice.

Mixture of Experts for Unmanned Aerial Vehicle Motor Thrust Models

AbstractThis study aims to analyze three unmanned aerial vehicles’ static and dynamic motor thrust models and to compare them with flight test telemetry motor angular velocity data as ground truth. This comparison determines which model is the most accurate based on the root mean square of the difference between the motor models and the telemetry data values. The Burgers and Staples models are dynamic thrust models with nonlinear drag, while the Gibiansky model is a static thrust model with linear drag. The mixture of experts (MoE) architecture assigns weights to each thrust model through a gating network such that higher weights map to the more accurate models. Flight tests include two maneuvers: triangle and shoelace loop. The triangle maneuver uses the cornering option to stop and turn at each waypoint. In contrast, the shoelace loop maneuver uses the curved option to fly smooth curves at waypoints instead of stopping and turning at each waypoint. Simulation results of the motor thrust models use the telemetry data to compute the motor angular velocities with tuned model parameters tailored to the type of maneuver. Regarding motor angular velocity estimation, the Gibiansky and Burgers models tend to overestimate, while the Staples model tends to underestimate. The combination of the simulation results of the models with the mixture of experts allows us to achieve higher accuracy than any of the individual models.

FAS-UNet: A Novel FAS-Driven UNet to Learn Variational Image Segmentation

Solving variational image segmentation problems with hidden physics is often expensive and requires different algorithms and manually tuned model parameters. The deep learning methods based on the UNet structure have obtained outstanding performances in many different medical image segmentation tasks, but designing such networks requires many parameters and training data, which are not always available for practical problems. In this paper, inspired by the traditional multiphase convexity Mumford–Shah variational model and full approximation scheme (FAS) solving the nonlinear systems, we propose a novel variational-model-informed network (FAS-UNet), which exploits the model and algorithm priors to extract the multiscale features. The proposed model-informed network integrates image data and mathematical models and implements them through learning a few convolution kernels. Based on the variational theory and FAS algorithm, we first design a feature extraction sub-network (FAS-Solution module) to solve the model-driven nonlinear systems, where a skip-connection is employed to fuse the multiscale features. Secondly, we further design a convolutional block to fuse the extracted features from the previous stage, resulting in the final segmentation possibility. Experimental results on three different medical image segmentation tasks show that the proposed FAS-UNet is very competitive with other state-of-the-art methods in the qualitative, quantitative, and model complexity evaluations. Moreover, it may also be possible to train specialized network architectures that automatically satisfy some of the mathematical and physical laws in other image problems for better accuracy, faster training, and improved generalization.