Physics-informed machine learning models for ship speed prediction

Virtual sensors, or soft sensors, have greatly contributed to the evolution of the sensing systems in industry. The soft sensors are process models having three fundamental categories, namely white-box (WB), black-box (BB) and gray-box (GB) models. WB models are based on process knowledge while the BB models are developed using data collected from the process. The GB models integrate the WB and BB models for addressing the concerns, i.e., accuracy and intuitiveness, of industrial operators. In this work, various design aspects of the GB models are discussed followed by their application in the process industry. In addition, the changes in the data-driven part of the GB models in the context of enormous amount of process data collected in Industry 4.0 are elaborated.

Thermal models of buildings are helpful to forecast their energy use and to enhance the control of their mechanical systems. However, these models are building-specific and require a tedious, error-prone and time-consuming development effort relying on skilled building energy modelers. Compared to white-box and gray-box models, data-driven (black-box) models require less development time and a minimal amount of information about the building characteristics. In this paper, autoregressive neural network models are compared to gray-box and black-box linear models to simulate indoor temperatures. These models are trained, validated and compared to actual experimental data obtained for an existing commercial building in Montreal (QC, Canada) equipped with roof top units for air conditioning. Results show that neural networks mimic more accurately the thermal behavior of the building when limited information is available, compared to gray-box and black-box linear models. The gray-box model does not perform adequately due to its under-parameterized nature, while the linear models cannot capture non-linear phenomena such as radiative heat transfer and occupancy. Therefore, the neural network models outperform the alternative models in the presented application, reaching a coefficient of determination R2 up to 0.824 and a root mean square error down to 1.11 °C, including the error propagation over time for a 1-week period with a 5-minute time-step. When considering a 50-hour time horizon, the best neural networks reach a much lower root mean square error of around 0.6 °C, which is suitable for applications such as model predictive control.

One promising approach to reduce carbon foot print of internal combustion engines (ICEs) is using alternative fuels like hydrogen, particularly by converting medium and heavy-duty diesel engines to dual-fuel hydrogen-diesel engines. To minimize elevated NOx emissions from hydrogen-fueled engine, fast and accurate emission models are essential for engine model-based control and for engine calibration and optimization using hardware-in-the-loop (HIL) setups. In this study, a fast-response NOx emissions sensor is used to measure the transient NOx emissions from a dual-fuel hydrogen-diesel engine. Subsequently, steady-state models (SSMs), quasi steady-state models (QSSMs), and transient sequential models (TSMs) in the form of black-box (BB) and gray-box (GB) models are developed for transient NOx emissions prediction. GB models utilize both information from a one dimensional (1D) physical engine model and experimental data for training, while BB models only use experimental data. SSMs are optimized artificial neural networks (ANNs) trained using steady-state data, QSSMs are optimized ANNs trained using transient data, and TSMs are time-series networks trained using transient data. Long short-term memory (LSTM) and gated recurrent unit (GRU) networks are used as the time-series deep learning networks. The results showed that the 1D physical model has the poorest performance with successive model performance improvement from SSM to QSSM and from QSSM to TSM. The developed BB TSM model in this study can predict transient NOx emissions with an R2 value greater than 0.96 at 89,000 predictions per second which makes this model suitable for real-time engine model-based control where computational efficiency is crucial. The developed GB TSM model can predict transient NOx emissions with an R2 value greater than 0.97 but it is computationally more expensive. The extra accuracy of the GB TSM models makes them the best choice for HIL setups where more computational power is available, and accuracy is more crucial.

• Enhanced SSV airflow rate estimation of empirical models via data-driven modeling • Achieved high accuracy by customizing models individually for each room • Compared gray-box (physics-informed) with black-box (purely data-driven) models • Assessed data efficiency and model generalization using learning curve analysis • Revealed model robustness and key factors affecting airflow by sensitivity analysis Empirical models are widely used in building design and simulation to estimate airflow rates for single-sided natural ventilation (SSV). However, these models often suffer from limited accuracy. Data-driven modeling offers a viable solution to improve model accuracies by customizing models for various practical scenarios. This study investigates the feasibility and performance of two data-driven modeling approaches for SSV: gray-box modeling (physics-informed using polynomial regressions) and black-box modeling (purely data-driven using machine learning regressions). Taking SSV through a solitary bottom-hung window as a case study, training data were acquired from a full-scale test bench via tracer gas measurements conducted over 11 days. The results indicate that both modeling approaches significantly improve accuracy compared to existing empirical models from the literature. Notably, individual models tailored for each test room generally outperform generalized models developed for all rooms. This finding highlights the importance of customizing models on-site in real-world applications. When comparing the two modeling approaches, gray-box models require lower training ratios to achieve good generalization, demonstrating higher data efficiency than black-box models. Furthermore, sensitivity analysis reveals that gray-box models are more robust and reliable during interpolation and extrapolation, making them particularly suitable for scenarios with limited data availability. Incorporating opening area corrections and selecting root mean square error (RMSE) as the objective function are crucial for achieving optimal gray-box model performance. Overall, this study highlights the potential of physics-informed gray-box data-driven modeling as an advanced strategy for enhancing SSV airflow rate estimation, providing valuable insights to guide future research.

In the present paper the roll damping decay is investigated for the KRISO Container Ship in various working conditions. For this purpose flow is simulated by solving numerically the unsteady three dimensional equations of fluid motion in which closure to the turbulence is achieved through the DES-SST model. The governing equations are solved by using the finite volume method and the free surface elevation is determined by using a VOF technique. Comparisons with the experimental data are provided to validate the numerical approach in terms of the time history of the roll angle variation for different roll motions and ship speeds.

The prediction of the added resistance and attainable ship speed under actual weather conditions is essential to evaluate the true ship performance in operating conditions and assess environmental impact. In this study, a reliable methodology is proposed to estimate the ship speed loss of the S175 container ship in specific sea conditions of wind and waves. Firstly, the numerical simulations are performed to predict the added resistance and ship motions in regular head and oblique seas using three different methods; a 2-D and 3-D potential flow method and a Computational Fluid Dynamics (CFD) with an Unsteady Reynolds-Averaged Navier-Stokes (URANS) approach. Simulations of various wave conditions are compared with the available experimental data and these are used in a validation study. Secondly, following the validation study in regular waves, the ship speed loss is estimated using the developed methodology by calculating the resistance in calm water and the added resistance due to wind and irregular waves, taking into account relevant wave parameters and wind speed corresponding to the Beaufort scale, and results are compared with simulation results obtained by other researchers. Finally, the effect of the variation in ship speed and therefore the ship speed loss is investigated. This study shows the capabilities of the 2-D and 3-D potential methods and CFD to calculate the added resistance and ship motions in regular waves in various wave headings. It also demonstrates that the proposed methodology can estimate the impacts on the ship operating speed and the required sea margin in irregular seas.

Hybrid emission and combustion modeling of hydrogen fueled engines

Self-induced stochastic resonance (SISR) is the emergence of coherent oscillations in slow-fast excitable systems driven solely by noise, without external periodic forcing or proximity to a bifurcation. This work presents a physics-informed machine learning framework for modeling and predicting SISR in the stochastic FitzHugh–Nagumo neuron. We embed the governing stochastic differential equations and SISR-asymptotic timescale-matching constraints directly into a Physics-Informed Neural Network (PINN) based on a Noise-Augmented State Predictor architecture. The composite loss integrates data fidelity, dynamical residuals, and barrier-based physical constraints derived from Kramers’ escape theory. The trained PINN accurately predicts the dependence of spike-train coherence on noise intensity, excitability, and timescale separation, matching results from direct stochastic simulations with substantial improvements in accuracy and generalization compared with purely data-driven methods, requiring significantly less computation. The framework provides a data-efficient and interpretable surrogate model for simulating and analyzing noise-induced coherence in multiscale stochastic systems. • A physics-informed neural network that learns stochastic dynamics directly. • Noise-Augmented State Predictor (NASP) for data-efficient multiscale modeling. • Predicts noise-driven oscillations and coherence from short trajectory data. • Shown on self-induced stochastic resonance in FitzHugh–Nagumo neuron model.

We explore the data-parallel acceleration of physics-informed machine learning (PIML) schemes, with a focus on physics-informed neural networks (PINNs) for multiple graphics processing units (GPUs) architectures. In order to develop scale-robust and high-throughput PIML models for sophisticated applications which may require a large number of training points (e.g., involving complex and high-dimensional domains, non-linear operators or multi-physics), we detail a novel protocol based on h-analysis and data-parallel acceleration through the Horovod training framework. The protocol is backed by new convergence bounds for the generalization error and the train-test gap. We show that the acceleration is straightforward to implement, does not compromise training, and proves to be highly efficient and controllable, paving the way towards generic scale-robust PIML. Extensive numerical experiments with increasing complexity illustrate its robustness and consistency, offering a wide range of possibilities for real-world simulations.

Snow depth or snow water equivalent in mountainous region is crucial for hydrology, water resources management, meteorological and climate research. Remote sensing can be used for snow depth monitoring in regional scale or global scale. However, the spaceborne remote sensing of snow depth in mountain is challenging because of the sensor sensitivity and spatial resolution problems. Recently, time series Sentinel-1 is used for snow depth retrieval in mountains, which shows encouraging accuracy. In this study, an algorithm to estimate the snow depth in mountainous regions using optical and passive microwave remote sensing observations is proposed, which can be applied in periods before and after the launch of Sentinel-1. The optical and passive microwave remote sensing observations and the Sentinel-1 derived snow depth in 2016-2021 are used to train the snow depth retrieval algorithm using the Extreme gradient boosting (XGBoost) machine learning algorithm. Validations are performed using cross validation method and independent <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">in-situ</i> snow depth data. The cross validation shows correlation coefficient of 0.81 and mean absolute error (MAE) of 0.17m. The correlation coefficient and MAE of predicted snow depth and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">in-situ</i> snow depth in 2002-2016 are 0.61 and 0.33 m, respectively, which shows significant higher accuracy compared with AMSR-E/AMSR2 snow depth products. The site dependence of the machine learning method is also discussed. The machine learning based snow depth retrieval presented in this study can be applied to mountains globally to the optical and passive microwave remote sensing era.

Soil moisture is vital for studying hydrological processes, vegetation growth, and climate change. Recently, Global Navigation Satellite System-Reflectometry (GNSS-R) technology has shown great potential for soil moisture monitoring. However, most existing GNSS-R algorithms rely empirically on multiple land surface parameters, which increases model complexity and the risk of overfitting. This study investigates underlying surface influencing factors for soil moisture retrieval and developes a new soil moisture retrieval with considering the factors that influence the relationship between GNSS-R measurements and surface soil moisture, using Cyclone Global Navigation Satellite System (CYGNSS) data in the middle and lower reaches of the Yangtze River (23.5°N-35°N, 110°E-123°E). Monthly CYGNSS reflectivity and SMAP soil moisture data from 2019 to 2024 are analyzed with consideration of influencing factors such as elevation, slope, biomass, land use, tree height, and Normalized Difference Vegetation Index (NDVI). Results show the following key findings: (1) Reflectivity exhibits a significant correlation with soil moisture, with 84.05% displaying a positive correlation. (2) Over the six-year period from 2019 to 2024, the correlation between SMAP soil moisture and CYGNSS reflectivity shows clear seasonal variability, with lower correlations in spring and autumn and higher correlations in summer and winter, peaking in summer. (3) Geodetector analysis revealed that land use had the most significant effect on soil moisture in the single-factor analysis, while biomass and reflectivity had the strongest interaction in the two-factor analysis, characterized by nonlinear enhancement. (4) A retrieval scheme was designed based on the influencing factors, and its accuracy was validated using the eXtreme Gradient Boosting (XGBoost) machine learning algorithm. The root mean square error (RMSE) of the soil moisture retrieval using all selected features was 0.0526 cm<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup>/cm<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup>. The major feature group achieved an RMSE of 0.0527 cm<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup>/cm<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup>, while the minor feature group yielded an RMSE of 0.0611 cm<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup>/cm<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup>. Spatial RMSE analysis further demonstrated the effectiveness of Geodetector in guiding feature selection. The retrieval results were further validated using the SSM V2.0 high-resolution soil moisture product, which has been extensively evaluated against in situ measurements, demonstrating strong consistency and confirming the robustness and practical applicability of the proposed method.

Based on some typical complex network analysis methods and machine learning techniques, a general multiple-measures composited strategy-guided epilepsy diagnosis analysis framework is proposed in this brief paper. Five typical network analysis methods for biology time series analysis are utilized for real applications, including the classical visibility graph (VG), horizontal visibility graph (HVG), the limited penetrable visibility graph (LPVG), the modified frequency degree method (MFDM), and the quantity graph (QG). By using the aforementioned typical transformation methods, the EEG signal sets to be classified are transferred into graph network object sets. The main network features and related indicators are calculated and extracted as features for classification tasks. Some key features are selected via variance analysis, and the eXtreme Gradient Boosting (XGBOOST) machine learning algorithm is used for related binary and five-class classification tasks for electroencephalographic time series. Numerical experiments demonstrate that, through ten-fold cross-validation on the entire dataset, the classification accuracy for two-class classification consistently reaches 97.8% (with a specificity of 97.5%), while for five-class classification, the accuracy stably reaches 82.4% (with a specificity of 95.6%). Therefore, our classification framework can be effectively used to assist hospital doctors and medical specialists in diagnosing related diseases, especially to help accelerate the treatment of epilepsy patients.

Predictive models for the effect of environmental factors on the abundance of Vibrio parahaemolyticus in oyster farms in Taiwan using extreme gradient boosting

Diabetic osteoporosis exhibits heterogeneity at the molecular level. Ferroptosis, a controlled form of cell death brought on by a buildup of lipid peroxidation, contributes to the onset and development of several illnesses. The aim was to explore the molecular subtypes associated with ferroptosis in diabetic osteoporosis at the molecular level and to further elucidate the potential molecular mechanisms. Integrating the CTD, GeneCards, FerrDb databases, and the microarray data of GSE35958, we identified ferroptosis-related genes (FRGs) associated with diabetic osteoporosis. We applied unsupervised cluster analysis to divide the 42 osteoporosis samples from the GSE56814 microarray data into different subclusters based on FRGs. Subsequently, FRGs associated with two ferroptosis subclusters were obtained by combining database genes, module-related genes of WGCNA, and differentially expressed genes (DEGs). Eventually, the key genes from FRGs associated with diabetic osteoporosis were identified using the least absolute shrinkage and selection operator (LASSO), Boruta, support vector machine recursive feature elimination (SVM - RFE), and extreme gradient boosting (XGBoost) machine learning algorithms. Based on ROC curves of external datasets (GSE56815), the model's efficiency was examined. We identified 15 differentially expressed FRGs associated with diabetic osteoporosis. In osteoporosis, two distinct molecular clusters related to ferroptosis were found. The expression results and GSVA analysis indicated that 15 FRGs exhibited significantly different biological functions and pathway activities in the two ferroptosis subclusters. Therefore, we further identified 17 FRGs associated with diabetic osteoporosis between the two subclusters. The results of the comprehensive analysis of 17 FRGs demonstrated that these genes were heterogeneous and had a specific interaction between the two subclusters. Ultimately, the prediction model had a strong foundation and excellent AUC values (0.84 for LASSO, 0.84 for SVM - RFE, 0.82 for Boruta, and 0.81 for XGBoost). IDH1 is a common gene to all four algorithms thus being identified as a key gene with a high AUC value (AUC = 0.698). As a ferroptosis regulator, IDH1 is able to distinguish between distinct molecular subtypes of diabetic osteoporosis, which may offer fresh perspectives on the pathogenesis of the disease's clinical symptoms and prognostic heterogeneity.

In recent years, Jas’s Larch Inchworm (Erannis jacobsoni Djak, EJD) outbreaks have frequently occurred in forested areas of Mongolia, causing significant damage to forest ecosystems, and rapid and effective monitoring methods are urgently needed. This study focuses on a typical region of EJD infestation in the larch forests located in Binder, Khentii, Mongolia. Initial super-resolution enhancement was performed on Sentinel-2 images, followed by the calculation of vegetation indices and first-order spectral derivatives. The Kruskal–Wallis H test (KW test), Dunn’s multiple comparison test (Dunn’s test), and the RF-RFECV algorithm were then employed to identify sensitive features. Using support vector machine (SVM), random forest (RF), and extreme gradient boosting (XGBoost) machine learning algorithms, along with field survey data and UAV remote sensing data, multiple models were developed to assess the severity of EJD infestation and the corresponding spatial distribution characteristics. Seven sensitive combined features were obtained from high-quality super-resolution Sentinel-2 images. Then, a high-precision monitoring model was constructed, and it was revealed that the areas prone to EJD infestation are located at elevations of 1171–1234 m, on gentle slopes, and in semi-shady or semi-sunny areas. The super-resolution processing of Sentinel-2 satellite data can effectively refine monitoring results. The combination of the first-order spectral derivatives and vegetation indices can improve the monitoring accuracy and the discrimination of light and moderate damage. D8a and NDVIswir can be used as important indicators for assessing the severity of EJD infestation. EJD has an adaptive preference for certain environments, and environmental factors directly or indirectly affect the diffusion and distribution of EJD.