Numerical Machine Learning Research Articles

Machine Learning‐Guided Design of 10 nm Junctionless Gate‐All‐Around Metal Oxide Semiconductor Field Effect Transistors for Nanoscaled Digital Circuits

In this paper, we introduce an innovative design approach based on combined numerical simulations and machine learning (ML) analysis to investigate the design key parameters of ultra‐low scale junctionless gate‐all‐around (JLGAA) field‐effect transistor (FET) devices. To this end, precise 3D numerical models that incorporate quantum effects and ballistic transport are employed to simulate the current–voltage (I–V) characteristics of 10 nm‐scale JLGAA FET devices. The influence of design parameter variations and high‐k dielectric material on the subthreshold characteristics is thoroughly examined. Various ML algorithms were employed to analyze and classify the key design parameters influencing the subthreshold figures‐of‐merit (FoMs), the subthreshold swing (SS) factor and ION/IOFF ratio. The obtained results highlight that channel radius and channel doping design parameters are particularly important for affecting swing factor behavior. Similarly, these features also play a significant role in predicting and affecting ION/IOFF current ratio values. Additionally, machine learning is used to determine the optimal design parameters for each figure of merit (FoM) output value. In this context, the models effectively predicted both ION/IOFF current ratios and SS classification, with Naive Bayes achieving an accuracy of 90.8% for ION/IOFF and 92.6% for SS, showcasing the model's robustness in these classification tasks.

Compressive behavior of elliptical concrete-filled steel tubular short columns using numerical investigation and machine learning techniques

This paper presents a non-linear finite element model (FEM) to predict the load-carrying capacity of three different configurations of elliptical concrete-filled steel tubular (CFST) short columns: double steel tubes with sandwich concrete (CFDST), double steel tubes with sandwich concrete and concrete inside the inner steel tube, and a single outer steel tube with sandwich concrete. Then, a parametric and analytical study was performed to explore the influence of geometric and material parameters on the load-carrying capacity of elliptical CFST short columns. Furthermore, the current study investigates the effectiveness of machine learning (ML) techniques in predicting the load-carrying capacity of elliptical CFST short columns. These techniques include Support Vector Regressor (SVR), Random Forest Regressor (RFR), Gradient Boosting Regressor (GBR), XGBoost Regressor (XGBR), MLP Regressor (MLPR), K-nearest Neighbours Regressor (KNNR), and Naive Bayes Regressor (NBR). ML models accuracy is assessed by comparing their predictions with FE results. Among the models, GBR and XGBR exhibited outstanding results with high test R2 scores of 0.9888 and 0.9885, respectively. The study provided insights into the contributions of individual features to predictions using the SHapley Additive exPlanations (SHAP) approach. The results from SHAP indicate that the eccentric loading ratio (e/2a) has the most significant effect on the load-carrying capacity of elliptical CFST short columns, followed by the yield strength of the outer steel tube (\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\:{f}_{yo}$$\\end{document}) and the inner width of the inner steel tube (\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\:2{a}_{ii}$$\\end{document}). Additionally, a user interface platform has been developed to streamline the practical application of the proposed ML.

The Impact of Climate Change and Urbanization on Compound Flood Risks in Coastal Areas: A Comprehensive Review of Methods

Many cities worldwide are increasingly threatened by compound floods resulting from the interaction of multiple flood drivers. Simultaneously, rapid urbanization in coastal areas, which increases the proportion of impervious surfaces, has made the mechanisms and simulation methods of compound flood disasters more complex. This study employs a comprehensive literature review to analyze 64 articles on compound flood risk under climate change from the Web of Science Core Collection from 2014 to 2024. The review identifies methods for quantifying the impact of climate change factors such as sea level rise, storm surges, and extreme rainfall, as well as urbanization factors like land subsidence, impervious surfaces, and drainage systems on compound floods. Four commonly used quantitative methods for studying compound floods are discussed: statistical models, numerical models, machine learning models, and coupled models. Due to the complex structure and high computational demand of three-dimensional joint probability statistical models, along with the increasing number of flood drivers complicating the grid interfaces and frameworks for coupling different numerical models, most current research focuses on the superposition of two disaster-causing factors. The joint impact of three or more climate change-driving factors on compound flood disasters is emerging as a significant future research trend. Furthermore, urbanization factors are often overlooked in compound flood studies and should be considered when establishing models. Future research should focus on exploring coupled numerical models, statistical models, and machine learning models to better simulate, predict, and understand the mechanisms, evolution processes, and disaster ranges of compound floods under climate change.

Detection of solar internal flows with numerical simulation and machine learning

Detection of solar internal flows with numerical simulation and machine learning

Advancing photovoltaic panel temperature forecasting: A comparative study of numerical simulation and machine learning in two types of PV power plant

Photovoltaic (PV) panel temperature dynamic monitoring and forecasting is important for managing and maintaining of PV power plant. However, it is uncommon to use a variety of methods to predict and evaluate the panel temperature of different types of PV power plants. Therefore, this study aims to advance PV panel temperature forecasting through a comparative analysis of numerical simulation and machine learning models in two types of PV power plants: land- and water-mounted. The results indicate that PV panel temperature condition for two types of PV power plants can be well captured by the numerical simulation (NS) and machine learning, except for the NS in water-mounted PV power plant (R2 with 0.66). Models perform better in land-mounted PV power plants, with Random Forest Regression (RFR) and ResNet models demonstrating superior accuracy across both environments. While all models exhibit some degree of deviation from normal distribution in residuals, RFR and ResNet show the least deviation and lowest prediction errors, highlighting their robustness in forecasting PV panel temperatures. This study can provide valuable insights for predicting PV panel temperatures and for the management and maintenance of PV power plants.

Modeling of a metal hydride energy storage tank dynamics using hybrid numerical, experimental, and machine learning methods

This study presents an integrated analysis combining numerical simulations, experimental investigations, and machine learning models to simulate the performance of metal hydride systems for hydrogen storage under various conditions by using a LaNi5 metal hydride cylindrical tank of 500 NL capacity, with a focus on PCM thermal enhancements and surface water heating and cooling. Numerical simulations offer insights into thermal and reaction dynamics, while experimental data are used for model validation and supplemented with additional numerical data for training a FFNN model to predict dynamic hydrogen flow, surface temperature, and surface heat flux. The obtained mean RMSEs for charging/discharging experimental data is 2.8 SOC% for the state of charge and 0.3 °C for surface temperatures. The study compares the effectiveness of water cooling and heating in metal hydride surface across different temperatures, highlighting their impact on the efficiency of hydrogen absorption and desorption processes. Energy consumption is shown to be non-linear; during charging, heat flux peaks before decreases rapidly within around 0.2 h. During discharging, heat flux rises steadily. The study also explores the impact of adding copper fins to the PCM covering metal hydride system, revealing significant improvements in charging and discharging efficiency, specifically, the integration of 16 copper fins led to a notable decrease in complete charging time from approximately 6.5 h to just over 5.2 h, and in complete discharging time from around 4.5 h to under 3.96 h. Using PCM, the highest heat flux is 5 times less during charge and 2 times less during discharge than that for water cooling mode.

High-risk nuclide screening and parameter sensitivity analysis based on numerical simulation and machine learning

During nuclear accidents, large quantities of radionuclides will be released into the environment, posing serious health hazards to local residents. The screening of high-risk nuclides is critical for the development of subsequent nuclear emergency response measures. In order to overcome the shortcomings of traditional screening methods, a machine learning method was proposed to screen high-risk nuclides and predict their contamination to groundwater more effectively. The performances of Support Vector Machine (SVM), Random Forest (RF) and Back Propagation Neural Network (BPNN) algorithms were compared, and sensitivity analyses of the initial leakage concentration ratio (C0/Cp), distribution coefficient (Kd) and decay coefficient (λ) on the model outputs were performed. Results showed that RF classification model achieved the highest prediction accuracy for screening high-risk nuclides. The contribution of the input parameters ranked as Kd > λ > C0/Cp. BPNN regression model was found to be the best for predicting when high-risk nuclides would pollute groundwater. The output was negatively correlated with C0/Cp and positively correlated with Kd and λ, with the parameter influence ranking as Kd > C0/Cp > λ. The contribution of Kd mainly came from itself, and the contribution of C0/Cp and λ mainly due to their interaction with other parameters.

Message-Passing Monte Carlo: Generating low-discrepancy point sets via graph neural networks

Discrepancy is a well-known measure for the irregularity of the distribution of a point set. Point sets with small discrepancy are called low discrepancy and are known to efficiently fill the space in a uniform manner. Low-discrepancy points play a central role in many problems in science and engineering, including numerical integration, computer vision, machine perception, computer graphics, machine learning, and simulation. In this work, we present a machine learning approach to generate a new class of low-discrepancy point sets named Message-Passing Monte Carlo (MPMC) points. Motivated by the geometric nature of generating low-discrepancy point sets, we leverage tools from Geometric Deep Learning and base our model on graph neural networks. We further provide an extension of our framework to higher dimensions, which flexibly allows the generation of custom-made points that emphasize the uniformity in specific dimensions that are primarily important for the particular problem at hand. Finally, we demonstrate that our proposed model achieves state-of-the-art performance superior to previous methods by a significant margin. In fact, MPMC points are empirically shown to be either optimal or near-optimal with respect to the discrepancy for low dimension and small number of points, i.e., for which the optimal discrepancy can be determined.

Performance Analysis and Rapid Optimization of Vehicle ORC Systems Based on Numerical Simulation and Machine Learning

The organic Rankine cycle (ORC) system is an important technology for recovering energy from the waste heat of internal combustion engines, which is of significant importance for the improvement of fuel utilization. This study analyses the performance of vehicle ORC systems and proposes a rapid optimization method for enhancing vehicle ORC performance. This study constructed a numerical simulation model of an internal combustion engine-ORC waste heat recovery system based on GT-Suite software v2016. The impact of key operating parameters on the performance of two organic Rankine cycles: the simple organic Rankine cycle (SORC) and the recuperative organic Rankine cycle (RORC) was investigated. In order to facilitate real-time prediction and optimization of system performance, a data-driven rapid prediction model of the performance of the waste heat recovery system was constructed based on an artificial neural network. Meanwhile, the NSGA-II multi-objective algorithm was used to investigate the competitive relationship between different performance objective functions. Furthermore, the optimal operating parameters of the system were determined by utilizing the TOPSIS method. The results demonstrate that the highest thermal efficiencies of the SORC and RORC are 6.21% and 8.61%, respectively, the highest power outputs per unit heat transfer area (POPAs) are 6.98 kW/m2 and 8.99 kW/m2, respectively, the lowest unit electricity production costs (EPC) are 7.22 × 10−2 USD/kWh and 3.15 × 10−2 USD/kWh, respectively, and the lowest CO2 emissions are 2.85 ton CO2,eq and 3.11 ton CO2,eq, respectively. The optimization results show that the RORC exhibits superior thermodynamic and economic performance in comparison to the SORC, yet inferior environmental performance.

Optimization of Supercritical CO2 Fracturing Based on Random Forest-Particle Swarm Optimization Model and Pre-existing Fracture Network

Summary On a global scale, shale oil/gas has become an important alternative energy source for conventional oil and gas. The potential advantages of supercritical CO2 (ScCO2) make it an ideal alternative to hydraulic fracturing, used for shale reservoir transformation and production increase while also promoting the geological storage of CO2, which is in-line with today’s carbon capture, utilization, and storage technology and helps to address the challenges of global climate change. To further study the fracture propagation and optimization of a complex fracture network (CFN) in ScCO2 fracturing under complex geological conditions using the cohesive module of ABAQUS to establish a fluid structure coupling model and completing indoor and field experimental verification, we introduce the global embedded cohesion zone model (CZM) combined with Python to generate two natural-fracture (NF) distribution models, conjugate and power law, to establish a dispersed mesh model. Based on this model, we studied the fracture propagation problem of ScCO2 fracturing under different engineering and geological conditions. The simulation results will be used as data-driven data to establish an optimization model of the random forest-particle swarm optimization algorithm (RF-PSO) and optimize the CFN. Research has shown that (1) ScCO2 is more inclined to pass through NFs and propagate in the rock matrix, and hydraulic fractures (HFs) combine better with NFs. Compared with hydraulic fracturing, ScCO2 fracturing has significant advantages (only the fracture width is lower than hydraulic fracturing, its initiation pressure and fracture length are much better than hydraulic fracturing, and there are more small fractures, making it easier to form a CFN). (2) During the process of fracture propagation, once dominant fractures form, the trend of the “Matthew effect” is inevitable. The process of fracture propagation is influenced by multiple factors, especially the distribution of NFs; the larger the reservoir filtration coefficient is, the more ScCO2 fracturing fluid that is lost, which is more unfavorable for fracturing construction. While maintaining the same amount of fracturing fluid injection, as the displacement increases, the fracture complexity increases, and the fracturing control range expands. Compared with other parameters, the effect of fracturing fluid temperature (FFT) on the expansion of ScCO2 fracturing fractures is not significant. (3) The established RF-PSO optimization model has an error of 2.89%, which can well adapt to CFN optimization problems under complex NF conditions and reduce uncertainty. We propose in this article a research method for fracture network optimization from fracture modeling, dynamic simulation, and optimization modeling. By combining numerical simulation and machine learning, the CFN optimization design of ScCO2 fracturing under CFN conditions is achieved, providing a research approach for the optimization of fracturing in fractured reservoirs.

Generic Models for Predicting Coseismic Displacements of Earth Slopes Based on Numerical Analysis and Machine Learning Algorithm

Generic Models for Predicting Coseismic Displacements of Earth Slopes Based on Numerical Analysis and Machine Learning Algorithm

Prediction of Steam Turbine Blade Erosion Using Computational Fluid Dynamics Simulation Data and Hierarchical Machine Learning

Abstract The information of the degree of blade erosion is vital for the efficient operation of steam turbines. However, it is nearly impossible to directly measure the degree of blade erosion during operation. Moreover, collecting sufficient data of eroded cases for predictive analysis is challenging. Therefore, this paper proposes a blade erosion prediction method using numerical simulation and machine learning. Pressure data of several blade erosion cases are collected from the numerical turbine simulation. The machine learning approach involves training on collected simulation data to predict the degree of erosion for the first-stage stator (1S) and the first-stage rotor blade (1R) from internal pressure data. The proposed erosion prediction model employs a two-step hierarchical approach. First, the proposed model predicts the 1S erosion degree using the k-nearest neighbor (k-NN) regression. Second, the proposed model estimates the 1R erosion degree with linear regression models. These models are tailored for each of the 1S erosion degrees, utilizing pressure data processed through fast Fourier transform (FFT). The evaluation shows that the proposed method achieves the prediction of the 1S erosion with a mean absolute error (MAE) of 0.000693 mm and the 1R erosion with an MAE of 0.458 mm. The evaluation results indicate that the proposed method can accurately capture the degree of turbine blade erosion from internal pressure data. As a result, the proposed method suggests that the erosion prediction method can be effectively used to determine the optimal timing for maintenance, repair, and overhaul (MRO).

Neural networks for emergent behavior in biological microstructures

PurposeThis paper develops a neural network surrogate model based on a discrete lattice approach to investigate the influence of complex microstructures on the emergent behavior of biological networks.Design/methodology/approachThe adaptability of network-forming organisms, such as, slime molds, relies on fluid-to-solid state transitions and dynamic behaviors at the level of the discrete microstructure, which continuum modeling methods struggle to capture effectively. To address this challenge, we present an optimized approach that combines lattice spring modeling with machine learning to capture dynamic behavior and develop nonlinear constitutive relationships.FindingsThis integrated approach allows us to predict the dynamic response of biological materials with heterogeneous microstructures, overcoming the limitations of conventional trial-and-error lattice design. The study investigates the microstructural behavior of biological materials using a neural network-based surrogate model. The results indicate that our surrogate model is effective in capturing the behavior of discrete lattice microstructures in biological materials.Research limitations/implicationsThe combination of numerical simulations and machine learning endows simulations of the slime mold Physarum polycephalum with a more accurate description of its emergent behavior and offers a pathway for the development of more effective lattice structures across a wide range of applications.Originality/valueThe novelty of this research lies in integrating lattice spring modeling and machine learning to explore the dynamic behavior of biological materials. This combined approach surpasses conventional methods, providing a more holistic and accurate representation of emergent behaviors in organisms.

Enhanced predictive modeling of Nusselt number in boiler tubes: numerical simulations and machine learning for water and SiO2/water

Enhanced predictive modeling of Nusselt number in boiler tubes: numerical simulations and machine learning for water and SiO2/water

Wellbore salt-deposition risk prediction of underground gas storage combining numerical modeling and machine learning methodology

Strengthening the construction of underground gas storage (UGS) is significant for securing national energy and achieving carbon neutrality. However, salt deposition occurs in the wellbore affecting the safety of UGS seriously. How to predict the salt-deposition risk in UGS wells accurately and quickly is still challenging. This work aims at establishing a fast and convenient framework of wellbore salt-deposition risk prediction in UGS wells. Firstly, the influence of the production parameters on wellbore salt-deposition is analyzed by an in-house simulator. Further, an intelligent diagnostic model of salt-deposition risk is proposed based on the machine learning algorithm. The effect of salt deposition on reservoir is analyzed by reservoir salt-deposition simulation model. Finally, downhole video survey is used to validate the modeling accuracy of diagnostic model. Results indicate that the contribution ratio of daily gas production and water-gas-ratio to wellbore salt deposition is 3:2. After the salt deposition, the reduction in water saturation and the permeability of near wellbore areas are 77.14 % and 98.91 %. The modeling accuracy of diagnostic model is 97 %, which is applied in the Y UGS successfully. The proposed framework can predict the salt-deposition risk of UGS wells in real time to help engineers control salt clogging issues.

CViewer: a Java-based statistical framework for integration of shotgun metagenomics with other omics datasets

BackgroundShotgun metagenomics for microbial community survey recovers enormous amount of information for microbial genomes that include their abundances, taxonomic, and phylogenetic information, as well as their genomic makeup, the latter of which then helps retrieve their function based on annotated gene products, mRNA, protein, and metabolites. Within the context of a specific hypothesis, additional modalities are often included, to give host-microbiome interaction. For example, in human-associated microbiome projects, it has become increasingly common to include host immunology through flow cytometry. Whilst there are plenty of software approaches available, some that utilize marker-based and assembly-based approaches, for downstream statistical analyses, there is still a dearth of statistical tools that help consolidate all such information in a single platform. By virtue of stringent computational requirements, the statistical workflow is often passive with limited visual exploration.ResultsIn this study, we have developed a Java-based statistical framework (https://github.com/KociOrges/cviewer) to explore shotgun metagenomics data, which integrates seamlessly with conventional pipelines and offers exploratory as well as hypothesis-driven analyses. The end product is a highly interactive toolkit with a multiple document interface, which makes it easier for a person without specialized knowledge to perform analysis of multiomics datasets and unravel biologically relevant patterns. We have designed algorithms based on frequently used numerical ecology and machine learning principles, with value-driven from integrated omics tools which not only find correlations amongst different datasets but also provide discrimination based on case–control relationships.ConclusionsCViewer was used to analyse two distinct metagenomic datasets with varying complexities. These include a dietary intervention study to understand Crohn’s disease changes during a dietary treatment to include remission, as well as a gut microbiome profile for an obesity dataset comparing subjects who suffer from obesity of different aetiologies and against controls who were lean. Complete analyses of both studies in CViewer then provide very powerful mechanistic insights that corroborate with the published literature and demonstrate its full potential.EDLr-6-tH-9CdVCxoa6z_8Video

Design optimization of groundwater circulation well based on numerical simulation and machine learning

The optimal design of groundwater circulation wells (GCWs) is challenging. The key to purifying groundwater using this technique is its proficiency and productivity. However, traditional numerical simulation methods are limited by long modeling times, random optimization schemes, and optimization results that are not comprehensive. To address these issues, this study introduced an innovative approach for the optimal design of a GCW using machine learning methods. The FloPy package was used to create and implement the MODFLOW and MODPATH models. Subsequently, the formulated models were employed to calculate the characteristic indicators of the effectiveness of the GCW operation, including the radius of influence (R) and the ratio of particle recovery (Pr). A detailed collection of 3000 datasets, including measures of operational efficiency and key elements in machine learning, was meticulously compiled into documents through model execution. The optimization models were trained and evaluated using multiple linear regression (MLR), artificial neural networks (ANN), and support vector machines (SVM). The models produced by the three approaches exhibited notable correlations between anticipated outcomes and datasets. For the optimal design of circulating well parameters, machine learning methods not only improve the optimization speed, but also expand the scope of parameter optimization. Consequently, these models were applied to optimize the configuration of the GCW at a site in Xi’an. The optimal scheme for R (Q = 293.17 m3/d, a = 6.09 m, L = 7.28 m) and optimal scheme for Pr (Q = 300 m3/d, a = 3.64 m, L = 1 m) were obtained. The combination of numerical simulations and machine learning is an effective tool for optimizing and predicting the GCW remediation effect.

From simulation to reality: CFD-ML-driven structural optimization and experimental analysis of thermal plasma reactors

Thermal plasma reactors offer an environment of high temperature, enthalpy, and reactivity, making them highly efficient for solid waste treatment and promising for clean energy production from municipal and industrial waste. Optimal geometrical parameters of the reactor can enhance waste treatment and reactor performance. This study presents a comprehensive analysis of 11 key geometrical parameters of a thermal plasma reactor. Utilizing CFD Fluent software, numerical simulations were conducted to generate a dataset. Subsequently, a predictive model focusing on the average temperature in the core melt zone was trained using six Machine Learning (ML) algorithms. The Particle Swarm Optimisation (PSO) algorithm optimized the hyperparameters of the Gradient Booster Regression (GBR) model, which was combined with a Genetic Algorithm (GA) to identify the reactor's optimal geometrical parameters. A DC arc plasma torch-solid waste thermal plasma reactor treatment system was established on this basis. The study also explored the effects of gasification coefficient, reaction temperature, and thermal plasma jet mode on system performance. Findings indicate that the PSO-GBR model achieved the highest prediction accuracy, with the temperature in the core reaction zone reaching 3621 K. The deviation between numerical simulations and machine learning predictions was a mere 1.3%. Enhancing syngas yield and energy efficiency is achievable by controlling reaction temperature and increasing the gasification coefficient. A laminar plasma jet mode, at equal power, provides a more effective reaction environment. The accuracy and reliability of the machine learning-driven regression model and optimization results are significant in guiding the optimal design of plasma reactors and advancing waste-to-energy conversion processes.

Combined effect of random porosity and surface defect on fatigue lifetime of additively manufactured micro-sized Ti6Al4V components: An investigation based on numerical analysis and machine learning approach

Surface defects and internal porosities resulting from the additive manufacturing process contribute to a scatter in fatigue lifetime, increasing uncertainty in applications such as aerospace engineering. This study proposes a combined approach of machine learning and finite element modelling to explore the interaction between random porosity distribution and high surface roughness on the fatigue lifetime of micro-sized additively manufactured parts. To achieve this, a high-fidelity finite element model is constructed with randomly distributed porosities and surface defects. A fatigue lifetime estimation model, developed using a combination of continuum damage mechanics and the theory of critical distance, is then applied to predict lifetime using effective stress extracted from simulations. A total of 1200 simulations are conducted using these models to examine the interaction effects of subsurface porosity and surface defects on localized stress distribution and fatigue lifetime. Subsequently, two machine learning algorithms, Support Vector Regression and Kernel Ridge Regression, are employed to estimate fatigue lifetime, accounting for the effects of randomly distributed subsurface porosities and surface defects. Eighty percent of the developed finite element models are employed to construct an input dataset, incorporating geometrical features of pores, and to train the algorithms, while the remaining 20% are reserved for verification purposes. The observed results demonstrate that the combined effect of random porosity and surface defects increases fatigue scatter behaviour, leading to greater uncertainty in lifetime prediction. Furthermore, the machine learning algorithms exhibit promising prediction performance.

Bidirectional and nested model for numerical simulation and machine learning: a case study of long-term settlement prediction of tunnel

Numerical simulation and machine learning are commonly adopted research methods in engineering. This paper proposes a bidirectional and nested model for numerical simulation and machine learning (BNNM). This model permits numerical simulation methods and machine learning methods to participate in each other’s calculation process. It helps overcome the obstruction of unclear mechanisms and inaccurate parameters in numerical simulation methods, and avoid overfitting problem caused by too many features in machine learning methods. Moreover, BNNM frees machine learning methods from the dependence on a specific set of labels. The BNNM helps train machine learning models using obtainable labels, and output results that cannot be easily obtained using field, experiment, and numerical simulations. To illustrate its construction method and performance, a representative BNNM model is constructed using BPNN, in addition to a simple numerical simulation model. This model predicts the long-term settlement of shield tunnel. The results show that the representative model effectively reduces the modelling difficulty associated with numerical simulation and improves prediction accuracy of BPNN model. The model also derives long-term constitutive models of various soils with only the tunnel settlement data set. Although a simplified constitutive model was used, the main advantages of the BNNM model have been highlighted.