Data-driven Model Research Articles

Efficacious application of data-driven machine learning models for predicting and optimizing the flexural tensile strength of fiber-reinforced concrete

In the literature, a limited number of studies have utilized the random forest (RF) technique to model the flexural tensile strength (FTS) of steel fiber reinforced concrete (SFRC). This study established modeling of the FTS of SFRC containing mono-fibrous (exclusively crimped and hooked) systems. A comprehensive database of 193 records was used from 18 published studies (including 30 documents sampled by authors). Machine-learning numerical models were developed and evaluated. Partial dependence plots were constructed to visualize the relationship between the model parameters and rank the parameters based on their importance. It has been found that the RF model can be rationally used to predict and optimize the FTS of SFRC, as most of the predicted–tested results were close to the ±80% accuracy range. The model showed of underestimation nature for SFRC with high FTS values. The Gini index-based importance analysis showed that the fiber’s content and tensile strength had the highest significance among the other variables. A drastic increase in the FTS by an increase in the fibrous dosage was achieved. However, no major SFRC’s tensile strength enhancement for dosages was more than 1.5%, while its optimum diameter was 0.5–0.7mm. The results also illustrated that the effect of the fiber’s length is quite similar to that of the fiber content. However, no significant increase can be achieved by increasing the fiber’s length from 40 to 60mm.

Wind Energy Harvesting with Building-Integrated Ducted Openings: CFD Simulation and Neural Network Optimization

Building-integrated wind energy harvesting offers a promising local and decentralized energy generation solution, reducing transmission losses and infrastructure expenses. However, the complex urban wind patterns demand accurate aerodynamic optimization for optimal performance. This study seeks to optimize the geometric design of building-integrated ducted openings to maximize wind energy harvesting. The research aims to create a data-driven model predicting the openings' performance accurately, explore the inter-linked impacts of different geometrical variables, find optimal configurations across different building heights, and understand underlying aerodynamic mechanisms. The proposed design comprises a nozzle, throat, and diffuser. To increase power density, a data-driven design optimization method is employed. The optimization process involves determining the optimum opening height from the ground, along with five geometrical variables: inlet diameter, throat diameter, outlet diameter, nozzle length, and diffuser length. A comprehensive test matrix is developed using design of experiments (DoE) to build an artificial neural network (ANN) model. This neural network is developed using computational fluid dynamics (CFD) simulations validated with experimental data. Subsequently, a gradient-based global optimization method fine-tunes the neural network and identifies optimal designs for various heights. The research concludes that nozzle and diffuser opening angles within the range of 19° ≤ ΦNoz ≤ 23°and 7.5° ≤ ΦDif ≤ 8.2°, respectively, result in the optimal converging-diverging ducted opening geometry. This optimal opening can yield a power density increase of 31.5 to 35.5 times when compared to freestream flow and 17.5 to 22.4 times in comparison to cylindrical openings, depending on the opening location along the building height.

Separate-layer injection scheme optimization based on integrated injection information with artificial neural network and residual network

Abstract Separate-layer injection technology is a highly significant approach for enhancing oil recovery in the later stages of oilfield production. Both separate-layer and general injection information are crucial parameters in multi-layer oilfield injection systems. However, the significance of general injection information is usually overlooked during the optimization process of separate-layer injection. Moreover, conventional optimization schemes for separate-layer injection fails to meet the immediate and dynamic demands of well production. Consequently, a separate-layer injection optimization method based on ANN-Res model was proposed. Firstly, the primary controlling factors for production were identified through grey correlation analysis and ablation experiments. Then, a data-driven model was established with artificial neural network (ANN), in which the residual block was utilized to incorporate general injection information, eventually formed an ANN-Res model that integrates separate-layer and general injection information. Finally, a workflow for separate-layer injection optimization was designed in association with the ANN-Res model. Analysis of primary controlling factor for production shows that the combination of separate-layer and general injection information for production prediction leads to redundancy. The results of injection-production prediction demonstrate that the ANN-Res model is significantly better than that of the ANN model which only inputs separate-layer or general injection information. Furthermore, the result of optimization proves the proposed method can be successfully applied to injection optimization, realizing the purpose of increasing oil production and decreasing water cut, thereby improving oilfield development.

Digital twin-enabled robust production scheduling for equipment in degraded state

Technological advancements are leading to a world where digital twins will become integral to manufacturing operations management. While wide-ranging applications of digital twins are being researched, robust production scheduling remains an enduring challenge, especially considering the numerous sources of uncertainty in a complex manufacturing system that can affect the validity of the obtained solution. Thus, the article proposes a Prognostics and Health Management (PHM)-enabled digital twin framework to perform job scheduling for a flow shop scheduling problem considering real-time equipment health state. The framework combines the adoption of a Genetic Algorithm optimizer with data-driven modelling leveraging algorithms like Principal Component Analysis and models like Discrete Event Simulation with the purpose to solve the engineering task of scheduling at hand. Building on such a technological blend, the framework incorporates the degradation and fault detection and diagnosis of failure modes across multiple components. The effect of degradation and faults on job processing times is learned as a distribution from the field data. The proposed framework is then validated in a laboratory environment where degradation is produced by inducing degradation and faults in the equipment. By means of various experiments, the optimized makespan is compared for the output schedules in different configurations. When equipment is degraded, production scheduling with PHM-enabled field-synchronized digital twin results in better makespan estimation, if compared to the digital twin without PHM. This shows the superiority of the framework in terms of more realistic makespan estimations, which finally corresponds to improved production schedule optimization, sensitive also to degraded states.

Offset-Free Koopman Model Predictive Control of Thermal Comfort Regulation for a Variable Refrigerant Flow-Dedicated Outdoor Air System-Combined System

Abstract Variable refrigerant flow (VRF) system has been an appealing solution of air conditioning for residential and commercial buildings, due to its flexibility and cost effectiveness, while lack of ventilation capability is a major drawback. Incorporation of dedicated outdoor air system (DOAS) is a typical practice. However, good coordination between DOAS and VRF is critical for achieving desired thermal comfort is challenging due to the possible complexity of mixed sensible and latent heat exchanges. In this paper, to handle the nonlinear dynamic characteristics of VRF-DOAS system, we propose an offset-free Koopman model predictive control (MPC) strategy for thermal comfort regulation, in which the MPC design is computationally more efficient due to the convex problem formulation and the use of reduced-order Koopman models, and the offset-free MPC structure enhances the robustness to model uncertainties and unmeasured disturbances. A control-oriented model is obtained by hybridizing the first-principle and data-driven modeling approach. The proposed controls strategy is evaluated with a Modelica simulation model of a VRF-DOAS system. A Dymola-Python cosimulation platform is developed via the functional mockup interface (FMI), for which the MPC algorithms are implemented in Python. Simulation results show significantly better performance of the offset-free Koopman MPC in thermal comfort regulation.

Revolutionizing Cybersecurity with Machine Learning: A Comprehensive Review and Future Directions

In the realm of computing, data science has revolutionized cybersecurity operations and technologies. The key to creating automated and intelligent security systems lies in extracting patterns or insights from cybersecurity data and building data-driven models. Data science, encompassing various scientific approaches, machine learning techniques, processes, and systems, studies real-world occurrences through data analysis. Machine learning techniques, known for their flexibility, scalability, and adaptability to new and unknown challenges, have been applied across many scientific fields. Cybersecurity is rapidly expanding due to significant advancements in social networks, cloud and web technologies, online banking, mobile environments, smart grids, and more. Various machine learning techniques have effectively addressed a wide range of computer security issues. This article reviews several machine learning applications in cybersecurity, including phishing detection, network intrusion detection, keystroke dynamics authentication, cryptography, human interaction proofs, spam detection in social networks, smart meter energy consumption profiling, and security concerns associated with machine learning techniques themselves. The methodology involves collecting a large dataset of phishing and legitimate instances, extracting relevant features such as email headers, content, and URLs, and training a machine learning model using supervised learning algorithms. These models can effectively identify phishing emails and websites with high accuracy and low false positive rates. To enhance phishing detection, it is recommended to continuously update the training dataset to include new phishing techniques and employ ensemble methods that combine multiple machine learning models for improved performance

Expert knowledge data-driven based actor–critic reinforcement learning framework to solve computationally expensive unit commitment problems with uncertain wind energy

With the expansion of power grid, unaffordable computational cost and time will pose serious challenges of time-efficient scheduling in unit commitment problem (UCP). However, existing optimization methods, i.e., mathematical programming methods and meta-heuristic algorithms, are powerless and time-consuming to handle computationally expensive UCP (CEUCP). Thus, reinforcement learning methods with strong inference and time-saving performances are motivated to solve the computationally expensive challenges in tackling CEUCPs. In this paper, a novel expert knowledge data-driven based actor–critic (AC) reinforcement learning methodology is proposed for solving CEUCPs. Specifically, in the proposed AC reinforcement learning methodology, expert knowledge, data-driven surrogate model, and improved meta-heuristic algorithm are integrated for further performance enhancement. Firstly, a novel action selection mechanism (based on the expert knowledge of thermal units characteristic) is integrated into AC to improve the efficiency of network training. Secondly, an improved extreme learning machine (ELM) data-driven surrogate model is proposed to build reward function in AC. In detail, original cost function in reward is replaced by a lightweight ELM model. Shape distance is integrated into ELM for enhancing accuracy. Finally, original marine predators algorithm (MPA) is improved for obtaining optimal dispatching decisions and rewards of AC method quickly and correctly. Original search pattern is replaced by quantum based representation for boosting convergence. The excellent performances of the proposed AC framework are verified by simulations of 10-units, 100-units, and 100-units with wind energy test systems.

Application of deep learning through group method of data handling for interfacial tension prediction in brine/CO2 systems: MgCl2 and CaCl2 aqueous solutions

Capillary/residual CO2 trapping is one of the main mechanisms of CO2 storage in underground formations. Therefore, it is required to estimate the brine/CO2 interfacial tension under different conditions. Although many methods have been proposed so far, the error of estimation is still high. This paper proposes a novel deep learning method to estimate the brine/CO2 interfacial tension at various temperatures, pressures, and salinities. The proposed method is a neural network with the Group Method of Data Handling (GMDH) learning method. The GMDH has the advantage of handling the structural and parametric optimization of the network automatically. The proposed method is tested on an experimental dataset of brine/CO2 interfacial tension with CalCl2 and MgCl2 salts. The results of the proposed method were compared with four of the best performing methods in the literature. The Average Absolute Percentage Error (AAPE) of the method on the training, testing and all data was 1.3 %, 2.95 %, 1.73 %, respectively, while the best method from the literature could reach an AAPE of 8.16 % on all data. Therefore, the proposed method performs far better than the existing methods. Also, a sensitivity analysis was done to determine the most influential inputs to estimate the output. The contribution of this work is to show the applicability of the GMDH method to construct more optimal data-driven models to estimate the brine/CO2 interfacial tension. Also, the utilized dataset is collected under a wide range of pressure, temperature and salinity conditions that increases the generality of the model.

Importance of social inequalities to contact patterns, vaccine uptake, and epidemic dynamics.

Individuals' socio-demographic and economic characteristics crucially shape the spread of an epidemic by largely determining the exposure level to the virus and the severity of the disease for those who got infected. While the complex interplay between individual characteristics and epidemic dynamics is widely recognised, traditional mathematical models often overlook these factors. In this study, we examine two important aspects of human behaviour relevant to epidemics: contact patterns and vaccination uptake. Using data collected during the COVID-19 pandemic in Hungary, we first identify the dimensions along which individuals exhibit the greatest variation in their contact patterns and vaccination uptake. We find that generally higher socio-economic groups of the population have a higher number of contacts and a higher vaccination uptake with respect to disadvantaged groups. Subsequently, we propose a data-driven epidemiological model that incorporates these behavioural differences. Finally, we apply our model to analyse the fourth wave of COVID-19 in Hungary, providing valuable insights into real-world scenarios. By bridging the gap between individual characteristics and epidemic spread, our research contributes to a more comprehensive understanding of disease dynamics and informs effective public health strategies.

A Case Study of a Tiny Machine Learning Application for Battery State-of-Charge Estimation

Growing battery use in energy storage and automotive industries demands advanced Battery Management Systems (BMSs) to estimate key parameters like the State of Charge (SoC) which are not directly measurable using standard sensors. Consequently, various model-based and data-driven approaches have been developed for their estimation. Among these, the latter are often favored due to their high accuracy, low energy consumption, and ease of implementation on the cloud or Internet of Things (IoT) devices. This research focuses on creating small, efficient data-driven SoC estimation models for integration into IoT devices, specifically the Infineon Cypress CY8CPROTO-062S3-4343W. The development process involved training a compact Convolutional Neural Network (CNN) and an Artificial Neural Network (ANN) offline using a comprehensive dataset obtained from five different batteries. Before deployment on the target device, model quantization was performed using Infineon’s ModusToolBox Machine Learning (MTB-ML) configurator 2.0 software. The tests show satisfactory results for both chosen models with a good accuracy achieved, especially in the early stages of the battery lifecycle. In terms of the computational burden, the ANN has a clear advantage over the more complex CNN model.

Ensemble Machine Learning Classification Models for Predicting Pavement Condition

Forecasting pavement performance condition is essential within the pavement management system to optimize decisions with regard to planning maintenance and rehabilitation projects. Accurate forecasts facilitate timely interventions and assist in formulating cost-effective asset management plans. Data-driven machine learning models that utilize historical data to improve forecasting precision have gained attention in the field of asset management. Although numerous studies have employed regression-based models to forecast pavement condition, transportation asset management often operates according to condition index ranges rather than exact values. Therefore, classification models are suitable for predicting pavement condition grades and determining the appropriate maintenance type for pavement assets. This research focuses on developing five machine learning classification models to predict pavement condition: random forest; gradient boost; support vector machine; k-nearest neighbors; and artificial neural network. To enhance prediction performance, these models are integrated using ensemble methods, including voting and stacking. The classification models are developed using a dataset from the Georgia Department of Transportation that documented the condition of asphalt pavements for predefined maintenance sections between 2017 and 2021. A voting ensemble model constructed with the two best-performing individual classification models reached the highest accuracy rate at 83%. Although the performance of individual models fluctuates, ensemble models consistently produce a top-tier performance regardless of the variations in data sampling. Therefore, ensemble methods are recommended for developing pavement condition prediction models to improve accuracy and achieve a more consistent quality of predictions. The findings of this research will provide transportation agencies with information to help them strengthen their forecasting practices in relation to pavement condition, thereby improving their maintenance planning and cost savings.

Application of a hybrid algorithm of LSTM and Transformer based on random search optimization for improving rainfall-runoff simulation

Flood forecasting using traditional physical hydrology models requires consideration of multiple complex physical processes including the spatio-temporal distribution of rainfall, the spatial heterogeneity of watershed sub-surface characteristics, and runoff generation and routing behaviours. Data-driven models offer novel solutions to these challenges, though they are hindered by difficulties in hyperparameter selection and a decline in prediction stability as the lead time extends. This study introduces a hybrid model, the RS-LSTM-Transformer, which combines Random Search (RS), Long Short-Term Memory networks (LSTM), and the Transformer architecture. Applied to the typical Jingle watershed in the middle reaches of the Yellow River, this model utilises rainfall and runoff data from basin sites to simulate flood processes, and its outcomes are compared against those from RS-LSTM, RS-Transformer, RS-BP, and RS-MLP models. It was evaluated against RS-LSTM, RS-Transformer, RS-BP, and RS-MLP models using the Nash–Sutcliffe Efficiency Coefficient (NSE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Bias percentage as metrics. At a 1-h lead time during calibration and validation, the RS-LSTM-Transformer model achieved NSE, RMSE, MAE, and Bias values of 0.970, 14.001m3/s, 5.304m3/s, 0.501% and 0.953, 14.124m3/s, 6.365m3/s, 0.523%, respectively. These results demonstrate the model's superior simulation capabilities and robustness, providing more accurate peak flow forecasts as the lead time increases. The study highlights the RS-LSTM-Transformer model's potential in flood forecasting and the advantages of integrating various data-driven approaches for innovative modelling.

Serological evidence of zoonotic filovirus exposure among bushmeat hunters in Guinea

Human Ebola virus (EBOV) outbreaks caused by persistent EBOV infection raises questions on the role of zoonotic spillover in filovirus epidemiology. To characterise filovirus zoonotic exposure, we collected cross-sectional serum samples from bushmeat hunters (n = 498) in Macenta Prefecture Guinea, adjacent to the index site of the 2013 EBOV-Makona spillover event. We identified distinct immune signatures (20/498, 4.0%) to multiple EBOV antigens (GP, NP, VP40) using stepwise ELISA and Western blot analysis and, live EBOV neutralisation (5/20; 25%). Using comparative serological data from PCR-confirmed survivors of the 2013-2016 EBOV outbreak, we demonstrated that most signatures (15/20) were not plausibly explained by prior EBOV-Makona exposure. Subsequent data-driven modelling of EBOV immunological outcomes to remote-sensing environmental data also revealed consistent associations with intact closed canopy forest. Together our findings suggest exposure to other closely related filoviruses prior to the 2013-2016 West Africa epidemic and highlight future surveillance priorities.

Status, Challenges, and Trends of International Research on Roadside Safety

Roadside safety refers to the assessment and improvement of safety measures related to roadside environment, design, management, and objects. It encompasses factors such as road design, signage, markings, traffic control devices, and roadside features, and its goal is to reduce accident risk, minimize injuries, and enhance overall safety and comfort for road users. To comprehensively summarize roadside safety research progress, this review retrieved 1637 English papers published between 2000 and 2022, using the Web of Science Core Collection database. VOSviewer software was utilized to visualize and analyze the literature, conduct a situational analysis of publication, create knowledge maps of the main research hotspots and trends, and summarize research status, methods, systems, challenges, and trends in this field. Results showed an overall increasing trend in relevant research. The countries, institutions, and journals contributing most are the United States, the University of Nebraska, and the Transportation Research Record, respectively. Current research hotspots include evaluation of roadside safety and risk levels, factors influencing roadside safety and driving risks, drunk and drug-impaired driving in relation to roadside traffic accidents, frequency and severity of roadside accidents, and roadside safety assurance techniques and improvement strategies. Current modeling methods mainly consist of mathematical statistical analyses and data-driven modeling based on machine learning. Future research should focus on comprehensive quantitative mapping of influencing factors and evaluation criteria, establishing an active-guidance-based evaluation system and optimization strategy, improving the accuracy of computational problems and model construction, and exploring theories and technologies of intelligent transportation for roadside safety management and improvement.

NGC 1856: Using Machine Learning Techniques to Uncover Detailed Stellar Abundances from MUSE Data

We present the first application of the novel approach based on data-driven machine learning methods applied to Multi-Unit Spectroscopic Explorer (MUSE) field data to derive stellar abundances of star clusters. MUSE has been used to target more than 10,000 fields, and it is unique in its ability to study dense stellar fields such as stellar clusters providing spectra for each individual star. We use MUSE data of the extragalactic young stellar cluster NGC 1856, located in the Large Magellanic Cloud (LMC). We present the individual stellar [Fe/H] abundance of 327 cluster members in addition to [Mg/Fe], [Si/Fe], [Ti/Fe], [C/Fe], [Ni/Fe], and [Cr/Fe] abundances of subsample sets. Our results match the LMC abundances obtained in the literature for [Mg/Fe], [Ti/Fe], [Ni/Fe], and [Cr/Fe]. This study is the first to derive [Si/Fe] and [C/Fe] abundances for this cluster. The revolutionary combination of integral-field spectroscopy and data-driven modeling will allow us to understand the chemical enrichment of star clusters and their host galaxies in greater detail expanding our understanding of galaxy evolution.

A data-driven robust decision-making model for configuring a resilient and responsive relief supply chain under mixed uncertainty

A data-driven robust decision-making model for configuring a resilient and responsive relief supply chain under mixed uncertainty

Modeling of Vehicle Mobility in Shallow Water With Data-Driven Hydrodynamics Model

Abstract In this study, a data-driven hydrodynamics model is proposed to enable quick prediction of vehicle mobility in shallow water, considering the effect of tire–soil interaction. To this end, a high-fidelity coupled vehicle–water interaction model using computational fluid dynamics (CFD) and multibody dynamics (MBD) solvers is developed to characterize the hydrodynamic loads exerted on a vehicle operated in shallow water, and it is used to generate training data for the data-driven hydrodynamics model. To account for the history-dependent hydrodynamic behavior, a long short-term memory (LSTM) neural network is introduced to incorporate effects of the historical variation of vehicle motion states as the input to the data-driven model, and it is used to predict hydrodynamic loads online exerted on vehicle components in the MBD mobility simulation. The impacts of hydrodynamic loads on the vehicle mobility capability in shallow water are examined for different water depths and incoming flow speeds using the high-fidelity coupled CFD-MBD model. Furthermore, it is demonstrated that the vehicle–water interaction behavior in scenarios not considered in the training data can be predicted using the proposed LSTM data-driven hydrodynamics model. However, the use of non-LSTM layers, which do not account for the sequential variation of vehicle motion states as the input, leads to an inaccurate prediction. A substantial computational speedup is achieved with the proposed LSTM-MBD vehicle–water interaction model while ensuring accuracy, compared to the computationally expensive high-fidelity coupled CFD-MBD model.

Enhancing long short-term memory (LSTM)-based streamflow prediction with a spatially distributed approach

Abstract. Deep learning (DL) algorithms have previously demonstrated their effectiveness in streamflow prediction. However, in hydrological time series modelling, the performance of existing DL methods is often bound by limited spatial information, as these data-driven models are typically trained with lumped (spatially aggregated) input data. In this study, we propose a hybrid approach, namely the Spatially Recursive (SR) model, that integrates a lumped long short-term memory (LSTM) network seamlessly with a physics-based hydrological routing simulation for enhanced streamflow prediction. The lumped LSTM was trained on the basin-averaged meteorological and hydrological variables derived from 141 gauged basins located in the Great Lakes region of North America. The SR model involves applying the trained LSTM at the subbasin scale for local streamflow predictions which are then translated to the basin outlet by the hydrological routing model. We evaluated the efficacy of the SR model with respect to predicting streamflow at 224 gauged stations across the Great Lakes region and compared its performance to that of the standalone lumped LSTM model. The results indicate that the SR model achieved performance levels on par with the lumped LSTM in basins used for training the LSTM. Additionally, the SR model was able to predict streamflow more accurately on large basins (e.g., drainage area greater than 2000 km2), underscoring the substantial information loss associated with basin-wise feature aggregation. Furthermore, the SR model outperformed the lumped LSTM when applied to basins that were not part of the LSTM training (i.e., pseudo-ungauged basins). The implication of this study is that the lumped LSTM predictions, especially in large basins and ungauged basins, can be reliably improved by considering spatial heterogeneity at finer resolution via the SR model.

A FEM-guided data-driven machine learning model for residual stress characterization in ultrasonic surface rolling of lightweight alloys

A FEM-guided data-driven machine learning model for residual stress characterization in ultrasonic surface rolling of lightweight alloys

Data-driven Discrete Simulation-based Dynamic Modeling and Continuous Optimization for Comprehensive Carbon Efficiency of Batch Hobbing

Data-driven Discrete Simulation-based Dynamic Modeling and Continuous Optimization for Comprehensive Carbon Efficiency of Batch Hobbing