Input Variables Research Articles

A Comparative Analysis of the Prediction of Gas Condensate Dew Point Pressure Using Advanced Machine Learning Algorithms

Dew point pressure (DPP) emerges as a pivotal factor crucial for forecasting reservoir dynamics regarding condensate-to-gas ratio and addressing production/completion hurdles, alongside calibrating EOS models for integrated simulation. However, DPP presents challenges in terms of predictability. Acknowledging these complexities, we introduce a state-of-the-art approach for DPP estimation utilizing advanced machine learning (ML) techniques. Our methodology is juxtaposed against published empirical correlation-based methods on two datasets with limited sizes and diverse inputs. With superior performance over correlation-based estimators, our ML approach demonstrates adaptability and resilience even with restricted training datasets, spanning various fluid classifications. We acquired condensate PVT data from publicly available sources and GeoMark RFDBASE, encompassing dew point pressure (the target variable), as well as compositional data (mole percentages of each component), temperature, molecular weight (MW), and specific gravity (SG) of heptane plus, which served as input variables. Before initiating the study, thorough assessments of measurement quality and results using statistical methods were conducted leveraging domain expertise. Subsequently, advanced ML techniques were employed to train predictive models with cross-validation to mitigate overfitting to the limited datasets. Our models were juxtaposed against the foremost published DDP estimators utilizing empirical correlation-based methods, with correlation-based estimators also trained on the underlying datasets for equitable comparison. To improve outcomes, pseudo-critical properties and artificial proxy features were utilized, leveraging generalized input data.

Experimental study and statistical analysis of system performance parameters of a household freezer

The present research comprehensively examines the influences of different input variables, such as cabinet load level, ambient temperature, relative humidity, door opening time and day, on the output parameters of frosting amount, recovery time, and total energy consumption of the upright domestic freezer following the door opening operation. Moreover, a holistic statistical methodology, which is known as the GLM-ANOVA, was implemented for determining the parametric experimental results. Meanwhile, the effects of binary interactions between the input factors on the output parameters were extensively evaluated using statistical methods. As a consequence, the amount of frost increases with an upward gradient as the load level within the cabinet escalates from 25 °C to 32 °C. Both the duration of door openings and the relative humidity level have a double impact on the frosting. Moreover, the duration required for the system to recover from half load to full load more than doubles with an increase in ambient temperature from 25 °C to 32 °C. On the other hand, at an outdoor temperature of 25 °C, the recovery time demonstrates a close to linear relationship with the load level of the cabinet. Furthermore, the duration of door openings and the load capacity within the cabinet are considered two important factors that simultaneously influence the daily energy consumption. The infiltration of ambient air into the freezer compartment and the presence of moisture in the air substantially increase the energy consumption, especially when the relative humidity fluctuates between 30 % and 65 % and the door opening duration stretches from 10 to 20 s, respectively.

Denoising of Photon-Counting LiDAR Bathymetry Based on Adaptive Variable OPTICS Model and Its Accuracy Assessment

Spaceborne photon-counting LiDAR holds significant potential for shallow-water bathymetry. However, the received photon data often contain substantial noise, complicating the extraction of elevation information. Currently, a denoising algorithm named ordering points to identify the clustering structure (OPTICS) draws people’s attention because of its strong performance under high background noise. However, this algorithm’s fixed input variables can lead to inaccurate photon distribution parameters in areas near the water bottom, which results in inadequate denoising in these areas, affecting bathymetric accuracy. To address this issue, an Adaptive Variable OPTICS (AV-OPTICS) model is proposed in this paper. Unlike the traditional OPTICS model with fixed input variables, the proposed model dynamically adjusts input variables based on point cloud distribution. This adjustment ensures accurate measurement of photon distribution parameters near the water bottom, thereby enhancing denoising effects in these areas and improving bathymetric accuracy. The findings indicate that, compared to traditional OPTICS methods, AV-OPTICS achieves higher F1-values and lower cohesions, demonstrating better denoising performance near the water bottom. Furthermore, this method achieves an average MAE of 0.28 m and RMSE of 0.31 m, indicating better bathymetric accuracy than traditional OPTICS methods. This study provides a promising solution for shallow-water bathymetry based on photon-counting LiDAR data.

Investigating the ability of deep learning to predict welding depth and pore volume in hairpin welding

To advance quality assurance in the welding process, this study presents a deep learning (DL) model that enables the prediction of two critical welds’ key performance characteristics (KPCs): welding depth and average pore volume. In the proposed approach, a wide range of laser welding key input characteristics (KICs) is utilized, including welding beam geometries, welding feed rates, path repetitions for weld beam geometries, and bright light weld ratios for all paths, all of which were obtained from hairpin welding experiments. Two DL networks are employed with multiple hidden dense layers and linear activation functions to investigate the capabilities of deep neural networks in capturing the complex nonlinear relationships between the welding input and output variables (KPCs and KICs). Applying DL networks to the small numerical experimental hairpin welding dataset has shown promising results, achieving mean absolute error values of 0.1079 for predicting welding depth and 0.0641 for average pore volume. This, in turn, promises significant advantages in controlling welding outcomes, moving beyond the current trend of relying only on defect classification in weld monitoring to capture the correlation between the weld parameters and weld geometries.

Sensitivity Analysis of Fatigue Life for Cracked Carbon-Fiber Structures Based on Surrogate Sampling and Kriging Model under Distribution Parameter Uncertainty

The quality and reliability of wind turbine blades, as core components of wind turbines, are crucial for the operational safety of the entire system. Carbon fiber is the primary material for wind turbine blades. However, during the manufacturing process, manual intervention inevitably introduces minor defects, which can lead to crack propagation under complex working conditions. Due to limited understanding and measurement capabilities of the input variables of structural systems, the distribution parameters of these variables often exhibit uncertainty. Therefore, it is essential to assess the impact of distribution parameter uncertainty on the fatigue performance of carbon-fiber structures with initial cracks and quickly identify the key distribution parameters affecting their reliability through global sensitivity analysis. This paper proposes a sensitivity analysis method based on surrogate sampling and the Kriging model to address the computational challenges and engineering application difficulties in distribution parameter sensitivity analysis. First, fatigue tests were conducted on carbon-fiber structures with initial cracks to study the dispersion of their fatigue life under different initial crack lengths. Next, based on the Hashin fatigue failure criterion, a simulation analysis method for the fatigue cumulative damage life of cracked carbon-fiber structures was proposed. By introducing uncertainty parameters into the simulation model, a training sample set was obtained, and a Kriging model describing the relationship between distribution parameters and fatigue life was established. Finally, an efficient input variable sampling method using the surrogate sampling probability density function was introduced, and a Sobol sensitivity analysis method based on surrogate sampling and the Kriging model was proposed. The results show that this method significantly reduces the computational burden of distribution parameter sensitivity analysis while ensuring computational accuracy.

Predicting High-Strength Concrete's Compressive Strength: A Comparative Study of Artificial Neural Networks, Adaptive Neuro-Fuzzy Inference System, and Response Surface Methodology.

Machine learning and response surface methods for predicting the compressive strength of high-strength concrete have not been adequately compared. Therefore, this research aimed to predict the compressive strength of high-strength concrete (HSC) using different methods. To achieve this purpose, neuro-fuzzy inference systems (ANFISs), artificial neural networks (ANNs), and response surface methodology (RSM) were used as ensemble methods. Using an ANN and ANFIS, high-strength concrete (HSC) output was modeled and optimized as a function of five independent variables. The RSM was designed with three input variables: cement, and fine and coarse aggregate. To facilitate data entry into Design Expert, the RSM model was divided into six groups, with p-values of responses 1 to 6 of 0.027, 0.010, 0.003, 0.023, 0.002, and 0.026. The following metrics were used to evaluate model compressive strength projection: R, R2, and MSE for ANN and ANFIS modeling; R2, Adj. R2, and Pred. R2 for RSM modeling. Based on the data, it can be concluded that the ANN model (R = 0.999, R2 = 0.998, and MSE = 0.417), RSM model (R = 0.981 and R2 = 0.963), and ANFIS model (R = 0.962, R2 = 0.926, and MSE = 0.655) have a good chance of accurately predicting the compressive strength of high-strength concrete (HSC). Furthermore, there is a strong correlation between the ANN, RSM, and ANFIS models and the experimental data. Nevertheless, the artificial neural network model demonstrates exceptional accuracy. The sensitivity analysis of the ANN model shows that cement and fine aggregate have the most significant effect on predicting compressive strength (45.29% and 35.87%, respectively), while superplasticizer has the least effect (0.227%). RSME values for cement and fine aggregate in the ANFIS model were 0.313 and 0.453 during the test process and 0.733 and 0.563 during the training process. Thus, it was found that both ANN and RSM models presented better results with higher accuracy and can be used for predicting the compressive strength of construction materials.

Chemical Fractions and Magnetic Simulation Based on Machine Learning for Trace Metals in a Sedimentary Column of Lake Taihu

In this study, the chemical fractions (CFs) of trace metal (TMs) and multiple magnetic parameters were analysed in the sedimentary column from the centre of Lake Taihu. The sedimentary column, measuring 53 cm in length, was dated using 210Pb and 137Cs to be 124 years old. Surface layers of the column were found to contain significantly higher concentrations of Cd, Co, Cu, Pb, Sb, Ti, and Zn than the middle and bottom layers. The sedimentary core contained a substantial amount of ferrimagnetic minerals. Most of the TMs were present in the residual state, except for Mn and Pb. The chemical fractions of Cd exhibited the most significant variation with depth. The pollution load index (PLI) indicated moderate TMs pollution levels in the region, whereas the risk assessment code (RAC) classified Mn as being heavily polluted. Multiple linear regression (MLR) and random forest (RF), support vector machine (SVM), and XGBoost (1.7.7.1) machine learning models were used to simulate the RAC and total concentration of TMs, using physical and chemical indicators and magnetic parameters of the sediments as input variables. The MLR model outperformed RF, SVM, and XGBoost in simulating the CFs and total concentrations of most TMs in the sedimentary column, with R2 up to 0.668 and 0.87. The SHapley Additive exPlanations (SHAP) method reveals that χarm/χ is the dominant factor influencing the RAC of As in the XGBoost models. For the RAC of Co and Cu in RF models, C% and N% exhibit greater contributions.

An experimental study of knock analysis of HCNG fueled SI engine by different methods and prediction of knock intensity by particle swarm optimization-support vector machine

Hydrogen and its careers have a lot of potential in transportation industry due to lesser emissions and better performance. Performance of hydrogen enriched compressed natural gas (HCNG) fueled engine is limited by knock. The purpose of this study is to evaluate and predict knock at different operating conditions by varying (0 %–40 %) hydrogen amount in HCNG, by varying (25 %–100 %) load on engine and by varying (700 rpm–1700 rpm) speed of engine. Knock is evaluated by knock ratio, knock intensity, exhaust temperature and in-cylinder heat transfer rate. Quasi-dimensional combustion model (QDCM) MATLAB program is used to calculate in-cylinder pressure theoretically. By increasing (0 %–40 %) hydrogen in HCNG 32.7 %, 7.6 %, 33.5 %, 21.5 % & 6.4 % increment observed in, in-cylinder pressure, knock intensity, knock ratio, in-cylinder heat transfer rate and exhaust temperature respectively. By increasing (25 %–100 %) load on engine 75.8 %, 69.8 %, 88.2 %, 77.8 % & 46.7 % increment observed in, in-cylinder pressure, knock intensity, knock ratio, in-cylinder heat transfer rate & exhaust temperature respectively. By increasing speed (1100–1700) rpm of engine aforementioned parameters decreases by 12.19 %, 38.9 %, 36.06 %, 47.2 % & 14.1 % respectively. Knock intensity is predicted by particle swarm optimization-support vector machine (PSO-SVM) algorithm effectively. To minimize the prediction error, 31 combination of different (1–5) inputs applied to predict knock intensity and found combination of 5 input variable is 67.2 % & 71.9 % more accurate than 1 input variable in-terms of mean squared error and mean absolute error respectively. Findings of this study can be used to train electronic control unit of engine and in the development of HCNG fueled engine.

Aerodynamic Drag Coefficient Prediction of a Spike Blunt Body Based on K-Nearest Neighbors

Spike blunt bodies are a method to reduce drag when a body moves at speeds above sound. Several numerical works based on computational fluid dynamics (CFD) have deeply studied fluid performance and highlighted its advantages. However, most documentation focuses on modifying spike physical properties while keeping constant supersonic or hypersonic flow conditions. In recent years, machine learning models have emerged as viable tools to predict values in almost any field, including aerodynamics. In the case of CFD, many models have been explored, such as support vector regression, ensemble methods, and artificial neural networks. However, a simple and easy-to-implement method such as k-Nearest Neighbors has not been extensively explored. This work extrapoled k-Nearest Neighbors to predict the drag coefficient of a spike blunt body for a range of supersonic and hypersonic speeds based on drag data obtained from CFD analysis. The parametric study of the spike blunt body was performed considering body diameter, spike length, and freestream Mach number as input variables. The algorithm presents proper predictions, with errors less than 5% for the drag coefficient and considering a minimum of three neighbor nodes. The k-NN was compared again Kriging model and k-NN presents a better accuracy. The above validates the flexibility of the method and shows a new area of opportunity for the calculation of aerodynamic properties.

A Techno-Economic Analysis of a Hybrid Microgrid System in a Residential Area of Bangladesh: Optimizing Renewable Energy

In the face of a significant power crisis, Bangladesh is turning towards renewable energy solutions, a move supported by the government’s initiatives. This article presents the findings of a study conducted in a residential area of Pabna, Bangladesh, using HOMER (Hybrid Optimization of Multiple Energy Resources) Pro software version 3.14.2. The study investigates the feasibility and efficiency of a grid-connected hybrid power system, combining photovoltaics (PV), a biomass generator, and wind energy. The simulation produced six competing solutions, each featuring a distinct combination of energy sources. Among the configurations analyzed, the grid-connected PV–biomass generator system emerged as the most cost-effective, exhibiting the lowest COE at USD 0.0232, a total net present cost (NPC) of USD 321,798.00, and an annual operating cost of USD 6060.59. The system presents a simple payback period of 9.25 years, highlighting its economic viability. Moreover, this hybrid model significantly reduces CO2 emissions to 78,721 kg/year, compared to the 257,093 kg/year emissions from a solely grid-connected system, highlighting its environmental benefits. Sensitivity analyses further reveal that the system’s performance is highly dependent on solar irradiance, indicating that slight variations in solar input can significantly impact the system’s output. This study underscores the potential of integrating multiple renewable energy sources to address the power crisis in Bangladesh, offering a sustainable and economically viable solution while also mitigating environmental impacts.

Optimized ensemble deep random vector functional link with nature inspired algorithm and boruta feature selection: Multi-site intelligent model for air quality index forecasting

Air quality index (AQI) forecasting is complex due to its variability, instability, and inconsistent trends resulting from dynamic atmospheric conditions, various contaminants, and interactions between environmental factors. Advanced modeling techniques are needed to accurately forecast AQI values to capture subtle patterns and variations in air quality data. Thus, a new forecasting model is suggested in this study to improve the accuracy of AQI forecasting. The model integrates three-phase decomposition technique, a feature selection approach, and ensemble Deep Random Vector Functional Link (EDRVFL), optimized using adaptive teaching-learning-based optimization and differential evolution (ATLDE). The AQI series was first broken down into a group of intrinsic mode functions (IMFs) with different frequencies using multivariate variational mode decomposition (MVMD). Subsequently, a feature selection method based on the Boruta technique was applied to identify the most significant input variables. Finally, for daily AQI levels forecasting, ATLDE optimized the EDRVFL model (EDRVFL-ATLDE). Three daily AQI series gathered from Chengdu, Wuhan, and Taiyuan in China from January 1, 2018, to December 30, 2022, were used to test and confirm the proposed model via empirical research. Based on the results, the proposed model can yield the superior results for three cities (Chengdu: correlation coefficient (R = 0.987), root mean square error (RMSE = 5.583), Wuhan: (R = 0.987), (RMSE = 3.299), and Taiyuan: (R = 0.996), (RMSE = 4.521)) in China. The experimental findings demonstrated the feasibility of the three-phase hybrid methodology, outperforming all other models regarding forecast accuracy.

Improved maize leaf area index inversion combining plant height corrected resampling size and random forest model using UAV images at fine scale

ContextAccurate monitoring of leaf area index (LAI) is conducive to timely and targeted management measures. Unmanned aerial vehicle (UAV) remote sensing provides an important way for non-destructive monitoring of crop leaf area index. ObjectiveIn this study, visible light (RGB) and multispectral remote sensing data from the UAV and ground-measured LAI data from the Plant Canopy Analyzer LAI-2200 C were used to conduct inversion of maize LAI on a fine scale. MethodsTo address the problem of spatial scale mismatch between the spatial resolution of UAV images and the ground-measured LAI, the scale difference between UAV image data and ground-measured data was reduced by removing the outermost ring data measured by the LAI-2200 C instrument, calculating the spatial resolution of the UAV images after resampling based on the height of the plant, and the resampling method based on the circle. Finally, through the above method to resample the UAV images, we extract the vegetation index and canopy height features as the input variables of the random forest model to build the maize LAI inversion model in vegetative stages and reproductive stages respectively, which is referred to as the Vis_H+RF method. Results and conclusionsThe Vis_H+RF method of Tongliao experimental station has an R2 of 0.96 in the vegetative stages and a R2 of 0.61 in the reproductive stages, both of which perform well and have certain migration capabilities. SignificanceThe LAI inversion model constructed based on the method in this study is basically consistent with the actual situation and can provide data support for maize growth monitoring.

Improved determination of the S-wave velocity of rocks in dry and saturated conditions: Application of machine-learning algorithms

The determination of S-wave velocity (Vs) is of significant importance in various engineering disciplines, including mining, civil, and geotechnical engineering. It is beneficial to indirectly determine Vs under both dry and saturated conditions and to understand its relationship with influencing input variables: coring depth (H), durability index (DI), water content (Wa), dry density (ρd), saturated density (ρs), and porosity (n). In this study, we evaluate these relationships using three multiple machine-learning algorithms (MLAs): artificial neural network (ANN), fuzzy inference system (FIS), and gene expression programming (GEP), alongside a linear regression method (LRM) and predict both dry S-wave velocity (Vs-dry) and saturated S-wave velocity (Vs-sat) of rocks. The research involves the analysis of 90 datasets derived from samples of schist, phyllite, and sandstone rocks collected from Azad and Bakhtiari dam sites in Iran. The diversity of these datasets is a key advantage of this study, providing a solid foundation for models training and testing while enhancing the models’ generalizability. Model optimization techniques are employed in the Python, MATLAB, GenXProTools, and SPSS environments to identify the most effective versions of ANN, FIS, GEP, and LRM models, respectively. The prediction performance analysis reveals that all applied models yield acceptable levels of accuracy for predicting Vs-dry and Vs-sat. However, GEP emerges as the best model for predicting both Vs-dry and Vs-sat. The ANN and FIS models also achieve high levels of accuracy, while LRM performs comparatively less well. Additionally, sensitivity analysis conducted using the cosine amplitude method (CAM) highlights the influence of different variables on Vs-dry and Vs-sat. The ρd is found to be the most influential parameter on Vs-dry, whereas DI exhibits the least impact. Conversely, the ρs significantly affects Vs-sat, while Wa shows the lowest impact. The exceptional performance of these proposed MLAs confirms their applicability in real-world rock engineering and geotechnics projects, offering precise determination of Vs. The diversity of studied rock types and datasets, along with the use of cost-effective and easy measurable inputs, the determination of Vs in both dry and saturated status, and the application of robust MLAs for Vs determination are the main novelties of this study. However, further researches involving additional datasets and more rock types are required to validate these findings.

Predicting prognosis for epithelial ovarian cancer patients receiving bevacizumab treatment with CT-based deep learning

Epithelial ovarian cancer (EOC) presents considerable difficulties in prognostication and treatment strategy development. Bevacizumab, an anti-angiogenic medication, has demonstrated potential in enhancing progression-free survival (PFS) in EOC patients. Nevertheless, the identification of individuals at elevated risk of disease progression following treatment remains a challenging task. This study was to develop and validate a deep learning (DL) model using retrospectively collected computed tomography (CT) plain scans of inoperable and recurrent EOC patients receiving bevacizumab treatment diagnosed between January 2013 and January 2024. A total of 525 patients from three different institutions were retrospectively included in the study and divided into training set (N = 400), internal test set (N = 97) and external test set (N = 28). The model’s performance was evaluated using Harrell’s C-index. Patients were categorized into high-risk and low-risk group based on a predetermined cutoff in the training set. Additionally, a multimodal model was evaluated, incorporating the risk score generated by the DL model and the pretreatment level of carbohydrate antigen 125 as input variables. The Net Reclassification Improvement (NRI) metric quantified the reclassification performance of our optimal model in comparison to the International Federation of Gynecology and Obstetrics (FIGO) staging model. The results indicated that DL model achieved a PFS predictive C-index of 0.73 in the internal test set and a C-index of 0.61 in the external test set, along with hazard ratios of 34.24 in the training set (95% CI: 21.7, 54.1; P < 0.001) and 8.16 in the internal test set (95% CI: 2.5, 26.8; P < 0.001). The multimodal model demonstrated a C-index of 0.76 in the internal test set and a C-index of 0.64 in the external test set. Comparative analysis against FIGO staging revealed an NRI of 0.06 (P < 0.001) for the multimodal model. The model presents opportunities for prognostic assessment, treatment strategizing, and ongoing patient monitoring.

Structural reliability analysis with parametric p-box uncertainties via a Bayesian updating BDRM

The parametric probability-box model, often abbreviated as “p-box” is frequently used to characterize epistemic uncertainties. However, structural reliability analysis with p-box uncertainties can often be computationally intensive. This paper presents an efficient method to accurately compute the bounds of failure probabilities within this context. The method’s key innovation lies in its ability to achieve high efficiency with only a single round of model evaluations. First, the Fractional Exponential Moments-based Maximum Entropy Method (FEM-MEM) with Bivariate Dimension Reduction Method (BDRM) is employed for precise reliability assessment, where a single-round of model evaluations are carried out. Subsequently, Bayesian Updating (BU) is applied to adjust the weights obtained from BDRM in response to changes in input variables, while ensuring the invariance of integration points. Following this adjustment, the FEM-MEM is once again employed to compute the failure probability after the information change, without necessitating additional model evaluations. To compute the bounds of failure probabilities, a Kriging model is employed to construct a surrogate relationship between the interval variables and failure probabilities. The accuracy and efficiency of the proposed method are demonstrated through numerical examples, with comparisons made against pertinent double-loop method.

Determination of the safe operating area in potentially destructive measurements using probability of feasibility

In this paper, a novel method of constraining the space of input variables is introduced in case of potentially-destructive electrical measurements. Contrary to traditional sweep-based measurement execution, the method enables safe evaluation of samples lying on arbitrary grids. To achieve that, the constraints are modeled as Gaussian processes, which allow to predict the constraint value and estimate the model uncertainty at the given sample. The further a candidate point lies from the samples, for which measurement results are available, the higher the uncertainty. Since the initial models are usually inaccurate, we update them and their uncertainty predictions each time new measurement results are available. Transistor measurements on several uniform random experimental designs prove that the proposed method can efficiently identify the safe operating area with the error rate not exceeding 2.5%. Few constraint violations found were not severe and no degradation in transistor performance was observed after the measurement campaign.

A Partial Domain Generalization Method for Modeling Multiple Multistage Manufacturing Processes

In complex manufacturing systems, materials are processed by machines sequentially before final products are get, forming a multistage manufacturing process (MMP). In a modern massive production factory, it is common that multiple MMPs work simultaneously to fulfill productivity needs. Multiple MMPs may contain machines with different running time, processing the same types of products but with different specifications. In such a Multiple MMP(MMMP) system, it is critical for machines to learn from each other to gain optimal parameter settings of each. However, traditional machine learning methods usually fail to consider both similarities and unique features among multiple processes at the same time. To address this problem, a partial domain generalization method is proposed based on the thought of transfer learning to combine useful information from historical processes while maintaining features from the in-production process. The proposed method also suits for problems with the assumption that only partial input variables are available. Studies reveal that the proposed method has superior prediction performance over traditional machine learning methods and some widely-used transfer learning methods.

Enhancing streamflow simulations with gated recurrent units deep learning models in the flood prone region with low-convergence streamflow data

Highly accurate streamflow simulations are essential for water resources and river management. However, there are longstanding challenges for streamflow modeling in the hydrology field. As a result, deep learning methods have been introduced as novel tools and are being used in simulation. This paper proposes a novel approach that utilizes unconventional data preprocessing techniques to improve the accuracy of daily streamflow predictions in flood-prone regions. We applied this approach to a flood-prone area in western Iran and simulated daily streamflow using a deep learning Gated Recurrent Unit (GRU) method under various data selection scenarios. These scenarios focused on identifying the optimal combination of input variables, time steps, and outlier removal techniques. The outlier removal methods investigated in this study include Mahalanobis distance, critical section removal, Z-score, and no removal. The average rainfall of the area, data driven precipitation, Normalized Difference Vegetation Index (NDVI), surface soil moisture, groundwater baseflows, and streamflow in the hydrometric station were evaluated using correlation control method, and inputs with the lowest correlation were removed. Based on the results obtained from the deep learning models produced in the research, it was found that the GRU model, with various modified inputs using Z-score removal, had the best performance. The model had an average of Root Mean Squared Error (RMSE) at 5.24 mm and R2 (Coefficient of Determination) at 0.91 during training, while during validation, it had an RMSE of 7.74 mm and R2 of 0.83. Considering using relatively low convergence data in streamflow simulation in this study, it can be said that the listed scenarios showed an appropriate result in dealing with the data and recognizing the complex pattern of the daily streamflow, and the future studies will show the improvement of the GRU models if higher convergence data will be used.

Exploring explainable ensemble machine learning methods for long-term performance prediction of industrial gas turbines: A comparative analysis

In today's modern life, where electricity demand is one of the fundamental necessities, gas turbines play a pivotal role in meeting this demand. As such, it is imperative to address the challenges faced in the field. Current models often rely on simplifying assumptions, neglecting the intricate relationships between variables. This limitation leads to reduced accuracy and reliability, ultimately affecting the overall efficiency of gas turbine systems. Furthermore, the complexity of gas turbine behavior, coupled with the scarcity of comprehensive datasets, exacerbates the problem.To address these challenges, this research aimed to develop an advanced model capable of accurately forecasting real gas turbine behavior. The proposed approach leveraged ensemble decision trees, robust preprocessing techniques, and rigorous evaluation using an extensive dataset spanning from 2011 to 2015. The training and validation phases were conducted on data from 2011 to 2014, with the 2015 dataset reserved for evaluation.The results demonstrated that the bagging structure outperformed the boosted structure, exhibiting lower complexity and higher reliability. Remarkably, the bagging approach with only 30 estimators achieved a superior root mean square error of 1.4176, outperforming the boosted trees with 200 learners. The model effectively captured the overall gas turbine performance, though it encountered limitations in certain specific operating ranges.To further investigate the model's behavior, an evaluation was conducted to assess the effects of the input variables on the output power. While the interpretability of the results posed some challenges, the overall findings were deemed acceptable and provide valuable insights for optimizing gas turbine performance. The significance of this research lies in its potential to inform decision-making and enhance the efficiency of gas turbine systems, ultimately contributing to the reliable and sustainable supply of electricity.

Comprehensive resilience assessment of bridge networks using ensemble learning method

The assessment of seismic resilience in bridge networks holds significant importance for urban disaster prevention and mitigation efforts. Unlike individual bridges, there has been limited efficiency in assessing bridge networks. A seismic resilience assessment methodology for bridge networks using ensemble learning methods is proposed in this paper. Initially, a comprehensive resilience index is proposed, integrating both structural and functional aspects of bridge networks. Using 3 ensemble learning methods, 9 parameters related to network structure and traffic characteristics are chosen as input variables for predicting the seismic resilience index. Finite element models of 18 bridges are constructed and combined to generate 3500 sets of virtual bridge networks for model training. The predictive accuracy of models trained using the 3 ensemble methods exceeds 89 %, and the expected values of peak ground acceleration (PGA) and functional loss rate are the most influential features. The methodology offers insights into the application of ensemble learning for bridge network seismic resilience assessment.