The injection of carbon dioxide into depleted oil and gas reservoirs has emerged as a promising method for enhanced oil recovery (EOR) and safe carbon storage. This process significantly influences the density of CO2-hydrocarbon mixtures, a critical property that directly affects EOR efficiency and storage integrity. However, accurately predicting mixture density remains challenging due to the complex and nonlinear behavior of hydrocarbons, and many existing empirical models show limited reliability under varying conditions. This study primarily aims to develop robust tools leveraging advanced algorithms, Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), Bayesian Neural Network (BNN), Adaptive Boosting (AdaBoost), and Random Forest (RF) to accurately estimate the density of CO2-hydrocarbon mixtures, utilizing a rich and extensive databank (17,081). The model inputs included temperature (T), pressure (P), molecular weight (Mw), pseudo-critical temperature (Tpc), pseudo-critical pressure (Ppc), and the mole fraction of CO2. The results were evaluated against several equations of state (EoSs), including the Cubic-Plus-Association (CPA) model developed in this study, as well as Peng-Robinson (PR), Soave-Redlich-Kwong (SRK), and Redlich-Kwong (RK) models for representative CO2-hydrocarbon mixtures (n-Butane, Propane, and Hexadecane). For these specific mixtures, the CPA and PR models showed the best performance among the examined EoSs, while the XGBoost model achieved the highest overall accuracy, with an R2 of 0.9968 and an average absolute percent relative error (AAPRE) of 0.7647. High R2 values (0.97-0.99) indicate that LightGBM, BNN, AdaBoost, and RF all achieved strong predictive performance. Furthermore, trend analysis confirmed that the XGBoost model can accurately capture density variations in response to changes in pressure variable. Moreover sensitivity analysis indicated that pressure with relevancy factor of (0.2045) has the most significant effect on the output. The leverage technique showed that over 96% of the dataset is statistically reliable. In the final stage, Shapley additive explanations (SHAP) analysis demonstrated that both Tpc and Ppcpositively influence the output parameter. These results demonstrate that the XGBoost model provides a reliable and accurate alternative to experimental methods for predicting CO2-hydrocarbon density across a wide range of operating conditions.
For radiative coolers that are not fully exposed to the sky, thermal radiation from parasitic hot objects, such as the ground surface under direct sunlight, poses a significant challenge and can severely impact performance. Here, we present a special configuration of reflectors for radiative coolers, suitable for building walls or other parts of buildings not fully oriented towards the sky. This configuration has the potential to reduce visible-light disturbance when a visibly transparent infrared-reflecting material, such as ITO, is used, while providing angular selectivity in the mid-infrared spectrum to reduce the impact of parasitic thermal radiation from objects below the horizon line and enhance radiative cooling to the cosmic background through the atmospheric transparency window. By using this method, the calculated net radiative exchange within the atmospheric window reaches up to 65.1 W/m2 under the idealized optical and environmental assumptions adopted in the model. This value represents an upper bound for the calculated net radiative exchange under these assumptions. Additionally, to minimize visual complications, we calculated and presented the limitations on the smallest size of the structure configuration. The calculations show that the reflectors size should be greater than seven times the maximum wavelength of the atmospheric window. Finally, experimental measurements conducted on a prototype of the proposed structure show good agreement between the calculations and theoretical predictions.
The iron and steel industry is one of the most energy-intensive industrial sectors, accounting for approximately 8 % of global final energy use and nearly 10 % of annual greenhouse gas emissions. Within this value chain, direct reduction iron (DRI) production is among the most energy-demanding stages and a key focus for decarbonization efforts. In this study, the conventional syngas supply for DRI, typically produced via steam methane reforming (SMR), is replaced by a high-temperature solid oxide electrolyzer (SOE)-based system. The required electrical energy and partial thermal energy are provided by solar energy and biomethane, respectively, while waste heat recovery from other steelmaking units is implemented to enhance overall efficiency. The main objectives are to reduce energy consumption, emissions, and production costs while improving exergy efficiency and maintaining compatibility with existing industrial infrastructure. Comprehensive energy, exergy, environmental, economic, and exergo-economic analyses are performed. A multi-objective optimization using a genetic algorithm is then applied to determine optimal operating conditions regarding the four main objectives. The optimized results are benchmarked against conventional DRI routes and literature data. The findings show a 12 % reduction (from 3.06 to 2.72 MWh per ton of DRI) in specific energy consumption and annual energy consumption can be reduced by up to 680 GWh, while CO2 emissions decrease by approximately 1 MtCO2 per year (from 0.6 to 0.2 tonCO2 per ton of DRI). Additionally, syngas production efficiency improves by nearly 30 %, and natural gas consumption is reduced by about 50 %. These results demonstrate the potential of the proposed SOE-based configuration for advancing low-carbon, energy-efficient ironmaking.
Abstract Performance assessment of tall buildings against blast loading is considered a crucial aspect of safe structural design. In this regard, the selection of the optimum positions for structural elements plays an important role. Achieving the best locations of outriggers by dynamic analysis of all possible cases will lead to excessive computational effort. As an alternative approach, using genetic algorithms (GAs), which is proposed in this paper, demands fewer structural dynamic analyses to achieve the optimum configuration. The mathematical formulation is presented to predict the optimum positions of outriggers against blast loading. In the GA algorithm, the positions of outriggers in tall buildings with 20, 30, and 40 stories were optimized against 4 blast scenarios. The sampled structures were modeled using OpenSees considering material and geometry nonlinearity. The GA code was developed using MATLAB and linked with the OpenSees tool as the structural analysis core. To affect the blast loading on sampled structures, the empirical formulation of blast-induced pressure is used. The maximum structural interstory drift ratio (MIDR) is considered as the objective function. To present the results in a practical manner, the formulations to predict the optimum positions are proposed. In the proposed formulation, the optimum relative positions of second and third outriggers are considered as outputs ( h 1 / H total and h 2 / H total ), and the blast scaled distance ( Z ) and number of stories ( N ) are considered as input variables. The nonlinear regression, based on simulated data, calculated using the bootstrap technique, is applied to present the final formulation with minimum prediction error. Based on the results, linear combination of N , Z 2 , and N × ln ( Z ) is proposed as the most suitable relationship. The proposed formulation was tested for 2 new sample buildings with 25 and 35 stories. The accuracy of the proposed formulation was verified through comparison of the predicted best locations of outriggers with those calculated using the nonlinear dynamic analysis.
Metaheuristic algorithms have become a widely adopted approach for addressing feature selection problems in high-dimensional datasets. Among these methods, Particle Swarm Optimization (PSO) has received attention due to its simple structure, efficient search capability, and adaptability to different optimization scenarios. As a result, numerous PSO-based feature selection methods have been proposed in recent years, each introducing various modifications to improve search performance and subset quality. Despite this rapid development, a structured analysis that highlights the strengths, limitations, and practical implications of these approaches remains necessary. This survey provides a systematic examination of prominent PSO-based feature selection algorithms reported in the literature. The reviewed methods are analyzed and compared with respect to several important aspects, including search behavior, strategies used to balance exploration and exploitation, design of fitness functions for evaluating feature subsets, and commonly used evaluation criteria such as classification accuracy, dimensionality reduction rate, and computational cost. The analysis highlights the main limitations of PSO-based feature selection, including a tendency to premature convergence, sensitivity to parameter settings, and scalability issues in high-dimensional environments. Based on these observations, several open research challenges are identified and potential directions for future work are outlined in order to improve the applicability of PSO-driven feature selection methods.
