This paper presents an early version of the primary results on fault detection in multilevel inverters through artificial intelligence techniques. Multilevel inverters, while advantageous in terms of efficiency and waveform quality, are inherently prone to failures due to their large number of semiconductor components. Faults in switching devices, if not identified in time, are seen to cause severe instability and device breakdown. The original study assessed four deep learning (DL) models for reference voltage prediction and subsequent fault identification. In this conference version, the discussion is limited to the convolutional neural network (CNN) method. Localized temporal and spatial features in voltage signals are extracted by CNNs, by which precise recognition of abnormal switching behavior is achieved. These preliminary results confirm that the CNN approach is promising and are used as a basis for extended journal research, which is planned to include broader comparisons and comprehensive validation. Experimental investigations also confirm that CNN-based methods achieve reliable fault detection in real time.

Modeling the suspended sediment load (SSL) in rivers is of great importance in various fields, e.g. hydrological sciences, water resources management, and dams engineering. In this study, the daily SSL time series at two distinct stations are modeled. The machine learning methods of random forest (RF) and long short-term memory (LSTM) are implemented. The outcomes state that the RF yielded better SSL predictions compared with LSTM. Besides the aforementioned individual models, this study improves the SSL forecasts through the development of hybrid versions of RF and LSTM. To achieve this, an optimizer, i.e. the immune system algorithm (ISA), was initially coupled to the RF. Next, recurrent neural networks (RNN) were hybridized with an LSTM. Therefore, the hybrid forms of RF and LSTM, i.e. RF-ISA and RNN-LSTM, were proposed. The findings reveal that both hybrid models, specifically RNN-LSTM, outperformed their relevant individual forms. The RF-ISA model, with its optimized hyperparameter selection, demonstrated better generalization compared to the baseline RF model, while the RNN-LSTM model effectively captured temporal trends in SSL fluctuations. The values of evaluation error metrics, namely root mean square error (RMSE), normalized RMSE (NRMSE), correlation coefficient (R), Nash-Sutcliffe efficiency (NSE), Willmott’s index (WI), and percent bias (PBIAS) in the test phase for the best model of RNN-LSTM were achieved as follows: R = 0.9434, RMSE = 105567 ton/day, NRMSE = 0.0291, NSE = 0.8893, WI = 0.9701, PBIAS = −0.7135% (first station), and R = 0.9824, RMSE = 57238 ton/day, NRMSE = 0.0311, NSE = 0.9632, WI = 0.9912, PBIAS = 0.0271% (second station). The outcomes of SHapley Additive exPlanations (SHAP) explainer exhibited that one-day delayed SSL and river discharge data represented the greatest and least impacts on the models' output, respectively.

Saltwater intrusion in estuarine systems is intensifying due to climate change and human disturbances, which poses increasing risks to water security and ecological stability. Reliable salinity prediction is therefore essential for adaptively to manage saltwater intrusion and ecological protection. However, conventional approaches often suffer from limited predictive accuracy, short forecast horizons, and a lack of uncertainty assessment. Consequently, in this study, a hybrid deep learning is developed to improve multi-step salinity prediction. The Temporal Convolutional Networks (TCN) and Multi-Head Self-Attention (MHSA) methods are coupled within a Deep Autoregressive Recurrent (DeepAR) probabilistic. This integration strengthens temporal feature extraction and dependency identification across multiple time scales, while an embedded Gaussian Process (GP) module quantifies predictive uncertainty. The results show that the proposed hybrid model outperforms the baseline models in the test period, with improvements of up to 42.9% in R2 and 43.7% in Nash-Sutcliffe efficiency (NSE), and reductions of up to 27.6% in root mean square error (RMSE) and 28.4% in mean absolute error (MAE). The long-term predictions are more stable and effectively capture key salinity dynamics, including peak magnitudes and phase shifts. Furthermore, the proposed model achieves the best Mean Interval Score (MIS) across all forecast durations at the three prediction horizons. It also yields more uniformly distributed residuals, which suggests improved reliability and generalization. Thus, this study introduces an innovative methodology that supports high-resolution, reliable forecasting for estuarine ecosystem management and protection.

Various fluctuations were observed in the reported deaths and confirmed cases of COVID-19. It is, however, essential to predict lower record rates of morbidity and mortality based on the available data. In this study, we use a Bayesian setting to obtain the point and interval predictions of the future lower records of confirmed cases and deaths. To do this, we adopt a Markov Chain Monte Carlo (MCMC) algorithm for computing Bayes predictors and explain all the tricks required to implement them in detail. We also provide real examples of applications of the proposed algorithm to predict future lower records. The results demonstrate the proximity of the predicted values to the true lower records of datasets.

Predicting river water quality is important in hydrological, environmental, and water resources management fields. In this context, the concentration of dissolved oxygen (DO) in rivers is one of the most important water quality indicators that needs to be forecasted accurately. This study firstly developed two deep learning (DL) models, including Bidirectional Long Short-Term Memory (BiLSTM) and Temporal Fusion Transformer Neural Network (TFTNN) for predicting daily DO concentrations at two river stations in United States (Beaverton Creek and North Umpqua River) in timeframe from 2016 to 2023. Then, a hybrid model was implemented via hybridizing TFTNN and BiLSTM to establish the TFTNN-BiLSTM model. The findings revealed that the developed TFTNN-BiLSTM performed better compared with their baseline models. Finally, a novel two-stage hybrid model named RBMO-TFTNN-BiLSTM was proposed by coupling TFTNN, BiLSTM, and an optimizer named Red-Billed Blue Magpie Optimization (RBMO). The outcomes denoted the superior performance of proposed RBMOTFTNN-BiLSTM than TFTNN, BiLSTM, and TFTNN-BiLSTM. During test stage, root mean square error of best RBMO-TFTNN-BiLSTM relative to baseline TFTNN and BiLSTM was reduced by 51.81 % and 47.66 % (Beaverton Creek), 40.18 % and 41.48 % (North Umpqua River). A Categorical Boosting (CatBoost) was also developed for a comparison, and the results indicated its lower accuracy than DL methods. Assessing the outcomes of Shapley Additive exPlanations (SHAP) illustrated that river water temperature and pH presented the highest and least impacts on the output results of models, respectively. The proposed hybrid methodologies in this study offer reliable predictive tools for accurate prediction of river DO concentration

Multilevel power inverters have a complex semiconductor structure that elevates the risk of switch faults. Furthermore, voltage drops across floating capacitors, which are integral components of multilevel power converter structures, can disrupt accurate system status assessment and lead to incorrect or delayed fault detection. This article proposes a novel approach for short-circuit fault detection and location in multilevel power converters using artificial intelligence with a focus on reliability prioritization. Five reference voltage prediction methods were analyzed including a switching algorithm and four deep learning-based techniques i.e., convolutional neural networks, gated recurrent units, long short-term memory networks, and a hybrid model combining convolutional neural networks with long short-term memory networks. Fault location was performed through a reliability-based strategy prioritizing components with higher failure probabilities, significantly improving the fault identification speed. Our method reduced the duration of fault detection compared to similar methods and included a novel fault location method based on prioritizing fault detection according to the lifetime of fundamental components. We predicted and verified online voltage references using four different deep learning methods and compare the outcomes in an experimental setup. Simulation and experimental results demonstrated the effectiveness and practicability of the proposed method in detecting and locating faults in various types of multilevel power inverters.

This paper presents a comparative analysis of machine learning methods for Alzheimer’s disease diagnosis. The limitations of these methods are discussed and their potentials for early detection are explored as well. In comparing machine learning models for Alzheimer’s diagnosis, gradient-boosting techniques emerged as top classifiers. XGBoost (XGB) classifier, LightGBM (LGBM) classifier and CatBoost consistently achieved an acceptable accuracy which outperforms other models. The Gradient boosting classifier, slightly outperforms random forest and the AdaBoost. Decision trees and logistic regression were less reliable, while the Support Vector Machines (SVM) remained moderate. Crucial biomarkers included Mini-Mental State Examination (MMSE), Activities of Daily Living (ADL), memory complaints, behavioral problems, and functional assessments, all pivotal predictors.

Despite the devastating effects of floods, the concept of resilience is still not fully considered in the assessment and management of flood risk. To study how resilience can lower the risk of floods and further enhance disaster response, this research aims to close this knowledge gap. With a focus on the Kashkan watershed in Iran, the study combines the extended catastrophe progression method with the pressure-state-response model. Three catastrophe models, namely the cusp, swallowtail, and butterfly, are applied. According to the findings, southern regions, i.e., Pol-Dokhtar city, have the highest risk of floods and the lowest resilience. Resilience and flood risk have a complementary relationship, according to the analysis, and resilience is a helpful metric for risk assessment. The results emphasize the necessity to incorporate resilience-focused pre-disruption planning and post-disaster recovery into flood risk management strategy. This work offers a foundation to incorporate resilience into future flood policies and strategies.

This study presents an Electroencephalography (EEG) emotion recognition using a long short-term memory (LSTM)-based method. Our proposed method selects window sizes and overlaps to divide the EEG data into segments, which optimally captures subtle signal changes. A Bidirectional LSTM (BiLSTM) layer is added to standard LSTM layers to better detect forward and backward patterns in the data. By using this dual-layer setup, we aim to improve both the feature extraction and the classification accuracy. The model was tested on the Database for Emotion Analysis using Physiological signals (DEAP) dataset and showed acceptable accuracy across emotional dimensions: arousal (94.0%), liking (98.9%), dominance (95.3%), and valence (99.6%). Our results suggest that the model better supports emotion recognition and has potential for mental health monitoring and adaptive therapy.

Ground level ozone is considered a major air pollutant. It is formed when nitrogen oxides and volatile organic compounds react under sunlight. It harms human health and damages plants and materials. It also contributes to climate change. It is a photochemically formed compound that is extremely hazardous to the environment and human health. Proper forecasting of the boundary layer ozone has been a challenge due to the nonlinearly related to both meteorological and chemical conditions and the scarcity of fine-scale vertical ozone patterns. This study uses the OMPROFOZ ozone profile product which is a product of the Ozone Monitoring Instrument (OMI) on the Aura satellite to estimate the ozone concentrations in the boundary layer. A set of deep learning models, i.e., RNN, CNN, GRU, LSTM, and hybrid forms i.e., GRU-CNN and LSTM-CNN, is evaluated to benchmark forecasting accuracy. The first, ConvBiGRU-AttentionNet, integrates attention mechanisms within a convolutional gated recurrent structure. The second, EMD-ConvBiGRU-AttentionNet, adds Empirical Mode Decomposition to extract multi-scale temporal features before modeling. The proposed models outperform conventional methods across metrics such as RMSE, MAE, R2, and skill scores. EMD-ConvBiGRU-AttentionNet achieves the highest prediction accuracy. Visual analyses, i.e., residual plots, cumulative error distributions, and attention maps, confirm the capacity of the model to capture spatio-temporal patterns in atmospheric data.

The ability to predict multiaxial fatigue life of Al-Alloy 7075-T6 under complex loading conditions is critical to assessing its durability under complex loading conditions, particularly in aerospace, automotive, and structural applications. This paper presents a physical-informed neural network (PINN) model to predict the fatigue life of Al-Alloy 7075-T6 over a variety of multiaxial stresses. The model integrates the principles of the Geometric Multiaxial Fatigue Life (GMFL) approach, which is a novel fatigue life prediction approach to estimating fatigue life by combining multiple fatigue criteria. The proposed model aims to estimate fatigue damage accumulation by the GMFL method. The proposed GMFL-PINN combines this physics-based approach with data-driven neural networks. Experimental validation demonstrates that GMFL-PINN outperforms FS, Smith-Watson-Topper (SWT) and Li-Zhang (LZH) fatigue life prediction methods which provides a reliable and scalable solution for structural health assessment and

Groundwater pollution presents a serious concern in arid and semiarid regions, where water resources are already limited. In such contexts, reliable and efficient methods for assessing groundwater vulnerability are critical. Without adequate knowledge of the vulnerability, groundwater is at greater risk of severe contamination. This not only threatens the availability of clean water but also demands significant time and financial resources for remediation and restoration. Modelling groundwater vulnerability is even more demanding and complex in data-scarce regions. Consequently, this study investigates and predicts spatial variations in groundwater quality and vulnerability within a data-scarce area by applying efficient machine learning methods that compensate for the limited availability of quality groundwater data. Supportive machine learning approaches such as bagged adaptive boosting (BAB), averaged neural network (avNNet), heteroscedastic discriminant analysis (HAD), rotation forest (RotationF), and an ensemble method were applied to assess groundwater vulnerability using k-fold cross-validation. The results demonstrate that the BAB model achieved the best performance, with both accuracy and precision exceeding 85%. Furthermore, the stacking ensemble approach, specifically the BAB model combination, increased precision by 4% and reduced false alarms by 6%. The most influential variables affecting groundwater quality include groundwater depth, precipitation, proximity to waterways and roads, topographic humidity, and the percentage of fine-grain material. The results also show that variability in the data significantly impacts the modelling performance.

The environmental aspect of combustion can be considered one of the main factors in achieving sustainable power production. While various studies emphasised the positive aspect of propylene glycol as an additive for diesel-biodiesel fuel composition on emission and performance metrics of a diesel engine, an unknown element of the role of this additive in sustainable power generation cycles is still not achieved. The role of this additive in the sustainable power generation cycle is unclear, and this is considered a research gap. In this study, an environmental assessment of engine performance and emission specifications was conducted in the propylene glycol, diesel, and biodiesel fuel combination. Propylene glycol was mixed with B2 and B5 fuels at 3, 5, and 7% ratios. Fuel samples were examined in a diesel engine. Environmental assessment was conducted using life cycle assessment according to IMPACT2002+ life cycle inventory analysis instruction. According to the results, the lowest midpoint and endpoint indicators of environmental impacts were related to B5PG7, i.e., 18% lower than the control. Accordingly, B5PG7 was selected as the best fuel sample from an environmental and efficiency point of view.

We present a detailed review and evaluation of machine learning (ML) methods for modeling and predicting stress evolution in various materials and systems. Stress evolution is considered a fundamental phenomenon in materials science, structural engineering and biomechanics. It is frequently modeled with deterministic methods, which struggle to handle high-dimensional, complex and non-linear data. A promising substitute is Machine Learning (ML), which offers instruments to enhance predictive accuracy and more effectively capture complex patterns. We used the Scopus database to find relevant literature and the PRISMA framework for systematic screening for creating an extensive database for this review. Based on how well supervised, unsupervised and deep learning approaches apply to stress modeling, under various loading and environmental circumstances, we present a new taxonomy of machine learning approaches. Furthermore, we critically evaluate these approaches' advantages and disadvantages, and further highlight the significance of feature engineering, data quality and model interpretability. The review ends by outlining potential future directions, especially with regard to deep and hybrid models that combine ML with traditional techniques to improve prediction of stress evolution in a variety of applications.

Explaining the decision-making behavior of deep neural networks (DNNs) can increase their trustworthiness in real-world applications. For natural language processing (NLP) tasks, many existing interpretation methods split the text according to the interactions between words. Also, the evaluation of explanation capability focuses on justifying the importance of the divided text spans from the perspective of interaction contribution. However, the prior evaluations are misled by extra interactions, making the evaluation unable to acquire accurate interactions within the text spans. Besides, existing research considers only absolute interaction contribution, which causes the evaluation to underestimate the important text spans with lower absolute interaction contribution and to overestimate the unimportant text spans with higher absolute interaction contribution. In this work, we propose a metric called Text Interaction Proportional Score (TIPS) to evaluate faithful interpretation methods. More specifically, we use a pick scheme to acquire the interactions within the divided text span and eliminate the influence of the extra interactions. Meanwhile, we utilize the relative interaction contribution between the divided text span and whole text to measure the importance of the acquired interactions. The proposed metric is validated using two interpretation methods in explaining three neural text classifiers (LSTM, CNN and BERT) on six benchmark datasets. Experiments show that TIPS outperforms a baseline method in three ways consistently and significantly (i.e., acquiring interactions within the text span, measuring importance of interaction, and distinguishing the important and unimportant text spans).

Solar energy is one of the renewable and clean energy sources. Accurate solar radiation (SR) estimates are therefore needed in solar energy applications. Firstly, two deep learning models, including gated recurrent unit (GRU) and long short-term memory (LSTM), were developed in this study. Next, a data pre-processing technique named multivariate variational mode decomposition (MVMD) was used to construct the MVMD-GRU and MVMD-LSTM hybrid models. To better test the performance of proposed simple and hybrid models, four stations located in the Illinois State of the USA (i.e., Dixon Springs, Fairfield, Rend Lake, and Carbondale) were considered as the study sites. Whole the simple and hybrid models were established under two different strategies, i.e., local and external. In the local strategy, SR of each location was estimated using the minimum and maximum air temperatures from the same station. While, minimum and maximum air temperatures as well as SR data from the nearby station were utilized in external strategy to estimate SR time series of any target site. Root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R2) metrics were used when evaluating the models performances. The overall results revealed that the proposed MVMD-GRU and MVMD-LSTM hybrid models illustrated better SR estimates compared to the simple GRU and LSTM in both the local and external strategies. The values of error metrics obtained for the superior hybrid models (i.e., MVMD-LSTM) during the testing period were as: RMSE = 2.532 MJ/m2.day, MAE = 1.921 MJ/m2.day, R2 = 0.916 at Dixon Springs; RMSE = 2.476 MJ/m2.day, MAE = 1.878 MJ/m2.day, R2 = 0.921 at Fairfield; RMSE = 2.359 MJ/m2.day, MAE = 1.780 MJ/m2.day, R2 = 0.924 at Rend Lake; RMSE = 2.576 MJ/m2.day, MAE = 1.941 MJ/m2.day, R2 = 0.914 at Carbondale. Therefore, the coupled models proposed in this study can be possibly recommended as suitable alternatives to the simple deep learning models with a reliable precision in estimating SR time series.

Global adoption of wind energy continues to increase, while improving the efficiency of turbine settings requires reliable wind speed (WS) models. The latest models rely on artificial intelligence (AI) optimizations which constructs tests on a range of novel hybrid models to examine the reliability. Gradient Boosting (GB), Random Forest (RF), and Long Short-Term Memory (LSTM) are used in new combinations for data pre-processing. A Time Varying Filter-based Empirical Mode Decomposition (TVFEMD) model is coupled with the GB and LSTM standalone models, to create TVFEMD-GB and TVFEMD-LSTM hybrids, which are run in competition with each other. Eventually, a preferred hybrid form is established, simultaneous hybridization of TVFEMD with GB and LSTM. This study is the first to hybridize these fundamental systems, and create a TVFEMD-GB-LSTM model that can forecast WS. This study finds that the novel hybrid models exhibit superior performance to standalone GB and LSTM models, opening the pathway to alternative WS prediction techniques.

Progress in non-destructive methods to detect the characteristics of fruits is a new and attractive process for researchers and specialists in this field. On the other hand, these researchers move toward identifying their impacts on their surroundings in line with diagnostic efficiency. One of these essential impacts is the environmental impact of the non-destructive detection process of fruits. Navel oranges are one of the most popular and widely consumed fruits, whose maturity indices such as soluble solids contents (SSC) values and acidity are considered as parameters in determining the quality of this product. This study used the hyperspectral method in the vis-NIR range to evaluate and measure navel oranges' SSC and acidity values. In the following, by applying the life cycle assessment method, the environmental impacts of measuring and evaluating these two parameters of the characteristics of navel oranges were investigated. The Impact2002+ method was used to evaluate the impact of the life cycle list. Based on the findings, the environmental impacts of SSC measurement are about 40, 42, 20, and 18 % higher than those of the environmental impacts of pH measurement from the point of view of endpoint impacts for Human Health, Ecosystem quality, climate change, and resources, respectively. The random forest modeling results showed a suitable and acceptable correlation and relationship (over 90 %) between the wavelengths selected from the feature selection stage and environmental impacts.

This paper presents an innovative approach to skin cancer classification based on Dense Convolutional Network 121 (DenseNet-121), a state-of-the-art deep convolutional neural network. By automated analysis of dermatological imagery, this paper aims to enhance diagnostic precision significantly. Given a dataset of dermatological image, this study compares the performance of six leading architecture models: EfficientNet, ResNet-50, VGG16, GoogleNet, DenseNet-121, and MobileNet V3 Large. During a series of extensive training and validation phases, it has been established that DenseNet-121 is an extremely effective tool for identifying diverse types of skin cancers with high accuracy indicating a potent for early skin cancer detection and aiding in the advancement of medical diagnostic procedures.

Capacitor banks (CBs) play a crucial role in energy storage and frequency control within autonomous microgrids. However, the impact of internal capacitor configurations, varying in terms of equivalent series resistance (ESR), capacitance, and rated voltage, on CB degradation, reliability, and peak current remains an understudied aspect. Moreover, the absence of a capacitance degradation coefficient in the standard MIL-HDBK-217 equations for predicting the reliability of electrolytic capacitors poses a significant challenge. To address these issues, this study examines a microgrid composed of diverse renewable energy systems, featuring nine distinct CB arrangements. The design of CBs considers both capacitance and peak output current individually. An evaluation is conducted to compare construction costs, lifetimes, and peak output currents across all layouts. Additionally, a novel formula is introduced to estimate the reliability and lifetime of CBs, while an existing formula for calculating CB peak output current is enhanced. The research explores the impact of ambient temperature and capacitor voltage on the reliability of various capacitor designs, proposing a novel framework for assessing CB reliability based on MIL-HDBK-338B, which accounts for both short-circuit and open-circuit faults. The practicality of these findings is confirmed through a comparison of experimental and simulation results. The inverter operation video, simulation, and all production data including PCB and processor codes are also attached.