Search Results (2,764)

Industrial fed-batch fermentation generates coupled, time-dependent process data that are difficult to interpret using isolated alarms. This study developed a retrospective, batch-aware machine learning framework integrating operational safety margins, anomaly characterization, 30 min critical event prediction, and 60 min product concentration forecasting. Complete fermentation batches were separated into training, validation, and locked test sets to prevent within-batch information leakage. Linear and ensemble classifiers and regressors were evaluated against temporal and feature ablation baselines. On the locked test set, gradient boosting was the most selective classifier (accuracy = 0.901; sensitivity = 0.556; false alarm rate = 3.1%), random forest prioritized sensitivity (accuracy = 0.479; sensitivity = 0.857; false alarm rate = 59.5%), and logistic regression showed an intermediate operating point (accuracy = 0.878; sensitivity = 0.571; false alarm rate = 6.2%). The ridge increment model achieved an RMSE = 1.173 g/L and R² = 0.9984 versus RMSE = 4.072 g/L and R² = 0.9807 for persistence; removing the current product concentration increased the RMSE to 10.078 g/L and reduced R² to 0.8819. The archived process risk and CCP state formulas were recovered and verified against all source records, enabling a fully auditable rule-based retrospective scenario for alarm burden and downtime. The contribution is therefore workflow integration and traceable batch-level validation rather than a new learning algorithm. The framework is intended for operator-oriented decision support and requires prospective, cross-campaign and cross-reactor validation before deployment. Full article

To address the strong load rigidity and high energy consumption of large-scale broiler breeding, as well as the lack of precise load-characteristic support and efficient scheduling methods for breeding-oriented energy supply and consumption optimization, this paper conducts load-characteristic analysis for large-scale broiler houses and proposes an energy management optimization method based on the improved dream optimization algorithm (IDOA). First, year-round field monitoring was performed on a large-scale broiler farm in Laiyuan, Hebei, to analyze the load characteristics across seasons and breeding cycles and reveal the load evolution rules. Second, a comprehensive operational cost objective is established, considering the renewable operation and maintenance cost, the time-of-use power trading cost, and the carbon emission cost. Third, adaptive weight, dynamic mutation, and opposition-based learning strategies are embedded into the IDOA to better balance global exploration and local exploitation, and the improved algorithm outperforms other meta-heuristic algorithms on the CEC2017 benchmark functions. Simulation tests covering the four-season typical days and key breeding stages demonstrate that, compared with the rule-based dispatch strategy, the proposed method lowers the daily operating cost and effectively smooths the grid power profile, while the ESS state of charge is always maintained within the preset limits. Sensitivity analyses on the ESS capacity and the load and renewable forecasting errors further verify the robustness of the dispatch results, and the single-dispatch runtime of about 23 ms amply satisfies the real-time requirement of field implementation. This study offers theoretical support and practical reference for energy saving and carbon reduction in large-scale livestock breeding and breeding-park energy system scheduling. Full article

The energy transition poses substantial challenges for all actors in modern power systems, where the output of key renewable technologies is weather-dependent and electricity prices are increasingly volatile. This work presents a decision-support algorithm that optimizes the day-ahead dispatch of a parabolic-trough concentrating solar power (CSP) plant with two-tank molten-salt thermal energy storage (TES) connected to the Spanish grid. The algorithm couples a deep neural network (DNN) that forecasts hourly day-ahead electricity prices—trained on Spanish market data for 2016–2021 using only predictors available before day-ahead market closure: the previous-day natural-gas index, calendar variables, lagged hourly price profiles, and the previous-day generation mix—with a genetic algorithm (GA) that maximizes expected market revenues subject to the technical constraints of the plant, using a 10 min discretization of the TES operating trajectory. Under a strictly chronological evaluation, the forecasting module achieved a mean absolute error of 12.31 EUR/MWh on the held-out year 2021 and 3.57 EUR/MWh on 2020, outperforming persistence benchmarks by 22.1% and 32.0%, respectively. Across a 32-scenario benchmark spanning seasons, gas-price regimes, day types, and irradiance patterns, the optimizer adopted solutions within 3.25% of a perfect-foresight exact optimum (range: 0.70–6.94%), with five random seeds, increasing gross revenue by 20.5% over a no-storage baseline but by only 1.1% over a simple rule-based dispatch strategy, with no evidence of storage carry-over ahead of adverse meteorological days. The results illustrate how embedding machine-learning price forecasts in plant-control algorithms can increase gross market revenue for dispatchable solar generation. Full article

The high predictive performance of machine learning models in complex classification tasks often comes with limited transparency, particularly for black-box models whose decision processes cannot be directly inspected. In this study, a two-stage Random Forest–SPEA2 hybrid rule inference and optimization architecture is proposed to achieve high predictive performance while retaining an explicit rule-based decision structure. In the first stage, root-to-leaf decision paths extracted from Random Forest trees are converted into interval-based IF–THEN rules, and a class-balanced candidate rule pool is constructed. In the second stage, the Strength Pareto Evolutionary Algorithm 2 (SPEA2) simultaneously maximizes rule-level accuracy and weighted F1-score using Pareto-based multi-objective optimization. The resulting rules preserve an explicit IF–THEN representation whose feature conditions and decision boundaries can be directly inspected. The proposed framework was evaluated on three benchmark datasets representing different classification characteristics: Breast Cancer Wisconsin for binary classification, Dry Bean for multi-class classification, and ISOLET for high-dimensional multi-class classification. Under stratified 5-fold cross-validation, RF-SPEA2 achieved mean accuracies of 96.77%, 98.08%, and 95.54% on Breast Cancer Wisconsin, Dry Bean, and ISOLET, respectively, with relatively low variability across folds. Comparative experiments showed that the proposed method achieved competitive predictive performance against conventional machine learning models while outperforming several traditional rule-based approaches. Representative IF–THEN rules obtained for the three datasets demonstrate that the resulting decision structure remains directly inspectable, although the practical interpretability of individual rules depends on the dimensionality and semantic meaning of the input features. These findings indicate that RF-SPEA2 provides a promising framework for combining competitive predictive performance with an explicit rule-based representation across datasets with different dimensionalities and class structures. Full article

Accurate prediction of pest density and population levels in almond production is of great importance for crop yield, the prevention of quality losses, and the effectiveness of integrated pest management (IPM) strategies. Complex feature interactions in agricultural field data can affect both the predictive performance and interpretability of machine learning models. In this study, an explainable hybrid machine learning framework combining Mutated Particle Swarm Optimization (MPSO), XGBoost classification, and rule extraction was developed to classify pest insect density levels. The proposed approach was evaluated using two complementary validation settings. First, it was compared with 16 classification algorithms, including ensemble, deep learning, and rule-based methods, under stratified 5-fold cross-validation, achieving a mean accuracy of 79.63% and a mean Weighted F1-Score of 79.79%. In addition, nested stratified 5-fold cross-validation was employed to separate MPSO-based hyperparameter optimization from outer-fold performance evaluation. Under this more rigorous evaluation, the proposed approach achieved a mean accuracy of 79.27 ± 1.66% and a mean Weighted F1-Score of 79.29 ± 1.38%. Furthermore, interpretable IF–THEN decision rules were derived from the trained XGBoost decision structures to provide a transparent representation of the feature–threshold combinations associated with the classification decisions. Full article

The Korean Cryptographic Module Validation Program (KCMVP) requires the joint review of source code, submission documents, and their traceability. This study presents an evidence-gated decision-support prototype combining rule-based candidate generation (L1), rule-bound evidence retrieval (L2), program-fact verification, and constrained LLM-assisted review (L3). Its 166 YAML rule assets, comprising 97 code rules, 65 document rules, and four traceability rules, were developed with reference to official KCMVP procedures and National Intelligence Service guidance on submission preparation and cryptographic algorithm implementation. On an LEA-centered development corpus, L1 matched 115 of 128 author-annotated positive file–rule pairs (89.84%). However, because these annotations were created during system development, independent expert validation is required before interpreting this value as detection accuracy. In nine source-derived mutation groups, the baseline and improved-evidence conditions each produced six binary dispositions and three holds (exact McNemar test, $p=1.0$). This result defines the current role of L2 as supplying traceable evidence, validating citations, and withholding unsupported decisions rather than independently determining compliance. L3 reviews only residual candidates supported by validated evidence and program facts under a structured output contract; failure to satisfy input, evidence, citation, or output-validation conditions results in a hold. Overall, the results demonstrate a feasible and auditable architecture that preserves evidence provenance and explicitly holds unresolved cases instead of forcing unsupported decisions. Full article

Population aging is placing growing pressure on elderly care systems, and artificial intelligence (AI) is increasingly viewed as a potential solution. In community-based elderly care, multimodal AI can integrate sensing, data fusion, risk prediction, explanation, and intervention. Such systems may enable a shift from post-event response to early warning and proactive care. However, existing reviews have often examined these components separately. As a result, the dependencies between different stages—particularly the transition from risk prediction to closed-loop intervention—remain insufficiently explored. To address this gap, this narrative review organizes the literature around an end-to-end framework comprising “sensing–fusion–prediction–explanation–intervention–feedback.” This review includes 94 publications up to 30 June 2026. Each publication was classified as providing either direct or indirect evidence relevant to community-dwelling older adults. This review compares major algorithmic approaches to multimodal fusion and temporal risk prediction across several deployment-related dimensions, including temporal modeling, robustness to missing modalities, calibration, interpretability, validation design, and computational cost. It also examines explainable AI and intervention-generation methods, ranging from post hoc attribution and tiered rule-based alerts to reinforcement learning policies. This review also identifies key challenges at the data, model, intervention, and system levels. It highlights the need for uncertainty-aware, privacy-preserving, and closed-loop solutions. By treating the entire care loop, rather than any single algorithm, as the unit of analysis, this review bridges the gap between component-focused research and the integrated systems required for effective community-based elderly care. Full article

This work addresses the challenge of performing distributed online convex optimization in resource-constrained systems, including IoT networks and wireless sensor networks, where communication resources are limited. We develop an edge-based event-triggered (EBET) distributed inexact gradient descent algorithm that departs from the conventional requirement of full state exchange at every iteration. In the proposed scheme, each communication link is equipped with an independent triggering rule; an agent forwards its current estimate to a neighbor only after the state deviation on that link surpasses a user-specified threshold, thereby eliminating redundant transmissions. The algorithm additionally tolerates bounded errors in the gradient evaluation, making it applicable to settings where exact gradients are either unavailable or too costly to compute. Through a Lyapunov-like analysis that couples the consensus update with the non-expansive projection operator, we show that the static regret of every agent grows at a sublinear rate of $\mathcal{O}\left(\sqrt{T}\right)$ provided that the step size, gradient error bound, and triggering threshold are chosen appropriately. Simulation results on regularized linear regression and logistic regression tasks validate the theoretical findings, confirming that sublinear regret is attained with far fewer communication rounds than a fully connected baseline, thus achieving a favorable balance between solution quality and communication cost. Full article

Modeling under uncertainty remains a fundamental challenge in engineering and nonlinear system identification, particularly when both interpretability and computational efficiency are required. Interval type-2 adaptive neuro-fuzzy inference systems (IT2ANFIS) have demonstrated strong capabilities in handling uncertainty; however, the considered approaches typically introduce uncertainty either in the center or in the width of membership functions, limiting their representational flexibility. To address this problem, this paper proposes a novel IT2ANFIS model that simultaneously incorporates uncertainty in both the centers and widths of Gaussian membership functions, enabling a more expressive yet compact representation of uncertainty. First, a formulation of the proposed membership function is developed. This allows the application of gradient-based learning with explicitly derived update rules for both premise and consequent parameters. Second, the proposed approach avoids explicit type-reduction procedures by directly aggregating lower and upper firing strengths. This eliminates the need for computationally intensive iterative algorithms and improves inference efficiency. Third, a local first-order gradient analysis is used to define a practical reference for learning-rate scaling. In addition, multiple learning-rate strategies, including fixed, switching, and moment-based strategies, are investigated to evaluate their influence on convergence and model performance. Furthermore, the proposed approach is validated through benchmark systems and experimental evaluation on a DC motor system. The results demonstrate improved modeling accuracy and robustness compared to conventional ANFIS-based approaches. The proposed framework provides a practical and computationally efficient approach to uncertainty-aware nonlinear system modeling and related engineering applications. Full article

In evacuation analysis for underground stations, Dijkstra’s algorithm is widely used to estimate the maximum evacuation time on the assumption that it returns a unique and reproducible route. This assumption does not hold in structurally symmetric networks, where many routes share an identical cost and the route actually returned depends on an arbitrary, implementation-dependent tie-breaking rule. The aim of this study is to quantify how much the maximum evacuation time varies solely as a result of this tie-breaking rule, and to determine how many repeated executions are required before that variability is adequately characterized. A six-level underground station network of 720 nodes and 2222 edges was used. A random tie-breaking rule was implemented within Dijkstra’s algorithm, and the evacuation simulation was repeated under independently seeded runs of 1, 10, 25, 50, 100, and 1000 iterations while the station layout, movement speeds, congestion thresholds, and evacuee distribution were held fixed. A single execution produced a maximum evacuation time of 782 s, whereas the 1000-iteration case yielded a range of 671–864 s. The mean and median stabilized within 10–25 iterations, but the observed minimum and maximum continued to widen through 1000 iterations. These results show that a single Dijkstra execution can reasonably estimate typical evacuation performance but may underestimate the upper-tail evacuation times that govern life-safety design. For practical application to underground station design and evacuation-time verification, it is recommended that Dijkstra-based route generation be repeated across multiple tie-breaking realizations and that upper-percentile evacuation times be reported alongside the deterministic single-run result, so that the required safe egress time used in life-safety assessment reflects the variability inherent in equal-cost path selection. Full article

Harbour bar entrances present severe maritime safety risks, particularly under low-light and adverse weather conditions where conventional visible-spectrum RGB (Red, Green, and Blue) cameras lose effectiveness. This Communication presents an advanced, 24/7 thermal monitoring system deployed at the Port of Nazaré, Portugal, developed within the framework of the European BRIGHTER project. Combining Long-Wave Infrared (LWIR) sensors with hybrid Artificial Intelligence (AI)—integrating computer vision, heuristic rules, and YOLOv7-based Convolutional Neural Networks (CNNs)—the system achieves continuous real-time detection and classification of fishing vessels, recreational boats, people, and birds. The hardware architecture links two remote camera sites via a hybrid network (fibre optics and long-range Wi-Fi) to a centralised Communication Centre powered by a high-performance Central Processing Unit (CPU) and Graphics Processing Unit (GPU). Over a two-year operational period, the system collected a domain-specific dataset of over 48,000 annotated thermal images. Results demonstrate high target precision, robust persistent tracking using the Hungarian algorithm, and automated JavaScript Object Notation (JSON) metadata generation for triggering intelligent geofenced alarms. Full article

Post-harvest sorting is essential for converting the biological variability of chili peppers into consistent commercial grades, efficient processing, and higher market value. Rapid advances in machine vision, multimodal sensing, deep learning, and intelligent actuation are transforming sorting equipment from rule-based classifiers into integrated perception-to-execution systems. However, existing studies often evaluate isolated algorithms or components, leaving limited evidence that recognition accuracy translates into reliable sorting during continuous operation. We conducted a structured narrative review of English-language studies published from January 2008 to July 2026 using Web of Science, Scopus, AESC, and PubMed. Evidence was synthesized along a perception-to-execution chain, distinguishing pepper-specific sorting systems and component studies from cross-crop engineering evidence. Direct system-level evidence was concentrated in a small number of chili and bell pepper studies, whereas much of the technical discussion drew on apple, tomato, potato, sweet potato, and onion research. Visible, spectral, fluorescence, and multimodal features can distinguish ripeness, color grades, and surface defects, although performance remains sensitive to cultivar, illumination, pose, and dataset design. A small number of pepper sorting studies support the feasibility of integrating vision, conveying, and physical separation under specific operating conditions. Cross-crop studies inform the discussion of localization, pneumatic actuation, and system integration, but do not establish the performance of these approaches in chili pepper sorting. Reported performance is difficult to compare because studies rarely standardize latency, target association, false and missed rejections, product damage, energy use, and long-term reliability. Robust deployment therefore requires cross-batch datasets, synchronized target-level traceability, and evaluation protocols linking perception outputs to final physical destinations. This review identifies the boundaries of current pepper-specific evidence and proposes a system-level evaluation framework and research priorities for testing the transferability of cross-crop engineering approaches to pepper sorting. Full article

Clozapine is the only evidence-based treatment for treatment-resistant schizophrenia (TRS); however, up to 70% of patients have an inadequate response, leading to clozapine-resistant schizophrenia (CRS) and, in its most severe form, ultra-treatment-resistant schizophrenia (UTRS). Because evidence-based treatment algorithms after clozapine failure remain limited, management requires an individualized, multimodal approach. We present a UTRS complex case alongside a scoping review conducted according to the PRISMA-ScR framework to summarize the current evidence on the diagnosis, neurobiology, and treatment of CRS and UTRS. A 24-year-old man with schizophrenia required 11 months of psychiatric hospitalization due to persistent psychosis, severe aggression, disorganized thinking, prominent negative symptoms, and profound functional impairment. Extensive investigations, including neuroimaging, cerebrospinal fluid analysis, metabolic studies, and genetic testing, ruled out alternative diagnoses. Despite sequential treatment with multiple antipsychotics, optimization of long-acting injectable therapy, clozapine, pharmacological augmentation, two courses of electroconvulsive therapy (ECT), and intensive multidisciplinary rehabilitation, the patient continued to exhibit severe positive, negative, and cognitive symptoms and functional impairment, meeting the criteria for UTRS. This review emphasizes the importance of excluding pseudoresistance by confirming adherence, assessing clozapine exposure when therapeutic drug monitoring is available, and conducting diagnostic reassessment before establishing a diagnosis of CRS or UTRS. Among available interventions, ECT remains the strongest evidence-based augmentation strategy following clozapine failure, while pharmacological augmentation and other neuromodulatory techniques are supported by low-certainty evidence. These findings underscore the need for individualized multimodal management that integrates pharmacological, biological, and psychosocial interventions, and support consideration of precision psychiatry and palliative psychiatry principles as potential future frameworks when evidence-based therapeutic options have been exhausted. Full article

A cooperative interval game assigns a range of possible worths, rather than a single number, to each coalition. For the Shapley-like rule based on Moore subtraction, existing axioms determine the midpoint of each player’s allocated interval but need not determine its lower and upper endpoints. We close this identification gap on the full class of interval games. In midpoint–radius coordinates, the center is the ordinary Shapley value, whereas each Moore marginal adds the radii of its predecessor and successor coalitions. We introduce two requirements for this radius component. Total Moore exposure fixes the aggregate radius generated along random permutation chains, and boundary exposure balance divides each proper coalition’s contribution equally, in aggregate, between its members and nonmembers. The familiar center axioms together with these two conditions uniquely recover the Moore–Shapley rule, including both endpoints. Without boundary exposure balance, one insider-share parameter remains for each proper coalition size, yielding an $(n-1)$-dimensional family. We also derive an aggregate-envelope identity and the sharp Hausdorff–Lipschitz constant, establish logical independence of the six axioms for $n\ge 3$, and give an exact $O\left(n{2}^{n-1}\right)$ algorithm together with an unbiased permutation-sampling alternative. A stylized, non-empirical four-partner storage example separates total exposure from its allocation across players. All results concern the original Moore-based rule rather than other interval solution concepts. Full article

The second-life utilization of retired power batteries is critically constrained by high cell-to-cell variability in capacity and internal resistance, which severely reduces the usable capacity of repacked modules. To overcome this challenge, this paper presents a multi-stage screening and grouping strategy that balances accuracy with practical efficiency. The method comprises four progressive steps: static Euclidean distance-based pre-screening, 0.1C low-rate reference capacity calibration, 0.5C operating-condition re-screening, and serpentine sorting for final grouping. A total of 389 retired 18650 ternary lithium-ion batteries from a single batch were studied. First, 89 cells were pre-screened using voltage–internal resistance Euclidean distance, from which 16 cells were selected for 0.1C calibration to establish a low-rate reference capacity baseline. Subsequently, 52 cells were re-screened from the remaining 300 and tested at a 0.5C rate. Finally, the 52 cells were assembled into 13 groups via serpentine sorting and uniformly calibrated to 50% SOC. A benchmark conversion coefficient β, defined as the ratio of the mean 0.5C capacity to the mean 0.1C capacity, and a comprehensive consistency index ($CQI$) were established for evaluation. Results show that the mean 0.1C capacity is 2835.2 mAh with β = 0.9681. After serpentine grouping, the capacity range across the 13 groups is only 16.69 mAh, with a coefficient of variation of 0.0407%—significantly outperforming random grouping—and the $CQI$ reaches 0.985. The proposed method reduces the total capacity testing time from approximately 21.9 days to about 2.5 days, improving efficiency by approximately 88%. In contrast to prior work focusing solely on algorithmic improvements, this study, for the first time, integrates static outlier exclusion, small-sample-rate mapping, and serpentine balanced grouping into a closed-loop engineering workflow, providing a deterministic, rule-based solution for the entire screening-to-grouping pipeline in second-life applications. The method requires neither complex instrumentation nor sophisticated algorithms and exhibits strong robustness against common measurement errors, offering an economical, reliable, and easily replicable engineering solution for retired battery second-life utilization. Full article