Abstract In Sub-Saharan Africa, maize is a dietary staple. However, maize crop pests and diseases threaten harvests and undermine food security. Machine learning (ML) and deep learning (DL) have been widely used for the classification of these diseases. With sufficient data, DL methods generally outperform ML approaches. However, feature engineering can yield better results with limited data for models trained from scratch. Pre-trained DL models and data augmentation address data scarcity, yet these models often lack flexibility and can be resource-intensive. This paper proposes a dual-axis integration of feature learning and feature engineering for the classification of maize crop diseases and pests under conditions of severe class-level data scarcity. Validation was conducted using real-world datasets from Burkina Faso, Namibia, Uganda, and Tanzania. Two model variants were developed: an unpruned model incorporating a complete set of engineered features and a pruned model retaining only the most significant engineered features. Their performance was compared against 28 pre-trained models commonly used for crop disease classification in the literature, using accuracy, F1-score, parameter efficiency, and computational efficiency. Normalised, unweighted average scores showed the three highest-ranking models as the proposed unpruned model (0.9742), MobileNetV3Small (0.9736), and MobileNetV3Large (0.9731). EfficientNetV2L had the lowest score of 0.4772. Cross-regional evaluation of the proposed models using an independent maize-disease dataset from Bangladesh indicated that their convolutional representations, learned from the Sub-Saharan African dataset, retained useful discriminative information after retraining only the classification head. Deployment-relevant metrics revealed that MobileNetV3Small offers the fastest inference (2.01 ms) and lowest latency-derived energy estimate (30.09 mJ), while the pruned model offers the best storage efficiency (2.24 MB). Thus, in resource-constrained settings, hybridising feature engineering and feature learning can yield more effective, deployable crop disease classification models.
Abstract This evidence-based critical review examines how emerging artificial-intelligence (AI) capabilities can strengthen resilient smart-grid fault intelligence. Rather than comparing algorithms primarily by reported accuracy, the review organizes recent fault-detection evidence around operational capabilities: spatial/topology awareness, temporal context, label efficiency, physical consistency, trust and uncertainty, privacy-preserving collaboration, and deployment feasibility. A structured primary-study evidence matrix was developed to extract task, system or dataset, topology, temporal context, robustness, trust, privacy, physics, deployment evidence, and principal limitations. The extracted evidence shows a clear asymmetry in the maturity of emerging approaches. Recent Transformer work demonstrates real-time temporal fault detection on test-system waveforms, self-supervised methods reduce dependence on manually labeled events, physics-informed approaches can narrow the simulation-to-field gap, and graph-based models increasingly address topology and cross-topology generalization. However, external validation, uncertainty quantification, reproducibility, and standardized deployment evidence remain uneven. Based on this evidence, the paper introduces a fault-intelligence evidence maturity profile that avoids reducing readiness to a single generic score. The synthesis leads to an evidence-derived roadmap progressing from trustworthy and data-efficient intelligence to topology- and physics-aware learning, distributed cyber-physical intelligence, and finally integrated safety-bounded decision support. The review deliberately distinguishes demonstrated evidence from plausible future capability and therefore avoids unsupported claims of algorithmic superiority.
Globally, there is a significant transformation involving conventional power systems, such transformations are driven by concerns over climate change, energy security, and sustainability. In recent decades, more distributed generation (DG) systems and distributed energy resources (DER) systems have penetrated the conventional power grids, particularly those based on wind and solar technologies. The rapid technological developments of DER systems are largely governed by advancements in power electronics technologies and related fields, which form the core of the characteristics of modern power systems. However, the increasing relevance of DER systems has transformed traditional unidirectional power flow into bidirectional power flow, introducing new challenges to system stability, reliability, and security. As a result, various international standards, grid codes (GCs), and inverter control algorithms that shape key power systems parameters; such as active power/ frequency control (APC/FC) and reactive power/voltage control (RPC/VC), are evolving. In addition, emerging concept known as fault ride-through (FRT) capabilities together with modern power electronics-based protection logics are been developed to facilitate secure and reliable integration of DER into modern power grids. The paper reviews the key technical, regulatory, and policy challenges associated with DER integration, including issues related to control algorithms, grid interoperability, and operational flexibility. Furthermore, it presents an overview of power systems supportive functions provided by DER inverters, which are categorized into reactive power/voltage control (RPC), voltage ride-through (VRT), and frequency ride-through (FRT) capabilities. The advantages, limitations, and evolution of these control algorithms and research gaps in both policy challenges and technical challenges are analyzed, along with emerging trends that will continue to shape the future of modern power systems.
Abstract The operating temperature of photovoltaic (PV) modules is a key factor affecting electrical output because heat accumulation reduces voltage, power generation, and conversion efficiency. This study experimentally evaluated a passive thermal-management method using perforated aluminium composite panel (ACP) plates installed on the rear side of a 100 Wp polycrystalline PV module under outdoor conditions. Three perforation diameters, 20, 22.5, and 25 mm, were compared with an uncooled reference panel. Solar irradiance, PV surface temperature, lower-panel/ACP-plate temperature, voltage, current, output power, and efficiency were monitored during a 14-day outdoor campaign, with the detailed profile reported at 15 min intervals. At the peak irradiance of 1009 W/m², the 25 mm configuration recorded a PV panel surface temperature of 42.2 °C and a perforated ACP plate temperature of 38.5 °C, while producing 65.16 W at 16.25 V and 4.01 A, corresponding to an efficiency of 11.75%. The same configuration reached a maximum instantaneous efficiency of 13.29% and the highest mean performance ratio for the reported profile. These findings show that increasing perforation diameter enhanced rear-side heat removal and improved PV electrical performance without auxiliary cooling power.
Abstract The rapid proliferation of 5G networks has enabled a wide range of data-intensive applications, including video streaming, online gaming, metaverse platforms, and real-time communication services. Understanding the behavior and characteristics of such heterogeneous traffic is essential for efficient network management, resource allocation, and quality of service (QoS) optimization. However, most existing studies rely on simulated or synthetic datasets, limiting their applicability to real-world scenarios. This paper presents a comprehensive data-driven analysis of 5G traffic using a large-scale real-world dataset collected from commercial mobile networks. The dataset consists of over 328 h of traffic captured from diverse applications, including Netflix, YouTube Live, Zoom, Roblox, and cloud gaming platforms, using mobile-based packet capture tools without specialized hardware. The collected data is processed into a time-series format with packet-level features, enabling detailed traffic characterization and analysis. We investigate the effectiveness of machine learning and large language models (LLMs) in classifying and analyzing 5G traffic patterns across multiple application categories. Experimental results reveal that traditional structured machine learning approaches outperform LLM-based models in handling network traffic classification tasks, particularly in multi-class and imbalanced scenarios. Furthermore, LLM-based methods exhibit high rates of inconsistency and prediction bias, highlighting their limitations when applied to structured network data. The findings of this study provide practical insights into the suitability of different AI approaches for real-world 5G traffic analysis. The proposed framework contributes to the development of more reliable and efficient traffic classification systems, supporting advanced network optimization, intelligent traffic management, and future 5G/6G deployments.
