ABSTRACT. Real-time visual intelligence at the industrial edge is increasingly required for occupational safety, online quality control, and rapid operational intervention. In practice, however, edge deployment is limited by constrained memory, restricted power envelopes, and heterogeneous embedded hardware. This paper presents a compact comparative study of lightweight deep learning models for two industrially relevant tasks: personal protective equipment monitoring for worker safety and visual defect recognition for product inspection. To ensure reproducibility and practical relevance, the study is grounded on two public datasets: the CHV PPE dataset for safety detection and the NEU-DET steel surface defect dataset for defect recognition. Within a unified edge-oriented protocol, the paper compares YOLOv8-nano and YOLO11-nano for object-level safety monitoring, and MobileNetV3 and EfficientNet variants for lightweight defect recognition. The analysis centers on the latency-memory-accuracy trade-off rather than predictive quality in isolation, using a unified measurement protocol that reflects edge deployment constraints. The results indicate that compact YOLO detectors remain the most suitable choice for real-time safety localization, with YOLO11-nano offering the strongest overall balance. For fixed-view defect inspection, EfficientNet provides the highest recognition accuracy, whereas MobileNet yields the most favorable operating point under stricter response-time and memory budgets. The study shows that practical edge-AI selection in industrial environments should be treated as a constrained systems decision involving task formulation, acceptable error, inference deadline, and runtime footprint rather than as a single-model ranking problem. In addition, the study reports deployment-oriented measurements under fixed inference settings, enabling a more transparent comparison of compact models for industrial edge scenarios.

ABSTRACT. Modern energy management systems, even within advanced energy internet (EI) infrastructures, remain fundamentally reactive, optimization-bound, and incapable of reasoning about context, intent, or uncertainty. While the EI paradigm has established a powerful cyber-physical architecture for interconnecting distributed energy resources via software-defined packetized networks, the question of how such systems should think, adapt, and govern energy decisions intelligently remains an open challenge. This paper introduces cognitive energy management (CEM); a new conceptual framework that addresses this gap by redefining how energy systems perceive, reason, learn, and act within complex operational environments. Grounded in the EI cyber-physical foundation, CEM extends beyond conventional optimization by embedding goal-directed reasoning and continuous adaptation into the energy management loop, positioning itself as the cognitive governance layer of EI-based infrastructures. We formally define CEM, distinguish it from rule-based and optimization-based paradigms through structured comparison, and articulate its core architectural layers. To demonstrate the framework's practical value, we develop a toy problem grounded in smart port energy management; one of the most operationally demanding EI node environments in modern infrastructure. Specifically, we model a predictive vessel turnaround scenario in which a CEM-enabled system plans energy procurement, storage pre-charging, and load scheduling across a six-hour operational horizon. The demonstration illustrates how CEM moves the EI beyond feasibility-seeking toward intelligent, anticipatory energy governance.

ABSTRACT. Countries all over the world face persistent grid instability driven by underinvestment, ageing infrastructure, and rapid demand growth. Kenya, for example, records an average of 9.42 hours of power outages and 3.57 interruptions per customer monthly—figures that exceed national and regional reliability benchmarks. Conventional backup solutions such as diesel gen- erators and lithium-ion battery systems remain inadequate due to high operating costs, emissions, and performance degradation under tropical temperatures. The proposed concept in the paper presents an energy storage and supply system that integrates Flywheel Energy Storage (FESS) with Molten Carbonate (MCFC) hydrogen fuel cell technology to enhance renewable integration and grid resilience. The system combines a high-speed carbon-fibre flywheel (50,000 RPM, 0.0005 N·m friction loss) for sub-second frequency response with a hydrogen subsystem for multi-day energy storage. A hier- archical IoT-based monitoring and predictive control framework employs an LSTM neural network to forecast grid anomalies 15 seconds in advance with 93.7% accuracy, enabling optimal power allocation between subsystems. Experimental hardware- in-loop testing over 1,000 hours demonstrates frequency stability within ±0.05 Hz during 90% of renewable fluctuation events and 72+ hours of hydrogen-powered backup. The results confirm the feasibility of AI-optimised hybrid storage as a scalable pathway toward resilient, low-carbon smart grids across Sub-Saharan Africa.

ABSTRACT. Reliable, low-carbon energy supply in Nordic climates requires coordinated use of multi-carrier energy systems (MESs) and advanced storage technologies. In the Finnish context, seasonal hydrogen storage (HSS) and sand battery–based high temperature thermal storage (SBS) are particularly important, as they enable shifting surplus renewable generation from windy and sunny periods to long, cold winters while reducing dependence on fossil heating fuels. This paper presents an energy hub (EH) optimization framework for such MESs with high renewable penetrations. The two-level model co-optimizes long-term planning and short-term operation of generation, hybrid daily/seasonal storage, energy conversion units, demand response, and vehicle-to-grid–enabled plug-in electric vehicles. Additionally, it explicitly includes battery degradation costs and assesses the coordinated use of seasonal HSS and SBS under uncertainty in renewable generations, demand, and prices. The case studies demonstrate operating cost and emission reductions of 4.8% and 6.3%, respectively, while achieving a renewable energy penetration of 91%.

ABSTRACT. This paper presents Pelagia as a simulation-based Decision-Intelligence framework linking an initial 3D concept design, a reproducible GitHub-based simulation workflow, and comparative evaluation across deterministic and stochastic studies. The 3D design defines a dock-adjacent near-shore system with a three-petal folding structure and coordinated body motion, from which motion-derived states and signal-informed features are derived for analysis. The Decision-Intelligence framework organises these features into an enterprise-oriented pipeline in which local behaviour is encoded, classified, reviewed locally, and prepared for privacy-preserving global sharing in reduced visual form. The simulated framework evaluation includes a 12-test deterministic baseline, a 120-test deterministic campaign, a 120-test stochastic campaign, and a repeated 120-stochastic x 100 study. Comparative confusion matrices, repeated-run accuracy, and macro-F1 analysis indicate strong state separability under deterministic conditions and its persistence under stochastic variation. The framework, therefore, provides a reproducible and interpretable basis for simulation-driven decision support in near-shore energy systems.

ABSTRACT. The increasing frequency of extreme meteorological events poses a critical challenge for the planning of renewable energy systems that depend on reliable forecasts. Green hydrogen produced through solar photovoltaic (PV) electrolysis emerges as a clean and decentralized energy vector; however, numerical weather prediction (NWP) models present systematic biases that are amplified under atypical conditions, compromising production estimation. This work proposes a two-pass bias-correction scheme applied to Global Forecast System (GFS) model forecasts using climate-regime-differentiated XGBoost, with explicit detection of heat and cold waves, evaluated using data from the solarimetric station of the Federal University of ABC (UFABC) in Santo André, Brazil. The results show reductions in temperature root mean square error (RMSE) of up to 1.662 °C and reductions in irradiance errors of up to 111.55 W/m², leading to improvements exceeding 18% in green hydrogen production estimation compared with raw GFS model forecasts.

ABSTRACT. An interactive workshop focused on career development, leadership, and professional growth in engineering. Through practical advice and shared experiences from women leaders in academia and industry, participants will explore how to build confidence, identify opportunities, overcome challenges, and shape a meaningful career path.

ABSTRACT. This paper presents a compact, solid-state switch-based repetitive high-voltage nanosecond pulse generator (NSPG) to deliver repetitive high-voltage pulses with fast rise/fall time. The NSPG consists of an AC-DC conversion stage, a pulse generation stage, and a high-voltage pulse boosting circuit. The developed NSPG is experimentally tested up to 1 kHz pulse repetition frequency (PRF) across various capacitive-resistive (RC) loads chosen to match the dynamic electrical characteristics of an electron beam source. The results demonstrate that the NSPG efficiently generates pulses up to 8 kV with approximately 100 ns pulse duration, maintaining consistent performance across different RC loads. Additionally, it has been observed that the effect of circuit parasitics on the output pulse is minimal, even when increasing from lower to higher pulse voltage levels.

ABSTRACT. Abstract—Wireless Body Area Networks (WBANs) are critical enablers of sustainable digital healthcare systems, supporting continuous physiological monitoring through wearable and implantable sensors. However, limited battery capacity and stochastic energy harvesting introduce significant challenges in maintaining long-term reliability, delay performance, and energy efficiency. Ensuring resilient and low-carbon healthcare infrastructures require intelligent resource allocation mechanisms capable of operating under uncertainty. This paper proposes an AI-enabled graph-based reinforcement learning framework for optimizing energy harvesting of WBANs. The network is modeled as a dynamic graph capturing topology, communication links, and energy dynamics. The resource allocation problem is formulated as a stochastic control process aiming to maximize long-term energy efficiency while maintaining queue stability and battery sustainability. A multi-objective reward function integrating energy efficiency maximization, delay minimization, and battery protection is introduced. The proposed framework is benchmarked against Grey Wolf Optimization (GWO). Simulation results demonstrate improved energy efficiency, reduced queue delay, and enhanced battery sustainability, contributing to resilient and sustainable digital healthcare systems aligned with clean transition objectives.

ABSTRACT. Industrial chemical and gas processing facilities are under increasing pressure to make emissions of greenhouse gases more efficient without sacrificing operational safety and efficiency. Process faults, such as disturbances to feed, temperature and equipment degradation, are a direct cause of excess energy consumption, unplanned flaring and off specification product wastage. Traditional machine learning methods have high accuracy in binary fault diagnosis (95.5%) but accuracy drops dramatically in fine-grained multi-class diagnosis (53.6% among 21 types of faults), which limits their usefulness in targeted emission reduction interventions. This article proposes a framework for Retrieval-Augmented Generation (RAG) that combines traditional anomaly detection with open-source Large Language Models (LLMs) to provide more interpretable and more natural language fault diagnoses with emission impact assessments and recommendations for corrective actions. The framework is evaluated using the Tennessee Eastman Process (TEP) benchmark, where the RAG-LLM system demonstrates a diagnostic agreement of 71.4% with expert annotations, a relative improvement of 65% compared to a non-augmented LLM baseline, with a concurrent decrease in hallucination rate from 34.7% to 8.2%. The proposed methodology utilizes a vector-embedded process knowledge base with a locally deployed LLM through a process of retrieval orchestration, offering decarbonization advice for fault categories where the classifier is most challenged. The Retrieval Precision@3 is found to be 0.847, and the BERTScore F1 is found to be 0.863, indicating high-quality knowledge grounding.

ABSTRACT. Photovoltaic (PV) converters deployed in the Arabian Peninsula are subject to abrupt irradiance attenuation due to Shamal-driven airborne dust, with field-measured reductions of 40–85% occurring within minutes and aerosol optical depths exceeding 1.5 at peak storm intensity. These transients drive insulated-gate bipolar transistors (IGBTs) through power-cycling thermal excursions that dominate converter wear-out. This paper proposes a dust-storm-aware dynamic switching-frequency modulation (DSFM) strategy for a two-phase interleaved boost converter (IBC). A closed-form constrained control law modulates the switching frequency within a 20–50 kHz window, subject to a 5 kHz/ms slew limit and analytically verified small-signal stability margins (≥6 dB gain, ≥45° phase). A four-layer Cauer RC thermal network provides 1-ms online estimates of junction temperature; Coffin–Manson with rainflow counting and Palmgren–Miner summation provide offline lifetime quantification. PI gains are tuned offline using three population-based metaheuristics (WOA, HHO, and MFO), minimizing a weighted thermal and voltage cost. DSFM is benchmarked in MATLAB/Simulink against fixed-frequency PWM, MPPT with fixed switching, MPC, and SMC across three scenarios, with a 36-point parametric sensitivity sweep. Under the moderate Shamal profile, DSFM reduces peak junction-temperature swing by 36.3% and extends predicted IGBT service life by a factor of 2.18, with DC-link overshoot held to 1.2 % and settling time to 58 ms.

ABSTRACT. Shipboard power systems have traditionally relied on centralized power management logic to balance generator loading and restore nominal frequency after changes in loading or generation capacity. In inverter-dominated DC architectures, similar expectations often persist, with DC-link voltage deviation assumed to require explicit supervisory correction. This paper presents a system-level interpretation of exception handling in variable-voltage DC shipboard power systems, where coordination emerges from voltage-dependent local device behavior. Rather than treating DC-link voltage as a control error, voltage is used as the execution variable of coordination. During source loss, overload beyond droop capability, excessive current injection, and large load rejection events, DC-link voltage evolves according to instantaneous power balance. Voltage-responsive loads autonomously substitute their controlled variable when activation thresholds are reached, supporting voltage-forming sources and enabling progressive system response through local interaction. Voltage regulation is defined as active control of DC-link voltage from the prevailing operating point, independent of power flow direction; thus, load power reduction constitutes voltage support when driven by voltage-controlled operation. This behavior differs fundamentally from conventional AC frequency restoration, where nominal operating conditions are actively enforced. In the variable-voltage paradigm, controlled voltage deviation is an integral part of system operation and enables decentralized coordination without supervisory control.

ABSTRACT. The increasing adoption of converter-dominated DC microgrids has introduced new challenges related to source overloading and voltage stability during abnormal operating conditions. If overload progression is not resolved by power management or load-side response, voltage-supporting sources may eventually reach equipment protection boundaries such as current limitation or protective shutdown, which may reduce system survivability under severe loading conditions. This paper investigates the system-level behavior emerging when converter interfaced loads participate in DC-link voltage support during source overload situations. A modeling-based study is performed using a droop-controlled DC source supplying two converter interfaced loads. Two operating scenarios are compared: a case without load participation and a case in which one load exhibits voltage-responsive behavior during undervoltage conditions while the second load remains non-participating. The analysis focuses on the evolution of DC-link voltage, source loading, load-current behavior, and the resulting operating-point progression during overload conditions. The results demonstrate that participating loads can reduce source stress by decreasing their current demand as the DC-link voltage falls below a predefined participation threshold. Compared to the non-participating case, the voltage-responsive behavior reduces source stress and contributes to improved DC-link voltage stabilization. The study highlights how locally acting load participation can influence the overall operating behavior of DC microgrids during source overload events without requiring centralized coordination or communication between devices. The presented work focuses on the observed system-level behavior associated with load participation during overload conditions and provides a modeling-based demonstration of its influence on DC microgrid operating-point evolution.

ABSTRACT. The transition toward low-carbon energy systems requires advanced solutions capable of integrating multiple energy sources while ensuring efficiency, reliability, and sustainability. Artificial intelligence (AI) has emerged as a transformative technology for optimizing multi-energy systems by enabling intelligent forecasting, control, and decision-making. This paper explores the role of AI in enabling low-carbon and multi-energy systems, focusing on intelligent optimization, predictive analytics, and integrated energy management. A conceptual framework is proposed to demonstrate how AI techniques such as machine learning, deep learning, and reinforcement learning can enhance energy efficiency, reduce carbon emissions, and improve system resilience. The study highlights key challenges, opportunities, and future research directions for AI-driven sustainable energy systems.

ABSTRACT. Kenya’s annual electrical energy consumption of 14,472 GWh continues to increase due to industrial expansion, technological advancement, and population growth. This rising demand necessitates the integration of renewable energy sources such as solar, wind, geothermal, hydropower, and bioenergy into the national grid. Safe integration requires strict synchronization of voltage, frequency (50 Hz), phase sequence, and phase angle to maintain grid stability and prevent system faults.

This paper presents an intelligent dual-layer PCB architecture-based grid protection and monitoring system incorporating multi-parameter sensing, smart grid communication, and predictive analytics. The system continuously monitors voltage, current, frequency, and phase variations across three-phase networks using true-RMS sensors and pulse-width modulation (PWM) frequency simulation for testing. Upon detecting deviations beyond synchronization thresholds, the controller executes automatic islanding protection, logs the event, and performs safe reconnection after stability restoration.

Real-time data is transmitted via Modbus RTU, LoRa, MQTT and Wi-Fi hybrid protocols through the dual-layer PCB architecture used to a cloud-based IoT dashboard for visualization, alerts, and trend analysis. A machine learning module processes historical data for fault prediction and adaptive threshold adjustment. Experimental results demonstrate accurate monitoring, predictive fault detection, and automated protection, offering a scalable pathway toward Kenya’s smart grid modernization and reliable renewable energy integration.

ABSTRACT. This paper presents a data-driven framework for optimizing energy resource usage in smart homes by combining advanced electricity price forecasting with adaptive battery scheduling. The proposed system integrates a hybrid prediction model—merging Facebook’s Prophet and Transformer-based deep learning—to capture both seasonal patterns and complex temporal dependencies in dynamic electricity pricing. Forecasts are then utilized in a tolerance-aware decision-making module that strategically controls battery charging and discharging operations. This module employs quantile-based thresholds and real-time price deviation checks to ensure cost-effective energy management under uncertainty. Simulation results using realworld data from April 2025 demonstrate the system’s ability to reduce electricity costs while respecting operational constraints, such as battery capacity and efficiency. The modular design supports customization for different battery configurations and pricing conditions, offering a scalable solution for intelligent, autonomous energy control in future smart home environments.

ABSTRACT. Battery energy storage systems (BESS) are increasingly commercialized through contracts that allocate operational control and merchant-revenue risk between asset owners and offtakers, including full tolling agreements, revenue floors with upside sharing, and profit-sharing structures.

In its simplest form, a BESS tolling agreement can be characterized by two parameters: a fixed payment ($P$) and a revenue-sharing coefficient ($\alpha$). The fixed payment is made by the offtaker to the battery owner to guarantee a minimum income and support project financing, whereas $\alpha$ determines the share of merchant revenues allocated to each party. These parameters depend on market characteristics and participants' risk preferences.

In this context, we propose a Stackelberg formulation to analyze representative BESS tolling agreements. For illustration, the offtaker acts as the leader, offering contract terms while anticipating the battery owner's decision under merchant-revenue uncertainty. The results show how revenue uncertainty and the risk aversion of both parties jointly shape contract design, with contracts shifting from pure tolling agreements to hybrid revenue-sharing structures as market volatility and risk preferences change.

ABSTRACT. Containerized Battery Energy Storage Systems re quire cooling systems to maintain battery temperatures within safe operating limits. However, the operation of these systems represents a significant share of auxiliary energy consumption, while reducing cooling may increase battery temperature and accelerate battery degradation, creating an operational trade off between cooling energy consumption and battery lifetime. To address this trade-off, this work presents an optimization model for the operation of a containerized battery energy storage system, aiming to minimize electricity procurement costs and battery degradation while simultaneously meeting hydrogen production requirements. A case study based on meteorological data from the metropolitan region of São Paulo, Brazil, shows that calendar aging dominates battery degradation. The results also indicate a transition from thermally constrained operation under hot conditions to energy-driven operation under colder conditions, while higher photovoltaic generation reduces grid dependence with similar cooling demand.

ABSTRACT. Climate change and the growing demand for clean energy solutions have driven interest in green hydrogen as a key energy vector in the transition toward low-carbon systems. In this context, residential energy communities with distributed photovoltaic generation represent a favorable environment for local hydrogen production, as they generate solar surpluses that can be valorized through electrolysis rather than exported to the grid at reduced compensation rates. This work proposes a linear programming model for the optimal sizing of hydrogen production systems fed by photovoltaic surpluses in energy communities, minimizing the Annual Cost of the System as the objective function. The model integrates distributed photovoltaic generation, electrolysis, compression, and storage into a single linear optimization problem, validated over eleven real residential communities in Santo André, Brazil. The results provide valuable insights for the techno-economic planning of such systems, demonstrating that the community scale represents an effective intervention level for integrating green hydrogen production into the urban energy transition.

ABSTRACT. In Europe, sulphur hexafluoride (SF₆) remains essential for medium- and high-voltage switchgear due to its ex-ceptional dielectric and arc‑quenching properties. At the same time, its extremely high global warming potential places SF₆ under one of the most stringent regulatory regimes worldwide. With the adoption of Regulation (EU) 2024/573, the EU has clearly defined the transition to-wards closed-loop SF₆ lifecycle management, prioritising emission reduction, reclamation, and re‑use over contin-ued reliance on virgin gas. European and international standards, including IEC 60480, IEC 60376, and CIGRE TB 914, establish a com-prehensive technical framework for the reuse, purifica-tion, and quality control of SF₆. Within this framework, modern online gas drying and reclaiming technologies, applicable even during normal operation of gas‑insulated switchgear (GIS), enable compliance with regulatory requirements while significantly reducing outages, opera-tional costs, and environmental impact. A key focus is placed on humidity control, which is recognised as a critical factor influencing gas quality, insulation performance, and long‑term asset reliability. Periodic monitoring of SF₆ purity, humidity, and decom-position products supports predictive maintenance strat-egies aligned with European asset management and sus-tainability objectives. In addition, off‑site reclamation technologies based on membrane, pressure - cryogenic separation allow used SF₆ to be purified to a quality equal to or exceeding vir-gin gas, with substantially lower energy consumption and negligible emissions compared to new gas production. Together, these approaches demonstrate that compliance with European climate regulation, high grid reliability, and reduced lifecycle costs can be achieved simultaneous-ly, establishing closed‑loop SF₆ management as the new operational standard for the European power sector

ABSTRACT. To meet the European Green Deal’s goal of a 90% reduction in transport emissions by 2050, a significant modal shift from road and air toward rail and sustainable public transport is required. However, the current fragmentation of booking systems and the absence of a unified "travel contract" framework remain critical barriers to the cross-border adoption of Mobility as a Service (MaaS). This article explores the development of an integrated software platform designed to serve as the digital backbone for multimodal journeys, bridging the gap between long-distance rail and local "last mile" transit. By analyzing the current policy landscape and technical requirements, the study identifies that the primary bottleneck is increasingly policy-driven rather than technological. Furthermore, the article examines the shift in financial sustainability models, contrasting the early commercial pioneers with the public-sector-led initiatives. We conclude that the viability of such digital infrastructure must be evaluated through a holistic cost-benefit analysis that prioritizes decarbonization and public health over traditional ticket margins. By resolving local policy and financial hurdles, this integrated approach can finally empower European citizens to adopt sustainable mobility as a reliable travel option.