Zheng Jun Chew

Spatiotemporal graph neural networks have become widely adopted for multi-station wind speed forecasting, yet most rely on a single, symmetric graph with shared node representations. These designs treat inter-station relationships as instantaneous and reciprocal, ignoring the directional, time-delayed propagation that characterizes atmospheric influence. We propose Lag-Aware Cross-correlation Graph Neural Network (LACGNN), a multi-view framework that integrates three complementary graphs: a learned adaptive graph for latent dependencies, a dynamic cross-correlation graph capturing peak-lag directional propagation within each input window, and a static cross-correlation graph that links stations by the similarity of their full-lag correlation profiles, capturing persistent structural similarities that a single correlation value cannot express. LACGNN further introduces a dual embedding mechanism with per-station learnable weights that separates each station's representation into self and influence components, allowing different meteorological variables to contribute to local prediction and spatial message passing in distinct proportions. Experiments on three geographically diverse datasets (USA, UK, and Ireland) demonstrate that LACGNN consistently outperforms 12 advanced baselines across both MAE and RMSE, with improvements confirmed by statistical tests, while maintaining competitive inference latency among baselines. The source code is available at https://github.com/asarm/ccf-wind-speed.

During train operation, railway traction-returning current generates a power-frequency magnetic field around the rail, offering a potential energy source for self-powered trackside monitoring nodes. The H-shaped-core magnetic field energy harvester (MFEH) is attractive because it can be installed beneath the rail without enclosing the conductor, yet its output is strongly affected by the coupled rail-core-coil system. To clarify these effects, a three-dimensional electromagnetic-circuit-coupled finite-element model of an experimentally validated laminated-silicon-steel H-shaped-core MFEH was established to examine core and coil parameters. Increasing the center-leg and side-leg lengths weakens demagnetization but intensifies eddy-current losses, causing output power to approach saturation. Under a 50 Hz, 300 A current in a 54E1 rail and series-tuned matching, output power approaches 5.1 W beyond a center-leg length of 1000 mm and 3.25 W beyond a side-leg length of 700 mm. Within the investigated ranges, center-leg and side-leg lengths of approximately 800 and 400 mm provide the best power–volume performance, respectively. Increasing side-leg height or width also improves output. A larger coil span improves output by reducing internal resistance, whereas more turns yield diminishing gains because of higher winding and eddy-current losses. These findings provide a quantitative basis for parametric design of H-shaped-core MFEHs in railway environments.

This study evaluates the potential of integrating floating photovoltaic (FPV) systems with green hydrogen production on UK reservoirs to support decarbonization across electricity, heating, and transport sectors. PVsyst was used to simulate annual electricity generation for monofacial and bifacial systems at Killington reservoir and Drift reservoir, while HOMER Pro was used to model hydrogen production via electrolysis and its potential applications. Results indicate that maximum FPV deployment could generate approximately 61 GWh/year at Killington and 20 GWh/year at Drift. Surplus electricity during peak production enables PEM electrolysis, producing up to 869,149 kg/year and 185,277 kg/year of hydrogen for the bifacial systems, respectively. This hydrogen could alternatively deliver up to 9.216 GWh/year and 1.977 GWh/year of electricity or 26.071 GWh/year and 5.558 GWh/year of heat, or support approximately 1,225,808 km/year and 454,550 km/year of hydrogen-powered transport. Additional co-location benefits include significant reductions in reservoir evaporation, estimated at 1.96 million m³/year for Killington and 452,037 m³/year for Drift. Overall, the findings demonstrate that hydrogen integrated FPV systems represent a promising system configuration under idealized deployment conditions, with location-specific modeling providing a UK-specific multi-sector assessment of the low-carbon potential of reservoir-based energy systems. The hydrogen use cases presented are alternative applications of the total hydrogen produced and are not intended to occur simultaneously.

This article presents a self-configurable and self-poweredactive rectifier that operates from 0.25–20 V for energy harvesting applications. The proposed circuit self-startups from a low voltage using a charge pump and amplifies the voltage with a voltage doubler topology to provide succeeding circuits such as boost converters with a higher voltage. When the voltage of the energy harvester reaches a high threshold, the circuit switches its topology to a full-wave rectifier that does not amplify the voltage. The start-up circuit can limit its voltage intake to prevent boosting the high voltage, which may damage the whole circuit. Comparators with a maximum operating voltage of 5.5 V used in the implementation of the rectifier are protected by a diode and resistor based circuit. A piezoelectric energy harvester (PEH) that has a wide open-circuit voltage of 0.4–15 V under the acceleration of 0.04–0.3 g was used to test the circuit. The experiment results showed the rectifier can startup from 0.25 V and switch its topology according to the PEH voltage. The voltage and power conversion efficiencies are over 90% in most cases.