Abstract

This study numerically investigates the multiphaseflow characteristics inside a hollow rotating shaft undernear-vacuum conditions, aiming to clarify thefundamental flow mechanisms governing rotor cooling inflywheel energy storage systems (FESS). Computationalfluid dynamics (CFD) simulations were conducted tocompare the stationary and rotating configurations,focusing on the influence of shaft rotation on phasedistribution, pressure evolution, and velocity fieldcharacteristics. In the stationary case, the impinging jetinduces a strong local pressure peak and downward flow,accompanied by pronounced air entrainment,incomplete gas removal, and localized cavitation neargeometric transitions. When rotation is introduced,centrifugal force dominates the flow dynamics, leadingto a stable annular flow structure with liquid adhering tothe wall and air confined to the shaft core. The air phasedevelops a helical backflow path driven by the low-pressure core region, while near-wall backflow andcavitation are significantly suppressed. The findingsreveal a transition from a gravity- and inertia-driven,chaotic two-phase regime to a rotation-dominated,stratified annular flow regime. This transition enhancesflow stability and liquid film continuity, offeringimportant insights into the coupled flow-phaseinteractions under vacuum and providing practicalguidance for the thermo-fluid optimization of high-speedflywheel rotors.

Abstract

Deep Q-network (DQN) has shown significantpotential in industrial control, yet its high explorationcosts and slow convergence during early deploymenthinder practical adoption. To address this, this studyproposes a data-driven pretraining framework for DQN,based on a multi-step temporal difference (TD)algorithm. Historical operation data and well-designedreward functions are employed to pretrain the agent,significantly reducing post-deployment exploration. Theproposed framework is demonstrated through a casestudy on boiler combustion control system. The agentbuilt on an enhanced recurrent neural network (RNN)architecture regulates coal feeding and air supply tomeet dynamic steam load demands. The effects ofdifferent reward formulations and TD steps onpretraining efficacy are analyzed. Results show thatagents pretrained with linear and nonlinear rewardsboth outperform historical control strategies. The 3-stepTD pretraining strategy outperforms single-step TD,achieving 83.4% consistency with expert actions andrealizing a 5.8% improvement in energy efficiencycompared to manual operation.

Abstract

Machine learning (ML)-based Model Predictive Control (MPC) has shown promise for smart buildings, with online ML model adaptation often integrated to enhance long-term performance. However, abrupt building thermodynamic changes, such as opening/closing windows, can hardly been captured by the conventional online ML model adaptation. An Adaptive Transfer-Learning-Assisted Modular Learning (ATML) framework combined with dataset truncation is proposed by this paper to address the challenge. A simulation case study using a residential building was conducted for evaluation. Results show that, compared with conventional online ML model adaptation, ATML reduces prediction errors by 32.4% after 2 hours of the systemic change and by 25.4% after 22 hours, thereby demonstrating its effectiveness in coping with abrupt systemic changes.

Abstract

In recent years, the Model Predictive Control (MPC) for building systems has received extensive attention, but few studies have applied machine learning (ML)-based MPC to split air conditioners. This paper presents a field implementation of an iterative ML-based MPC with a cloud-based framework to demonstrate its feasibility and energy-saving potential for split air conditioners. The measured results show that, compared with conventional PID control, MPC achieved an average energy-saving rate of 22.9%. In addition, iteratively updating the predictive model enables MPC to achieve enhanced thermal comfort. Hence, this paper demonstrates the energy-saving capability of ML-MPC for split air conditioners and potential scalable cloud-based deployment in residential buildings.

Abstract

Solar ground source heat pump systems offer asustainable solution that reduces energy consumptionand environmental impact. This study develops anintegrated solar ground source heat pump system,evaluating its exergy and environmental impacts. Resultsshow that the exergy efficiency is 49.16%, with a totalexergy destruction of 111.43 kW. The environmentalimpact is greatest during construction, followed byusage, disposal, and maintenance. This study providesvaluable insights for optimizing renewable energyintegration in building systems and serves as a guide forthe design of similar energy-efficient projects.

Abstract

Evaporative cooling using hygroscopic compositehydrogels and radiative cooling are widely studiedpassive thermal management strategies for photovoltaic(PV) modules. This work explores a synergistic approachthat integrates evaporative cooling through ahygroscopic hydrogel on the back side with radiativecooling on the front side of PV modules. A highlyaccurate heat and mass transfer model is developed andvalidated to simulate three configurations: radiativecooling only; evaporative cooling only on the back side;and the combined use of both techniques. Resultsdemonstrate that the hybrid configuration achieves thebest cooling performance, yielding an averagetemperature reduction of 3.2 °C and power generationefficiency (PCE) increase by 1.2%, which corresponds toa 500% and 26% relative improvement in PCE comparedto the radiative-cooling-only and evaporative-cooling-only cases, respectively. Notably, radiative cooling hasonly a minor influence on the hydrogel’s absorption anddesorption dynamics, increasing nighttime waterabsorption by about 1%, while the enhanced radiationextends the duration of the positive temperaturereduction effect by 10%. Radiative cooling can beconsidered as a supplementary method coupled with PVmodules, where evaporative cooling serves as theprimary passive cooling mechanism.

Abstract

How to maximize the use of renewable energy toachieve the most efficient utilization of energy resourcesis increasingly drawing attention. Against this backdrop,this study proposes a rule-based energy managementstrategy optimized by the Sparrow Search Algorithm fora novel diesel-hydrogen hybrid power system,investigating the impact of varying power source limitsthrough simulation. The findings demonstrate thatcontrolling the diesel engine power limit to 40–45 kWand the fuel cell power limit to 5–10 kW maximizes therenewable energy proportion and achieves significantenergy conservation, establishing an optimal balancebetween sustainability and dynamic performance

Abstract

The wake interactions between wind turbines in largewind farms significantly impact optimal layout andpower generation. Therefore, it is crucial to accurateprediction of turbine wake development. In this study,we utilize Large Eddy Simulation (LES) coupled with theactuator line method to investigate the flow of wakesdownstream of a scaled wind turbine model and theNREL 5MW turbine. To improve the accuracy of wake simulations, we propose a novel difference schemecalled the linear upwind stabilized transport (LUST) scheme, which we compare to the second-order centraland upwind schemes. Our results demonstrate that the LUST scheme provides proper numerical dissipation forwind turbine wake simulations, leading to improve accuracy in predicting mean velocity and turbulent intensity when compared to experimental measurements. Furthermore, the LES results indicate that the dissipation introduced by the LUST scheme stabilizes numerical simulations while still capturingsmall scales in the flow field. These findings highlight the effectiveness of the LUST scheme in accurately simulating wind turbine wakes and emphasize the importance of proper numerical dissipation in obtaining reliable results.

Abstract

Ambitious renewable expansion in ecologically fragile regions must reward not only scale but also match with environmental limits. We present a general coupling–coordination framework that evaluates clean-energy build out and ecological integrity on the same decisionfooting. For each location we derive energy efficacy 𝑆and ecological efficacy 𝐸. Match quality is measured by asymmetric, bounded coupling function 𝐶; overall development uses an ecology-prioritised index 𝑇; theirgeometric blend 𝐷 yields a single coordination score that enforces explicit scale thresholds (𝐷2 ≤ 𝑇) and achievesits upper envelope only along the balance line 𝑆= 𝐸. Weclassify cells into coordination tiers (D1–D5) and label the dominant subsystem (ecology-led, balanced, energy-led) to generate rank–driver codes that translate analysis intoaction. Applied to the Qinghai–Tibet Plateau, historical assets (n=1,606) cluster in the mid-to-high corridor—91.90% well coordinated (D4), 6.04% high-quality (D5),2.05% barely coordinated (D3)—while internal imbalance is common (ecology-led 60.77%, balanced13.88%, energy-led 25.34%), implying that unilateral expansion would depress coupling. A resource-based assessment shows a resilient, ecology-led landscape (median 𝐷 =0.777, mean 𝐷 =0.767, mean 𝐶 =0.959; mean𝑆 =0.499, mean 𝐸 =0.695), with province-scale contrasts: Gansu high- 𝐶 but single-driver; Qinghai high-level andmore balanced; Tibet tri-modal with ridge-adjacent optima and energy-pressured pockets; Xinjiang potential constrained by grid/flexibility; Yunnan and Sichuan coordination limited by engineering accessibility. The diagnostics point to three robust pathways—close the lagging subsystem, expand near the balance ridge, and pace build intensity with grid, storage and ecological restoration—providing a generalisable, auditable instrument for high-quality clean-energy growth under ecological redlines.

Abstract

To address the challenges in distribution network protection caused by high-penetration distributed generation (DG) integration, this study proposes a multi-source information fusion-based fault location algorithmfor distribution networks. A dynamic current matrix forfault zones and a topology-adaptive node differential matrix are constructed, and the fault location matrix isderived through matrix multiplication. Grounded in the current differential principle for distribution networkfault localization, logical analysis is performed on the fault discrimination matrix. PSCAD simulation demonstrated that the criterion is capable of accurately locating fault nodes. The algorithm exhibits advantages such as simplicity in principle, low memory consumption, high computational efficiency, and superior accuracy, demonstrating strong practicality. It provides a robustsolution for active distribution networks with high-penetration renewable energy.