Abstract

Seasonal meteorological variations significantly degrade the off-design performance and exergy efficiency of compressed air energy storage (CAES) systems. To address this challenge, this study investigates the dynamic charging process of a 10 MW advanced adiabatic CAES (AA-CAES) system and evaluates its impact on system-level storage performance. A system-level dynamic simulation model is developed in Modelica, capturing the nonlinear coupling of thermal, volumetric, and rotational inertias. The coupled response mechanism between inlet guide vane (IGV) aerodynamic regulation and proportional-integral-derivative (PID) control of the cooling water flow is examined, revealing the dynamic evolution of exergy efficiency under cross-seasonal operating conditions. Results indicate that seasonal disturbances significantly shift the system’s operating boundaries. While IGV regulation exhibits inherent mechanical delay and the discharge pressure lags due to the heat exchanger’s thermal capacitance, the cooling-water PID controller robustly tracks the pinch-point temperature difference under both winter and summer conditions, demonstrating effective dynamic disturbance rejection. During the sliding-pressure charging process, the dynamic exergy efficiency varies nonlinearly. The coordinated IGV–PID control strategy yields the greatest efficiency improvement in the medium-pressure range, effectively minimizing irreversible thermodynamic losses. Notably, this efficiency gain exhibits marked seasonal asymmetry, with a substantially larger improvement in winter (1.51%) compared to summer (0.34%). Furthermore, by thermodynamically balancing inter-stage power consumption, the multi-stage coordinated architecture restricts the cross-seasonal variation in energy storage density to merely 2.6%, demonstrating exceptional thermodynamic stability for reliable grid integration.

Abstract

Occupant-centric control (OCC) can reduce heating, ventilation and air-conditioning (HVAC) and lighting waste, but vision-based occupancy sensing is still often assessed as a computer-vision task rather than as building energy infrastructure. This paper reviews recent camera-, thermal- and edge-AI occupancy detection studies and uses a compact meeting-room benchmark to examine trade-offs that constrain scalable deployment. Prior work shows strong potential for estimating presence, counts, localisation and activity, yet evidence remains fragmented across bespoke rooms, camera views, privacy assumptions, short tests and accuracy metrics weakly linked to control decisions. To ground the review, real-time object detectors were deployed on Raspberry Pi edge devices, using person detections as occupancy counts. Externally metered power, sustained throughput, compute use and count error were assessed together. The case study is illustrative rather than exhaustive, showing that device power was broadly similar across models while processing speed and energy per processed frame varied substantially. YOLO-family models generally gave the most reliable counts, while lighter detectors improved computational efficiency but required post-processing or higher undercounting risk. The review argues that occupancy-vision research should move beyond model ranking toward decision-oriented benchmarks covering occlusion, domain shift, privacy, thermal resilience, duty cycling and net energy benefit. Vision sensing can support intelligent buildings, but only when its energy and governance costs are explicitly considered.

Abstract

In energy systems, load curves are where natural electricity demand meets power-system supply. Under rural energy-transition policies in China, village collectives, local governments, and cooperatives are showing stronger willingness to participate in energy-transition planning, while local electricity-use patterns are being reshaped by distributed photovoltaics, electric-vehicle adoption, and changing seasonal demand. However, user-side information usable for constructing load profiles is often fragmented across actors and difficult to translate directly into power-system planning variables. This paper proposes an AI-assisted workflow for non-utility actors to generate reviewable residential living-load representations from fragmented rural evidence. Questionnaires, scenario-recall responses, appliance records, and electricity bills are first assigned source-based credibility weights and then structured by large language models into three auditable input libraries. Appliance-attribute-based trigger rules are used in a bottom-up Monte Carlo simulation to generate K=300 residential living-load candidates. Weak validation using limited local references — transformer capacity, seasonal electricity-use benchmarks, and intraday-shape plausibility checks — selects a representative baseline profile with bounded plausibility. Applied to a representative rural transformer area in central China, the workflow produces an interpretable residential living-load representation with clear seasonal patterns and morning-evening peaks. Illustrative S0/S1/S2 scenario traces further show that stated response willingness under time-of-use information and direct-incentive assumptions can shift selected schedulable loads toward midday and slightly improve local PV supply-to-load alignment. However, residential-side flexibility remains limited and cannot by itself substantially reshape transformer-area source-load matching.

Abstract

With the rapid development of electric aircraft, thermal management of high power-density motors is increasingly limited by compact structures, high heat flux, and lightweight constraints. This study proposes a passive thermal management strategy for motor housings using form-stable composite phase change materials (FS-PCMs). The FS-PCM exhibits leakage-free phase transition, a melting range of 50–55 °C, latent heat of ~149 J g⁻¹, and good thermal stability. Two identical motor housings, with and without FS-PCM embedded in fin gaps, were fabricated and tested under different heat source temperatures, ambient conditions, and pulsed heating loads. Results show that FS-PCM integration reduces peak temperature by 15–20 °C, delays temperature rise, and induces a stable phase change plateau. Under pulsed heating, the performance strongly depends on the matching between duty cycle, ambient temperature, and phase transition kinetics. Thermal resistance and energy analyses indicate that mechanical compaction reduces interfacial thermal resistance to below 0.1 K W⁻¹, with negligible mass penalty. Overall, the proposed structure provides an effective lightweight passive thermal management solution for aerospace motor applications.

Abstract

Deep reinforcement learning (DRL) has demonstrated its potential in multi-objective HVAC control, enabling simultaneous optimization of energy consumption, thermal comfort, and indoor air quality. However, DRL policies often produce high-variance, unstable control signals, limiting their deployment in real-world BMS. Policy distillation offers a solution toward stable and efficient controllers, while conventional imitation learning suffers from distribution shift, leading to performance degradation in closed-loop operation. Moreover, HVAC control systems exhibit slow, history-dependent dynamics, rendering the Markov assumption insufficient and motivating the use of sequence modeling. This paper proposes a robust sequence-based policy distillation framework with dataset aggregation for deployable HVAC control. A stability-aware multi-objective loss is introduced to enforce smooth and reliable control actions, while dataset aggregation mitigates distribution mismatch. The approach is validated across diverse weather conditions and occupancy scenarios in EnergyPlus simulation, demonstrating improved stability and consistent multi-objective performance compared to standard distillation baselines.

Abstract

To address issues in existing demand response research, such as oversimplified industrial process modeling, insufficient multi-timescale coupling, and lack of multi-production line coordination, this paper proposes a demand response control strategy for automotive manufacturing parks that accounts for material coupling and process constraints. First, a refined power model for each production line is developed, utilizing Resource Task Network (RTN) to characterize inter-line material coupling, thereby establishing a park-level load adjustability modeling method tailored for demand response. Second, a revenue settlement mechanism and effective response criteria are designed, and a “boundary-capacity” collaborative optimization method is introduced. The first stage determines the optimal production baseline and power adjustable boundaries, while the second stage optimizes the economically optimal response capacity within these boundaries. Simulation results demonstrate that, while ensuring order fulfillment, the proposed method reduces operating costs by 8.92% compared to the default scheduling scheme, validating its effectiveness in enhancing demand response capability and economic performance.

Abstract

Data centers (DCs) serve as vital infrastructure of the global digital economy, accompanied by explosive growth in computing demand, massive energy consumption, and intensive carbon emissions. Currently, existing DC energy systems struggle to synergistically improve operational economy, environmental performance and energy efficiency while guaranteeing stable service quality. Against global decarbonization and China’s East Data, West Computing strategy, this study constructs a solar-biomass integrated energy system for a DC in Aksu, Xinjiang, exploiting local abundant solar energy and cotton straw biomass resources. The system integrates photovoltaic power generation, biomass power units, proton exchange membrane electrolysis, energy storage, hybrid cooling, and green methanol synthesis for long-duration hydrogen storage. Day-night differentiated operation rules and flexible workload scheduling are adopted to realize compute-power-cooling collaborative regulation. Based on 8760-hour dynamic simulations, a surrogate multi-objective optimization model is established to optimize Energy Reuse Effectiveness (ERE), Average Delay Rate (ADR) and unit area cost (ACTA). Results show that the proposed system achieves PUE of 1.132, ERE of 1.082, ADR of 0.174 and ACTA of 149.44 $/m². The carbon utilization efficiency (CUE) is reduced by 29.05% compared with conventional unscheduled scenarios.

Abstract

Short-term load forecasting (STLF) is fundamental to
power system operation and optimal dispatch.
Traditional machine learning methods suffer significant
accuracy degradation when training and test data exhibit
distribution shift or when historical data is extremely
scarce. This paper systematically investigates the
performance of a supervised fine-tuned large language
model (GPT-4o-mini) for point-by-point STLF under three
representative scenarios with increasing difficulty:
stationary distribution, distribution shift, and few-shot
cross-distribution prediction. Experimental results on
real-world load data reveal a clear performance
crossover: while traditional methods (MLP, LSTM,
BiLSTM, RF) maintain competitive accuracy under
stationary conditions, the fine-tuned LLM achieves a
mean percentage error (MPE) of only 6.69% in the most
challenging few-shot cross-distribution scenario,
compared to over 42% for all traditional baselines. This
demonstrates that pre-trained knowledge enables LLMs
to achieve effective cross-distribution transfer with
minimal task-specific data.

Abstract

The proliferation of electric vehicles demands user-friendly remote control interfaces that empower EV owners to participate in vehicle-to-grid (V2G) programs. This paper presents a user-centric intelligent remote control system integrating mobile application interfaces, T-Box edge intelligence, and software-defined networking (SDN)-based in-vehicle QoS management. A hierarchical architecture is proposed featuring a React Native mobile application with real-time feedback; an MPC-based T-Box controller for intelligent charging scheduling; and a three-tier end-to-end QoS chain comprising T-Box egress HTB/tc priority scheduling, cellular modem DSCP-based air-interface scheduling, and operator QCI-guaranteed dedicated bearer — validated on a production T-Box platform. Simulation under 200-vehicle-scale scenarios combined with hardware-level QoS chain verification demonstrates 92% user satisfaction, 1.2-second average command response time (79% improvement over baseline), 31.5% charging cost reduction, and zero HAD video frame drops.

Abstract

Against the background of global low-carbon energy transition, the storage and transportation of hydrogen remain challenging. Ammonia has become an ideal hydrogen carrier due to its high hydrogen density and easy liquefaction. However, conventional medium-temperature ammonia decomposition reactors suffer from low conversion efficiency, poor heat transfer performance, and thermodynamic equilibrium limitations. To address these issues, this study proposes a novel reactor design that integrates metal foam for heat transfer enhancement and hydrogen-selective permeable membrane to break the thermodynamic equilibrium. Experimental results show that metal foam significantly improves the heat transfer performance of the reactor. The 40 PPI copper foam reduces the maximum radial temperature difference to 61.55 °C and greatly increases the overall heat transfer coefficient. In the temperature range of 450-600 °C, the metal foam-packed membrane reactor exhibits much higher conversion than conventional reactors. It achieves nearly complete conversion at 600 °C, and reaches 65.18% at 500 °C, which is 36.93% and 20.43% higher than that of the traditional packed-bed reactor and single membrane reactor, respectively. After parameter optimization, the conversion is further improved to 82.89%, and the reactor shows better stability under high gas hourly space velocity. This work provides a feasible scheme and experimental support for efficient medium-temperature ammonia decomposition for hydrogen production.