Abstract

Efficient temperature regulation is essential formaintaining the sensitivity and noise performance ofspaceborne infrared detectors under varyingobservation modes. This paper proposes an ActiveDisturbance Rejection Controller (ADRC) with dynamicparameter tuning based on the Soft Actor-Critic (SAC)reinforcement learning algorithm, aiming to optimize thetemperature control performance and disturbancerejection capability of spacecraft systems. A thermaltransfer model was established, and the Soft Actor-Critic(SAC) algorithm was introduced to dynamically tuneparameters (system gain, controller bandwidth andobserver bandwidth) by maximizing a weighted sum ofthe reward and policy entropy. This approach enablesreal-time estimation and compensation of totaldisturbances. Simulation results demonstrate that, intemperature tracking tasks, the proposed methodreduces the integral of absolute error (IAE) by 14.61%and 14.87%, and shortens the settling time by 36.36%and 58.47%, compared to fixed-parameter ADRC and PIDcontrollers, respectively. Under external periodicdisturbances, the proposed controller improves controlaccuracy by 15.6% and 13.6%. Monte Carlo robustnesstests further show that, under ±5% parameterperturbations, the method exhibits small fluctuations inIAE, settling time, and overshoot. The proposed SAC-ADRC strategy provides a promising solution for rapidand high-precision thermal regulation of infrareddetectors in aerospace applications.

Abstract

The increasing energy demand for air conditioningsystems presents substantial economic andenvironmental challenges, making effective airconditioning load monitoring crucial. Traditionalintrusive load monitoring techniques were costly andchallenging to implement, whereas non-intrusive loadmonitoring (NILM) technology offered a cost-effectivealternative. This study constructed the CN-AC-NILMdataset, customized for the characteristics of Chineseelectrical appliances and cultural usage patterns, andsubsequently developed an RNN-attention model for airconditioning systems, with a focus on evaluating theimpact of the attention mechanism on the model’sperformance. Initially, the optimal input feature set forthe air conditioning load disaggregation predictionmodel was determined using the forward featureselection method based on the dataset, followed by anevaluation of performance differences among variousRNN models. Subsequently, it was observed that in thecomparative analysis of baseline models applied to airconditioning system for NILM, LSTM exhibited higheraccuracy than GRU, and bidirectional modelsoutperformed unidirectional models. Furthermore,after introducing the attention mechanism in thebaseline models, the average accuracy of RNN modelsincreased by 21.09%, accompanied by a 5.33% increasein average training time. Finally, the effects of attentionmechanisms and bidirectional structure on the modelswere compared, revealing that the introduction of theattention mechanism outperformed the bidirectionalstructure, and the LSTM-Attention exhibited the bestoverall performance for NILM of the air conditioningsystems. Notably, Attention mechanisms enhancedmodel interpretability through the visualization ofattention weights. By improving load disaggregationaccuracy and interpretability, this attention-enhancedmodel enables more precise energy management for airconditioning systems. It reduces electricity costs,optimizes energy use, and supports demand-sidemanagement, offering substantial economic savings andlowering environmental impact, especially in highenergy-demand settings.

Abstract

Charging stations, as essential infrastructure for smart scheduling and energy management in smart cities, significantly improve the operational efficiency and intelligence of urban energy systems. This study proposes a load forecasting model that integrates LSTM and Transformer networks, introducing the Mamba module as a critical bridge to fuse temporal features extracted by both models, thereby constructing a superior knowledge representation to enhance prediction accuracy. First, LSTM and Transformer networks are used separately to extract latent temporal features from the load data. Subsequently, the Mamba module fuses and enhances these features to capture finer-grained information. Comprehensive experimental results demonstrate that the proposed method achieves an average prediction accuracy of 93.4% on data from multiple electric vehicle charging stations, without substantially increasing computational overhead.

Abstract

Combined heating and power integrated energy system (CHP-IES) enables efficient coordination and dispatch of multiple energy sources, enhancing renewable energy utilization and overall efficiency, thus serving as a key technical approach to support the low-carbon transition under the dual carbon goals. First, the paper develops a detailed model of key components in the system, including gas turbines, gas-fired boilers, electrical energy storage, and thermal energy storage, thereby capturing the multi-energy flow coupling relationships within CHP-IES. Second, a pricing mechanism based on the Stackelberg game is established, in which the distributed energy resource (DER) operator acts as the leader and the users serve as followers. The CHP-IES dynamically determines electricity and heat prices based on market signals, while users adjust their flexible electricity and heat loads in response to price signals. Finally, the proposed method is validated through a case study.

Abstract

In response to the frequent faults and low efficiency caused by response delay and power imbalance in dual-unit intercooled reheat gas turbine/solid oxide fuel cell (IRGT/SOFC) hybrid systems under multiple operating scenarios, this study proposes a multi-input multi-output (MIMO) long short-term memory (LSTM) neural network for dynamic identification and prediction. By integrating physics-based modeling with data-driven training, the proposed framework captures the transient behaviors of a marine parallel IRGT-SOFC system across varying fuel scenarios. Sensitivity analyses on sampling rate, hidden layer configuration, and learning rate demonstrate that a single-layer LSTM with 50 neurons and 0.01 s sampling interval provides the optimal balance between prediction accuracy and computational cost. The trained model achieved R² values exceeding 0.998 across all key outputs, with MAPE consistently below 0.79% except T_FCavg, effectively predicting power output, SOFC temperature, and overall system efficiency. The proposed model effectively captures the transient response patterns of dual-unit systems under varying fuel input, laying a high-precision modeling foundation for predictive power control of next-generation marine propulsion systems in complex operational scenarios.

Abstract

Geothermal development in abandoned reservoirs with CO₂ recirculation aligns with energy-transition goals, yet more efficient system designs are needed. This study reconfigures multi-stage fractured horizontal well patterns into a fractured closed-loop geothermal system (FCLG), in which injection-production fractures interconnect to create multiple rows of high-conductivity channels for heat extraction. We build numerical models to compare a conventional closed-loop system (CLG), a conventional multi-stage fractured horizontal well system (MSFHW), and the proposed FCLG, and we obtain dynamic fracture conductivity from core displacement experiments under geothermal conditions. The multiphysics simulations reveal three characteristic production stages for FCLG: an initial plateau, a decline, and a secondary plateau. Sensitivity analysis indicates injection temperature governs heat-extraction efficiency most strongly, followed by injection flow rate and temperature difference; lower injection temperatures, lower flow rates, and larger temperature differences improve performance. Relative to MSFHW, FCLG delivers ~1.7× higher heat-extraction efficiency, ~3× longer stable production, a maximum produced-fluid temperature increase of 12.5-19.9 K, and a maximum net thermal-power gain of 8.9-9.4 MW. Even a single-fracture FCLG (SFCLG) outperforms CLG, achieving 2.27× higher efficiency, a 5.5-16.4 K rise in peak produced temperature, and a 2.25-3.6 MW increase in maximum net heat power. These results demonstrate that the FCLG concept and accompanying models can substantially enhance geothermal recovery from abandoned reservoirs, offering a robust framework for high-efficiency energy applications.

Abstract

Data centers, as rapidly growing contributors to
global electricity demand, play a critical role in deep
decarbonization. Meanwhile, regions rich in renewable
energy, exemplified by Qinghai province, China, face high
curtailment rates, with wind and solar curtailment
exceeding 7.3% and 8.9% respectively in 2022. This paper
proposes a strategic solution: integrating data centers
into industrial parks rich in renewable energy to form a
localized green energy consumption ecosystem. This
approach will not only increase the utilization rate of
renewable energy, but also open up new ways for
attracting high-quality industrial investment to
underdeveloped regions with abundant renewable
energy, like Qinghai. The core difficulty lies in designing
mechanisms to incentivize data center investments,
while capturing the nonlinear feedback between policy
choices and market adoption. To investigate the strategic
interactions among key stakeholders, including the
government, state grid corporation, and data center
investors, this study develops a three-dimensional
evolutionary game model that includes carbon penalties,
renewable energy subsidies, and consumer incentives.
This study sets up a case of building data centers in
Qinghai province for calculation and analysis. The results
show that subsidies for investors significantly accelerate
the adoption of renewable energy by state grid
corporation, while modest upfront support for data
center investment can effectively stimulate enterprise
participation and accelerate green uptake on the
demand side. This study offers policy framework and
institutional design path for regions with high power
curtailment rates, such as Qinghai province, emphasizing
the important role of these regions in China’s national
energy transformation and highlighting their potential to
attract green digital infrastructure investment.

Abstract

This study proposes a modeling framework and power system selection process for the deployment of high-payload drones in timber extraction operations, with the objective of promoting sustainable forestry practices. The introduction of such drones is expected to improve labor efficiency and operational productivity by reducing hazardous and physically demanding tasks, and lowering the labor and cost requirements associated with forest road development. To support optimal power system selection, evaluation criteria were established from the perspectives of economic efficiency, operational feasibility, and environmental impact. Simulations were conducted for four candidate power systems, battery-powered systems (BPS), BPS with off-grid charging, series hybrid systems (SHS), and tethered power supply systems (TPS), across payload capacities of 100 kg, 200 kg, and 300 kg. The results indicate a general trend of increasing overall performance scores with larger payloads. While both the BPS and TPS exhibited favorable performance, the operational advantages of the BPS were found to be limited in remote forest environments due to challenges in securing high-voltage grid access and operational interruptions associated with battery replacement. In contrast, the TPS demonstrated superior scalability and operational stability at higher payloads in all weighting scenarios, benefiting from a higher payload-to-airframe ratio and a continuous power supply. These findings suggest that the TPS offers strong potential for large-scale deployment, particularly in the transport of heavy thinning and clear-cut timber.

Abstract

Recycled concrete aggregates and recycled bricks, collectively known as construction and demolition wastes, are renewable resources. Considering the poor performance of recycled aggregate, geosynthetics are generally used to improve its roadability, and the geosynthetics with different strengths have an impact on the interface of the reinforced soil. The monotonic direct shear test on the interface of geosynthetics and reclaimed aggregate are studied based on large-scale direct shear instrument, and consider the influence of normal stress on the interface shear stiffness and damping ratio. Tests showed that: The peak shear stress at the recycled material / geosynthetic interface increases with the increase of the normal stress. The dilatancy of the unreinforced test is stable in the late shearing period, while the dilatancy of the reinforced test is increasing. The peak strength and residual strength of the interface are obviously improved with the increase of the strength of the glass fiber reinforced material, but the integrity of the soil is reduced to a certain extent, which affects the cohesion.

Abstract

Since 2001, China’s photovoltaic (PV) industry has evolved from a latecomer to the global leader. Grounded in innovation ecosystem theory, this paper examines the development of China’s PV industry through four stages characterized by technology, market, finance, and policy dynamics: Pre-commercialization R&D accumulation stage (prior to 1995), Initial commercialization stage driven by rapid international market expansion (1996-2008), Technology and industrial chain integration stage accelerated by domestic market growth (2009-2017), Intense competition and phase out stage under widely technological diffusion and overcapacity (post-2018). This study reveals that policy tools should strategically coordinate technological, market, and financial elements to guide innovation ecosystem participants, thereby facilitating the evolution of the whole innovation ecosystem in different development stages. This coordinated governance mechanism proves critical for achieving leapfrog development in emerging industries.