With the rapid development of electric vehicles and energy storage systems, efficient thermal management of lithium-ion batteries has become crucial. This study presents a systematic experimental investigation on a refrigerant-based direct-cooling battery thermal management system, focusing on the interdependence between control parameters and performance indicators. A comprehensive experimental platform was established to examine the effects of compressor frequency, expansion valve opening, and condenser fan speed on multiple performance metrics including energy efficiency, cooling capacity, temperature distribution, and system stability time consumption. Through multi-dimensional sensitivity analysis, the complex relationships between control variables and system indicators were revealed, providing guidance for optimizing control strategies. Under optimal operating conditions, the system achieved a calculated COP (COPcal)of 8.4, a cooling capacity of 4.04 kW, a maximum temperature difference of 4.0 degrees C, and reaching stability within 220 s. The findings contribute to a deeper understanding of parameter interdependencies and to the development of efficient control schemes for direct-cooling battery thermal management systems.

The reliable and efficient operation of lithium-ion batteries is crucial for electric vehicles. Temperature nonuniformity within a battery pack can affect its capacity, lifespan, and safe operation. In current battery thermal management systems, the temperature distribution typically exhibits the characteristic of lower temperature on the inlet side and higher temperature on the outlet side. For this reason, this paper employs a single-phase immersion battery thermal management system for 70 cylindrical 4680 cells and proposes a novel distribution manifold structure. By guiding the coolant to the rear of the batteries, this design enhances the heat exchange efficiency between the coolant and the rear cells, thereby improving the temperature uniformity of the battery pack. The research results show that, compared with the original model, Scheme 2 reduces the maximum temperature to 40.86 degrees C, a reduction of 9.64% and the maximum temperature difference to 6.97 degrees C, a reduction of 59.82% at the cost of a 12.21% increase in pump power. Furthermore, this paper investigates key influencing factors such as the direction of holes and the position of holes through numerical simulation. Under the condition of a coolant inlet flow rate of 0.2 m/s, Case 3 has achieved control of the maximum temperature difference of the battery pack within 5 degrees C at the cost of a 15.7% increase in pump power. Meanwhile, under the condition of 0.25 m/s, Case 3 exhibits the optimal performance. The proposed liquid distribution manifold design not only optimizes the coolant flow path but also significantly enhances the temperature control capability for batteries, providing a practical solution for improving the overall performance and reliability of the battery system.

The Tesla turbine provides a promising and more economical alternative for small-scale Compressed Air Energy Storage (CAES). However, due to significant differences in its structure compared to traditional turbines, it lacks comprehensive performance characterization in terms of thermodynamic performance, which has become an obstacle in the modeling and simulation process of this turbine in different thermal systems. In this study, an experimental rig was first built to characterize the Tesla turbine's performance, obtaining operating data under diverse conditions to serve as a validation benchmark. To overcome experimental limitations and extend the investigation to broader operating ranges, a three-dimensional CFD model was subsequently developed and validated, achieving a power prediction error of less than 4.3%. Building on this, a novel correction factor was derived from experimental data to calibrate the simulation. This calibrated, high-fidelity dataset was then used to train a three-layer feedforward neural network. The resulting surrogate model predicts isentropic efficiency with high accuracy (R2=0.84, MAPE=4.58%) in milliseconds, achieving a speedup of over 107 times compared to the multi-hour ANSYS CFX baseline. The model successfully captured the critical thermo-mechanical coupling, revealing a thermal activation mechanism that enables high-efficiency operation in the high-parameter domain. By bridging the gap between high-fidelity physical characterization and computational efficiency, this work delivers a validated, dynamically responsive predictive tool, laying a robust foundation for system-level optimization and intelligent control of Tesla turbines in future AA-CAES applications.

In low-grade waste heat recovery and small-scale organic Rankine cycle (ORC) systems, the Tesla turbine is a promising expander due to its simplicity and adaptability. However, existing theoretical models often rely on idealized smooth-wall assumptions and simplified velocity profiles, failing to accurately predict turbine performance under realistic manufacturing conditions. This discrepancy leads to design uncertainties, creating significant difficulties for the precise sizing and widespread application of Tesla turbines in ORC systems. To bridge the gap between theoretical idealization and engineering reality, this study develops a universal nine-parameter dimensionless system that integrates working fluid properties, geometric constraints, and operating conditions. The methodology employed in this study accounts for changes in effective viscosity caused by surface roughness, thereby enabling precise characterization of flow evolution. The model’s reliability is validated against data from a small-scale ORC test bench, showing excellent agreement with experimental trends. Furthermore, through multi-parameter sensitivity analysis and sequential quadratic programming (SQP) multi-objective optimization, the optimal dimensionless configuration maximizing isentropic efficiency and power was determined. The optimized efficiency is 62% higher than the reference condition, and the output power is 83% higher. These findings provide a reliable computational tool and direct design guidance for selecting the optimal Tesla turbine geometry under specific heat sources and working fluids.

The application of hydrogen in heavy-duty transportation is significantly hindered by onboard storage challenges. Methylcyclohexane (MCH), as a promising liquid organic hydrogen carrier (LOHC), presents a compelling alternative due to its high safety and compatibility with existing fueling infrastructure. This paper establishes a comprehensive, steady-state thermodynamic model of an onboard integrated MCH-hydrogen system, introducing a novel waste heat recovery strategy that simultaneously utilizes internal combustion engine (ICE) exhaust and cooling water jacket heat. Thermodynamic simulations demonstrate that under baseline operating equilibrium conditions, the MCH conversion rate reaches 99.80%. At this reference point, the ICE achieves a thermal efficiency of 42.2%, contributing to a net system efficiency of 31%. Comprehensive parametric analyses are conducted to investigate the coupled effects of reactor conditions, purification pressures, and the excess air ratio on the system performance. Furthermore, the vehicle's driving range and potential payload penalties are evaluated across various fuel tank configurations. Finally, an economic assessment reveals that the proposed MCH-fueled truck achieves clear cost-competitiveness against traditional diesel heavy-duty trucks when the hydrogen infrastructure price drops below 2.0 USD/kg and diesel prices exceed 3.5 USD/gallon. This study establishes a robust thermodynamic benchmarking and offers critical design insights for the deployment of LOHC-powered heavy-duty transportation.

With the rapid development of multi-agent collaborative decision-making in smart grids, electricity trading systems increasingly rely on large-scale user load data to optimize trading strategies. However, heterogeneous agents exhibit diverse privacy requirements, while existing unified privacy-preserving mechanisms struggle to balance personalized privacy demands with overall system efficiency. To address this challenge, we propose a dynamic differential privacy scheme tailored for multi-agent game environments. Specifically, we first design a personalized privacy risk assessment metric based on weighted mutual information to quantify the privacy leakage risk of diverse users. Then, we formulate the electricity trading problem as a multi-objective optimization that captures the trade-off between utility and privacy, and establish a Markov decision process model to maximize system economic efficiency under privacy-preserving constraints. Finally, we develop an adaptive noise-injection reinforcement learning framework integrated with a denoising network, and theoretically prove that the proposed noise mechanism satisfies ε-differential privacy. Simulation results demonstrate that, compared with benchmark methods using fixed-noise and personalized-noise mechanisms, our approach enhances privacy protection effectiveness by 79.11% and 77.76%, respectively. Moreover, the incorporation of the denoising network eliminates 50.97% of the cost increase induced by noise injection, thereby achieving a coordinated optimization of privacy preservation and operational efficiency.

Accurate monitoring of State of Charge (SOC), State of Health (SOH), and State of Temperature (SOT) is indispensable for ensuring the operational safety and efficiency of Battery Management Systems (BMS) in electric vehicles and large-scale energy storage. However, conventional data-driven approaches often isolate these states, overlooking their intrinsic physical coupling, which inevitably leads to suboptimal estimation accuracy. To address this challenge, this paper introduces a novel integrated co-estimation framework, the Attention Mechanism-Parallel Temporal Convolutional Neural Network (AM-PTCN). By leveraging a parallel feature extraction structure combined with a physically interpretable attention mechanism, the model dynamically identifies and re-weights influential factors within multivariate time-series inputs, effectively disentangling the complex interdependencies among battery states. Furthermore, estimation uncertainty is quantified to provide a probabilistic assessment of system reliability. The optimized model achieves Mean Absolute Errors (MAE) of 1.3323% for SOC, 1.8901% for SOH, and 0.2208 for SOT. Crucially, rigorous ablation studies and comparative experiments validate the specific contributions of the attention-based parallel architecture, demonstrating superior accuracy and robustness over existing machine learning approaches. Validated on noise-contaminated datasets, this novel joint estimation framework provides a valuable reference for advanced battery monitoring.

The direct refrigerant cooling battery thermal management system (DRC-BTMS) has attracted significant interest owing to its superior heat transfer efficiency. However, unlike conventional liquid cooling, direct cooling relies on phase-change heat transfer, resulting in a higher sensitivity of pressure-temperature coupling compared to liquid cooling, which increases the control accuracy requirements and practical regulation difficulty. To address this, this study constructed an experimental platform and conducted targeted investigations to analyze the effects of control parameters including compressor speed and expansion valve opening on system thermodynamic parameters and battery pack temperature distribution. With the expansion valve fixed at 3.5 turns, increasing the compressor speed from 4800 rpm to 6000 rpm maintained the maximum temperature at 45 °C, while the maximum temperature difference decreased from 7 °C to 5 °C. Among the compressor speeds, the maximum coefficient of performance (COP) of 3.46 was achieved at 4800 rpm, with a cooling capacity of 4.57 kW. Increasing the expansion valve opening further enhanced the mass flow rate. When the compressor operated at a constant speed of 4800 rpm, increasing the valve opening from 3 to 4 turns raised the cooling capacity from 4.15 kW to 5.25 kW. At 4 turns, the COP reached a maximum of 3.86. Simultaneously, the maximum battery temperature decreased from 49 °C to stabilize at 45 °C, and the maximum temperature difference decreased significantly from 11 °C to 5 °C. Furthermore, comparative analysis reveals that the system superheat and battery temperature during the charging process were higher than those during the discharging process under identical conditions, indicating greater cooling demands during charging.

As urban transportation systems increasingly focus on environmental impact, ports, as a crucial part of urban transportation, become particularly important in their integration of automation and green energy. Electric Autonomous Guided Vehicles (E-AGVs) can be used as energy storage equipment to assist the consumption of renewable energy and reduce carbon emissions in automated Ports. However, when confronted with the intricate challenges of AGV charge/load coordination, conventional manual scheduling and optimization methods struggle to effectively manage multi-action scenarios, multiple intelligent agents, indeterminate durations, and the unpredictability of external environmental factors. To this end, this paper proposes a Dynamic Multi-task Coordination Approach with Action Filtering Reinforcement Learning to address the charging/loading scheduling problem of AGVs. Subsequently, a Prohibited Action Filtering Algorithm is devised, which leverages internal network improvements to effectively restrict and constrain AGV actions. Based on the Proximal Policy Optimization (PPO) algorithm framework, a multi-objective reward mechanism is set up. Through a multi-objective reward mechanism and iterative updates of the policy and value networks, this approach leads to a significant improvement in the training efficiency of reinforcement learning. Simulation experiments demonstrate that the proposed algorithm outperforms conventional methods and other reinforcement learning algorithms in critical metrics such as transportation time, total electricity consumption, and carbon footprint. Specifically, our algorithm achieves that the transportation time is only 25% of manual scheduling, 39% of DQN and 41% of the SA. Moreover, it reduces carbon footprint to just 64% of the SA and 20% of manual scheduling.

Current energy architectures are ill-suited for decarbonizing high-altitude cold regions, as they lack the capability to sustain a rigid thermal supply amidst the inherent volatility of renewable generation under harsh climatic constraints. A Thermal-Dominated Island Integrated Energy System (TD-IIES) is proposed in this paper. The TD-IIES differs by establishing a Carnot Battery as the central energy hub to decouple heterogeneous energy flows, unlike traditional electricity or hydrogen-dominated architectures. An optimization framework is developed that considers surrogate-assisted multi-objective optimization by integrating Deep Neural Networks (DNN) with the Non-Dominated Sorting Genetic Algorithm-II (NSGA-II) for optimization of thermodynamic parameters and capacities of the equipment while executing Event-Driven Hierarchical Control Strategy (ED-HCS). A case study using real data obtained from Naqu, Tibet, is presented to compare the configurations with and without a Compound Parabolic Concentrator (CPC). The results show that the system provides a maximum Carnot Battery round-trip electrical efficiency of 64.93% and a Heat Pump COP of 4.75. The techno-economic optimum is the No-CPC scheme with a Levelized Cost of Electricity (LCOE) of 0.176 $/kW & sdot;h. The cost is 4.9% lower than the cost when CPC is included. And it also represents a reduction of around 43% compared to the mainstream off-grid design benchmarks. Resilience analysis reveals a "wind-dominated" anisotropy acting on the reliability of systems. The No-CPC scheme has better economic adaptability to the "strong wind/weak sun" scenario, while the scheme with CPC inclusion has better fault tolerance (N-1 redundancy) and can still achieve survival supply when the core heat pump is faulty. This study demonstrates that the TD-IIES framework, by leveraging the thermally coupled Carnot Battery as the central energy hub, achieves cost-competitive and resilient off-grid operation in high-altitude cold regions. The techno-economic advantage of the No-CPC scheme originates from redirecting capital from solar thermal collectors toward wind and solar generation, while the fault-tolerance advantage of the CPC-included scheme derives from its thermally autonomous backup capability.

With the explosive growth of e-commerce, efficient last-mile delivery has emerged as a critical challenge. Truck-drone collaborative delivery has garnered significant attention as a promising solution to this problem. In this work, we formulate the truck-drone collaborative delivery problem as a collaborative optimization problem, aiming to minimize the total completion time for delivering packages to customers by leveraging the complementary strengths of the drone's speed and the truck's endurance. We propose a two-stage framework to address this challenge. In the first stage, the Lin-Kernighan-Helsgaun (LKH) algorithm is used to generate a high-quality initial traveling salesman problem (TSP) solution, serving as a robust starting point. In the second stage, a sequence allocate policy (SAPPO), based on proximal policy optimization (PPO), refines the TSP solution by optimizing the truck-drone collaborative path using a specially designed action space. Extensive experiments conducted on both random dataset and TSPLIB benchmarks demonstrate that our method significantly outperforms existing algorithms regarding delivery time, while exhibiting improved scalability and less training time.

Advanced Adiabatic Compressed Air Energy Storage (AA-CAES) is promising for large-scale storage, but the capital cost of conventional turbomachinery hinders distributed deployment. We propose an AA-CAES system integrating a low-cost bladeless Tesla turbine and a unified techno-economic framework. The framework couples thermodynamic calculations, a condition-dependent feedforward neural network turbine surrogate, and hybrid component cost models. Two-stage particle swarm optimization first screens candidate topologies and then updates turbine efficiency at each operating point. The converged design is evaluated against a conventional-turbine baseline. This coupling links condition-dependent component performance to the scale-dependent economic assessment. The optimized design raises RTE from 27.6% for direct turbine substitution to 39.3% and reduces initial investment by about 49%. At full scale, its levelized cost of storage (LCOS) is 0.208 USD/kWh versus 0.184 USD/kWh for the conventional baseline. Under the modeled costs, duty cycle, financing, and scaling assumptions, the LCOS curves cross near 4500  kW; below this case-specific boundary, the Tesla-turbine system has lower modeled CAPEX and LCOS. These results define a bounded distributed-storage niche rather than a universal commercial threshold.

Battery pack capacity estimation is crucial for improving the performance, safety, and reliability of energy storage systems. However, the inherent cell inconsistency and unobservable internal electrochemical reactions make this task challenging, especially when the battery behavior is only partially observed. To address this challenge, this article proposes a novel framework, UniCap, to estimate the capacity of series battery packs by utilizing partial behavior observed near the lower voltage boundary. Within this framework, the traditional estimation process is reformulated into two tractable subtasks. First, a hybrid K-nearest neighbors method is developed to nondestructively quantify unreleased capacity caused by cell inconsistency. Second, a cross-stream cell-wise transformer is proposed to predict the cell behavior and corresponding capacity at the unknown upper voltage boundary. Based on the above, the lower bound of the pack capacity can be inferred. Without requiring full charge or discharge cycles, the proposed framework provides an efficient solution for battery pack capacity estimation. Validated on a large-scale industrial dataset comprising 1757 battery packs, the framework achieves state-of-the-art performance with a mean absolute error of 0.0968 Ah. Notably, compared with existing industrial workflows, the proposed method significantly reduces test duration with only marginal accuracy degradation, thereby improving automation and production efficiency.

Large-scale electric vehicle charging behavior impacts power balance, load fluctuations, and operational safety in power distribution networks. To mitigate these risks while maintaining grid stability, the constraint satisfaction problem faced by scheduling strategies urgently needs to be addressed. Traditional RL methods indirectly characterize constraints by processing them through reward and penalty functions, hindering the explicit representation of key physical constraints, thereby compromising policy feasibility and stability. To address these issues, this paper proposes Physics-Informed Reinforcement Learning (PIRL) for EV charging scheduling. By formulating the scheduling problem as a Constrained Markov Decision Process (CMDP), PIRL embeds physical information into the policy optimization process via differentiable equations, thereby ensuring the rigorous satisfaction of complex constraints. Simulation results show that PIRL average return increased by 137% while keeping physical loss increased by only 0.6%, demonstrating good generalization and applicability in different datasets.

While the Tesla turbine's structural simplicity and potential low cost make it an attractive expander for Organic Rankine Cycle (ORC) waste heat recovery, current assessments are significantly restricted by the reliance on airbased experimental data and immature economic modeling methodologies that fail to capture real-gas behaviors and the cost dynamics of modular design. To bridge these critical gaps, this paper presents a novel data-driven techno-economic assessment framework grounded in experimental validation using an organic working fluid (R245fa). Unlike traditional analytical approaches, we develop a neural network-based efficiency (eta tur) prediction model directly from experimental data to capture non-linear characteristics and introduce a detailed cost model explicitly linking geometric parameters to manufacturing costs for MW-class applications. Subsequently, thermodynamic analysis, optimization, and economic comparison are performed for systems incorporating both Tesla and traditional expanders. Results indicate the turbine achieves an isentropic efficiency (eta tur) of 59% under optimal conditions, with its manufacturing cost being over 90% lower than that of traditional turbines of an equivalent scale. Economic analysis reveals that for systems exceeding 9 MW, the Levelized Cost of Electricity (LCOE) of the Tesla turbine ORC (TORC) is lower than that of the traditional system (BORC), with an LCOE of 0.0158 $/(kW & sdot;h) at 37 MW, which is approximately 10% lower than that of BORC.

Dynamic systems often encounter disturbances like sensor outliers, which violate the Gaussian noise assumption in traditional Kalman filters (KFs). While maximum correntropy KFs (MCKFs) address this issue by utilizing higher order statistical information, their performance critically depends on the manual selection of a kernel scale parameter. Existing methods with fixed or empirically adjusted kernel scales struggle to handle disturbances of varying intensities, limiting practical applications. This article presents a novel adaptive MCKF framework. The key contributions are as follows: 1) Adaptive Kernel Scale Optimization: The kernel scale is modeled as a probabilistic variable, and variational Bayesian inference is employed to jointly estimate the system state, enabling automatic kernel scale optimization during the recursive process. 2) Theoretical Analysis and Extension: The computational complexity of the proposed algorithm is analyzed in the context of linear systems, and its theoretical connection to traditional correlation entropy filters is established. Furthermore, the method is extended to nonlinear systems. 3) Performance Enhancement: Experimental evaluations on typical single-target tracking task and complex nonlinear scenario demonstrate that the proposed approach outperforms existing methods. Under various types of noise interference, the average position estimation error is reduced by 40%-60%, highlighting its superior adaptability and robustness. In addition, the algorithm is successfully applied to real-world battery state-of-charge (SOC) estimation in Internet of Things (IoT) scenarios, demonstrating its practical value in embedded energy management systems.

Communication encryption is crucial in computer technology, but existing algorithms struggle with balancing cost and security. We propose EncGPT, a multi-agent framework using large language models (LLM). It includes rule, encryption, and decryption agents that generate encryption rules and apply them dynamically. This approach addresses gaps in LLM-based multi-agent systems for communication security. We tested GPT-4o's rule generation and implemented a substitution encryption workflow with homomorphism preservation, achieving an average execution time of 15.99 seconds.

With the exponential increase in the adoption of lithium-ion batteries, reusing and recycling have become critical for extending the lifespan of retired batteries and reducing environmental impact. Recent developments in deep learning provide efficient solutions for the screening and reuse of massive retired batteries, as they can estimate multiple battery states in a short observation period. However, the existing methods ignore the timescale differences between battery states, causing the model to collapse in optimization conflicts. In this paper, we revisit the impact of this conflict and propose a dual-path deep method for fast estimation of the state of both charge (SOC) and health (SOH) in a short observation time of the discharge phase. Specifically, the shared lower layers capture local time-varying features, while the two specialized paths integrate them into global features each focusing on a different timescales. Furthermore, to solve the ensuing optimization conflict, we seek for Pareto-efficient to achieve the optimal estimation of the two states. Exhaustive experiments and analysis on 89 realistic retired batteries and 16 public batteries with different chemistries and working conditions show that our framework can obtain reliable estimation. Using only an observation time of 400 s, the average root mean square error of SOC and SOH estimations is 1.01% and 1.72%, improved by 16% and 33% compared with state-of-the-art methods. Notably, our framework only has a parameter size of 0.0542 MB and can be deployed on most edge devices, which significantly promotes the application of data-driven models in the real world.

Cooperative pursuit using multi-UAV systems presents significant challenges in dynamic task allocation, real-time coordination, and trajectory optimization within complex environments. To address these issues, this paper proposes a reinforcement learning-based task planning framework that employs a distributed Actor–Critic architecture enhanced with bidirectional recurrent neural networks (BRNN). The pursuit–evasion scenario is modeled as a multi-agent Markov decision process, enabling each UAV to make informed decisions based on shared observations and coordinated strategies. A multi-stage reward function and a BRNN-driven communication mechanism are introduced to improve inter-agent collaboration and learning stability. Extensive simulations across various deployment scenarios, including 3-vs-1 and 5-vs-2 configurations, demonstrate that the proposed method achieves a success rate of at least 90% and reduces the average capture time by at least 19% compared to rule-based baselines, confirming its superior effectiveness, robustness, and scalability in cooperative pursuit missions.