Xingwang Zhang

Deep reinforcement learning (DRL) has strong potential for real-time demand response (DR) in industrial energy management, but its practical deployment is often limited by slow and unstable training. To address this challenge, this work presents a constraint-oriented, reward-enhanced (CORE) framework and instantiates it with deep deterministic policy gradient (CORE-DDPG). The proposed approach enforces production constraints and reduces reliance on manual reward engineering by introducing a dual-level expert-guided reward structure derived from a quadratic programming (QP) model. In a real-world case study, CORE-DDPG consistently outperformed representative DRL baselines and QP-based benchmarks. Compared with the without-DR baseline, CORE-DDPG achieved an average daily saving of $599.12 (2.70%), outperforming the day-ahead QP's $476.35 (2.02%) and the rolling QP's $505.37 (2.16%) under real prices. These improvements were statistically significant as confirmed by one-tailed paired t-tests (p = 0.035 and p = 0.048, respectively). CORE-DDPG also maintained robust performance against intra-day temporal sequence perturbations in electricity prices. Overall, the proposed approach provides a state-driven and economical solution for real-time control in continuous industrial processes.

Methane dry reforming represents a promising approach to reduce greenhouse gas emissions and to utilize natural gas for value-added products. However, this process is challenged by limited reactivity and severe coking at low temperatures. Here, we report a dual-atom ruthenium catalyst anchored in a boron nitride nanolayer. The spatially distinct coordination of the Ru atomic pair gives rise to the asymmetric charge distribution of the active center. The synergy effect of Ru dimer achieves the CH₄ and CO₂ conversions of 69% and 76% at 650 °C, respectively, which are near thermodynamic equilibrium, while maintaining an exceptional stability of 1000 h at 600 °C with negligible coking. Mechanism studies reveal that an electron-deficient Ru site facilitates the C-H cleavage in CH₄ via an enhanced polarization effect, while another electron-rich Ru site promotes CO₂ activation with a bidentate adsorption configuration. Furthermore, this unique dual-atom Ru structure enables a Langmuir-Hinshelwood pathway with a reduced barrier of the coupling reaction between CH₄ and CO₂ to form *CH₂O intermediate and produce syngas rather than excessive C-C binding and coke deposition. This work provides a strategy for designing multifunctional catalysts to solve the seesaw issue of reactivity and stability for methane dry reforming.

Accurate battery lifetime prediction is critical for operational safety and battery management. However, limited data and poor model transferability make this task challenging. This paper proposes the transferable generative pretrained transformer using low-rank adaptation (TransLoRA-GPT) to address the challenges. Specifically, depending on whether the source and target domains are the same, a base model and a transferred model are fine-tuned from the large language model (LLM), respectively. In the evaluation phase, by applying feature engineering from the charging data of the first 20 cycles, the model can predict the state of health (SOH) aging trajectory for the subsequent 100 cycles. The results indicate that by fine-tuning only 3.33% of the total parameters, the model outperforms traditional models with an average root-mean-square error (RMSE) of 0.0128. Moreover, it maintains robustness against data scarcity and delivers reliable predictions across the entire battery life cycle. Furthermore, it demonstrates strong transferability across different charging/discharging strategies (average RMSE of 0.0171) and battery types (average RMSE of 0.0594). Overall, this work highlights the potential of LLM-based transferable predictive models.

The key challenge in the development of a target property polymer stems from the unclear structure–property relationship (SPR). A multimodal graph neural network (GNN) framework, GeoALBEF, is proposed to address this problem. GeoALBEF introduces a training process based on an “Align before Fuse” architecture that optimizes a three-stage loss, which deeply fuses polymer graph information and text information. Benchmark tests constructed based on 5,000 polymer samples showed that GeoALBEF reduced the mean relative RMSE by 8.6% compared to the suboptimal model in the prediction task of 24 key properties in 6 categories. It is especially worth pointing out that the model achieves functional group level interpretability through the attention mechanism and the functional group averaging strategy. The interpretability was mapped using visualization methods. Additionally, we used reinforcement learning to quantify this interpretability and successfully decoupled the SPR of the polymers. This multimodal system, which combines high-precision prediction capability and mechanism analysis, establishes an intelligent mapping paradigm from polymer structure to property and is expected to promote the accelerated optimization of target property polymers.

Fe and Cu co-doping induces abundant oxygen vacancies in MnO ₂ for efficient ozone catalytic oxidation of toluene: firstly, ozone is activated to generate reactive oxygen species (ROS), and secondly, ROS attack toluene for degradation.

Solution absorption is a straightforward and efficient method for ozone treatment, but waste from inactive absorption solutions poses a risk of secondary pollution and raises the operating cost. Therefore, developing a sustainable recycling process for the absorption solution is essential for green ozone removal. In this study, we constructed a novel I^–/IO₃ ^– cycling system induced by electrocatalysis and photoelectrocatalysis to facilitate the reduction of KIO₃ in KI/KOH ozone absorption solution, thereby enabling absorption solution recycling. The stable operation of this system relies on high-performance cathode materials. By adjusting the concentration of oxygen vacancies on TiO₂, we reduced the energy barrier for IO₃ ^– reduction, optimized IO₃ ^– adsorption on the electrode surface, and improved the band gap structure of the electrode material, resulting in a TiO_2–x cathode with good IO₃ ^– reduction reaction (IO₃RR) performance. Notably, this method achieves an ozone removal cost of $3.72 per kilogram, only one-third of the cost associated with conventional catalytic ozone decomposition. This approach provides a promising new direction for green and efficient ozone removal.