中国石油大学（北京）

Thermodynamic promoters can shift hydrate equilibrium to lower pressures and higher temperatures, but their application is often limited by dissolution, dilution, and hydrate cage occupation, which reduce gas storage capacity. In this work, a small amount of tetrabutylammonium bromide (TBAB) was locally attached to a capillary-active metal mesh to avoid direct mixing with the bulk liquid phase. This strategy alleviated mass transfer blockage caused by hydrate film formation at the gas-liquid interface in static systems, reduced the adverse effect of promoter cage occupation on storage capacity, and enhanced heat transfer. A 300 ppm tryptophan solution showed the best hydrate formation performance at 2 degrees C and 3.5 MPa for a natural gas mixture containing 90 mol% CH4, 7 mol% C2H6, and 3 mol% C3H8. The optimal mesh-to-solution height ratio was 2.6:1, and linear hydrates formed along the mesh facilitated upward water migration. When 0.04 g of 25 wt% TBAB solution was attached to the mesh edge 1 cm above the liquid surface, upward migration of most of the water in the 12 g solution was achieved within 10 min, and 90% of the maximum gas storage density was reached within 1 h. This work provides an effective route for rapid natural gas hydrate formation in static systems with minimal thermodynamic promoter usage.

Refinery process pipelines operate for long periods under high temperature and pressure, complex media, and coupled operating fluctuations, leading to blurred health-state boundaries, nonlinear degradation, and progressive uncertainty accumulation during long-horizon prediction. Conventional evaluation methods based on single thresholds or empirical rules often fail to simultaneously ensure state-discrimination consistency, model robustness, and forward degradation predictability. To address these issues, this paper proposes an integrated framework for corrosion health-state assessment and degradation-process modeling. First, a Gated Robust Autoencoder with Cumulative-ordinal Enhancement Multilayer Perceptron (GRACE-MLP) is developed to output four-level health-state probabilities and aggregate them into a continuous Health Index (HI), providing smooth, interpretable, and boundary-stable observations for subsequent degradation modeling. Then, a Multimodal-observation Hidden Semi-Markov Model (Multi-HSMM) is constructed to characterize the stage persistence of corrosion degradation through explicit sojourn times, while continuously correcting state posterior distributions through reliability-weighted multimodal observation assimilation, thereby suppressing uncertainty accumulation during long-horizon prediction. The proposed framework is validated in two representative scenarios: an alkylation refining unit and a crude distillation unit/ atmospheric–vacuum distillation unit (CDU/AVDU) overhead system. The results show that, in the health-state assessment stage, GRACE-MLP achieves Accuracy (Acc), Precision (P), Recall (R), and F1-score (F1) values of 93.93%, 94.64%, 93.93%, and 93.87%, respectively. In the uncertainty-suppression stage of degradation prediction, multimodal observations outperform single-modality observations, with the Uncertainty Reduction Ratio (URR) and Width Reduction Ratio (WRR) reaching 56.8% and 62.2%, respectively, accompanied by simultaneous improvements in the Information Gain Ratio (IGR = 80.5%) and Evidence Consistency Score (ECS = 79.5%). These results demonstrate that GRACE-MLP improves the robustness of corrosion health-state assessment, while Multi-HSMM reduces posterior state entropy through multimodal observations and effectively suppresses uncertainty diffusion in long-horizon degradation prediction.

Saline lacustrine shales commonly retain NaCl-bearing pore water that may influence kerogen conversion. However, the hydrocarbon generation behavior of saline lacustrine kerogen under saline aqueous conditions remains poorly constrained by existing studies. This study aims to quantify the hydrocarbon generation evolution of saline lacustrine kerogen and evaluate the role of NaCl-bearing pore water. Type-II kerogen isolated from immature Miocene saline lacustrine shales of the Qaidam Basin was subjected to confined gold-tube pyrolysis in distilled water and NaCl solution at two heating rates. The results demonstrate that liquid hydrocarbons are generated at lower maturity levels, with high yields of heavy hydrocarbons during the low-mature stage. The main generation window for liquid hydrocarbons precedes that of freshwater lacustrine and marine shales, peaking at an Easy%Ro of 0.8%. Compared with distilled water, the NaCl-bearing system advances the bitumen generation peak to lower maturity and enhances early heavy-product formation. Wet gas and light oil are progressively produced during the mature stage via the continuous thermal cracking of the high-molecular-weight bitumen generated earlier. Primary cracking of kerogen/bitumen during the mature stage and secondary cracking of light oil at high maturity account for the increase in gas yields. Overall, this study provides experimentally constrained hydrocarbon generation models under NaCl-bearing pore water, helping to clarify the origin of low-maturity oils in saline lacustrine basins.

Aqueous-phase hydrogenolysis of lignocellulosic polyols offers a sustainable route toward industrially relevant commodity chemicals. However, complicated series and parallel reactions often induce strong intermediate inhibition on solid catalyst surface. We reported a simple aqueous-phase H2 activation strategy to synthesize carbon-supported Ru catalysts, which alleviates the inhibition caused by C4/C5 intermediates during sorbitol conversion, and significantly enhances catalytic activity, selectivity and durability. The catalysts post-treated in water under H2 display defect-enriched carbon edges that modulate the electronic structure of Ru nanoclusters. It is further shown that these structural modifications increase H2 adsorption and spillover, facilitating rapid hydrogenolysis of glucose and sorbitol into glycols as the key products. The proposed Ru catalyst achieved 72.8% sorbitol conversion and 41.4% yield of target glycols with a record high activity of 441.0 h−1 at 200 °C. This simple approach demonstrated in this work provides an alternative catalyst preparation technique for aqueous-phase biomass valorization.

Multimodal emotion recognition plays a crucial role in human-computer interaction and sentiment analysis. However, existing methods for multimodal emotion recognition face challenges in handling Chinese data scarcity, insufficient utilization of acoustic-textual features, and inadequate dynamic cross-modal interaction. To address these limitations, we propose DA-ECAN, a DualAudio-EnhancedCNN Cross-Attention Network, which integrates dual-modality feature enhancement and adaptive fusion. First, we construct CH-SIMS v2.0-DA, an augmented dataset leveraging large language models (LLMs) for text diversification and controllable text-to-speech (TTS) synthesis for multi-style speech generation. For feature extraction, the audio branch employs a dual-stream architecture: wav2vec 2.0 captures phoneme-level semantic dependencies from raw waveforms, while MHSA-BiLSTM analyzes MFCC-based spectral-temporal patterns to model prosodic details (e.g., pitch, rhythm). The text branch adopts LLM-enhanced CNN, where Qwen embeddings provide context-aware representations, and multi-scale convolutional kernels extract phrase-level and syntactic patterns. Crucially, a Transformer-based asymmetric fusion module dynamically aligns audio and text features by using text as queries to guide attention over acoustic keys and values, balancing modality contributions and mitigating text dominance. Experimental results on MC-EIU, CH-SIMS v2.0 and its enhanced dataset CH-SIMS v2.0-DA demonstrate that DA-ECAN outperforms existing baseline models across all metrics, validating its effectiveness in emotion recognition tasks.