Loess slopes are highly susceptible to rainfall-induced erosion because of their high porosity, weak interparticle cementation, and strong water sensitivity. Although enzyme-induced carbonate precipitation (EICP) has emerged as a sustainable soil stabilization technique, its application in loess remains limited by insufficient nucleation sites, non-uniform CaCO₃ precipitation, and limited resistance to continuous water erosion. To address these limitations, this study proposes a biohybrid stabilization strategy by integrating Artemisia sphaerocephala Krasch gum (ASK gum), a natural polysaccharide, with EICP to enhance the water erosion resistance of loess. Laboratory tests were conducted to evaluate mineralization parameters, shear strength, water stability, microstructure, mineralogical composition, and thermal behavior, followed by a five-month field trial under natural rainfall conditions. The results showed that 1 mol/L urea–CaCl₂ cementation solution and 5 g/L ASK gum were selected as the recommended parameters for loess stabilization when both mechanical performance and material efficiency were considered. Compared with untreated and EICP-treated loess, the A-EICP-treated specimens exhibited improved shear strength and water stability. SEM, FTIR, XRD, and TG analyses suggested that ASK gum contributed to the formation of a more continuous polysaccharide–CaCO₃ network, which filled pores, strengthened interparticle bonding, and improved structural integrity. Field results further demonstrated that A-EICP treatment effectively mitigated rainfall-induced erosion, reducing cumulative erosion mass by 94.28% and decreasing erosion ditch width and length by up to 83.33% and 63.43%, respectively. These findings indicate that ASK gum-assisted EICP is a promising bio-mediated approach for improving the erosion resistance of loess slopes under the tested conditions.
Crystalline silicon (c-Si) photovoltaic (PV) modules are being retired at an accelerating rate globally. Pyrolysis has emerged as a promising recycling strategy for end-of-life (EoL) PV modules through simultaneous thermal decomposition of the ethylene-vinyl acetate copolymer (EVA) encapsulant and fluorine-containing backsheet (TPT). Hence, research on co-pyrolysis characteristics of EVA and TPT is essential for better development of PV pyrolysis technology. This study systematically investigated the pyrolysis characteristics and synergistic mechanisms of EVA, TPT, and their mixtures at three mass ratios. A multiscale analytical approach integrating TG-FTIR, Py-GC/MS, fixed-bed pyrolysis experiments, and ReaxFF-MD simulations was adopted to directly correlate macroscopic product distributions with microscopic species evolution. Co-pyrolysis exhibited notable synergistic interactions, most prominently at an EVA:TPT mass ratio of 5:5 (ET55), which increased gas product yield by 4.45% while reducing oil yield by 4.06%. The oil composition exhibited a selective improvement in quality, with alkene yield increasing by 4.64% and acidic compound yield decreasing by 4.75%. Results of ReaxFF-MD simulations revealed that H radicals generated during EVA decomposition selectively intervened in TPT pyrolysis pathways, promoting alkene formation and CO₂ release while suppressing carboxylic acid intermediate accumulation and inhibiting acidic product generation. This work will provide practical strategies for better utilization of the organic encapsulants and optimizing the pyrolysis process of EoL PV modules.
Multinational disaster response requires rapid decisions despite fragmented data, changing hazards, and distributed authority. This study proposes a process-safety framework that connects predictive key performance indicator monitoring, threshold-based warning, and adaptive inter-agency risk sharing under explicit human accountability. A structured review of 68 studies informed the architecture, model-selection criteria, safety-barrier mapping, and governance controls. Functional feasibility was examined using a reproducible synthetic flood-response simulation with chronological training, validation, and held-out test partitions. Gradient-boosted trees achieved a test mean absolute error of 1.19 h and R² of 0.85 for medical-supply delay prediction; split-conformal intervals achieved 89.4% coverage at a nominal 90% level. At a 17 h warning threshold, sensitivity was 91.1% and the false-alarm rate was 4.8%. The simulation also demonstrated constrained responsibility reallocation without displacing statutory accountability. These results support proof-of-concept feasibility rather than field effectiveness, which requires validation in multinational exercises.
This study develops an intelligent detection method for abnormal flue gas temperature difference at the Superheater cross-section of a 600MW tangentially-fired pulverized coal boiler based on on-site DCS operational data. Bilateral Superheater flue gas temperature difference is used as the anomaly criterion, and high/low-risk sample labels are divided by the 80th percentile after missing data imputation, outlier removal and feature standardization. A working-condition-aware heterogeneous ensemble model (CA-HISF) based on weighted soft voting is proposed for abnormal condition detection. The model reaches 94.37% precision, 92.96% recall, 93.27% F1-score and merely 2.84% missing alarm rate on the test set with sequential data split, exceeding twelve competing machine learning algorithms in ROC performance, Brier probabilistic calibration, DCA net benefit and risk stratification. Statistical significance is confirmed by DeLong and McNemar tests. SHAP-GAM and Cox-KM survival analyses are employed for model interpretability. Quantitative SHAP-GAM results indicate furnace oxygen content, main steam temperature and SOFA-L1 are the primary influential parameters with contribution rates of 35.3%, 13.6% and 9.8%, respectively, and Cox-KM analysis confirms the model-derived risk score is an independent risk factor for excessive Superheater temperature difference. Seven CFD cases configured with core parameters verify the above statistical conclusions from the perspective of in-furnace temperature and flow field evolution. The CFD model is validated against field-measured data with an average relative error below 6%. Simulation results prove the excess air coefficient-related oxygen condition dominates Superheater temperature variation, while SOFA L1 opening is another vital influencing factor within the whole SOFA air distribution system.
Data-driven models in water treatment effluent prediction often pursue autocorrelation-driven pseudo-high performance, neglecting non-ideal reactor hydraulics. This causes severe temporal misalignment between influent and effluent data, undermining the physical interpretability of the model. To address this, a Mechanism-Guided Prediction (MGP) framework is proposed to predict effluent residual chlorine in potable water treatment. Using the Tank-in-Series residence time distribution, clearwell influent features are accurately aligned with the effluent by inherently accounting for backmixing and short-circuiting. These aligned features are coupled with first-order chlorine decay kinetics to establish a mechanistic prediction baseline. Subsequently, these mechanistic predictions and aligned influent data are input into an XGBoost-based MGP model, deliberately discarding lagged autoregressive effluent residual chlorine features. Validated on nine months of data from a full-scale plant (effluent standard: 0.4mg/L), the MGP model demonstrated a root mean square error of 0.017mg/L and an R2 of 0.77. It substantially outperformed pure mechanistic and machine learning baselines, which exhibited low R² values. Furthermore, SHAP analysis confirmed that excluding autoregressive features effectively avoids pseudo-high performance and restores physical interpretability. The MGP method seamlessly integrates mechanistic interpretability with machine learning predictive accuracy, providing reliable support for precise dosing control and environmentally safe disinfection.
