With the widespread application of dimethyl carbonate (DMC) in lithium-ion battery electrolytes, the efficient separation of a DMC/methanol/water mixture has become a critical challenge. Addressing the issues of high volatility, low mass transfer efficiency, and impurities caused by traditional entrainers, this work proposes a novel mixed solvent with synergistic effects in thermodynamics and process economics, consisting of 1-butyl-3-methylimidazolium bis(trifluoromethylsulfonyl)imide ([BMIM][NTf2]) and propylene carbonate (PC). A comprehensive method is proposed for evaluating mixed solvents from molecular to life cycle perspectives. Quantum chemical calculations assess the interaction energy at the molecular level, revealing enhanced separation mechanisms. Synergistic effects are analyzed at the phase equilibrium level. Process optimization for extractive distillation is achieved using a multiparticle parallel swarm optimization method. Finally, the environmental impact across the solvent production, usage, and recovery stages is evaluated throughout the entire life cycle. The extractive processes using single [BMIM][NTf2], PC, and their optimal blend are compared in terms of mass transfer efficiency, energy consumption, economic performance, and life cycle environmental impact. The results indicate that the separation of the DMC/methanol is governed by both electrostatic interactions and van der Waals forces between [BMIM][NTf2] and DMC, with PC playing a synergistic role. The mixed solvent exhibits significantly higher interaction energies with DMC compared with either solvent alone, enhancing the selectivity of DMC to methanol. The direct extractive distillation process using a mixed solvent of 69% [BMIM][NTf2] and 31% PC consistently achieves the lowest total annual cost (TAC) and environmental impacts, with TAC and global warming potential savings 10.06-10.08%, and 9.91-17.33% compared to using either [BMIM][NTf2] or PC alone. The mass transfer coefficient increases by 2.04% compared with single [BMIM][NTf2].

Global industrial sludge production continues to increase annually. Currently, the primary method for sludge management is incineration. This study proposes a novel approach for the incineration of sludge in a rotary kiln. The present study analyzed the promotional or inhibitive effects on CO2 and NOx formation and established detailed combustion mechanism models for pulverized coal and dry sludge using sensitivity analysis. Subsequently, this mechanism model was employed to simulate combustion characteristics and pollutant concentration distributions within the rotary kiln under co-combustion conditions. The results indicate that reactions R14 and R46 exert the greatest influence on pollutant emissions. During the co-firing process, the flue gas temperature within the kiln exceeds 1400°C, making it highly suitable for cement production. The nitrogen oxide content in the exhaust gas is 37.64 mg/m3, meeting the ultra-low emission standard for rotary kilns (100 mg/m3). This study provides a theoretical basis for the application of dried sludge in cement rotary kilns.

Plant-scale industrial carbon accounting is critical for developing targeted emission-reduction policies. However, most assessments of carbon-intensive sectors rely on aggregate statistics, which obscure significant heterogeneity among individual plants. China's pulp and paper industry (PPI), the largest globally, encompasses diverse production processes, raw material inputs, and emission sources. Existing accounting frameworks rely on statistical data and average emission factors within poorly defined system boundaries, which prevents differentiation at the individual plant level. Here, we propose a multimodal data fusion framework that integrates high-resolution remote-sensing imagery with plant textual data to capture structural and operational characteristics undetectable by any single data modality. Applied to 720 pulping and papermaking plants across China, the framework achieves R2 values of up to 0.96 across five plant types and estimates total sectoral carbon emissions at 163.6 million tonnes of CO2 in 2022, with pronounced regional disparities concentrated in eastern coastal provinces. Analysis of functional-zone contributions further reveals that wastewater treatment areas are a consistent cross-category emission driver, and that just 5% of high-emission plants account for approximately 43% of sectoral emissions—a skewed structure that demands differentiated regulatory intervention. Incorporating regional solar radiation data, rooftop photovoltaic deployment is projected to reduce annual PPI emissions by up to 10.3%, with primary-fiber pulp plants offering the greatest mitigation leverage. Beyond China's PPI, this scalable, data-driven approach provides a transferable blueprint for granular, plant-level carbon accounting in other heterogeneous heavy industries.

Against rapid global urbanization and industrialization, WWTPs face pressures such as meeting discharge standards and controlling costs. Accurate influent water quality prediction is vital for reducing energy consumption, and ensuring stable effluent compliance. Although many prediction methods have been proposed, the water quality prediction task still faces challenges from the data side. In practical scenarios, factors such as sensor drift, equipment maintenance, and data transmission interruptions often lead to sparse monitoring records. In this study, low-frequency conditions were simulated by downsampling high-frequency data to approximate scenarios where infrastructure constraints prevent dense data collection. Since low-frequency data loses substantial temporal information, the model's prediction accuracy with such data is limited. This study proposes a temporal-feature enhancement model (TAM-BiLSTM) to address the limitations of low-frequency data on predictive accuracy. The model predicts the COD load at the next time step given a sliding window of historical observations. It employs a deep coupling of an enhanced temporal attention mechanism with BiLSTM. By introducing a time-distance decaying bias, it automatically re-weights time steps to reduce the impact of less informative steps while prioritizing those most relevant to the prediction target. By conducting comparative experiments simulating different data collection intervals, results indicate that when both temporal resolution and training sample size decrease simultaneously, BiLSTM's R2 falls to 0.2131, while TAM-BiLSTM maintains 0.8134, on the held-out test set. It should be noted that low-frequency conditions in this study were simulated by systematically downsampling the original 1-min series rather than by native collection at coarser intervals.

Paper drying consumes >60% of total papermaking energy and remains a major bottleneck to low-carbon production. However, steam-system optimization in full-scale paper mills is hindered by fluctuating operating conditions, nonlinear process coupling, and limited model interpretability. This study develops an explainable plant-scale data-driven framework that integrates operating-regime partitioning, MLP-enhanced T-PLS prediction, four-level energy-efficiency state identification, and SHAP-based root-cause diagnosis. Using real production data from a full-scale paper mill, the drying process was divided into five operating regimes. Across these regimes, the proposed model reduced the prediction RMSE by 70.27% relative to the T-PLS baseline, while the AdaBoost classifier achieved a mean accuracy of 84.14% in identifying energy-efficiency states. SHAP-informed parameter reconfiguration reduced specific steam consumption by an average of 1.37% across four validated regimes (up to 2.90% per regime). Scenario analysis suggests that implementation in technically comparable large-scale Chinese paper mills could save approximately 1.71 million metric tons of steam annually. By extending conventional soft-sensing from prediction to interpretable, mode-specific decision support, the framework provides a scalable pathway for low-carbon optimization in papermaking and other heat-intensive continuous processes.

The separation of 4N-grade isopropanol (IPA) from the mixture obtained from direct propylene hydration is of considerable importance, particularly for its applications in the semiconductor industry. In this work, we propose an integrated method combining multilevel screening of ionic liquids (ILs) with parallel computing toward the optimal design of a 4N-grade IPA separation process. Thermodynamic screening is performed to select several ILs with good separation abilities, followed by further screening for feasible ILs based on their thermal stability. Subsequently, a Python-based parallel optimization algorithm is proposed for the accelerated optimization of flash temperature and total annual cost (TAC). The stability and efficiency of parallel computing are systematically analyzed. A comparative study is carried out against the conventional ethylene glycol (EG)-based extractive distillation process. The results indicate that although [EMIM][Ac] and [EMIM][EtSO4] can also break the azeotropes of IPA/DIPE and IPA/water, they undergo thermal decomposition during the recovery process. Therefore, the only effective candidate ILs remaining are [EMIM][DEP], [EMIM][DMP], and [EMIM][SCN], with the [EMIM][DEP]-based extractive distillation process demonstrating the best performance, as its TAC is reduced by 31.19% relative to the EG-based process. The proposed parallel optimization algorithm significantly reduces computing time while ensuring the stability of optimization results, achieving a relative deviation in TAC of less than 0.8% and an 85.9% reduction in computing time.

Frequent changes in process conditions and production materials make the papermaking wastewater treatment process (PWTP) inherently nonlinear and dynamically uncertain. Meanwhile, the industry is under increasing pressure to achieve coordinated pollution control and carbon reduction. Therefore, ensuring compliance with wastewater discharge standards while simultaneously reducing operational costs, energy consumption, and greenhouse gas (GHG) emissions remains a critical challenge. To this end, this study proposes a multi-objective optimization approach that integrates Kriging and High-Dimensional Model Representation (HDMR) with multi-agent deep reinforcement learning (MADRL) for the papermaking wastewater treatment process (PWTP). In this work, the biochemical and sedimentation processes were modeled using the Benchmark Simulation Model No. 1 (BSM1). A Kriging-HDMR-based surrogate model was developed to estimate the GHG emissions of the process in real time by integrating biochemical mechanisms and data-driven models. This surrogate model was embedded within a reinforcement learning framework to construct a multi-agent "exploring-observing-employing" dynamic optimization system. The MADRL strategy enables collaborative multi-objective optimization for both pollution reduction and carbon mitigation. Simulation results demonstrate that, compared to the BSM1 benchmark control, the proposed policy achieves a 3.52% reduction in operational costs, a 26.38% reduction in energy consumption, and a 7.9% reduction in GHG emissions, while maintaining compliance with effluent quality standards and achieving robust performance.

Due to the growing market demand for product differentia, the household paper production has to adapt to be more flexible in recent years, while it is challenged by uncertainty and complexity in dynamic scheduling. The objectives of minimizing maximum completion time and energy consumption are equivalent to the household paper firms, while traditional tools relied on expert knowledge and human intervention with multiple objectives in this issue. This paper mathematically identified and summarized the household paper workshop scheduling based on its process characteristics of high requirement on flexibility and energy efficiency, and innovatively proposed a deep reinforcement learning-based optimization system via formulating the problem into a Markov game. The optimal solutions are attained through interactions between multiple agents, representing different objectives, in the production scheduling environment. To validate the proposed approach, case studies were conducted using industrial data derived from a household paper company. The results demonstrate the superiority of the proposed method in multiple aspects. Future study could explore deeper to strengthen the model’s generalizability, scalability, and handling uncertainty to validate its applicability in real-world environments.

The papermaking process is typically energy-intensive, with high carbon emissions, facing significant pressure for energy conservation and emission reduction. The paper drying process is the most energy-consuming stage in the papermaking process, and improving its energy efficiency is crucial for achieving energy-saving and emission-reduction effects in the papermaking industry. To address these challenges, four machine learning algorithms-Random Forest, Support Vector Machine, CatBoost, and Stacking-are employed to tackle the regression problem of predicting steam flow and the classification problem of multilevel energy efficiency states in the drying section. Additionally, the SHAP method is utilized to enhance the interpretability of machine learning models. The results demonstrate that machine learning models achieve an excellent predictive performance. Specifically, the CatBoost algorithm establishes robust accuracy in both steam flow prediction (R 2 = 0.874) and energy efficiency classification (accuracy = 0.9646). Meanwhile, the Stacking algorithm also shows superior performance in both steam flow prediction (R 2 = 0.873) and energy efficiency classification (accuracy = 0.9638). More importantly, through SHAP-based feature optimization on low energy efficiency samples, significant energy savings are achieved: every sample achieved an energy consumption reduction of at least 2.4%, with 39% of the samples showing energy consumption reductions exceeding 3%, with an average energy consumption reduction of 2.93% across all samples. SHAP analysis further identifies key operational parameters including winding car speed, end section inlet air flow, front section inlet air flow, basic weight, and steam pressure as critical factors for energy optimization. This data-driven, interpretable machine learning approach not only effectively predicts energy efficiency in paper drying processes but also provides substantial energy-saving potential, offering valuable insights for paper companies to optimize operations and advance the industry's green transformation.

As environmental protection standards become increasingly stringent, wastewater treatment plants (WWTPs) must precisely control aeration volumes and chemical additions to achieve improved effluent quality. The accurate and rapid prediction of water pollutant loads is becoming increasingly urgent. An efficient multi-input, multi-output water-quality prediction model has become a practical necessity for WWTPs. However, the multi-input multi-output model needs to capture the complex interactions between input and output variables, as well as the hidden temporal characteristics of the water quality sequence. It often requires a complex model design and a large amount of training data to achieve good prediction results. The complex model design and extensive data training entail high time and computational resource costs, which will limit the model's applicability. Based on this, this study proposes an active deep learning framework. This framework first queries high-value samples in the data using an active learning module, and then learns the hidden, complex relationships within them using a multi-module fusion deep learning architecture. While ensuring the accuracy of the model's predictions, it significantly reduces the cost of training and the model's computational resource usage. This study uses a municipal WWTP as a case study to predict influent COD and NH3-N loads. The results show that, compared with the traditional passive deep learning model, the active deep learning framework proposed in this study can achieve a prediction effect similar to that of passive learning while reducing the model's time cost by 39.8% and the model's computational resource usage by 18.4%.

The development of new materials is a time-consuming and resource-intensive process. Deep learning has emerged as a promising approach to accelerate this process. However, accurately predicting crystal structures using deep learning remains a significant challenge due to the complex, high-dimensional nature of atomic interactions and the scarcity of comprehensive training data that captures the full diversity of possible crystal configurations. This work developed a neural network model based on a data set comprising thousands of crystallographic information files from existing crystal structure databases. The model incorporates a self-attention mechanism to enhance prediction accuracy by learning and extracting both local and global features of three-dimensional structures, treating the atoms in each crystal as point sets. This approach enables effective semantic segmentation and accurate unit cell prediction. Experimental results demonstrate that for unit cells containing up to 500 atoms, the model achieves a structure prediction accuracy of 89.78%.

As a technology and knowledge-intensive industry, the process industry, central to sustainable manufacturing goals, faces challenges with large volumes of dispersed data, high integration of production units, and complex workflows. Existing methods struggle to analyze unstructured mechanism and experience knowledge, leading to information silos. To support cleaner production through enhanced fault diagnosis and prevention, this study leverages knowledge graph theory. An improved Hidden Markov Model for industrial text segmentation is proposed, demonstrating a 3.2 % accuracy increase over general tools. By utilizing this method to effectively process unstructured data and extract valuable knowledge, a dedicated fault knowledge graph framework and ontology model for process industries is constructed. This knowledge graph is then integrated with machine learning algorithms to build an industrial status diagnosis model; crucially, it enables intelligent feature selection, bypassing complex dimensionality reduction tasks common in previous approaches. Through a case study on tissue paper break faults, the framework is demonstrated by establishing a paper break fault knowledge graph and diagnosis model. This approach provides causal reasoning for proactive interventions that reduce scrap rates and optimize resource utilization, key drivers for improving eco-efficiency and advancing green, sustainable operations within the process industries.

Pulp and Paper Industry (PPI) is one of the key sectors emitting tremendous carbon emission and is widely distributed in the Belt and Road Initiative countries. This research analyzes the embodied carbon emissions (ECE) in this area with influence mechanisms. A system dynamics model was built to predict the effects of various scenarios on ECE by adjusting factors such as the economic growth rate of the PPI and carbon trading policies. The results indicate that the per capita output effect and indirect carbon emission intensity foremostly drive ECE in China’s PPI, and it is recommended to be collaborated with sectors involved in the production and supply of electricity, heat, gas, and water to develop policies. In scenario-based analysis, economic growth and carbon trading policies show varied positive effects in this regard, and the target can be achieved in the designed scenario of reducing carbon intensity to below 65 % of the 2005 level. The findings provide insights into the carbon reduction for PPI and similar industries in countries along the Belt and Road Initiative.

Industrial wastewater treatment (WWTP) systems are characterized by nonlinearity, multivariate nature, and strong coupling, posing significant challenges to optimization and control for pollution and carbon emission reduction. This study aims to reduce Greenhouse gas (GHG) emissions, total operational costs, and pollutant discharges through multi-objective optimization and control. Given the nonlinearity and time-varying nature of industrial WWTP systems, a multi-objective intelligent optimization algorithm is employed to optimize the setpoints of key variables. To address the difficulty of control systems in tracking dynamic optimized setpoints, this study proposes two strategies: Back Propagation Neural Network-Proportional Integral Differential (BP-PID) control and Nonlinear Model Predictive Control (NMPC). Both strategies can effectively achieve the goals of pollution reduction, carbon emission mitigation, and efficiency enhancement in WWTP. Compared to the open-loop scenario, NMPC reduces GHG emissions by 22.1%, total operational costs by 28.5%, and pollutant discharges by 8.4%, while ensuring that the treated effluent meets pollutant standards.

Separating hemicelluloses from black liquor and producing high-value products have been proposed as a means to utilize residues, reduce carbon emissions, and generate additional revenues for pulp mills. However, the combustion of black liquor is essential to provide the mills with low-cost energies without high input of fossil fuels, so whether the utilization of residual hemicelluloses would actually reduce the overall carbon emissions remains doubtful. In this study, comprehensive energy and environmental impact analyses of the hardwood kraft pulping process incorporating membrane extraction of hemicelluloses and their high-value utilizations were conducted by using life cycle assessment (LCA). A process simulation method was used to study the black liquor treatment and membrane separation process. The LCA indicates that incorporating hemicellulose extraction and utilization leads to a 9.97% reduction in overall greenhouse gas emissions without alternation of pulp production, considering the increase of fossil fuel and electricity usage. Furthermore, hemicellulose extraction and utilization can also enhance the pulp production by approximately 15%, because the recovery boiler can support more production load due to the hemicellulose separation from black liquor. Sensitivity analyses indicate that coal carbon factor, the use of calcium oxide, and high-value utilization of hemicellulose have a significant impact on the Global Warming Potential associated with kraft pulp production activities. These findings indicate the potential of commercial hemicellulose extraction to mitigate the environmental impact of kraft pulp mills.