Accurate state-of-charge (SOC) estimation is essential for lithium-ion battery management systems, but remains challenging under battery-to-battery variation. This paper proposes a physics-guided lightweight NAS-TCN-Transformer with coulomb-counting residual learning for cross-battery SOC estimation. A five-dimensional feature representation is constructed by integrating voltage, current, temperature, voltage change rate, and cumulative discharged ampere-hour. The model predicts the residual between a coulomb-counting baseline and the target SOC rather than directly regressing SOC. Experiments on the NASA battery dataset show that the proposed method outperforms LSTM, TCN, and NAS-TCN-Transformer baselines in both single-battery and cross-battery settings, while retaining a lightweight architecture with only 18.676K parameters. The results confirm its effectiveness in improving estimation accuracy, generalization, and deployment efficiency.

Accurate State-of-Health (SOH) estimation for lithium-ion batteries is crucial for energy storage system reliability, yet achieving a balance between electrochemical interpretability and temporal degradation modeling remains challenging. This paper introduces a Physics-informed Extended Long Short-Term Memory (PI-xLSTM) framework that integrates sequential modeling with battery degradation physics. The model leverages xLSTM's exponential gating and reversible residual connections to capture long-term dependencies, while a physics-guided loss function incorporates electrochemical relationships-such as those between internal resistance, temperature, and SOH decay. Evaluated on 124 commercial $\mathbf{L i F e P O}_{\mathbf{4}} /$ graphite cells under diverse cycling conditions, PI-xLSTM reduces root mean square error (RMSE) by 26.83% and mean relative error (MRE) by 28.99%, while improving $\mathbf{R}^{\mathbf{2}}$ by 1.91% compared to the baseline xLSTM. The framework demonstrates strong robustness in tracking nonlinear aging trajectories, particularly in later battery life stages.

Decisions based solely on rational factors may lead to outcomes that do not necessarily align with investors' true preferences. Emotional factors, such as investors' risk preferences, also play a crucial role. A common challenge is to incorporate decision-makers' diverse risk preferences into a decision-making framework and to enable that framework to provide appropriate candidate solutions tailored to those preferences. Hesitant fuzzy sets, characterized by membership degrees represented as discrete arrays, are particularly effective in capturing real-world indecisiveness.However, the complexity and diversity of hesitant fuzzy elements also reveal deficiencies in some previous definitions of the inclusion relationship for hesitant fuzzy sets. Recently, Lu et al. proposed a rigorous definition of the inclusion relationships for hesitant fuzzy sets based on the diversity of hesitant fuzzy elements.Based on the revised definitions and propositions of hesitant fuzzy set knowledge system, the new concept of hesitant fuzzy soft $\beta$-covering approximation spaces (HFS$\beta$CASs) is introduced, and some fundamental propositions concerning HFS$\beta$CASs are investigated. Finally, a multi-attribute decision-making method within HFS$\beta$CASs that explicitly accounts for investors' risk preferences is proposed, thereby recommending the optimal investment option tailored to each investor.

In industrial processes, fault samples are often sparse and imbalanced, and GAN based generation with few samples tends to suffer from gradient vanishing and distribution mismatch. This paper proposes a MAML CGAN method for small sample fault data generation. Monitoring data from the same process under normal and multiple fault conditions are organized as related tasks; in the meta training stage, the conditional GAN generator is optimized within the MAML framework to enable fast adaptation across tasks, and in the meta test stage it is fine tuned on the target fault data to match its distribution. The generated samples are then combined with the original few fault samples to construct an augmented fault dataset with a size comparable to that of normal data, and the method is evaluated on the Tennessee Eastman process. Experimental results show that, compared with conventional CGAN and improved I WACGAN, the proposed method produces samples that are closer to real fault data in distribution similarity and variable correlation, and yields clear gains in diagnostic accuracy and F1 score, demonstrating its effectiveness for small sample fault data generation and augmentation in industrial processes.

The impact of economic growth on carbon emissions plays a crucial role in shaping national development strategies, particularly in low-income Asian countries where economic transformation is driving employment, foreign exchange earnings, and overall development. This study explores the relationship between economic expansion and CO2 emissions in low-income Asian countries from 2001 to 2020, using advanced analytical methods including multiple regression, moment quantile regression, and wavelet analysis to identify threshold points for sustainable development. Key factors such as forest area, natural resource rents, foreign direct investment (FDI), population density, and GDP are analyzed for their influence on CO2 emissions. The moment quantile regression results show that forest area and natural resource rents have a significant positive effect on CO2 emissions, particularly at higher quantiles, indicating intensified environmental pressure with industrialization. The heterogeneous impacts of FDI, GDP, and population density across quantiles suggest that their influence varies by development stage. Further, wavelet transform coherence (WTC) analysis reveals strong co-movements between emissions and economic indicators, especially FDI, while cross wavelet transform (XWT) results confirm that increases in FDI and GDP often precede higher emissions, underscoring the growth-emissions trade-off in low-income Asian economies. To mitigate these effects, the study proposes strategies such as promoting green infrastructure, fostering eco-friendly development, and implementing carbon offset programs, complemented by robust regulations and digital technologies. These measures can help reduce environmental impact while supporting continued economic growth in these nations.

Climate-adaptive mining management has become an important priority in China, where mining-related emissions, acid mine drainage, and metal-contaminated water can intensify under climate variability. Selecting suitable control strategies remains difficult because environmental effectiveness, ecological co-benefits, monitoring readiness, implementation feasibility, and lifecycle costs often create competing priorities for regulators and mining practitioners. To address this challenge, this study develops a hybrid fuzzy multi-criteria decision-making framework to prioritize climate-adaptive strategies and nature-based solutions for emission control, environmental policy support, and sustainable mining decision-making. A three-level decision hierarchy was constructed with five main criteria, twenty-five sub-criteria, and eight strategy alternatives. Fuzzy Analytical Hierarchy Process (AHP) was applied to determine the relative importance of the criteria and sub-criteria, while fuzzy ViseKriterijumska Optimizacija I Kompromisno Resenje (VIKOR) was used to rank the alternatives. The fuzzy AHP results show that MRV readiness and regulatory compliance receive the highest priority, followed by pollution-control effectiveness under climate variability. The fuzzy VIKOR results show that constructed wetlands for acid mine drainage treatment as the top-ranked strategy, followed by nature-based stormwater control and biochar-amended vegetative cover systems. The findings provide practical guidance for integrating climate adaptation, nature-based pollution control, and regulatory monitoring into mining environmental governance in China.

Accurate prediction of lithium-ion battery (LIB) capacity degradation is critical for reliable health management in applications such as electric vehicles and grid storage systems. Existing methods often fail to adequately model multi-scale temporal dependencies and exhibit training instability in deep neural architectures. To address these challenges, we propose a hierarchical multi-scale temporal network (HMSTN), featuring a novel multi-scale temporal block (MSTB). The MSTB integrates parallel convolutional branches with varying kernel sizes to capture localized fluctuations and incorporates a multi-head attention mechanism to model global cross-cycle degradation patterns. The hierarchical encoder-decoder architecture progressively downsamples input sequences by downsampling in the encoder to abstract long-term aging trends while restoring temporal resolution via skip connections and upsampling in the decoder to preserve critical short-term details. Auxiliary prediction heads at each encoder layer stabilize training through multi-task loss optimization, which effectively improves the overall loss function design. This mechanism establishes multi-path gradient flows, significantly improving robustness against training instability. Extensive experiments on different datasets demonstrate that HMSTN achieves superior accuracy in capacity prediction and exhibits strong generalization capabilities across diverse LIB chemistries and operating conditions.

Fault diagnosis in industrial processes often faces zero-shot challenges due to the scarcity of fault samples. Most existing approaches overlook statistical dependencies among attributes, leading to inference bias. To address this issue, we propose a zero-shot fault diagnosis method based on attribute coupling embedding. Independent predictors are trained for each attribute, with prediction residuals used to remove confounding effects of shared inputs. The statistical relationships among residuals are then exploited to learn the attribute coupling topology. This topology is explicitly embedded into a MultiLabel Graph Convolutional Network (MultiLabel-GCN), improving attribute probability estimation through structured priors. The loss function further incorporates asymmetric focal loss, negative probability clipping, and dynamic weight adjustment to mitigate severe attribute imbalance. Category inference is completed by nearest-neighbor matching with attribute prototypes. Experiments on the Tennessee Eastman Process (TEP) show that the proposed method achieves Top-1 accuracy of 83.26%-99.10% (average 92.47%), outperforming representative zero-shot diagnosis methods. These results demonstrate that attribute coupling modeling and graph-structured embedding can effectively enhance zero-shot fault diagnosis and support the safe operation of complex processes.

Accurate State of Charge (SOC) estimation is critical for the safe and efficient operation of batteries in electric vehicles (EVs). While deep learning models like Transformers have shown promise, they often struggle with sensor noise and complex temporal dynamics. Similarly, hybrid approaches like VMD-Transformer rely on fixed basis functions that lack adaptability. To address challenges such as sensor noise, nonlinear dynamics, and complex temporal dependencies, this study proposes a novel VMD-Basisformer model that integrates Variational Mode Decomposition (VMD) with an enhanced Basisformer neural network. The key innovations of our approach include: (1) a battery-optimized VMD process for noise reduction and multi-scale feature extraction; (2) a dual-path basis generation mechanism tailored to battery temporal dynamics; (3) a hierarchical attention architecture for capturing both local and global temporal dependencies. Experimental results under Dynamic Stress Test (DST) and Urban Dynamometer Driving Schedule (UDDS) conditions show that the proposed model significantly outperforms existing methods such as Transformer, Basisformer, and VMD-Transformer. Experimental validation on LiFePO 4 battery and supercapacitor datasets under DST and UDDS conditions shows that the VMD-Basisformer outperforms benchmark models (Transformer, Basisformer, VMD-Transformer) in accuracy and robustness. Ablation studies confirm the critical role of InfoNCE loss in ensuring temporal consistency.

Severely imbalanced sample distributions in industrial processes significantly degrade fault diagnosis performance, while limited observations hinder the effective learning of generative models. To address these challenges, this paper proposes an Attribute-Conditioned Diffusion Generative Enhancement Method for smallsample industrial fault diagnosis. Unlike conventional conditional generation paradigms based on discrete fault labels, the proposed method introduces physically and mechanistically meaningful fault attributes into the diffusion process and uses shared fault attributes, rather than class labels, as the basis for conditional modeling, thereby overcoming the limitation of class-isolated modeling and providing a shared, transferable semantic representation for small-sample fault generation. Furthermore, fault attributes are encoded as structured semantic tokens and injected into the diffusion denoising network via a cross-attention mechanism, enabling dynamic interaction between attribute semantics and intermediate features to guide the generation process and improve sample stability and quality under small-sample conditions. To reduce the statistical discrepancy between generated and real samples, a class-wise distribution calibration strategy based on feature moment matching is further introduced, enhancing the physical consistency and diagnostic usability of the generated data. Experimental results show that the proposed method achieves an average Jensen-Shannon divergence of 0.0384 between generated and real samples and an average diagnostic accuracy of 95.98 % under small-sample and imbalanced conditions, while maintaining stable performance in extremely scarce-data scenarios. These results demonstrate the effectiveness of the proposed method for industrial fault diagnosis under extreme data imbalance.

Accurate capacity estimation from multivariate time-series data is crucial for intelligent monitoring and predictive modeling. However, existing data-driven methods often struggle with underutilizing frequency-domain information and inadequately modeling features across different frequency bands. To overcome these limitations, this paper proposes the Dynamic-Static Feature Fusion Transformer (DSFF-Trans) model for multiband time-series capacity estimation. First, Wavelet Packet Decomposition (WPD) is used to convert time-domain signals into frequency-domain representations, allowing the separation of low-frequency static features and mid- to high-frequency dynamic features. For the decomposed features, a parallel CNN module with differentiated convolutional paths is designed to extract deep features. A feature fusion module is then introduced to adaptively balance the contributions of static and dynamic features. Finally, a Transformer encoder is used to effectively capture long-term dependencies in the time series. Experimental results on the publicly available CALCE dataset show that the proposed model reduces Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) by 48.05% and 37.36%, respectively, compared with conventional methods. These results validate the effectiveness of the proposed machine learning-driven multiband time-series modeling framework, with lithium-ion battery (LIB) capacity estimation serving as a representative application.

Accurate estimation of the state of health (SOH) is critical for the safe operation and effective management of lithium-ion batteries. However, data-driven methods face challenges in modeling the complex correlations among health factors (HFs) and often lack physical interpretability. In this paper, we propose a learnable graph (LG) spatiotemporal attention physics-informed network for accurate and reliable SOH estimation. Specifically, HFs are first extracted from the raw charging and discharging data of the batteries. A LG construction module is then employed to adaptively learn the topological structure among features and the importance of different time steps. Subsequently, the spatiotemporal attention network designed with a dual-path input architecture maps the inputs to SOH. One path deeply fuses spatiotemporal features across multiple time steps through the novel spatiotemporal feature extraction module, while the other path retains precise information from the current time step via a multi-layer perceptron, thereby achieving complementarity between long-range dependencies and immediate features. Finally, temporal attention weights are employed to compute the feature-specific partial derivatives across multiple time steps, and a degradation dynamics learning network is utilized to integrate the fitting capability of the data-driven approach with the physical constraints of battery degradation, enabling precise SOH estimation. Extensive experiments on the MIT and HUST datasets validate the effectiveness of our method and demonstrate its considerable potential for practical SOH estimation.

Flow meter is one of the most essential sensors in industrial development, energy measurement and environmental protection. Monitoring of flow meter performance can help detect anomalies early and enable timely corrective actions for critical industrial equipment in harsh operating environments. However, flow meter diagnostic models are often prone to overfitting and low accuracy caused by class-imbalanced small-sample data. To address these problems, a reinforcement learning Mahalanobis Taguchi system (RLMTS) model is proposed in this paper, which primarily consists of three modules, namely Mahalanobis space (MS) construction, threshold determination, and sample classification. In the MS module, an initial MS is constructed by selecting variables through orthogonal array design and signal-to-noise ratio analysis. Reinforcement learning is then introduced to adaptively refine the MS which is verified by the Mahalanobis distance. In the threshold determination module, a neural network algorithm is proposed to replace the traditional quality loss function for optimal threshold determination. In the sample classification module, the fault diagnosis of unknown samples is performed using the valid MS and calculated Mahalanobis distance. Experimental results show that the proposed RLMTS is not only suitable for flow meter fault diagnosis under different class-imbalance ratios with different small sample sizes, but also demonstrates a better diagnostic performance, stronger robustness, and broader applicability compared to the 19 benchmark diagnosis models. The use of RLMTS therefore guarantees stable operation of the flow meters, contributing to energy savings and environmental protection.

Accurately predicting the remaining useful life (RUL) of lithium-ion batteries (LIBs) is a critical step towards ensuring their safety. To address the challenges of limited historical data and inconsistency between training and testing data distributions, this paper proposes a LIBs RUL prediction approach that combines data generation and transfer learning (TL) under small sample conditions. This research uses k-nearest neighbor mega-trend diffusion (KNNMTD) to generate high-dimensional health indicators (HIs) for the source domain battery. During the pre-training and training phases, the maximum mean discrepancy (MMD) loss and fine-tuning are employed, respectively, with the MMD loss weight transformed by dynamic time warping (DTW). Compared to other methods, even with a 20% prediction starting point, the proposed method demonstrates good prediction performance on both single-domain target batteries and cross-domain target batteries.

Soft sets and rough sets are useful tools for decision-making. The article introduces the concept of interval-valued fuzzy soft /3-covering approximation spaces (IFS/3 CASs), which combine the theories of soft sets, rough sets, and interval-valued fuzzy sets. The IFS/3 CASs leverage the benefits of both soft sets and rough sets, utilizing lower and upper approximations to address ambiguous phenomena that involve interval-valued fuzzy type data with multiple attributes. The article explores fundamental propositions related to interval-valued fuzzy soft /3-neighborhoods and soft /3-neighborhoods in IFS/3 CASs. It further investigates four types of interval-valued fuzzy soft /3-coverings based on fuzzy rough sets and explores their relationships. Finally, the study demonstrates the applications of IFS/3 CASs across two distinct decision-making scenarios: one focusing solely on rational factors, and another incorporating both emotional and rational factors. The proposed methodologies support both strictly rational decision-making and a hybrid approach that synthesizes perceptual and rational elements, thereby enhancing adaptability in real-world applications.

Accurately estimating the capacity of lithium-ion batteries (LIBs) is a critical step in ensuring battery safety. At the moment, the data-driven capacity estimation method is a straightforward and successful strategy. Nevertheless, the issue of missing feature information continues to plague the feature selection process. Additionally, the capacity regeneration phenomena and the mean square error (MSE) loss function's limitations restrict estimation accuracy. To address these drawbacks, this study suggests a novel weighted kernel MSE (WKMSE) loss function, namely AM-LSTM-WKMSE, which combines the attention mechanism (AM) and long short-term memory (LSTM) for LIBs capacity estimation. First, the health indicators (HIs) were extracted from the voltage, current, temperature, and charging/discharging time curves. Following the correlation analysis of HIs, the AM is utilized to generate feature weights, which are subsequently entered into the LSTM. WKMSE is employed as a backpropagation loss function in training. Furthermore, experiments are conducted on the NASA dataset and the CALCE dataset. Comparing the method to alternative approaches, the testing results demonstrate that it has a 0.1% improvement in average the root mean square error of the batteries and superior estimation accuracy. Moreover, the proposed model exhibits strong generalization.

This study develops a value co-creation-oriented analytical framework to evaluate the performance and evolutionary dynamics of China's national-level quality policies from 1979 to 2023. A comprehensive categorization and scoring system is established to measure policy intensity, coordination, and comprehensiveness. Policy texts are systematically coded through content analysis, and indicator weights are determined using the Analytic Hierarchy Process (AHP). The resulting composite effect values are further analyzed through punctuated-equilibrium testing, breakpoint analysis, and a Vector Autoregression (VAR) model to estimate the temporal lag of policy implementation. Based on 10,962 policy documents retrieved from the Peking University Law Database, the results reveal clear evolutionary stages and cyclical upward trends in policy performance since the reform and opening-up, while the insufficient supply of demand-side policies remains a long-term structural weakness. The overall evolution path shows a transition from unilateral government provision centered on public value to dual government-market regulation driven by mixed commercial value, and finally toward pluralistic quality governance under value co-creation. Empirical evidence also indicates that quality policies act as short-term stimulus instruments that generate positive but sectorally differentiated effects across the three major industries. These findings highlight the need to expand policy coverage, enhance coordination and comprehensiveness, and rebalance the supply structure. Strengthening short-term stimulus effects while promoting inclusive, co-governed, and sustainable quality policy systems can further improve long-term effectiveness and provide useful insights for international discussions on value co-creation-based governance.

Hesitant fuzzy sets find extensive application in specific scenarios involving uncertainty and hesitation. In the context of set theory, the concept of inclusion relationship holds significant importance as a fundamental definition. Consequently, as a type of sets, hesitant fuzzy sets necessitate a clear and explicit definition of the inclusion relationship. Based on the discrete form of hesitant fuzzy membership degrees, this study proposes multiple types of inclusion relationships for hesitant fuzzy sets. Subsequently, this paper introduces foundational propositions related to hesitant fuzzy sets, as well as propositions concerning families of hesitant fuzzy sets. Furthermore, this research presents foundational propositions regarding parameter reduction of hesitant fuzzy information systems. An example and an algorithm are provided to demonstrate the parameter reduction processes. Lastly, a multi-strength intelligent classifier is proposed for diagnosing the health states of complex systems.