Jian Zhang

Dynamic carbon emission factors (DCEFs) allocate emissions to nodal demand at a given operating state, whereas marginal carbon emission factors (MCEFs) estimate emission changes caused by load adjustments but can become unstable under renewable forecast errors and binding constraints. We propose a reliability-shrunk composite carbon emission factor (CCEF) for bilevel low-carbon demand response. A directional finite-response MCEF is evaluated over each executable nodal response interval with day-ahead commitment fixed. Uncertainty samples yield a risk-adjusted marginal estimate and tail dispersion; a mean-squared-error weight retains reliable marginal information and otherwise shrinks it toward DCEF. Heterogeneous industrial, commercial, residential, and electric-vehicle flexibility is coordinated through welfare clearing, while fixed-commitment security-constrained economic dispatch verifies feasibility and emissions. In a modified IEEE 39-bus system, CCEF reduced emissions by 4.863% (2479.44 t), versus 3.188% for DCEF and 4.474% for MCEF, while delivering CNY 2.4809 million in social net benefit and 91.983% renewable utilization. Across 20 holdout seeds, CCEF exceeded MCEF by 0.494 percentage points (95% confidence interval: 0.400-0.588). Mechanism, network, sensitivity, and ablation analyses attribute the gain to reallocating flexibility under a common activity cap rather than relaxing security constraints.

AI-augmented industrial decision-support systems, such as retrieval-augmented diagnosis engines and small-language-model monitors, now report an internal reliability signal alongside every inference. No maintenance optimization model uses that signal as an uncertainty parameter, and none hedges against its adversarial corruption. We treat an AI-generated diagnostic signal’s trustworthiness as a first-class, potentially adversarial variable. We introduce a trust-weighted contamination ambiguity set, a Wasserstein ball whose radius scales with the reported reliability, mixed with a budgeted adversarial component, which generalizes both Huber contamination and Wasserstein ambiguity sets. Embedding it in a two-stage distributionally robust maintenance model with priced verification recourse, we obtain a closed-form worst-case cost, a polynomial-size mixed-integer linear program, and a verification policy that thresholds the reliability score. We validate on the NASA C-MAPSS turbofan dataset with a real prognostic model whose confidence predicts its error (Spearman -0.70), under real false-data-injection, poisoning, and noise attacks. Naive trust is cheapest but leaves 7 to 14 critical engines unserviced per run, while the proposed policy attains near-complete failure prevention at a third to a half of blanket-verification cost. Crucially, among policies deployable under an unknown attack regime, no fixed verification heuristic matches its missed-failure rate for less than 44 percent additional cost; only an oracle that knows the attack can equal it. The premise generalizes to a second dataset (XJTU-SY bearings, Spearman -0.89). A language-model case study (retrieval-augmented deepseek-chat) confirms that prompt injection and corpus poisoning corrupt the signal while its faithfulness stays blind, so the contamination hedge supplies the protection.

Serial multi-echelon inventory systems must increasingly be operated without knowing the demand distribution, and that distribution rarely stands still. Learning-based inventory theory offers sharp regret guarantees only for single-stage systems, while existing network methods assume stationarity or provide no worst-case guarantee. We propose the Restarting Echelon-Gradient (REG) policy, a time-varying echelon base-stock rule that runs projected gradient steps at each echelon with a variation-adaptive restart schedule and a delayed-feedback correction. A variation-transfer lemma converts demand drift, measured in the Wasserstein metric, into per-echelon cost-variation budgets, yielding a finite-sample dynamic-regret bound that grows as the two-thirds power of the horizon, the one-third power of the variation budget, and linearly in the number of echelons and the lead time; the rate extends to the fully coupled system under a demand-density condition with an echelon-separation margin, and a budget-agnostic variant attains it without knowing the budget. A matching lower bound establishes minimax optimality in the horizon and the variation budget. Experiments on controlled instances, standard serial benchmarks, and the real Willems supply chains confirm the predicted scaling and yield an operational rule: track the drift when demand shifts frequently relative to the lead time; under longer lead times a robust static base-stock remains competitive.

This paper proposes a composite carbon emission factor (CCEF) demand response framework to address the limitations of single-factor carbon accounting and achieve economic–environmental synergy. The CCEF mechanism integrates the dynamic carbon emission factor (DCEF) and marginal carbon emission factor (MCEF) through an adaptive weight allocation based on the real-time generation mix. To ensure practical scheduling, the load shifting process is embedded in a co-optimization model that minimizes system generation costs under demand-side physical constraints and network security limits. This mechanism guides spatiotemporal load shifting from thermal-dominated evening peaks to high-renewable midday periods based on carbon potential gradients. Simulations on a modified IEEE 39-bus system show that the CCEF framework achieves a unit emission reduction efficiency of 0.5024 tCO₂/MW and a total reduction of 462.03 tCO₂. These results outperform individual DCEF and MCEF strategies, demonstrating feasible scheduling and an effective balance between carbon reduction and operational costs.

Action Quality Assessment (AQA) aims to automatically evaluate action execution quality from videos. In complex scenarios such as figure skating, scoring depends on multiple technical elements and their temporal organization, exhibiting a clear stage structure. However, existing methods typically rely on uniform temporal segmentation or global feature aggregation, making it difficult to align with key scoring stages and limiting the modeling of action execution.To address this issue, we propose SG-AQA, a structure-guided framework for action quality assessment. First, stage representations aligned with scoring semantics are constructed by using the temporal centers of technical elements as anchors and performing weighted aggregation over the sequence. A dynamic stage number mechanism is further introduced to adapt to varying structural complexity. Second, we propose a Stage-conditioned Temporal Attention (SCTA) mechanism, which uses stage features as conditioning signals to guide temporal feature weighting, enabling the model to focus on stage-relevant key actions. In addition, a joint optimization strategy integrating regression, ranking, and structural consistency constraints is adopted to improve prediction stability.Experiments on the FineFS dataset demonstrate that the proposed method achieves consistent improvements and outperforms multiple baselines in terms of Spearman’s rank correlation. Further analysis shows that these gains stem from effective utilization of stage structure, leading to more accurate representation of critical action stages.

Action Quality Assessment (AQA) aims to automatically evaluate action execution quality from videos. In complex scenarios such as figure skating, scoring depends on multiple technical elements and their temporal organization, exhibiting a clear stage structure. However, existing methods typically rely on uniform temporal segmentation or global feature aggregation, making it difficult to align with key scoring stages and limiting the modeling of action execution.To address this issue, we propose SG-AQA, a structure-guided framework for action quality assessment. First, stage representations aligned with scoring semantics are constructed by using the temporal centers of technical elements as anchors and performing weighted aggregation over the sequence. A dynamic stage number mechanism is further introduced to adapt to varying structural complexity. Second, we propose a Stage-conditioned Temporal Attention (SCTA) mechanism, which uses stage features as conditioning signals to guide temporal feature weighting, enabling the model to focus on stage-relevant key actions. In addition, a joint optimization strategy integrating regression, ranking, and structural consistency constraints is adopted to improve prediction stability.Experiments on the FineFS dataset demonstrate that the proposed method achieves consistent improvements and outperforms multiple baselines in terms of Spearman’s rank correlation. Further analysis shows that these gains stem from effective utilization of stage structure, leading to more accurate representation of critical action stages.

In recent years, China’s digital economy has become a key engine for high-quality development. Assessing the operational efficiency of leading digital enterprises is crucial for optimizing resource allocation and promoting sectoral growth. However, existing research largely remains at regional or industry levels and typically reports efficiency scores without diagnosing the root sources of inefficiency. To fill this gap, this study measures the operational efficiency of 99 firms selected from China’s Top 100 Digital Economy list (2017–2022) using the BCC-DEA model, and analyzes their dynamic evolution via kernel density estimation. The findings reveal a fluctuating upward trend in overall efficiency, and that the gap in overall technical efficiency primarily originates from scale efficiency rather than pure technical efficiency. The kernel density peak exhibits a “rise–decline–rise” pattern, indicating existing but narrowing efficiency differences among firms. By decomposing efficiency, this study further classifies firms into four types, revealing that inefficiency is heterogeneous. This paper makes three main contributions. First, it identifies scale efficiency as the main source of efficiency gaps. Second, it classifies firms into four types, revealing that inefficiency is heterogeneous. Third, it uses kernel density estimation to track the dynamic evolution of efficiency, showing a narrowing efficiency gap but a persistent superstar effect. Two policy implications follow: firms with low pure technical efficiency should focus on management training and technology adoption, while firms with low scale efficiency should pursue scale expansion through mergers or partnerships.

Urban innovation capacity is increasingly critical to city development, and quantitative assessments of innovative cities’ innovation capability can be achieved via the composite index method, which fully integrates multidimensional indicators. This study develops a hierarchical data envelopment analysis (H-DEA) method to establish a composite index evaluation model for innovation capacity, which features flexible and objective two-level indicators—an advantage that avoids subjective weight assignment and adapts well to the hierarchical structure of innovation evaluation indicators. The proposed H-DEA model is applied to evaluate 67 innovative cities in China, yielding composite scores and rankings that are further compared with those from the traditional weighting method. Sensitivity analysis is conducted by adjusting different upper and lower bounds of the H-DEA model to verify its robustness. Additionally, these 67 cities are divided into four regions, with region-specific weights assigned to the evaluation indicators in the model. The results show that the eastern region has the highest average innovation capacity (0.3783), where technological innovation (weight 0.27) serves as a key driving force; the western region has the lowest average innovation capacity (0.3235), and its innovative cities should prioritize improving outcome transformation capacity (weight 0.1357). Overall, technological innovation receives the highest average weight (0.2422), while outcome transformation capacity gets the lowest (0.1647).

Under uncertain environmental conditions, a nonlinear system may encounter problems like information conflicts, ambiguity, loss, and unclear interdependencies, which can result in low accuracy in predicting abnormal system states. System failures lead to negative impacts. To address these issues, this study proposes an analysis model that incorporates factor weight adaptive adjustment and integrates Dempster-Shafer’s theory (D-S theory) algorithms to construct an evaluation model, and to quantify uncertainty factors and correlation factors within the system. This approach reduces the impact of uncertainty in information features on analysis accuracy, enhances the precision of system reliability state probability assessment, and improves decision-making management levels. The results of the wind turbine system case study indicate that the proposed algorithm achieves a system state identification accuracy of 97% and a system reliability probability of 65%, with an overall accuracy improvement of 4.5% compared to traditional algorithms and a reliability probability assessment accuracy improvement of 5.22%, better aligning with the actual system’s state probability distribution.

The pervasive misuse of Duty-Free Quota Abuse Smuggling has seriously undermined fiscal and market order. This study breaks through the traditional model’s assumption of complete rationality and establishes a Multi-Phase Dynamic Decision-Making Model for Duty-Free Quota Abuse Smuggling Chain System, incorporating the risk avoidance preference of illegal actors to analyze strategic interactions within the smuggling chain system. Through theoretical deduction and simulation experiments, the evolution of the system during the decision-making phases of Decentralized Profit-Seeking, Localized Collusive, and Collaborative Profit-Seeking was analyzed, and key intervention points were identified. The study results indicate that smuggling chains will continuously gravitate toward localized collusive; the risk avoidance of illegal actors suppresses local alliance benefits and shortens accumulation cycles; strengthening cost constraints reduces the overall level of smuggling in the system, with Quota Sellers being the most sensitive. Therefore, we propose hierarchical regulation, credit supervision, and differentiated law enforcement to precisely target smuggling chains.