
Purpose This study develops an interpretable grey forecasting framework under limited information conditions. The framework supports air quality early warning across regions. It focuses on disturbed annual Air Quality Composite Index (AQCI) sequences and heterogeneous warning pressure across cities. Design/methodology/approach This study proposes S-HFGM(1,N), a Hausdorff fractional multivariable grey model with a single indicator buffer. The model integrates target sequence buffering, Hausdorff fractional accumulation and multivariable grey modelling. The framework uses AQCI association pressure for city screening and selects socioeconomic drivers. After validation, Monte Carlo simulations based on residuals quantify uncertainty. Warnings are then interpreted for specific pollutants. The application covers the Beijing–Tianjin–Hebei region. It uses observations from 2013–2022 for training and 2023–2024 for validation. The validated model produces projections for 2025–2030. Findings The model achieves an average validation mean absolute percentage error of 3.6861%, outperforming selected grey forecasting benchmarks. The baseline projections show different trends under gradual structural evolution. Beijing faces slight upward pressure. By contrast, Shijiazhuang and Tangshan continue improving. Handan and Xingtai follow a relatively stable trajectory despite persistent pressure. The 80% and 95% prediction intervals widen with the forecast horizon, with Handan and Xingtai showing the widest 95% intervals in 2030. Future warning pressure is mainly associated with O3 and particulate matter. Practical implications The framework suggests that regional governance should move beyond uniform control measures. Instead, policies should reflect the pollutant profile of each city. Coordinated monitoring should also be strengthened for nonadjacent priority cities. Originality/value The unified grey framework combines disturbance weakening, nonlinear time scale memory reconstruction and structural external driving. It extends grey forecasting beyond single city prediction. This extension supports warning interpretation across regions for environmental policymakers and regional governance bodies.
Purpose China is entering a moderately aging society, in which the demand for primary healthcare resources is characterized by small samples, multiple influencing factors and strong temporal dependence. This study proposes a novel grey forecasting approach to improve prediction accuracy and stability for primary healthcare resource demand under incomplete information. Design/methodology/approach Drawing on empirical data for primary healthcare institutions and beds in China, this study employs grey relational analysis to identify key driving variables and develops a grey convolutional multivariate self-memory combination model, denoted as GMC-SM(1,n), by integrating grey system theory with the self-memory principle and compares it with classical grey prediction models. Findings The proposed GMC-SM(1,n) model exhibits superior accuracy, with fitting and prediction MAPEs of 1.02% and 0.28% for institutions, and 1.93% and 0.82% for beds. The model also achieves lower root mean square error, mean absolute error and standard deviation of errors (SD) than comparative models, demonstrating strong robustness and an enhanced ability to capture multivariable coupling and temporal dependence with small samples. Practical implications The results reliably support dynamic resource allocation and long-term planning under population aging, offering practical reference value for policymakers. Originality/value This study enriches grey forecasting methodology by integrating a self-memory mechanism into the grey convolutional multivariate framework, effectively addressing the issues of historical dependence and initial value sensitivity inherent in traditional models.
Purpose Intermittent demand in distribution-center operations is difficult to forecast because long zero-demand periods are often interrupted by irregular demand bursts. This study develops TADGM–RGPL, a cost-aware gray forecasting framework that combines a temporally attenuated discrete gray model (TADGM) with reward-guided parameter learning (RGPL), to account for both statistical calibration deviation and the asymmetric inventory consequences of underforecasting and overforecasting. Design/methodology/approach TADGM introduces a bounded recency-sensitive driver directly into the DGM state-transition equation, while RGPL learns a scalar decision-calibration factor from pre-origin inventory-cost feedback. The framework is evaluated on 701 intermittent and lumpy DC-SKU series using 4,206 rolling-origin windows. Conventional intermittent-demand methods, gray models, XGBoost and LightGBM are included as benchmarks. Findings TADGM does not dominate the benchmark methods in point-forecast accuracy. Standard TADGM–RGPL significantly reduces inventory cost relative to the conventional intermittent-demand and gray benchmarks under the stockout-sensitive p/h = 10 setting, whereas XGBoost and LightGBM achieve lower overall cost when all series are pooled. ADI-stratified analysis shows that TADGM–RGPL attains the lowest average cost in the highest-intermittency quartile, although its advantage over the machine-learning benchmarks is not statistically significant within that subset. Originality/value The study differs from residual reweighting by introducing recency sensitivity as a time-varying gray input in the DGM state-transition equation and by separating statistical forecasting from downstream cost-aware decision calibration. The empirical analysis further identifies the demand conditions under which the proposed forecasting-decision framework is relatively more competitive.
Purpose - Solar power generation exhibits significant time-varying seasonal fluctuations and nonlinear trends, which pose a challenge to existing forecasting techniques. Therefore, this paper proposes a novel grey forecasting framework that integrates seasonal decomposition, nonlinear modeling, and time-varying adaptation, namely the seasonal time-varying discrete grey power model Design/methodology/approach - Firstly, seasonal dummy variables are introduced to separate periodic fluctuations from the trend. Secondly, the seasonal effect is innovatively decomposed into fixed and time-varying components, precisely capturing the temporal evolution patterns across different periods. Concurrently, the double-power index structure quantifies the nonlinear driving effects of both historical values and temporal factors on the current state. The hiking optimization algorithm is employed for hyper-parameter tuning Findings - The new model was validated using datasets from the three major economies of the U.S., India, and China, and compared with six competitive models representing different types. Results show that the new model exhibits superior predictive accuracy. Furthermore, robustness tests across multiple time windows further validate the stability of the model under varying scenarios Practical implications - Forecasts for 2025-2027 indicate sustained growth in solar power generation across all three nations, but the magnitude and patterns of growth exhibit significant differences. The research findings provide a reliable decision support tool for solar power planning and grid dispatch in the global energy transition. Originality/value - The new model effectively captures the complex dynamics in solar power generation data by integrating seasonal time trend terms, dummy variables, and power index terms into the discrete grey modeling system
Purpose This study examines relational closeness between university students' academic achievement, represented by cumulative grade point average (CGPA) and mental health, well-being, sociodemographic variables using a self-reported student survey. Anxiety, depression, stress, sleep quality, diet quality, physical energy level, age, gender, department and scholarship status were analyzed. Relational and predictive relevance was evaluated without assuming causality, with an ergonomic interpretation of well-being indicators. Design/methodology/approach Grey relational analysis (GRA) estimated relational degrees between CGPA and selected variables. Deng's, absolute, relative, comprehensive and negative GRA models were applied, with sensitivity analyses across distinguishing coefficient values. Cronbach's alpha and exploratory factor analysis (EFA) examined whether anxiety, depression and stress represented a common structure. They were combined into a risk construct defined as “mental health risk.” Seven machine learning algorithms classified this construct. Repeated randomized train-test experiments assessed accuracy, precision, F1-score and receiver operating characteristic (ROC) area under the curve (AUC). Findings Anxiety, depression and stress showed relatively high relational degrees with CGPA compared with the other variables. These findings suggest that academic achievement is related not only to demographic characteristics but also to psychological well-being and ergonomics-related conditions of students. According to the machine learning results, AdaBoost achieved the highest mental health risk classification performance, with an accuracy of 82.98%. Originality/value This study integrates comparative GRA modeling, statistical construction of a mental health risk indicator, repeated machine learning evaluation and ergonomic interpretation. The framework links relational analysis with predictive classification while distinguishing analytical risk from clinical diagnosis.