Offshore wind power forecasting is challenging because rapidly varying metocean conditions induce strong non-stationarity and heterogeneous power-generation dynamics. This study proposes RegimeMoE-PatchTST, a metocean-conditioned mixture-of-experts framework for multi-step offshore wind power forecasting. For each historical window, four statistical descriptors of seven metocean variables are extracted to form a 28-dimensional condition vector. Three independently parameterized PatchTST networks serve as forecasting experts, while a gating network maps the condition vector to sample-dependent weights and softly fuses their multi-step predictions. The entire framework is trained end-to-end without predefined regime labels. Experiments on two offshore wind turbine datasets at 2-, 4-, 8-, and 16-h horizons show that the proposed model achieves the lowest mean squared error (MSE) in all eight turbine–horizon settings and reduces average MSE by 3.69% relative to PatchTST. Ablation results verify the importance of both metocean-conditioned routing and dynamic expert fusion, while regime-wise evaluation shows MSE reductions of 7.7% and 8.1% under high-wind and high-turbulence conditions. These results demonstrate improved adaptability and forecasting accuracy under heterogeneous offshore operating conditions.
Artificial intelligence (AI) is transforming procurement through intelligent automation, predictive analytics, real-time decision support, and data-driven supplier management, creating new opportunities to advance sustainable supply chain performance. Yet, research on AI-driven sustainable procurement management remains fragmented across the AI, procurement, and sustainable supply chain domains, limiting theoretical integration and a systematic understanding of how AI-enabled capabilities can generate sustainability value. This study addresses this gap through a systematic literature review (SLR) of peer-reviewed studies published between 2011 and 30 June 2026. The review synthesizes the evolution and intellectual structure of the field across AI technologies and capabilities, procurement applications, sustainability dimensions, supply chain resilience, theoretical perspectives, methodological approaches, and contextual conditions. The findings show that existing research is primarily concentrated on technology adoption, process automation, predictive analytics, and operational efficiency, while comparatively limited attention has been directed toward the organizational, relational, and governance mechanisms that shape the translation of AI capabilities into sustained and multidimensional sustainability outcomes. The research-gap analysis identifies important deficiencies in end-to-end procurement integration, AI capability maturity, data governance and interoperability, predictive sustainability risk management, responsible supplier management, human–AI collaboration, explainability and accountability, contextual boundary conditions, and comprehensive sustainability measurement. In response, the study develops an integrative future research agenda that moves beyond technology adoption toward capability development, responsible AI governance, resilient procurement ecosystems, and measurable sustainability value creation. By integrating fragmented streams of research and clarifying the relationships among AI capabilities, procurement processes, and sustainable supply chain outcomes, this study advances the conceptual foundations of AI-driven sustainable procurement management. It provides a structured basis for cumulative scholarship and evidence-informed managerial practice.
Minimizing energy inputs and greenhouse gas (GHG) emissions is essential for achieving sustainable crop production. This study analyzed changes in energy use and GHG emissions associated with chemical fertilizer application in crop production in the middle reach of the Yangtze River from 2014 (one year before the implementation of zero-growth actions) to 2022. Results revealed significant reductions in both total energy inputs and energy inputs per unit area from chemical fertilizers, with annual declines of 10.72 PJ and 0.48 GJ·ha−1, respectively. The energy use efficiency of chemical fertilizer application increased significantly at an annual rate of 0.40. Concurrently, total GHG emissions, emissions per unit area, and emissions per unit of energy input associated with chemical fertilizer application decreased significantly, with respective annual reductions of 0.31 × 107 kg, 14.08 kg·ha−1, and 0.26 kg·GJ−1. These reductions were primarily due to decreases in inputs of straight nitrogen (N), phosphorus (P), and potassium (K) fertilizers, rather than changes in compound fertilizers, as the input of compound fertilizers saw a significant increase. This study suggests that zero-growth actions, with the substitution of compound fertilizers for straight N, P, and K fertilizers as a key approach, are associated with lower energy inputs and GHG emissions from chemical fertilizer application in the middle reach of the Yangtze River. However, given the descriptive nature of this analysis, these findings should be interpreted as correlational, and further quasi-experimental studies are warranted to confirm causality.
