Against the backdrop of the carbon peaking and carbon neutrality goals and the high penetration of renewable energy, new-type power systems face challenges including uncertainty, multi-actor coordination, data silos, and security and trustworthiness. This review organizes the literature into three enabling layers: foundation models and multimodal representation; retrieval-augmented generation (RAG) and knowledge graphs (KGs); and multi-agent collaboration with embodied intelligence. It further discusses the role of world models in the preview and prediction of physical operational consequences. Across four domains—markets and trading; dispatch operation and maintenance; secure operation and situational awareness; and energy storage and energy-use management, this paper summarizes the applicable tasks, constraint injection and engineering boundaries. LLMs are most credible as high-capability assistants or bounded execution components coupled with evidence retrieval, physical solvers, digital twins and human oversight, rather than unconstrained autonomous controllers. Progress will depend on a closed loop linking trusted evidence, physical consequence assessment and accountable responsibility boundaries.
Owing to the rapid development of large language model (LLM), Text-to-SQL tasks have progressed significantly in recent years. However, applying Text-to-SQL to specialized domains such as finance, healthcare, and power engineering remains challenging owing to complex user queries and highly specialized database schemas. This paper proposes a highly reliable Text-to-SQL parsing framework named SQL4Domain, with the power industry considered as a typical application scenario. The framework first utilizes domain-specific knowledge bases to implement user-query intent recognition and key entity extraction. Subsequently, it generates structured query language (SQL) statements combined with domain knowledge prompts. Finally, it improves the accuracy and reliability of SQL statements via multi-strategic knowledge verification. Experiments on the power-industry dataset reveal that SQL4Domain significantly outperforms existing methods in terms of execution accuracy, with a performance improvement of 21.65% compared with the mainstream DAIL method.
At the intersection of information systems and strategic management, implementing closed-loop intelligent decision-making in complex organizations remains a pivotal theoretical and practical challenge. Addressing the “growth dilemma”—wherein software enterprises suffer from weak economies of scale due to labor-intensive customization—this study draws on the dual perspectives of software economics and knowledge governance to analyze Palantir, a pioneer in decision intelligence. Specifically, it systematically deconstructs the co-evolutionary mechanism between Palantir’s “Ontology”-driven technical architecture and its “Forward Deployed Engineer” (FDE) organizational model. The findings indicate that Palantir transforms bespoke, high-cost services into durable, modularized R&D assets through “on-site service assetization and productization,” thereby overcoming the diminishing returns inherent in traditional IT services. However, the scalable generalization of this model hinges on stringent boundary conditions: a concentrated tier-one client base (“super-anchor” accounts), a steady pipeline of interdisciplinary talent, deep organizational embeddedness enabled by administrative leverage, and sovereign-level trust mechanisms. Finally, grounded in the empirical context of digital transformation across China’s industrial and public sectors—and addressing core bottlenecks such as data quality heterogeneity, multi-level governance friction, and the structural undervaluation of software assets—this paper outlines localized evolutionary pathways and strategic recommendations for constructing indigenous intelligent decision-making systems.
Language education technology (LET) is one of the most representative and important components of smart education. In the face of new learning environments and technological trends, addressing and resolving the two fundamental contradictions of smart education—large-scale education versus personalized learning, and fragmented knowledge versus systematic construction—is particularly crucial in language education. Starting from the nature of language learning and the characteristics of contemporary artificial intelligence, this paper analyzes the limitations of existing online education models and language learning tools, and proposes a methodological framework that resolves the first contradiction through data-driven personalized learning services and the second through scenario-based learning and content-sharing mechanisms. It further elaborates a closed-loop mechanism for delivering high-quality data-driven learning services. Drawing on two concrete scenarios—machine language learning (i.e., programming) and international Chinese education—the paper introduces the practices of two online learning platforms, Shuishan Online and Shuishan Chinese, and analyzes the feasibility of this technical approach. Finally, the paper briefly discusses and prospects the technological development trends and practical issues of language education technology.
中文核心期刊
