发电行业如何安全协同？全球首个千亿级发电大模型“擎源”发布

Core Insights - The National Energy Group launched the world's first trillion-level power generation large model, "Qingyuan," aimed at addressing key pain points in the power generation industry such as high safety risks and complex decision-making [1][3] Group 1: Model Features and Applications - "Qingyuan" is designed specifically for the power system, achieving breakthroughs in heterogeneous data fusion, cross-business intelligent collaboration, and autonomous intelligent decision-making [1][3] - The model's initial applications cover four business domains: safety and environmental protection, power trading, production scheduling, and equipment maintenance, encompassing 13 application scenarios and 41 intelligent agents [1][4] - The model is built on a high-quality industry dataset of 450G, resulting in 610 million sets of SFT Q&A pairs, outperforming base models by 19.6 percentage points in the power generation field [4] Group 2: Industry Challenges and Solutions - The power generation industry faces challenges in the intelligent transformation process, including technical barriers and a lack of understanding of large model applications among production personnel [3][4] - The National Energy Group employs a "dual-domain responsibility" model to bridge the gap between technology and business personnel, fostering collaboration to explore AI empowerment in the power generation sector [3][4] Group 3: Efficiency Improvements - In the safety and environmental protection domain, "Qingyuan" significantly enhances the efficiency of technical supervision evaluations, reducing the evaluation time from one week to one day for a power plant [4] - In the power trading domain, "Qingyuan" improves electricity price prediction accuracy by optimizing model combinations based on different environments [5] - The model also optimizes scheduling and predictive indicators in production scheduling and provides precise diagnostics and maintenance strategies in equipment maintenance [5] Group 4: Data Security and Decision Timeliness - The implementation of the large model faces challenges related to data security and decision-making timeliness, with measures in place to ensure data safety through strict transmission protocols and classified data management [6] - Future plans include "model distillation" to refine large model capabilities into smaller models for localized deployment, addressing specific scenarios requiring rapid response [6][7] - The National Energy Group aims to promote the "Qingyuan" model through pilot verification, large-scale promotion, and ecosystem co-construction [6][7]