Workflow
破解安全、预测、协同难题 “擎源”发电大模型初现成效
Xin Hua Wang·2025-08-12 05:45

Core Viewpoint - The launch of the "Qingyuan" power generation model marks a significant advancement in the digital transformation of China's energy sector, integrating multi-dimensional data for enhanced operational efficiency and safety [1][2]. Group 1: Model Development and Features - The "Qingyuan" model, developed by the State Power Investment Corporation, incorporates operational monitoring, equipment status, and meteorological data, boasting a scale of over one hundred billion parameters [1]. - The model is designed to address the challenges of the power generation industry, which include the need for specialized AI models tailored to specific business requirements and the difficulty in obtaining high-quality industry data [2][3]. - Over six months, the company collected and processed more than 700TB of industry data, resulting in a high-quality dataset of 450GB, which was annotated by 380 industry experts [3]. Group 2: Application Areas - The "Qingyuan" model has been successfully applied in four key business areas: safety and environmental protection, electricity trading, production scheduling, and equipment maintenance, covering 13 scenarios and deploying 41 intelligent agents [4][5]. - In the safety and environmental protection sector, the model enhances lifecycle supervision of equipment and improves safety management efficiency [4]. - In electricity trading, the model predicts weather changes and market conditions, improving price prediction accuracy by 6.2% compared to traditional methods, leading to a 0.3% reduction in production costs and a 2% increase in profitability for a 600MW generator [5]. Group 3: Data Security and Future Directions - The company is addressing data security challenges by implementing strict transmission protocols and establishing a trusted data space for effective data management [7]. - Future efforts will focus on overcoming real-time response challenges in industrial control scenarios, with plans to utilize model distillation techniques to create lightweight models for local deployment [8]. - The company aims to promote the "Qingyuan" model through pilot validation, large-scale promotion, and ecosystem building, collaborating with universities and research institutions to enhance key technologies [8].