电力学报2026，Vol.41Issue(1)：28-42,15.DOI:10.13357/j.dlxb.2026.004
基于物理信息机器学习算法的电厂空冷岛运行策略优化研究
Optimization Study of Power Plant Air-cooled Island Operation Strategy Using Physics-informed Machine Learning Algorithms
苏宇 1枚军 2宋晓瑜 2张蓉2
作者信息
- 1. 山西漳山发电有限责任公司,山西 长治 046021
- 2. 太原智博热电工程设计有限公司,太原 030001
- 折叠
摘要
Abstract
The air-cooled island(ACI)is a major power-consuming facility in thermal power plants.To reduce the coal consumption rate of power generation,further optimization of ACI operation strategies is essential.Due to the complex and variable operating conditions of ACIs,purely data-driven AI algorithms exhibit insuffi-cient generalization capabilities.Taking a thermal power plant in Shanxi for example,this study proposes a da-ta-physics co-driven optimization algorithm centered on the physics-informed neural network(PINN)frame-work.The algorithm employs the ACI mechanistic model as a skeleton model,integrating big data-driven deep learning to establish a ACI mechanistic-AI fusion model.With economic optimization as the goal,an intelligent ACI operation system is developed based on the turbine's optimal vacuum model and the ACI mechanistic-AI fusion model.This system intelligently optimizes fan rotational speed or current to maintain optimal vacuum conditions under specified constraints.Results from typical daily analyses show that the average coal consump-tion savings in spring,summer,and autumn reach approximately 1.0 g/(kW·h).Coal savings increase as am-bient temperature decreases,peaking at 3.66 g/(kW·h)with significant energy-saving effects.Physics-in-formed machine learning algorithms demonstrate strong generalization capabilities in optimizing ACI operation strategies,achieving substantial coal-saving benefits.关键词
火电厂/空冷岛控制/数据与物理联合驱动/机器学习/物理信息融合/机理模型/运行优化/PINNKey words
power plant/air-cooled island control/data and physics joint drive/machine learning/physics-in-formed fusion/mechanism model/operation optimization/PINN分类
信息技术与安全科学引用本文复制引用
苏宇,枚军,宋晓瑜,张蓉..基于物理信息机器学习算法的电厂空冷岛运行策略优化研究[J].电力学报,2026,41(1):28-42,15.
