Mechanism analysis is the main method for energy-consumption feature selection of thermal power systems. Although this method can select a few features with specific physical meanings as the input variables for energy-consumption modeling, it may ignore some important attributes which affect the fitting and generalization capabilities of the model. We propose a robust attribute reduction method with rough set theory and apply it to the above task. Taking a boiler combustion system of some 600MW power unit as experimental subject, an energy-consumption model with the selected key features is built. Several comparative experiments were carried out on the operational data to evaluate the performance of the energy-consumption model. The experimental results show that the robust attribute reduction algorithm has strong generalization ability, and the energy-consumption model built with key features has a strong fitting ability and high prediction accuracy, thus providing an effective method for energy-consumption prediction and optimization of the thermal power system.

机理分析是火电系统能耗特征选择的主要方法。尽管此方法可以选择一些具有特定物理意义的特征作为能耗建模的输入变量，但它可能会忽略一些影响模型拟合和泛化能力的重要属性。我们提出了一种基于粗糙集理论的鲁棒属性约简方法，并将其应用于上述任务。以600MW左右机组的锅炉燃烧系统为实验对象，建立了具有选定关键特征的能耗模型。在运行数据上进行了几次比较实验，以评估能耗模型的性能。实验结果表明，鲁棒的属性约简算法具有较强的泛化能力，