A data-based approach for benchmark interval determination with varying operating conditions in the coal-fired power unit

Abstract

The modern coal-fired power units in China are mostly operated in a flexible manner. However, flexible operation results in performance degradation, energy-efficiency penalties, and increased energy consumption, which necessitates the detection of performance degradation to save energy. This paper presents a model for detecting the performance degradation of coal-fired power units by determining the benchmark intervals of variables under varying operating conditions using data-mining methods. The K-means clustering method is employed to categorize the operating conditions according to the similarity of historical operational data. Gaussian mixture model is adopted to determine the benchmark interval with respect to the varying operating conditions by estimating the probability of historical runtime data. The methodology is validated using a feedwater heating system of an on-duty coal-fired power unit. The results indicate that in comparison with the design-based method, the proposed method can provide benchmark intervals for 225 operating conditions. In addition, the determined benchmark interval can detect performance degradation earlier than design-based values, thereby providing opportunities for energy-efficiency enhancement.