Coal consumption prediction in thermal power units: A feature construction and selection method. (15th June 2023)

• Record Type: Journal Article
• Authors: Zhou, Jian; Zhang, Wei
• Is Part Of: Energy. Volume 273(2023)
• Publication Date: 2023-06-15
• Subjects: Thermal power units -- Coal consumption prediction -- Regression analysis -- K-means algorithm -- Genetic algorithm; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2023.126996
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: Digitization and related facilities have enabled the thermal power generation enterprises to record real-time data of thermal power units. There are many data-driven applications based on real-time monitoring and operational data in power units, while limited studies lay on the operational improvements, especially on coal consumption prediction under all working conditions. We build an intelligent prediction model of coal consumption based on key features selection, working condition clustering, and regression analysis. We combine feature construction and feature selection methods to cope with the problem caused by directly specifying feature subset for model building of traditional prediction method, which may fall into the thinking pattern and miss potentially better feature subset. Besides, to cope with the different coal consumption under different working conditions, we apply cluster analysis to construct a sub-coal consumption prediction model for each cluster category. Numerical results show that compared with other methods, it has the advantages of lower regression error and moderate model complexity, which can provide efficient decision support for operational improvement in thermal power generation. Highlights: An intelligent method is proposed for thermal power units' coal consumption. The feature construction process is applied to create potentially better features. The feature selection process is used to obtain the ideal feature subset. Numerical results validate the lower regression error and moderate complexity.