A visual analytics approach to piecewise linear regression modeling for flow characteristic curve correction

Abstract

In large-scale process industries, such as thermal power plants, regularly correcting the flow characteristic curve of the main valves using historical data is essential for ensuring accurate flow control. Piecewise linear regression modeling is a practical and effective method for this correction. However, the large size of historical data, the uneven sampling distribution, and the complexity of the optimization process often lead to serious problems, such as long iteration cycles and low modeling quality in existing practices. Regarding these issues in the industrial applications, this paper presents a visual analytics approach to improve flow characteristic curve correction from three perspectives: (a) interactively extracting flow characteristic data from vast historical samples with long periods and multidimensional correlation; (b) iteratively optimizing the model resolution process through data segmentation and heuristic algorithms in multiple aspects; (c) comprehensively evaluating, comparing, and fusing multiple candidate models. On this basis, we developed an interactive visualization tool, PlReVA, to integrate user interaction and domain knowledge deeply into the modeling process. A real-world case study using this tool was demonstrated on the historical dataset of actual thermal power generation. The effectiveness and practicality of PlReVA were confirmed through quantitative evaluation and feedback from domain experts.