Visualization Analysis Framework Of Piecewise Linear Regression Supporting Flow Characteristic Curve Correction

Flow control valves are widely used in industrial production such as thermal power generation.In order to ensure the accuracy of the control process,it is necessary to periodically calibrate the flow characteristic curve.Compared with calibration on site,the data-driven method has been applied in this field in recent years because of its lower cost,that is,the adjustment accuracy of the valve can be improved by analyzing and modeling the historical data of flow control.However,due to the large scale of historical data and uneven sampling distribution,the existing methods have problems such as high model complexity,poor robustness and long iteration period.In order to solve the above problems,this thesis proposed a flow characteristic curve calibration method based on piecewise linear regression model,and introduced visual analysis technology into the data modeling process.A visual analysis framework Pl Re VA(Visualization Analysis Framework of Piecewise Linear Regression)was designed and implemented to improve the interactivity,intuitiveness and efficiency of iterative analysis of the calibration modeling process.Pl Re VA first designed interactive views for longspan and multidimensional data to support iterative multi-attribute combination filtering;Secondly,a partition method combining isometric,merging and regression tree strategies was proposed to determine the number and location of the data segmentation.Besides,a variety of interactive visualization methods were designed to adaptively partition and evaluate the data segmentation.Then,a variety of auxiliary views were designed to help users deeply understand the model parameters and optimization process,and support the iterative optimization of the model;Finally,a variety of evaluation indicators were introduced and combined with multi-sided visual views to evaluate the model from multiple perspectives.On the basis of the above research,the historical data of steam turbine regulating valve in real thermal power control was used to do case analysis and invite several experts to iteratively optimize the Pl Re VA.The effectiveness and practicability of Pl Re VA were confirmed by test validation of model results and evaluation feedback from domain experts.The work of this thesis provides a potential support method for regression modeling and iterative analysis of complex industrial control data.