Pattern Classification Model And Multi-modal Fuzzy Adaptive Control Of Heating Process In Annealing Furnace

As one kind of the typical high-end steel,the cold-rolled strip has increasing demand in recent years.The annealing heating process is an important step in the production of the cold-rolled strip.It is a nonlinear system with multi-variables,much interference,large inertias and time delays.Thus,it is difficult to establish the model of such a process.The traditional control methods are hard to achieve stable and precise control of the strip temperature.Moreover,the state parameters such as steel type,strip size,and strip transmission speed change frequently,resulting in complex patterns.The controllers aiming at single pattern are difficult to ensure the long-term stability of the strip quality.The mutli-mode intelligent controllers switch the controllers according the change of patterns,causing large switching disturbance.Moreover,there is no adjustment mechanism and strategy of the parameters of the intelligent controllers in a specific pattern with small range fluctuation.Therefore,according to the demand for high precise control of strip temperature in the annealing heating process,pattern classification model and multi-modal fuzzy adaptive control method are studied.In this research,the new design methods of data-driven intelligent control are provided to satisfy the requirements of the annealing heating process under complex patterns.The main research results and innovations of this thesis are as follows:(1)Fuzzy-subset-based pattern classification model for annealing heating processAiming at the problems of unintuitive description of patterns,low efficiency of classification method,and complexity of the classification model,we proposes a fuzzy-subset-based pattern classification model for the annealing heating process.This thesis describes the patterns with fuzzy subsets to intuitively show the differences and relations between various patterns.In order to mine the pattern fuzzy sets efficiently,we reduce the feature parameters after feature extraction and adopt a fast clustering algorithm based on density peak without repeated learning to mine the cluster centers.The pattern fuzzy subsets are established by Gaussian membership functions.This thesis also develops a method to simplify the classification model and improve the interpretability.The similarity and coverage measure methods of fuzzy sets are proposed to detect the redundant fuzzy subsets.The pattern classification model can be simplified effectively through fuzzy subset merging.This method can intuitively describe the patterns and classify the data efficiently.The simplification of the pattern classification model improves its interpretability.(2)A fast mining method of data-driven fuzzy rulesAiming at the difficulty and low efficiency of mining fuzzy rules,this thesis proposes an improved Wang-Mendel(WM)method based on conflicting rules and a fuzzy rule mining method based on outlier detection.Considering the influence of conflicting rules on fuzzy output accuracy caused by the complex data distribution in annealing heating process,we propose an improved Wang-Mendel(WM)method based on conflicting rules.On the basis of fuzzy partition,rule support degrees are calculated.Conflicting rules with high support degrees are selected to construct output fuzzy subsets,improving the accuracy of fuzzy rules.We also propose a fuzzy rule mining method based on outlier detection to suppress the influence of outliers.The Gaussian distribution functions of several groups of output data are established on the basis of data division by fuzzy partition.Then the weights of data is obtained to be utilized in calculation of output fuzzy subset centers.Therefore,the outlier detection in multi-dimensional input-output space is transformed into the outlier detection in one-dimensional output space,which simplifies the problem of outlier detection and improves the robustness of fuzzy rules.(3)Variable universe adaptive fuzzy controller with self-tuning parameters of the annealing heating processSince the parameters fluctuate in a small range under a pattern of annealing heating process,the existing methods are not adaptive enough and parameters are hard to determine.Therefore,a variable universe adaptive fuzzy controller with self-tuning parameters is proposed.In order to meet the reference change in annealing heating process,we define a contraction-expansion(C-E)factor on the unknown universe.A proportional C-E factor with a sign factor and a piecewise proportional one are designed.An optimization problem of C-E factor parameters is established by using industrial data.The chaotic particle swarm optimization algorithm are used to optimize the parameters to realize self-tuning of C-E factor parameters.This method improves the flexibility and practicality of the variable universe adaptive fuzzy controller.The parameter self-tuning strategy is effective and the controller has good characteristics of short setting time,little overshoot,very small stable error,and strong robustness.(4)Model-free adaptive control method with disturbance compensation for strip temperatureA model-free adaptive control(MFAC)method with disturbance compensation is proposed to cope with the problems of large system variation,insufficient adaptation of controllers,large disturbance caused by controller switching,and the difficulty in rapid adjustment of strip temperature in the transition process of patterns.In order to track the changes of the system in real time,the equivalent local dynamic linearization data model is established to estimate the strip temperature.A model-free adaptive controller is designed to regulate the gas flow to realize rapid adjustment of strip temperature.In order to suppress the disturbance caused by the change of strip type or strip size,we establish a simplified strip temperature model based on heating efficiency for annealing heating process.Combined with the MFAC model and the controller,the disturbance can be estimated and compensated.The proposed method realizes the rapid adjustment of strip temperature and accelerates the transition process of patterns.The disturbance compensation effectively reduces the strip temperature control error and improves the strip eligibility ratio.In this thesis,a complete multi-mode fuzzy adaptive control scheme based on data is designed by studying the pattern classification method of annealing heating process,the variable universe adaptive fuzzy control method for a single pattern and the model-free adaptive control method of transition process of patterns.This thesis provides a new data-driven intelligent control method to solve the the annealing heating process control problem under several patterns.