Research On High-Precision Prediction And Multi-Objective Optimization Control Strategy Of Strip Coiling Temperature

In the modern iron and steel industry, the laminar cooling process is to improve the microstructure and properties of the strip, and to increase the quality and yield of the strip. In the laminar cooling process, complex water cooling, air cooling heat exchanging heat and internal heat conduction process take place in strip, which has the dramatic changes in the working conditions, strong nonlinearity, parameter time-varying, it also is difficult to accurately describe the complex industrial characteristics by mathematical model. And that temperature measurement cank’t be installed point due to the severe environment of the cooling zone, which makes it difficult to continuously measure strip temperature, there are some problems with the existing control methods that are inappropriate for the frequent change working conditions and rely on too much the accuracy of the strip temperature model, which leads to low accuracy on coiling temperature control and poor effect on tracking a given cooling rate.In this thesis, the laminar cooling process of hot strip production line of a Steel Corp is studied, the goal is to improve the quality of steel strip. The study is start with two aspects of temperature prediction model optimization and multi-objective optimization control strategy research, the advanced control theory and the improved optimization algorithm are introduced into the practical production, and then the thesis proposes the optimized BP neural network coiling temperature prediction model based on CPA-REGA (Correlation Pruning Algorithm based on Re-evolutionary Genetic Algorithm) and optimization control strategies of laminar flow cooling based on TMOGA (Transgenic Multi Objective Genetic Algorithm). The simulation experiments are carried out by using the actual production data of the laminar cooling process; the results verify the effectiveness of the proposed temperature prediction model with high accuracy and multi objective optimization control strategy. The research works of the thesis are specified in the following aspects:1) Re-evolutionary Genetic Algorithm (REG A)Existing many improved genetic algorithm (GA) adopt various strategies only in the normal process of population evolution, in the design concept, these methods are obviously constrained by natural biological evolution thought, have obvious lacks for resolving degradation phenomenon caused by the blindness and random in the population evolution and limit abilities of overcoming slow convergence and easily trapped in local optima of GA. Therefore on the basis of evolutionary strategy, the thesis puts forward REGA based on thought of Re-evolutionary which is proposed for the first time. The "atavism" operation is used to figure out the loss optimum model that is coupled to the next generation of the population, which improves the convergence rates in great deal. The thesis also analyzes impact of "space spans of population solution" and "gene segment distance" on population diversity. "Eugenics" operation pushes algorithms forward from the plane to multi dimensions forming spatial search, in order to explore and excavate a broader, superior optimization interval, and in the later evolution process, highly drives algorithms converging on the global optimal.2) Coiling temperature prediction model based on BP neural network optimized by CPA-REGAThe thesis proposes CPA-REGA to optimize the coiling temperature prediction model based on BP neural network, in which the "stage span elimination method" mainly considers from the diversity of the population, the population distribution uniformity in the plane space is taken into account at any moment to expand the search space that makes the algorithm search optimization in a wider and better area. Moreover, "DNA identification method" primarily takes consideration of searching optimization ergodicity of the population in the global space from the point of view of multidimensional space, which provides intuitive and efficient method to judge distance between any two individuals in multi dimension space. The simulation results show that the convergence speed and high precision of the coiling temperature prediction model can meet the requirements of real-time and on-line control, and the prediction accuracy can be kept within the range of ± 10℃.3) Online application of the coiling temperature prediction model based on "random dynamic input mode"BP neural network coiling temperature prediction model optimized by using CPA-REGA in the off-line is trained as the major mode, which namely is applied to the online temperature prediction. In view of the laminar cooling system is a strong coupling, strong-nonlinearity large time delay with delay time varying system, master model weights and threshold has been fixed, several points on the accuracy may sometimes will be lower than the off-line training accuracy when on-line prediction of coiling temperature is conducted. Aiming at this problem, the temperature prediction model based on the "random dynamic input mode" is put forward to maximize prediction accuracy of on-line temperature prediction model within the range of ± 10℃,which can provide reliable reference data for laminar cooling of presetting and feedforward control, thereby providing a new way for further improving the coiling temperature control accuracy.4) Transgenic Multi Objective Genetic Algorithm (TMOGA)The TMOGA uses the population ages Pareto frontier intersection to extract the optimum model and the establish gene bank in which excellent genes transplant to the next generation population by "transgene", which ensures the population evolution is steadily imminently to the Pareto optimal solution set. Congestion distance strategy based on decision variables and competition mechanism of gene bank keep the population diversity, which makes the algorithm mine and explore a wider and better search space. Randomly selected gene model guarantees the successive population Pareto frontier has good spatial distribution uniformity. Gene bank memory and curing function form a strong driving mechanism, so that the algorithm is close to convergence quickly jump out of the local edge, fast approaching the real Pareto optimal solution set.5) Optimization control strategy based on the TMOGA in the laminar cooling system coarse adjustment areaFor temperature control precision of coiling temperature and how to accurately track a given cooling rate, optimization control strategy based on TMOGA is proposed in coarse zone of laminar cooling system for search optimal on-off mode set of coarse block tube (the set of Pareto optimal solutions). Simulation results show that the multi-objective optimal control strategy can obtain the global Pareto optimal solution set and uniformly distribution in space, decision variables can be selected for the rich, reasonable. Therefore the control system has wide control range, high precision and strong balance capability between multi-objective, which thereby provides a powerful technical tool for the development of new Iron and Steel and optimization of cooling process,at the same time as laid a solid foundation for developing high-end and high value-added strip steel.