带钢卷取温度高精度预报及多目标优化控制策略研究

发布时间：2018-04-08 11:29

本文选题：层流冷却　切入点：遗传算法　出处：《北京科技大学》2016年博士论文

【摘要】：在现代钢铁工业中,层流冷却工艺是通过轧后强制水冷来改善带钢的组织性能,提高带钢质量和产量的过程。带钢在层流冷却过程中发生复杂的水冷、空冷换热及内部的热传导过程,具有工况条件变化剧烈、强非线性、参数时变、数学模型难以精确描述的复杂工业特性,而且整个冷却区的恶劣环境不能逐点安装温度检测仪表,带钢温度难以连续检测,现有的控制方法存在不能适应变化频繁的工况条件、过于依赖带钢温度模型精度的问题,导致卷取温度控制精度不高、对给定冷却速率跟踪效果差。本文以某钢铁公司带钢热连轧生产线的层流冷却过程为研究对象,以提高带钢成品质量为目标,从温度预报模型优化和多目标优化控制策略研究两方面入手,将先进控制理论和改进的优化算法引入到生产实际中,提出了基于再进化遗传算法的相关性剪枝法(Re-evolutionary Genetic Algorithm-Correlation Pruning Algorithm,REGA-CPA)优化的BP神经网络卷取温度预报模型和基于转基因多目标遗传算法(Transgenic Multi Objective Genetic Algorithm, TMOGA)的层流冷却优化控制策略,并利用层流冷却过程实际生产数据进行了仿真实验研究,仿真结果验证了所提出温度预报模型的高精度和多目标优化控制策略的有效性。本文研究工作具体表现在以下几个方面：1)再进化遗传算法(REGA)现有诸多改进遗传算法(Genetic Algorithm,GA)终究只是在种群的正常进化过程中所采取各种策略,在设计理念上明显受到自然界生物自然进化思想的束缚,对由于种群进化过程中的盲目性、随机性而引起的退化现象明显应对措施不足,对克服GA收敛速度慢和易陷于局部最优等缺点的效果终究有限。基于此,本文在进化策略上另辟蹊径,提出了一种基于重新进化思想的REGA。其中,首次提出了重新进化的思想,用“返祖”操作找回丢失的较优模式并将其耦合至下一代种群中,极大的提高了算法的收敛速度；分析了“种群解的空间跨度”和“基因段距离”对种群多样性的影响,用“优生”操作来推动算法从平面到多维空间的立体式搜索,以勘探和挖掘出更广、更优的寻优区间,并在种群进化后期,强力驱动算法收敛于全局最优.2)基于REGA-CPA优化的BP神经网络卷取温度预报模型本文提出了一种基于REGA-CPA优化的BP神经网络卷取温度预报模型,“阶段性跨度淘汰法”主要是从保持种群多样性方面考虑,随时考量整个种群在平面空间的分布均匀性,以拓展搜索空间,使算法能够在更广、更优的区域寻优；“DNA鉴定法”从多维空间来考量种群在全局空间的寻优遍历性,为判断任意两个个体在多维空间的距离提供了直观、高效的方法。仿真结果表明：该卷取温度预报模型的收敛速度快、精度高,满足实时在线的控制要求,预报精度在±10℃范围之内,3)“随机动态输入模式”卷取温度预报模型的在线应用在离线方式下训练好的基于REGA-CPA优化的BP神经网络卷取温度预报模型为主模型,即可应用于在线的卷取温度预报。鉴于层流冷却系统是一个强耦合、强非线性、大滞后且滞后的时间时变的系统,因主模型权值、阈值、结构已固定,在线预报卷取温度时,若干点的精度有时可能会低于离线时训练的精度。针对此问题,提出了“随机动态输入模式”卷取温度预报模型,以最大限度的保证在线温度预报模型的预报精度在±10℃范围以内,能为层流冷却的预设定及前馈控制提供可靠的参考数据,从而为进一步提高卷取温度的控制精度提供了新的途径。4)转基因多目标遗传算法(TMOGA)提出了TMOGA,利用历代种群Pareto前沿面的交集来提取较优模式并建立基因库,库中的优秀基因通过“转基因”的方式移植到下一代种群,以保证种群进化稳步向Pareto最优解集迫近；基于决策变量的拥挤距离策略和基因库的竞争机制,保持了种群的多样性,使算法可以挖掘和勘探出更广、更优的搜索空间；随机抽取基因的模式保证了历代种群Pareto前沿面均具有良好的空间分布均匀性；基因库的记忆、固化功能形成强力驱动机制,使算法接近收敛时迅速跳出局部前沿,快速逼近真实的Pareto最优解集。5)基于TMOGA的层流冷却系统粗调区优化控制策略针对如何提高带钢卷取温度的控制精度和如何准确跟踪给定冷却速率的问题,提出了基于TMOGA的层流冷却系统粗调区优化控制策略,用于搜索粗调区集管的最佳开闭模式集合(Pareto最优解集)；仿真结果表明,该多目标优化控制策略可获取全局Pareto最优解集且在空间分布均匀,所提供的决策变量丰富、合理,因此控制系统的控制范围广、精度高,对多目标的均衡能力强,从而为新钢种的开发、冷却工艺优化提供了强有力的技术手段,同时为发展高端、高附加值的带钢产品打下了坚实的基础。
[Abstract]:In modern iron and steel industry, the laminar cooling process is through forced water cooling after rolling to improve the microstructure and properties of steel strip, improve the quality and yield of strip. The occurrence of complex water in the laminar cooling process, air heat transfer and internal heat conduction process, with dramatic changes in working conditions, strong nonlinear, parameter time varying, complex industrial characteristic mathematical model difficult to describe precisely, and the cooling zone of the harsh environment can not point installed temperature measuring instrument, the strip temperature control method to continuous detection, the existing can not adapt to the frequent changes in working conditions, the accuracy of the model is too dependent on the strip temperature problem, cause the volume is not high accuracy temperature control for a given cooling rate, tracking effect is poor. Based on the laminar cooling process of hot strip rolling production line of a Steel Corp as the research object, in order to improve the product quality for the purpose of a strip The subject, starting from the optimal temperature prediction model and multi-objective optimization control strategy based on two aspects, the optimization and improvement of the advanced control theory into the actual production, this paper presents a correlation pruning method evolved based on genetic algorithm (Re-evolutionary Genetic Algorithm-Correlation Pruning Algorithm, REGA-CPA) BP neural network optimization model and prediction of coiling temperature transgenic based on multi-objective genetic algorithm (Transgenic Multi Objective Genetic Algorithm, TMOGA) of the laminar cooling control strategy optimization, and using the laminar cooling process of actual production data are simulated. The simulation results verify that the proposed temperature prediction model with high precision and multi objective optimization of the effectiveness of the control strategy. The specific performance of this research work in the following aspects: 1) and genetic algorithm (REGA) to the much improved genetic algorithm (Ge Netic Algorithm, GA) are only in the normal population evolution process in the various strategies in the design concept was constrained by natural biological evolution thought, due to the blindness of the evolutionary process, the randomness caused by the degradation of inadequate measures, is limited to overcome the slow convergence speed of GA and easy to fall into local optimum effect. Based on this, this article on the evolution strategy and put forward a kind of the re evolution based on REGA. is proposed for the first time to the evolution of thought "is a good method to retrieve the lost progenitor return operation which is coupled to the next generation, great to improve the convergence speed of the algorithm; analyze the impact of" space "population solution and gene segment distance on population diversity, with" eugenics "operation to promote the algorithm from the plane to The three-dimensional space to search, exploration and mining more widely, better searching interval, and the population in the late stage of evolution, strong drive algorithm converges to the global optimal.2) BP optimization REGA-CPA neural network prediction of coiling temperature model this paper proposes a BP REGA-CPA neural network optimization based on coiling temperature based on the prediction model, the stage span elimination method "is mainly considered from the diversity of the population, the population distribution of any consideration in the plane space uniformity, to expand the search space, the algorithm can in a wider area, seeking better;" DNA identification method "from the multidimensional space to consider in population the global space searching ergodicity, for the judgment of any two individual distance in the multidimensional space provide an intuitive and efficient method. The simulation results show that the coiling temperature prediction model has fast convergence speed and high precision, real-time Online control requirements, forecast accuracy within the range of - 10 DEG C, 3) "random dynamic input mode volume online temperature prediction model for application in offline BP optimization REGA-CPA neural network prediction of coiling temperature model based on the model of training, can be applied to the online prediction of coiling temperature in laminar flow. The cooling system is a strong coupling, nonlinear, large delay systems and the lag time variable, because the main model weights, threshold, fixed structure, on-line prediction of coiling temperature, some accuracy may sometimes be below the line from the training precision. Aiming at this problem, put forward the" random dynamic the input mode of coiling temperature prediction model, in order to guarantee the maximum online prediction model for temperature prediction accuracy within the range of - 10 DEG C, can provide reliable reference data for pre setting and feedforward control in laminar cooling, In order to further improve provides a new way for.4 the control accuracy of the coiling temperature) transgenic multi-objective genetic algorithm (TMOGA) proposed by TMOGA, the population Pareto frontier intersection to extract the optimum model and the establishment of gene pool, good genes in the library through the "transgenic" way to transplant to the next generation the population, in order to ensure the evolution of population steadily approaching to the Pareto optimal solution set; decision variable crowding distance strategy and competition mechanism based on gene library, keep population diversity, the algorithm can mining and exploration of a broader, better search space; random gene model to ensure the population were Pareto frontier has a good spatial distribution uniformity; gene library memory, curing function to form a strong driving mechanism, the algorithm is close to convergence quickly jump out of the local frontier, fast approaching the true Pareto optimal solutions. Set.5 TMOGA) coarse area optimization control strategy on how to improve the accuracy of the coiling temperature and how to accurately track the given cooling rate of laminar cooling system based on the proposed coarse region optimization control strategy of laminar cooling system based on TMOGA, search for the coarse set tube opening and closing mode set (the best the Pareto optimal solution set); the simulation results show that the control strategy can obtain the global optimal solution set of Pareto and the spatial distribution of the uniform multi-objective optimization, the decision variables supplied by the rich, reasonable, so the control system wide control range, high precision, strong ability to balance multiple objectives, so as to develop new steel grades. It provides powerful techniques for optimization of cooling process at the same time, the development of high-end, high value-added steel products to lay a solid foundation.

【学位授予单位】：北京科技大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TG334.9;TP18

【相似文献】