节点文献
面向生产目标的铅锌烧结过程建模及优化研究
Modeling and Optimization of Lead-zinc Sintering Process for Product Index
【作者】 丁雷;
【导师】 吴敏;
【作者基本信息】 中南大学 , 控制科学与工程, 2009, 博士
【摘要】 作为铅锌火法冶炼的一个重要流程,密闭鼓风烧结过程所生产的烧结块直接影响到后续熔炼炉炉况和铅锌的产量质量指标。烧结生产过程中综合透气性、烧穿点是两个最重要的过程状态参数,直接反映出烧结过程进行的好坏。针对影响综合透气性、烧穿点的信息具有不确定性、不完整性等特点,分别从不同的角度出发进行描述,并建立相应的集成预测模型。在研究烧穿点和透气性的基础上,提出了烧结块的产量质量预测模型。为使得烧结过程运行在最优状态,将整个烧结过程的优化分为产量质量优化和状态参数优化两大部分。首先提出针对产量质量的优化算法,以获得过程状态参数的优化设定值。接着提出针对过程状态参数的优化算法,以获得过程操作参数的优化设定值。论文的主要研究成果包括:(1)过程状态参数的集成预测模型首先考虑到历史数据中含有工况稳态发展趋势的有用信息,结合综合透气性的变化特点,提出具有在线修正能力的灰色理论预测模型,其根据实际综合透气性数据序列的单调性变化情况及变化后的数据个数,给出相应的修正公式;接着考虑到工艺参数预测模型能及时反映工艺参数对工况的影响,建立基于工艺参数的神经网络预测模型;最后考虑到信息熵技术具有降低不确定性带来的影响以及信息融合的能力,建立基于信息熵技术的综合透气性集成预测模型。针对实际烧结过程烧穿点不能被直接测量的问题,建立烧穿点的软测量模型。考虑到烧穿点的波动比较大,利用T-S模糊模型来建立其预测模型,以充分有效利用历史信息;为降低不确定性带来的影响以及降低建模的难度,将综合透气性指数、台车速度和中部风箱废气温度作为影响烧穿点的主要因素;同样在建立工艺参数的预测模型之后,利用信息熵技术建立烧穿点的集成预测模型。(2)产量质量智能预测模型结合机理分析和数据关联性分析,选取过程状态参数以及一些关键的原料参数作为输入变量,并建立烧结块产量质量的神经网络预测模型。针对产量检测周期和过程状态参数的采样周期不一致,存在多尺度问题,引入了空间积分的思想,解决了时间周期不匹配的问题。针对传统神经网络训练算法是基于梯度信息的、具有容易陷入局部极值、收敛速度慢的缺点,将粒子群算法强大的全局搜索能力和共轭梯度法强大的局部搜索能力有机结合起来,提出了一种基于共轭梯度法的混合粒子群算法来进行神经网络的训练。(3)基于混合粒子群算法的产量质量优化建立了以提高产量为目标,以质量要求为约束条件的优化模型,并提出了基于改进线性搜索法的混合粒子群算法来实现产量质量的优化。该算法首先将所建立的优化函数转化为一个有两个目标的优化问题,一个是原目标函数,另一个是约束条件违反程度函数;接着为达到并行优化的目的,采用约束水平比较法来比较粒子群算法所搜索到的解;最后当粒子群算法收敛停滞时,通过引入改进的线性搜索法来提高粒子群算法的活性,从而获得综合透气性、烧穿点的优化设定值。(4)基于多目标粒子群协同算法的过程状态参数优化建立了以综合透气性、烧穿点等状态参数的优化设定值为目标,以生产边界条件及质量要求为约束条件,以过程操作参数为优化变量的多目标优化模型。针对该模型提出了一种多目标粒子群协同优化算法,采用改进的粒子极值选取法,以及多粒子群合作优化的方式,获得过程操作参数的优化设定值。最后根据实际工业过程的特点,进行了烧结过程的优化仿真实验。实验结果表明所提出的优化方法能在一定程度上提高烧结块的产量质量。因此所提出的烧结过程优化方法为以后烧结生产的全流程优化提供了技术手段,同时也为复杂工业过程的建模及优化提供一种实用的、值得借鉴的实现方法。
【Abstract】 The sintering agglomerate of the imperial updraft sintering process, which acts as an important flow process of Lead-Zinc pyrometallurgical smelter processes, has a direct influence on the subsequent states of the smelting furnace and the indexes of quantity and quality of Lead-Zinc. The synthetic permeability and the burn-through-point (BTP) are the two most important parameters of process states, and could directly reflect the performance of the sintering process. Considering that the information influencing the synthetic permeability and the burn-through-point (BTP) features uncertainties and incompletion, the thesis describes them from different aspects respectively, in order to build the corresponding integration prediction models for the synthetic permeability and the BTP.The prediction models, which are based on the study of the BTP and the synthetic permeability, for the quantity and the quality of the sintering agglomerate are proposed. To obtain good performance of sintering process with optimal states, the optimization of sintering process is divided into two parts. The one is the optimization of quantity and quality, and the other is the optimization of process states. First, to obtain the optimal set values of the process states, the optimization algorithm for the quantity and quality is devised. Then, to obtain the optimal set values of the operation parameters of the sintering process, the optimization algorithm for the parameters of process states is presented. The main achievements are as follows:(1) Integrated prediction models for parameters of the process statesFirst, the prediction model for the synthetic permeability based on the gray theory with the capability of modification on line is proposed, since the history data preserves the effective information that reflects the operation trend of future steady states. The corresponding formulas are proposed according to the monotonic changes of the datum sequence of the synthetic permeability and the data numbers after these changes. Then, the technological-parameter-based prediction model using neural networks for the synthetic permeability is established, which can timely reflect the influence brought by the technological parameters. Finally, the information entropy technology is applied to establish the integrated prediction model for the synthetic permeability, due to its abilities of decreasing the uncertainties and fusing the information.The soft measurement model for the BTP is developed, to solve the problem that the BTP can’t be measured directly: Considering that the fluctuation of BTP is relatively big, the T-S prediction model for the BTP is presented to make full use of the history information. To decrease the effects produced by the uncertainties and lower the difficulty of modeling, the synthetic permeability, the strand velocity and the gas temperature of middle bellows are selected as the main factors impacting the BTP. Similarly, the prediction model based on neural networks is present. And the information entropy technology is applied to integrate the prediction models mentioned above.(2) Intelligent prediction models for the quantity and quality of the Lead-Zinc sintering agglomerateThe parameters of process states and some key parameters of raw material are chosen to build the prediction models for the quantity and quality of the sintering agglomerate using neural networks. The problem of the different time scales between the measurement period of quantity of the product and the sample period of parameters of process states is solved using the spatial integral method. To avoid the slow convergence and easily falling into the local optima of BP training algorithm based on gradient information, a hybrid particle swarm optimization (PSO) is employed to train the neural networks, which integrates the global search ability of the PSO and the powerful local search ability of the conjugate gradient algorithm.(3) Optimization of quantity and quality of the sintering agglomerate based on hybrid PSO algorithmAn optimization model with the constraint conditions of quality requirements is established to enhance quantity, and a hybrid PSO algorithm based on the improved line searching algorithm is presented to realize the optimization of quantity and quality. First, the optimization model is converted to a two-objective optimization problem, one of them is the origin objective function, and the other is the degree function of constraint violation. Then, to realize the parallel optimization, a constraint comparison method is applied to compare the searched results of the PSO. Finally, to handle the problem of premature convergence frequently appeared in the PSO, an improved line searching algorithm is introduced to maintain the particle activation when PSO stagnates. So the optimal set values of the synthetic permeability and the burn-through point could be achieved by using the hybrid PSO algorithm.(4) Optimization of parameters of the process states based on multi-objective particle swarm cooperative optimization algorithmWith the optimal set values of the synthetic permeability and the BTP as the optimization targets, the production frontier and quality requirements as the constraint conditions, and the parameters of process states as the optimization valuables, a multi-objective optimization model is developed. It is optimized by a multi-objective particle swarm cooperative optimization algorithm, which is improved using a new principle of selecting the particles’ optima, and multiple swarms optimizing in a cooperative way, so that the optimal set values of process operation are obtained.Finally, the simulation experiments of the optimization of sintering process are implemented according to the characteristics of the practical production process. The results of experiments show the presented optimization method in the thesis improved the quantity and quality of the sintering agglomerate to some extent. So the presented optimization method provides technical means for the optimization of the whole sintering production process, and yields a practical and effective method for modeling and optimization of the complex industrial process.
【Key words】 Lead-Zinc sintering process; synthetic permeability; burn-through-point; grey theory prediction model; technological-parameter-based prediction model; information entropy technology; hybrid particle; swarm optimization algorithm; particle swarm cooperative optimization algorithm;