【作者】 丁进良；

【导师】 柴天佑；

【作者基本信息】 东北大学 ， 控制理论与控制工程， 2012， 博士

【摘要】 我国赤铁矿资源丰富,但其品位低、磁性弱、嵌布粒度细、矿物组成复杂,难以选别,因此采用焙烧-磁选的工艺进行选别。赤铁矿选矿过程主要包括竖炉焙烧、磨矿和磁选工序将有用矿物和脉石分离,使得有用矿物成分富集,从而获得品位合格的精矿和尾矿。选矿过程运行优化控制是根据竖炉磁选管回收率、磨矿粒度和磁选精矿品位与尾矿品位等运行指标的目标值确定各过程控制系统设定值,并使被控变量跟踪设定值,从而将运行指标控制在目标范围内,并且尽可能提高磁选管回收率、磨矿粒度和磁选精矿品位,尽可能降低尾矿品位。上述运行指标的目标值由选矿生产全流程的综合精矿品位和产量等综合生产指标来确定。因此,运行指标的优化决策对提高综合精矿品位和产量具有重要意义。选矿生产全流程运行指标决策要使综合精矿品位和产量指标在目标值范围内,而且使综合精矿品位和产量尽可能高。运行指标和综合精矿品位与产量等综合生产指标之间不仅涉及到过程机理,而且与具体工艺因素相关,机理不清,难以用精确的机理模型描述。选矿生产过程的外部环境和内部干扰动态不确定,导致选矿生产全流程运行指标决策的综合生产指标(综合精矿品位和产量等)目标与范围、约束条件边界(如原料成分波动范围、磨机等设备故障引起的运行时间变化以及最大处理能力等)的频繁变化。因此,运行指标优化决策是难以建立模型的多目标非线性动态优化问题,难以采用现有的优化方法实现运行指标优化决策。实际生产中运行指标决策往往由工艺工程师凭经验进行。由于人工调整不当或不及时常常不能保证生产全流程的综合生产指标在其目标范围内,难以实现全流程的优化运行,从而造成产品质量差、能耗高、资源消耗大等问题。因此,研究如何在动态环境下及时有效的对选矿生产全流程运行指标进行优化决策,从而实现生产全流程的全局优化具有重要理论研究意义和应用价值。本文针对上述问题,依托国家973计划项目课题“复杂生产制造全流程一体化控制系统整体控制策略与运行控制方法”和国家自然科学基金青年基金项目“动态环境下复杂工业全过程多工序工艺指标闭环优化决策方法”,开展选矿生产全流程运行指标优化决策方法的研究,主要研究工作如下：1)给出了选矿生产全流程运行指标决策问题的数学描述。其中,给出了以运行指标磁选管回收率、强、弱磁磨矿粒度、强、弱精矿品位和强、弱尾矿品位,以及原矿性质与生产工况条件强、弱磁入磨品位、强、弱磁球磨机台时处理量、强、弱磁球磨机运行时间和废石品位等可测干扰作为输入,综合精矿品位和产量输出的性能指标的描述；决策变量为上述七个运行指标,约束条件包括综合精矿品位和产量的上下限范围、各运行指标上下限范围、磨机的最大设备处理能力与原矿品位的下限等资源的约束,以及运行指标和综合生产指标之间关系的等式约束；运行指标决策的优化目标为综合精矿品位和产量指标在其目标范围内,并且尽可能的高。并且对上述运行指标优化决策问题难以采用已有优化方法的难点进行了分析。2)针对选矿生产全流程运行指标多目标优化决策、磁选管回收率与磨矿粒度等运行指标和综合精矿品位与产量等综合生产指标之间难以用精确的机理模型描述和优化决策目标与约束条件动态变化的问题,将优化方法与综合生产指标预报、运行指标的动态校正相结合,提出了由运行指标初值优化、综合生产指标(综合精矿品位和产量)指标预报模型、指标前验和后验评估与动态校正组成的运行指标优化决策的结构。将多目标进化算法与案例推理相结合,选矿运行过程数据与专家知识相结合,动态校正与规则提取相结合,提出了运行指标初值优化、综合精矿品位和产量预报模型、指标前验和后验评估与动态校正模块的设计方法。3)提出了由线性主模型和非线性误差补偿模型组成的综合精矿品位和产量预报模型的混合建模策略。其中非线性误差补偿模型采用最小二乘支持向量机(LS-SVM)来建立。针对具有非高斯干扰的过程,模型参数估计中均值和方差性能指标不适用的问题,将随机控制系统概率密度函数PDF控制的思想引入参数选择问题的性能指标中,即通过使模型输出误差概率密度函数跟踪一个给定的分布形状来调整模型内部可调参数来保证模型的精度。采用上述方法选择LS-SVM非线性误差补偿模型的参数,建立了以运行指标为输入的综合精矿品位的预报模型。另外,针对不同的生产工况条件对精矿产量的影响,提出了基于多模型的精矿产量预报模型。利用实际生产数据,进行了仿真实验研究,验证了所提出方法的有效性。4)提出了基于粗糙集规则挖掘的运行指标动态校正方法。首先,根据选矿过程记录的运行数据特征,构建了综合精矿品位和产量偏差与运行指标增量间的规则形式,然后利用基于粗糙集的规则挖掘从大量实际生产数据中挖掘出补偿规则。当综合精矿品位和产量实际值或预报值和其目标值发生偏差时,利用所挖掘的增量规则对运行指标进行补偿校正。通过反馈补偿的仿真实验验证了所提方法的有效性。5)利用选矿过程的实际生产数据,在实验室研发的选矿生产全流程运行指标优化决策的半实物仿真平台上开展了实验研究。采用所提出的方法,开展了当综合精矿品位和产量目标值和范围变化,同时原料成分如入磨原矿品位、因故障等原因引起的磨机运行时间、磨机处理量三个边界条件发生变化时,运行指标决策的实验。实验结果表明,在目标值与范围和边界条件变化时,系统能够及时对运行指标进行决策,从而使综合精矿品位和产量在目标范围内,与人工决策相比,磁选管回收率提高2%,强、弱磁磨矿粒度分别提高1.49%和1.98%,强、弱磁精矿品位分别提高0.57%和1.26%,尾矿品位分别降低0.31%和0.67%,使得全流程的日综合精矿品位和产量分别提高0.57%和132.37t/d,实现选矿生产全流程的优化。更多还原

【Abstract】 Although there are plenty of hematite resources in China, they are all difficult to be separated for the nature of low grade, weak magnetic, disseminated, complex mineral composition. Therefore, the roasting and magnetic separation technology is employed. Hematite beneficiation process consisting of a shaft furnace roasting unit, grinding unit and magnetic separation unit will separate the useful mineral and gangue and enrich the useful mineral composition, and thus produces the overall concentrate with qualified grade and tailings. Optimal operational control of a benefication process is to generate the setpoints of control systems of each unit according to their operational indices such as magnetic tube recovery rate, particle size, concentrate and tailings grade, etc. The control systems force the controlled variables to follow up the setpoints so as to control the operational indices into their targeted ranges, at the same time, improving the magnetic tube recovery rate of shaft furnace, particle size and concentrate grade as high as possible, decreasing the tailings grade as low as possible. In fact, the above targeted ranges of operational indices are determined by the production indices, i.e. overall concentrate grade and output, of the beneficiation process. Therefore, the optimal decision-making of operational indices is of great significant for improving the overall concentrate grade and output.The objective of decision-making of the operational indices is to make the overall concentrate grade and output into their targeted ranges and improve them as high as possible. The relationship between the operational indices and the production indices not only relates to the process mechanism, but also relates to the specific process technical factors. This fact leads to mechanism of the relationship is unclear. As a result, it is difficult to be described using exact mechanism model. The dynamic uncertainties, in terms of the external environment and internal disturbance of the benefication process, lead to frequent variations in the operational indices decision-making problem formulation in terms of the targeted ranges and limitations of production indices (overall concentrate grade and output), and the limitations of constraints (i.e. composition variation of raw ore, run time fluctuation and maximum processing capacity of grinding unit, etc.). Therefore, optimal decision-making of operational indices is a multi-objective nonlinear dynamic optimization problem for a process difficult to be modeled and it is difficult to solve using the existing optimization approaches. In practice, the decision-making of operational indices is usually carries out by the technical engineers with experience. As the improper or untimely manual adjustments often can not guarantee the production indices into their targeted ranges, it is difficult to achieve the optimal operation of the whole process, resulting in poor product quality, high energy consumption and resource consumption and other issues. Therefore, it has important theoretical significance and application value to carry out the research on how to realize the timely and effectively optimal decision-making of operational indices under dynamic environment in order to achieve global optimization of the whole production process.Subject to the above problem, suppoted by the973projects "the total control strategy and operational control approach for complex manufacturing processes" and the National Natural Science Foundation project "closed-loop optimal decision-making approach of technology index for complex industrial processes under dynamic environment", the research on the optimal decision-making of operational indices for benefication processes has been carried out. The detailed work has been summarized as follows:1) The mathematical formulation of the decision-making of operational indices for beneficiation process is presented. In this formulation, the performance is described taking the operational indices (the magnetic tube recovery rate, particle size of high-and low-intensity grinding unit, concentrate and tailings grade of high and low intensity magnetic separation unit) and the measurable disturbance such as the nature of raw ore and working condition (such as raw ore grade, capacity per hour and run time of high-and low-intensity grinding unit, and grade of waste ore, etc.) as the inputs, and the production indices (the overall concentrate grade and output) as the outputs. The decision variables are the seven operational indices. The constraints include upper and lower limitations of the overall concentrate grade and output and the operational indices, the lower limitation of the maximum processing capacity and the raw ore grade etc. and the equality constraints, i.e. the relationship between the operational indices and the production indices. The objective of optimal decision-making of operational indices is to make the overall concentrate grade and output into their targeted ranges and as high as possible. Moreover, the difficulties in solving the above problem using the existing optimization methods are analyzed.2) Subject to the problems that multi-objective optimal decision-making of operational indices of benification process, the relationship between the operational indices and the production indices difficult to be modeled using exact mechanism model, and variations on the objective targets and range and constraints, a structure of optimal decision-making is proposed combining the optimization, the prediction of production indices and the dynamic tuning of operational indices. The structure consists of four modules, namely optimization of initial value of operational indices, predictive model of production indices (overall concentrate grade and output), priori-and posteriori-evaluation of production indices and dynamic turning of operational indices. At the same time, combining multi-objective evolutionary algorithm and case-based reasoning, process operation data and expert knowledge, dynamic tuning and rule extraction, design approaches of the above four modules are proposed.3) The hybrid modeling strategy of overall concentrate grade and output is proposed which is composed of a linear model and a nonlinear error compensation model. The least-squares support vector machine (LS-SVM) is adopted to establish the nonlinear model. Subject to the problem that the mean and variance are unsuitable in model parameter estimation of processes with non Gaussian disturbance, the idea of probability density function (PDF) control is introduced into the performance of parameter selection problem. It is to turn model parameters so that the modeling error PDF is controlled to follow a target PDF to guarantee the model accuracy. The predictive model of overall concentrate grade is established adopting the above approach, where the inputs are the operational indices. Moreover, considering the effect of different production conditions on the concentrate output, the multi-model based prediction model of overall concentrate output is proposed. Finally, the simulation experiments are carried out using the process data collecting from real plant and the results show the effectiveness of the proposed approach.4) The dynamic tuning approach of operational indices based on rough set rule extraction is proposed. First, the form of increasement rules is constructed between the production indices and operational indices according to the characteristic of collected data. Then, the rules are mined from a large number operational data using rough set rule extraction. When the error between the actual/predictive value and the targeted value of overall concentrate grade and output occur, the mined rules are performed to compensate the operational indices. The simulation experiment of feedback compensation is carried out and the results show the effectiveness of the proposed approach.5) The experiment research of the proposed optimal decision-making of operational indices is carried out on the semi-physical simulation platform in our laboratory. Adopting the proposed approach, three experiments are carried out. These experiments validate that under the variation condition of raw ore grade, run time of grinding unit causing by fault and processing capacity of grinding unit, and at the same time the targets and ranges of overall concentrate grade and output are changing, the proposed system can produce the operational indices in time and make the production indices into their targeted ranges. Comparing with the manual adjustment, the proposed approach can force the magnetic tube recovery rate improved by2%; particle size of high-and low-intensity grinding unit increased by1.49%and1.98%, respectively; concentrate grade of high-and low-intensity magnetic separation unit raised by0.57%and1.26%, respectively and tailings decreased by0.31%and0.67%, respectively. Finally, the overall concentrate grade and output are enhanced by0.57%and132.37t/d, respectively, which means that the optimization of whole production line is realized.更多还原