钢铁生产过程能耗预测与调度优化研究

【作者】 董人菘；

【导师】 王华； 余正涛；

【作者基本信息】 昆明理工大学 ， 生产过程物流学， 2014， 博士

【摘要】 钢铁生产流程是一个典型的高温、离散和连续混合的物理化学变化过程,具有多因素、多工序、多工位、强耦合、非线性等特点。高能耗是钢铁冶炼生产的主要问题,开展节能降耗不仅有利于节约生产成本,也有助于实现低碳加工。钢铁生产物流能耗预测,可以为钢铁企业制定能源总体规划方案提供支撑作用；在钢铁生产过程中,合理高效的生产调度对降低能耗和物耗,提高产品的品质,加快循环生产,降低生产成本有着极其重要的意义；此外,对钢铁生产过程中进行在线检测,及时发现故障,为避免浪费,节约能耗和物耗成本有着重要作用。本文将围绕钢铁生产过程的物流能耗预测,生产过程调度优化和生产过程故障预测等关键问题开展研究。主要完成了以下研究工作：(1)针对钢铁生产流程中能耗预测模型建立困难、预测精度较低等问题,提出了一种基于蚁群优化的小波神经网络的钢铁生产物流能耗预测模型,首先对钢铁生产过程以及影响生产能耗的因素进行分析,确定输入参数构成特征空间,然后利用小波变换重构特征空间,接着利用神经网络模型建立能耗预测模型,最后采用蚁群算法对求解过程进行优化。在炼铁、炼钢以及轧钢工序的能耗预测实验表明,提出的方法具有较好的普适性的同时,提高了预测精度,为钢铁企业提前了解能耗需求提供了指导。(2)针对钢铁生产过程中,炼钢-连铸关键工序存在多目标、多约束和不确定性等调度难题,提出一种基于遗传粒子群算法的炼钢-连铸过程的调度优化算法,首先炼钢-连铸生产工艺的约束条件下,建立以节能降耗、质量可靠、生产有序的条件下的最小化加工时间为优化目标的调度优化模型,利用粒子群算法收敛速度快与遗传算法全局搜索能力强的特点进行优化设计和参数求解,建立优化调度模型。实验结果表明,该算法是一种有效的炼钢-连铸生产调度优化方法,能够编制出实现连浇的可执行的炼钢作业计划,并且可以降低炉次在各工序间等待造成的损失,减小加工流程时间,从而达到降低能耗成本目的。(3)为了解决在轧钢过程中进行故障检测或诊断存在困难的问题,提出了一种基于多核学习理论的钢铁生产轧钢过程在线故障检测模型。首先针对学习样本建立核主元分析与支持向量数据域描述模型,然后基于T2、Q统计量,以及数据域描述包络情况对轧钢过程进行初步识别,最后采用基于多分类多核最小二次支持向量预测模型对初识结果进行细分类,识别故障级别。本文利用上述模型对轧钢加热炉故障和机组故障进行了试验,结果表明,该方法能有效检测钢铁生产轧钢过程的故障。更多还原

【Abstract】 Steel production process is a typical high temperature, discrete and continuous mixing process of physical and chemical changes, with multi-factor, multi-process, multi-position, strong coupling and nonlinear characteristics. High energy consumption is the main problem in the process of steel smelting production. Energy conservation and consumption is conducive not only to reduce production costs, but also to achieve a low-carbon processing purpose. Prediction of energy consumption in iron and steel production logistics can provide support for the iron and steel enterprises to formulate an energy overall plan; In the process of steel production, rational and efficient production scheduling has an extremely important significance for reducing energy and material consumption, improving product quality, accelerating the production cycle and reducing production costs; In addition, it plays an important role to online testing, failures detecting, avoid wasting and save energy and material consumption cost in the process of steel production. This dissertation covers research on prediction of logistics energy consumption in the process of steel production, scheduling optimization in production process, failures prediction in production process and so on. Mainly including the following three aspects:(1) For the issues, such as it is difficult to establish energy forecasting model and prediction accuracy is low, this dissertation proposes a model of steel production logistics and energy prediction based on ant colony optimization wavelet neural network. At first, we analyze steel production process and the factors that impact production energy, determine the input parameters constitute feature space. Then reconstruct feature space using wavelet transform, model energy consumption prediction with neural network model. Finally optimize the solution process using ant colony algorithm. The energy consumption prediction experiments in iron-making, steel-making and rolling process show that the proposed method has better universality, at the same time it improves the prediction accuracy and provide guidance for the steel enterprises understanding energy needs in advance.(2) For the problems that in the process of steel production, there exists multi-objective, multi-constrained and uncertainty in key process of steelmaking continuous casting, we propose a scheduling optimization algorithm based on genetic particle swarm optimization in steel-making and continuous casting process. Firstly, under the constraint condition of steel-making and continuous casting production process, establish scheduling optimization model to minimize the processing time of the optimization target with this understanding of energy saving, reliable quality and production ordered. Then, utilize the characteristics of fast convergence of particle swarm and global search ability of genetic algorithm to optimize the design and solving parameters. And finally, establish optimal scheduling model. Experimental results indicate that the proposed algorithm is an efficient scheduling optimization method in steel-making and continuous casting production, which is capable of compiling the executable steel-making operation plan to achieve continuous casting. And it can also reduce the loss which was made by the oven waiting time between the various processes; reduce the processing flow time, to reduce energy costs.In order to solve the difficult problems of failures detection or diagnosis in rolling process, we propose a online failure detection model in rolling steel production process based on multi-core learning theory. First, for learning samples, build a nuclear principal component analysis and support vector data description model. And then, preliminary identify the rolling process based on T2, Q statistic and data domain description envelope. Finally, make fine classification for the first results based on multi-core minimum secondary support vector prediction model, and identify the failure categories. We utilize the above model to do experiments for rolling mill furnace failures and testing unit failures, and the results show that the method can effectively detect failures in the process of rolling steel production.更多还原