【作者】 刘颖；

【导师】 王伟；

【作者基本信息】 大连理工大学 ， 控制理论与控制工程， 2010， 博士

【摘要】 钢铁工业是我国国民经济的重要基础产业,提高其生产率、合理利用资源、节能减排是钢铁企业发展的关键。其中生产调度水平的提高是增加企业效益、降低生产成本的重要途径。而对钢铁企业能源运行情况的准确预测,也可达到减少能源损失,降低环境污染的目的。本文依托国家“863计划”课题,以上海宝钢股份有限公司为研究背景,针对罩式炉退火工序的优化调度问题和能源系统中副产煤气产消情况的预测问题,展开了深入系统的研究工作。本文主要研究内容如下：通过分析罩式炉退火工序调度过程的复杂性,研究了将解决大规模集成电路硬件建模问题的仿真平台SystemC应用于离散事件系统仿真问题的可行性,采用平台映射的思想建立了基于SystemC的罩式炉退火工序离散事件仿真模型,该模型能够快速有效的对退火过程进行模拟。在此基础上,将其与提出的自适应遗传算法相结合,解决罩式炉退火的作业优化调度问题。提出一种基于改进回声状态网络的时间序列预测方法,该方法通过预测器直接建立预测原点和预测时域之间的关系,不存在由于迭代而引起的误差积累问题,且预测器的稳定性可以预先保证,不依赖网络的训练结果。采取高斯过程回归方法求取网络的输出权值,替代了原始回声状态网络算法中线性回归的过程,克服了线性回归易导致病态解的问题,而且可以给出预测结果的概率估计,提高了预测精度。该方法为本文后续工作奠定了基础。研究了钢铁企业副产煤气系统产消量的实时预测问题。考虑到现场采集的过程控制系统流量数据中含噪较高的实际情况,提出一种基于经验模态分解的降噪方法。该方法将经验模态分解得到的小尺度固有模态函数经低通滤波器自适应去噪,再对数据重构以达到降噪的目的。在此基础上,对于煤气产生量和非调节用户煤气消耗量,采用本文前述提出的改进回声状态网络时间序列预测方法建立预测模型,并给出基于最小均方差准则的模型参数优化方法,在优化过程中采用奇异值分解方法求取网络输出权值。对于调节用户及转换用户采用平均值方法建立预测模型。选取现场生产数据对提出的方法进行了仿真验证,预测效果较好。基于上述的研究工作,结合软件工程的方法,开发了宝钢副产煤气预测系统。在上海宝钢股份有限公司能源中心试运行结果表明该系统可以对煤气发生源及各用户的产消量实时进行预测,并在此基础上对煤气柜柜位进行预测,预测精度较高,给出的建议调整方案具有一定的合理性,受到现场调度专家的认可。该系统达到了节约能源、减轻人工调度复杂度的效果。更多还原

【Abstract】 Steel industry is an important sector of national economy. The key issues to the development of iron and steel enterprises are about increasing the productivity, utilizing resources reasonably and reducing environmental pollution. It is well known that production scheduling is an effective way to make economic profits and reduce the production costs. Also, precisely forecasting energy generation and consumption can reduce energy waste and environment pollution. Based on our work concerning National High-Tech Research and Development Programs, this dissertation studies the optimal scheduling problem for bell-type batch annealing process and the prediction problem for generation and consumption of byproduct gas system. The research is mainly carried out at the Shanghai Baosteel Co. Ltd. The dissertation has mainly carried on the following research.Analyzing the scheduling complexity of bell-type batch annealing process and the feasibility to solve a discrete event system simulation by using SystemC originally treated as a simulation platform for designing very large scale integrated circuit, a discrete event simulation modeling based on SystemC for the batch annealing shop is proposed. The modeling method simulates the batch annealing process in a quick yet effective way. An optimal method of bell-type batch annealing production, combined with an adaptive genetic algorithm, is put forward in this dissertation.For the energy prediction problem, a time series forecasting method based on improved echo state network is proposed. This method can be used to establish a direct relationship between the prediction origin and prediction horizon, which can avoid the iteration error accumulation and guarantee the stability of the predictor in advance, rather than relying on the network training results. Also, instead of using linear regression, Gaussian process is adopted to obtain the relationship between the reservoir state and network output, thereby preventing the ill conditioned reservoir state matrix form occuring. This way, not only a better prediction result but also an accurate probability estimation of the result is achieved.The real-time prediction method is investigated for the generation and consumption of byproduct gas system in steel industry. Since the practical flow data in process control system typically includes a variety of noises, a noise reduction method is proposed based on the empirical mode decomposition. The main idea of this method is to decompose the time series signal into a group of independent intrinsic mode functions, and those with small-scale are de-noised by filter with adaptive threshold. Based on such method, a time series prediction procedure for gas generation and consumption of non-adjustable users is presented by exploiting an improved echo state network, where the network parameters are optimized based on the least mean square error criterion, and the output weights are obtained by singular value decomposition. In addition, an average method is used to predict consumption of adjustable users and converted users. Finally, the real production data from Baosteel are used to verify the proposed approach. The running results are shown to be satisfactory.Based on the reported above, the prediction software system for byproduct gas system is developed be resorting to the software engineering method. The application in Shanghai Baosteel Energy Center shows that the system can accurately predict the variation of gas generation and consumption, as well as the level of gasholder. The results demonstrate that the system exhibits high accuracy and provides with a rational guidance for balancing and scheduling of the byproduct gas system, which is capable of saving energy and reducing the complexity of scheduling workers.更多还原