火电机组锅炉燃烧系统建模与优化研究

发布时间：2018-05-13 01:22

本文选题：神经网络 + 锅炉燃烧优化　；参考：《北京交通大学》2014年硕士论文

【摘要】：随着“节能减排”政策要求加强,在保证安全生产和满足负荷需求前提下,应努力提高火电厂运行效率、降低煤耗,控制并减少污染排放。本文针对火电厂锅炉燃烧系统进行了基于数据驱动的建模和优化研究。本文以中国华电一300MW机组为研究对象,给出了电厂锅炉的数据采集、预处理、优化指标选取的步骤和神经网络稳态建模过程及其评价。以降低单位相对煤耗和降低氮氧化物(NOx)排放量两者兼顾做为锅炉优化运行的评价指标。在两个典型负荷处分负荷点建模。比较了不同的神经网络建模训练算法的异同并对其泛化能力进行了评价。对电厂锅炉运行这一非线性动态工业过程,讨论和研究了在离散状态下一种基于数据驱动的建模方式,通过对锅炉燃烧过程的稳态建模和动态建模区别对待,给出了稳态和动态相结合的建模思想。给出了稳态过程和动态过程的定义、辨识方法、增益计算方法。在前向神经网络只能构建稳态模型和现场采集得到的数据存在动态特性的情况下,给出一种前向神经网络和一阶自回归模型相结合的混合模型。一阶自回归模型的存在使该混合模型能够描述动态特性,并推导出了其训练算法。将该模型运用到了火电厂的氮氧化物排放量建模中并对其建模能力进行了评价。结果表明加入的一阶自回归模型能够使混合模型提高拟合精度,说明在氮氧化物的形成过程中存在动态特性。最后采用遗传算法得到了两个典型负荷下的以降低煤耗和降低氮氧化物为目标的优化结果,基于神经网络模型得到了优化结果。并在滤除氮氧化物排放量数据中存在的扰动后得到了新的优化结果。对基于不同的模型得到的优化结果进行了比较和分析。结合现场数据给出了氮氧化物排放量相对增益计算的分析结果。利用数据库和动态网页技术制作了优化结果发布的动态网页,用以指导电厂机组人员对机组进行优化运行。
[Abstract]:With the strengthening of "energy saving and emission reduction" policy, under the premise of ensuring safe production and satisfying load demand, it is necessary to improve the operation efficiency of thermal power plants, reduce coal consumption, control and reduce pollution emissions. In this paper, data-driven modeling and optimization of boiler combustion system in thermal power plant are studied. This paper takes Huadian-300MW unit as the research object, gives the steps of data acquisition, preprocessing, optimization index selection and neural network steady-state modeling process and its evaluation of boiler in power plant. Both reducing the relative coal consumption per unit and reducing the no _ x emissions are considered as the evaluation indexes for the optimal operation of the boiler. Modeling at two typical load disposal points. The similarities and differences of different neural network modeling training algorithms are compared and their generalization ability is evaluated. For the nonlinear dynamic industrial process of power plant boiler, a data-driven modeling method based on discrete state is discussed and studied. The steady state modeling and dynamic modeling of boiler combustion process are treated differently. The idea of combining steady-state and dynamic modeling is presented. The definition, identification method and gain calculation method of steady and dynamic processes are given. Under the condition that the feedforward neural network can only construct the steady model and the data collected in the field have dynamic characteristics, a hybrid model combining the feedforward neural network and the first order autoregressive model is presented. The existence of the first order autoregressive model enables the hybrid model to describe the dynamic characteristics, and its training algorithm is derived. The model is applied to the modeling of NOx emissions in thermal power plants and its modeling ability is evaluated. The results show that the first order autoregressive model can improve the fitting accuracy of the mixed model, which indicates that there are dynamic characteristics in the formation of nitrogen oxides. Finally, genetic algorithm is used to obtain the optimization results with the goal of reducing coal consumption and nitrogen oxides under two typical loads, and the optimization results are obtained based on the neural network model. The new optimization results are obtained after filtering the disturbance in the nitrogen oxide emission data. The optimization results based on different models are compared and analyzed. The results of calculation of the relative gain of NOx emissions are given based on the field data. Using database and dynamic web technology, the dynamic web page of the optimization result is made to guide the crew of the power plant to operate optimally.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TM621.2

【参考文献】