基于机器学习的锅炉主辅机状态监测研究

发布时间：2018-03-08 04:08

本文选题：锅炉监测　切入点：机器学习　出处：《华中科技大学》2015年硕士论文　论文类型：学位论文

【摘要】：近年来,我国清洁能源的崛起、新排放标准的颁布促使传统的火电机组担任更多的调峰任务、进行大量的环保改造,经常处于非设计工况运行,这样将对锅炉系统设备的安全可靠运行带来严峻考验。在此背景下,对锅炉的主机和辅机的运行状态进行在线实时监测,进而实现在线诊断和预测,对保障锅炉机组长期稳定运行具有重要意义,本文尝试引入机器学习方法建立锅炉主辅机的状态监测模型。首先采用了基于聚类的炉膛稳态判别法,对实时的炉膛工况进行判别。通过采集大量的运行历史数据,引入K-Means方法和快速搜索密度峰值方法,建立稳态判别模型,结果显示采用改进型的快速搜索密度峰值方法建模效果较好。炉膛稳态的判定为下文主机燃烧状态监测提供了状态基础,在稳态的基础上才能实施主机的燃烧状态监测。锅炉主机状态监测的关键问题是对炉膛内部燃烧状态的在线监测,本文提出一种基于数值模拟和机器学习的炉膛内部监测方法。首先,对炉膛进行模型构造、网格划分,尽量多地对变化参数设置模拟工况,然后,以模拟结果为数据源,采用ELM和SVM来构建炉膛燃烧的回归模型,根据温度、速度和氮氧化物浓度的回归结果表明该方法效果较好。基于前述的锅炉稳态判别,应用此模型对当前的工况进行计算,快速获得当前的温度、速度、成分浓度的分布,使运行人员实时掌握燃烧状态。锅炉辅机主要包括磨煤机、风机、给水泵等,根据对其故障树的分析可知振动是设备故障的主要原因之一。分析其振动的特征,本文提出了结合相空间重构和极限学习机的振动软测量方法,并以磨煤机的振动烈度为例,对该方法的结果进行验证分析,证明该方法具有较高的精度。对于其他重要的辅机参数,本文开发了基于ARMAX时序参数回归预测的软件工具,用户只需简单设置即可对参数进行趋势预测。
[Abstract]:In recent years, with the rise of clean energy and the promulgation of new emission standards, the traditional thermal power units take on more peak-shaving tasks, carry out a large number of environmental protection reforms, and often operate in off-design conditions. This will bring a severe test to the safe and reliable operation of boiler system equipment. Under this background, on-line real-time monitoring of the running state of the main engine and auxiliary machine of the boiler will be carried out, and then the on-line diagnosis and prediction can be realized. This paper attempts to introduce the machine learning method to establish the condition monitoring model of the boiler main and auxiliary machine. Firstly, the method of judging the steady state of the furnace based on clustering is adopted. Through collecting a large number of running history data, introducing K-Means method and fast searching peak density method, the steady-state discriminant model is established. The results show that the improved fast searching peak density method is effective in modeling, and the determination of furnace steady-state provides a state basis for monitoring the combustion state of the main engine below. Only on the basis of steady state can the combustion state of main engine be monitored. The key problem of boiler mainframe condition monitoring is the on-line monitoring of combustion state in furnace. In this paper, a method of furnace internal monitoring based on numerical simulation and machine learning is proposed. Firstly, the furnace is modeled, meshed, and simulated conditions are set up as much as possible, and then the simulation results are used as the data source. The regression model of furnace combustion is constructed by using ELM and SVM. According to the regression results of temperature, velocity and nitrogen oxide concentration, the method is proved to be effective. Based on the steady-state discrimination of boiler, the model is used to calculate the current working conditions. The distribution of current temperature, speed, composition and concentration can be obtained quickly, so that the operators can grasp the burning state in real time. The boiler auxiliary machinery mainly includes coal mill, fan, feed water pump and so on. According to the analysis of the fault tree, the vibration is one of the main reasons of the equipment fault. After analyzing the characteristics of the vibration, this paper puts forward the method of vibration soft sensing combined with the phase space reconstruction and the ultimate learning machine, and takes the vibration intensity of the coal mill as an example. The results of this method are verified and analyzed, and it is proved that the method has high precision. For other important auxiliary machine parameters, a software tool based on ARMAX time series parameter regression prediction is developed in this paper. The user can predict the trend of the parameters simply by setting them up.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TM621.2

【相似文献】