【作者】 郑立刚；

【导师】 岑可法； 周昊；

【作者基本信息】 浙江大学 ， 工程热物理， 2009， 博士

【摘要】 随着能源消耗的增加与环保需求的提高,控制污染物排放成为环境领域与能源领域的重要课题之一。氮氧化物(NOx)是电站燃煤锅炉排放的主要污染物之一,为了满足国家日益严格的排放标准,寻求各个层面上的控制手段成为当今研究控制NOx排放的焦点。本文在此背景之下,开展了两方面的研究,一是煤粉射流在高温空气中的燃烧特性,二是基于人工智能的燃煤锅炉低NOx排放燃烧优化。高温空气燃烧技术已被证明在燃气工业炉上能取得节能、降低污染物排放的效果,但对于以煤炭为燃料的高温空气燃烧研究相对缺乏。基于人工智能的燃煤锅炉低NOx燃烧优化能有效降低NOx排放,在建模方法和优化算法上也需要进一步研究。本文主要内容包括两部分,第一部分为煤粉射流在高温空气中的燃烧特性,第二部分为基于人工智能的300 MW双炉膛燃煤锅炉低NOx燃烧优化研究。主要研究内容包括：(1)高温空气的发生。本论文采用丙烷气体在燃料贫燃条件下的燃烧产物作为煤粉射流燃烧所需的高温空气。为此,设计了一种多孔材料火焰稳定的协流燃烧器,中心射流为煤粉射流。最小协流平均温度为850 K,最大协流平均温度为1138 K。在测试实验条件下,高温空气温度Tcoflow均匀的区域面积约为40 mm(径向)×100 mm(轴向),温度水平受丙烷流量影响,约1000-1100 K。该协流燃烧器可提供氧气浓度7-9%左右、高温空气温度Tcoflow为1000-1100 K左右的高温空气。(2)基于图像处理的火焰振荡频率与火焰高度的测量利用高速摄影对射流火焰振荡频率进行了测量。研究结论表明,火焰上部的振荡频率<火焰中部的振荡频率<火焰根部的振荡频率。火焰振荡频率随着一次风速度、给粉速度、煤粉粒径、随着煤阶的增加而增加,且根部区域增加的幅度要大于中部和上部区域的变化幅度；随化学当量比φ先增加,然后降低；火焰振荡频率基本上不随02/CO2比例变化。(3)煤粉射流在高温空气燃烧中碳燃尽特性、氮析出特性研究。利用水冷探针对煤粉射流采样,对灰样进行元素分析与灰含量测定。实验结果表明,随着高温空气温度降低、氧气浓度的升高,碳剩余率越低,即燃尽率越高。当一次风为02/CO2时,随着氧气浓度增加,烟煤和无烟煤的碳剩余率都降低。相同条件下,02/CO2燃烧环境对无烟煤的影响要大于烟煤,这可能是烟煤含有高灰分的阻碍了CO2与C的气化反应。同时,随着高温空气温度的降低、氧气浓度的增加,无烟煤元素氮的剩余率要低于元素碳的剩余率,但相差并不大。(4)煤粉射流高温空气燃烧初期NOx排放研究当一次风为空气时,在本文所研究的四组高温空气参数中,烟煤与褐煤的NOx变化规律相类似,即随着高温空气温度的降低、氧气浓度升高,NOx排放值随之降低。烟煤燃烧初期的NOx为389-299 mg/m3,褐煤燃烧初期的NOx为1313～1080mg/m3;SH无烟煤的燃烧初期NOx排放规律与烟煤和褐煤不同,随着高温空气温度的降低、氧气浓度升高,NOx排放值随之增加,为513-767 mg/m3。当一次风为CO2时,在距离喷嘴出口位置到射流下游300 mm位置内,褐煤燃烧初期的NOx排放特性与烟煤不同,褐煤燃烧初期的NOx提前得到还原。当一次风为02/C02时,随着一次风中氧气浓度的增加,烟煤燃烧初期的NOx排放也随之增加,无烟煤燃烧初期的NOx排放先降低然后再增加,其原因可能是着火延迟导致的火焰拉长与火焰峰值温度的降低。富氧燃烧能够有效降低燃烧初期NOx排放,根据煤种的不同,降低幅度为38.78%-59.87%当一次风为空气时,无烟煤煤粉越细,初期NOx的排放浓度越低,与煤粉常规空气燃烧呈现出相似的规律。(5)燃煤锅炉NOx排放预测模型的建立。针对一台300 MW双炉膛燃煤锅炉,通过交叉相关性分析了锅炉运行参数与NOx排放量之间的依赖关系。为了消除各锅炉参数之间可能存在的线性相关性,对锅炉参数进行了主元分析。结果表明,只需18个主元可解释21个参数的99.999%,最终的NOx排放预测模型的最终输入减少到19个锅炉运行参数。SVR模型预测准确度最高,验证数据上的平均相对误差为1.59%,执行时间164sec,GRNN模型其次,BPNN模型再次,线性模型最差。SVR模型的执行时间仅为GRNN的16.7%,BPNN的3.9%。(6)低NOx排放燃烧优化。基于燃煤锅炉的NOx排放模型,利用遗传算法GA、蚁群算法ACO、分布估计算法EDA和粒子群算法PSO分别对模型的输入参数(即锅炉运行参数,一次风速和二次风速)进行了寻优。四种算法的计算时间分别为：120.18sec、120.14 sec、84.68 sec、29.17 sec。更多还原

【Abstract】 As a backdrop for the present work, pollution control has and will become one of the most important subjects in environment field and energy field as the energy consumption and environmental preservation increase. NOx is the one of the main pollutants in emissions of coal-fired boilers. In order to meet the increasingly rigorous emission limit, the strategies for NOx control at all levels have currently been a hot research topic. Therefore, the present dissertation includes two subjects, namely combustion characteristics of coal jet in high temperature/low oxygen air and low NOx emission combustion optimization of a coal-fired boiler by using artificial intelligence.High temperature air combustion has been proved to achieve energy savings and pollution reduction on industrial furnaces. However, there are only few reports on applications of this technique to coal combustion. AI based low NOx emissions combustion optimization of coal-fired boilers can reduce effectively NOx emissions. Never the less, it is still necessary to conduct further study on modeling tools and optimization algorithms, which are key parts of combustion optimization.The dissertation consists of two parts. In part one, combustion characteristics of coal jet in high temperature/low oxygen air is investigated. In part two, low NOx combustion optimization of a 300MW dual-furnace coal-fired boiler is conducted by using artificial intelligence. Specific objectives of the dissertation work are as follows.(1) Investigation on the generation of high temperature/low oxygen air. In this work, the hot exhaust gas from the propane premixed flame combusted in the fuel-lean conditions serves as the high temperature/low oxygen air, which will supply enough oxygen and ignition energy to the coal jet. Therefore, a novel flame-stabilized gas burner is designed. In all cases, the minimum averaged coflow temperature is 850K, and the maximum averaged coflow temperature is 1138K. In the tested cases, the approximately homogenous temperature of about 1000～1100K, being dependent on the propane flowrate, is appeared in a region with 100mm height and 40mm radius.In conclusion, the currently-designed porous medium flame-stabilized gas burner can provide high temperature/low oxygen air with the oxygen concentration of 7-9% and the coflow temperature of 1000-1100K. (2) Measurement of flame flicker frequency and flame height based on flame image processingThe flame flicker frequency was measured by using RedLake high speed carema. It is shown that the frequency of the flame root was highest, followed by that of the middle part of the flame, then followed by that of the upper part. The frequency increased with the primary air velocity, the coal flowrate, the particle size of coal and coal rank. On the whole, the frequency increased first and then decreased when the stoichiometry of the coflow flame is increased. The frequency show no significant change with the oxygen concentration in O2/CO2, which serves as the primary air to tansport the pulverized coal.(3) The burnout of carbon and the release characteristics of nitrogen for coal combustion in high temperature air. The coal ash was sampled by using water-cooled probe. Then, the coal ash was analyzed in the laboratory to obtain the ultimate analysis and ash content. The burnout of coal is increased with the increasing oxygen and the decreasing temperature. The remaining rates for both anthracite and bituminous decreased with the oxygen concentration from 10% to 30% when the O2/CO2 mixture was serves as the primary air. However, the influence of O2/CO2 on the anthracite was larger than on the bituminous due to the lower gasification rate between CO2 and C resulted form high ash content.The remaining rate of anthracite varied from 72.521%-77.448%. The release of nitrogen was intensified with the increasing oxygen concentration.(4) Study on the NOx emission at the early stage of the coal jet combustion in the high temperature air. As the primary air was set as air, the effects of the four high temperature air properties on NOx emission with bituminous and the lignite were quite similar. Lower air temperature and higher oxygen concentration resulted in lower NOx emission. However, for anthracite, the trend of NOx emission was different from that of the former coals. Lower air temperature and higher oxygen concentration resulted in higher NOx emission, and the value was 513～767 mg/m3. When CO2 was used as the primary air, the peak value for NOx emission of lignite was achieved at 205 mm, and then this value was decreased, which indicated the NOx at the early combustion stage was reduced in advance. When the primary air was the mixture of O2/CO2, the NOx emission for bituminous at the early combustion stage increased with increasing oxygen concentration in the primary air until the value reached 30%. Further increasing oxygen concentration did not increase the NOx emission any more.The NOx emitted increased from 139.4mg/m3 to 184.5 mg/m3. With increasing oxygen concentration in the O2/CO2, the NOx emission for anthracite first decreased and then increased. The possible reason for this might be that the ignition delay elongated the length of the flame and decreased the flame front temperature. It is found that oxygen-riched combustion can effectively reduce the NOx emission at the early combustion stage. For different coals the magnitude of the NOx emission reduced ranged from 38.78%-59.87%.As air was used as the primary air, finer Shenhua coal resulted in lower NOx emission, a similar trend as that of ordinary coal combustion in air.(5) The establishment of NOx prediction models of a coal-fired boiler. The cross correlation analysis between the operating parameters of a 300MW dual-furnace coal-fired boiler and the NOx emissions is performed. In order to eliminate the potential co-linearity between the operating parameters of the studied boiler, the principal components analysis is performed. The result shows that 18 principal components can explain the 99.999% variance of the 21 operating parameters. It is concluded that the SVR model demonstrates the best prediction producing the mean relative error of 1.59% on the testing subset consisted of 224 cases, followed by neural network model. The computational time for the SVR model is about 164sec, which is only 16.7% of that for GRNN and 3.9% of that for BPNN.(6) Low NOx combustion optimization of a coal-fired boiler. Combined with NOx emission models, Genetic Algorithm (GA), Ant Colony Optimization (ACO), Estimation of Distribution Algorithm (EDA) and Particle Swarm Optimization are respectively employed to search the optimal inputs of the SVR model so as to achieve the minimum NOx emissions for a particular boiler load by regulating the operating parameters of the studied boiler. The computational times for four optimization algorithms are 120.18 sec,120.14 sec,84.68 sec and 29.17sec, respectively.更多还原