汽轮机滑压运行初压智能优化方法的研究

发布时间：2018-07-20 11:54

【摘要】：近年来,社会电力用电结构已发生了较大的变化,电网负荷昼夜峰谷差越来越大。大量超临界汽轮机组被要求深度调峰,机组利用小时数逐年降低,低负荷运行时间普遍增加,热经济性大大降低。同时,随着我国经济、能源和环保形势的发展,火电机组节能降耗已成为企业生存运行的客观需要。因此,如何提高机组在低负荷阶段的运行经济性成为一个亟待解决的问题。要确保汽轮机变工况运行时仍能保持最佳状态,就必须对汽轮机的运行初压进行优化,以降低机组的热耗率。群智能优化技术是人们受生物进化或自然现象启发而提出的新方法,能很好的处理复杂系统的建模和优化问题。针对传统方法很难描述超临界汽轮机的复杂非线性、多工况等热力特性模型,不易实现机组初压优化的不足,本文对人工智能领域中混合蛙跳算法(shuffled frog leaping algorithm,SFLA)、最小二乘支持向量机(least squares support vector machine,LSSVM)及基于聚类的多模型建模技术进行重点研究,并将它们应用于机组初压优化,以实现机组的经济运行。主要研究内容如下:首先,针对典型混合蛙跳算法寻优能力不足的问题,提出了一种改进SFLA算法(mixed search SFLA,MS-SFLA)。通过引入了混沌反学习策略、非线性自适应惯性权值以及一个新的局部扰动策略以提高算法优化能力。通过13个基准测试函数的仿真测试,验证了改进的混合蛙跳算法具有较好的优化性能。基于该算法对最小二乘支持向量机回归算法超参数进行优化,数值仿真实验验证了该算法建模时的有效性。然后,研究了模糊C均值聚类算法在数据聚类划分中的应用。为了改善模糊C均值聚类对噪声和孤立点的鲁棒性,提出采用基于RBF核函数的模糊C均值算法。同时,为了解决诸如聚类精度受数据分布影响、对初始聚类中心敏感、易陷入局部最优以及难以确定最优聚类数的不足,提出一种新的基于G-K算法的双层聚类算法,热耗率多模型建模仿真试验验证了该算法的可行性。另外,针对单模型难以精确描述具有复杂非线性特性汽轮机热耗率的问题,提出了一种基于双层聚类算法和LSSVM融合的热耗率多模型建模方法,并利用MS-SFLA算法进行模型参数的选择。随后,将其应用到某600MW超临界汽轮机热耗率的建模,仿真实验证明该多模型建模方法能高精度的预测机组的热耗率,具有良好的泛化能力。最后,在建立好的热耗率多模型的基础上,利用MS-SFLA算法在给定负荷的可行运行初压范围内,以热耗率最低为优化目标来确定汽轮机变工况运行时的最优运行初压。将得到的最优运行初压作为汽轮机自动运行时主蒸汽压力的设定值,能达到机组优化运行的目的,并据此给出优化后的滑压运行曲线,该曲线具有更为实际的指导意义。
[Abstract]:In recent years, great changes have taken place in the structure of social electric power consumption, and the difference between day and night peak and valley of power grid load is increasing. A large number of supercritical steam turbine units are required to deep peak shaving, the number of operating hours of the units is reduced year by year, the running time of low load is generally increased, and the thermal economy is greatly reduced. At the same time, with the development of economy, energy and environmental protection in our country, energy saving and consumption reduction of thermal power units has become the objective need for enterprises to survive and run. Therefore, how to improve the operating economy of units at low load stage becomes an urgent problem to be solved. In order to ensure that the steam turbine can maintain the best condition in the off-condition operation, it is necessary to optimize the operation initial pressure of the turbine in order to reduce the heat consumption rate of the unit. Swarm intelligence optimization is a new method inspired by biological evolution or natural phenomena. It can deal with the modeling and optimization problems of complex systems well. The traditional method is difficult to describe the complex nonlinear, multi-condition thermodynamic characteristic model of supercritical steam turbine, and it is not easy to realize the initial pressure optimization of the unit. In this paper, the hybrid leapfrog algorithm (shuffled frog leaping algorithm), least square support vector machine (least squares support vector machine) and multi-model modeling technology based on clustering in artificial intelligence field are studied, and they are applied to initial pressure optimization. In order to achieve the economic operation of the unit. The main contents are as follows: firstly, an improved SFLA algorithm, (mixed search SFLA-MS-SFLA, is proposed to solve the problem of poor optimization ability of the typical hybrid leapfrog algorithm. Chaotic inverse learning strategy, nonlinear adaptive inertia weight and a new local perturbation strategy are introduced to improve the optimization ability of the algorithm. The simulation results of 13 benchmark functions show that the improved hybrid leapfrog algorithm has better performance. Based on this algorithm, the super-parameters of the least squares support vector machine regression algorithm are optimized, and the effectiveness of the algorithm is verified by numerical simulation. Then, the application of fuzzy C-means clustering algorithm in data clustering is studied. In order to improve the robustness of fuzzy C-means clustering to noise and outliers, a fuzzy C-means algorithm based on RBF kernel function is proposed. At the same time, in order to solve the problem that clustering accuracy is affected by data distribution, sensitive to the initial clustering center, easy to fall into local optimum and difficult to determine the optimal clustering number, a new two-layer clustering algorithm based on G-K algorithm is proposed. The feasibility of the algorithm is verified by heat consumption rate multi-model modeling and simulation. In addition, aiming at the problem that it is difficult to accurately describe the heat consumption rate of steam turbine with complex nonlinear characteristics by single model, a multi-model modeling method of heat consumption rate based on two-layer clustering algorithm and LSSVM fusion is proposed. MS-SFLA algorithm is used to select the model parameters. Then, it is applied to the modeling of heat consumption rate of a 600MW supercritical steam turbine. The simulation results show that the multi-model modeling method can predict the heat consumption rate of the unit with high accuracy and has a good generalization ability. Finally, on the basis of establishing a good multi-model of heat consumption rate, MS-SFLA algorithm is used to determine the optimal initial operating pressure of steam turbine in off-condition operation with the minimum heat consumption rate as the optimization objective within the feasible initial operating pressure range of a given load. The optimal operation initial pressure is regarded as the set value of the main steam pressure during the automatic operation of the steam turbine, which can achieve the purpose of the optimal operation of the unit, and based on this, the sliding pressure operation curve after the optimization is given, which has more practical guiding significance.
【学位授予单位】：燕山大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP18;TM621

【相似文献】