【作者】 王善书；

【导师】 樊绍胜；

【作者基本信息】 长沙理工大学 ， 控制理论与控制工程， 2012， 硕士

【摘要】 冷凝器电力、化工、机械等行业的大型换热设备，是火力发电厂的主要辅助设备之一，它在汽轮机装置的热力循环中起冷源的作用。因此，冷凝器工作性能的好坏对整个汽轮机运行过程中的经济性以及安全性会产生直接的影响。然而，冷凝器一旦投入运行，冷凝管的内部总会产生一些对运行有害的污垢。为了解决这方面的问题并为清洗提供理论依据，近年来不少国内外学者提出了预测污垢的新技术。本文以冷凝器内的结垢为研究对象，主要研究了冷凝器污垢的软测量建模和预测方法：论文首先对冷凝器污垢产生的原因及污垢是怎样影响冷凝器工作效率的情况做了一定的阐述，指出了对冷凝器污垢预测的重要意义，介绍了冷凝器的工作原理以及污垢预测的研究现状。其次，根据冷凝器污垢的形成机理、分类以及形成的各个阶段，对冷凝器污垢预测模型给出了详细的说明。然后，将污垢分解为软垢和硬垢两部分，而软垢又分阶段进行预测，硬垢则由数理统计模型得到，最后叠加得到总的污垢系数值。常规的污垢测量方法虽然应用广泛，但从经济性与准确性的角度上来说，都有各自的弊端。针对该问题，本文设计了一种基于K均值算法及切比雪夫神经网络的在线污垢预测方法。传统的切比雪夫神经网络是动态的单输入单输出模型，为适应冷凝器污垢受多种工况参数影响的情况，本文对算法进行了改进，用多输入单输出模型取而代之。在冷凝器污垢系数的预测过程中，污垢系数受工况突变或大扰动的影响很大，为保持污垢预测的连续性，对采样进来的数据进行了预处理，在预先设定的门槛值的限定下，预测值能够在预设精度范围内与实际值逼近一致。灰色模型与人工神经网络各具特点，本文在最后介绍了一种以灰色模型为主，以人工神经网络为辅的灰色神经网络的污垢预测方法。仿真对比结果表明，灰色神经网络预测模型在不同工况下预测精度比其它常规预测模型在短期预测中的精确度要高。最后，将各算法用MATLAB进行仿真，以检测算法的有效性和可行性。结果表明算法是有效且可行的。更多还原

【Abstract】 Condenser a large transfer auxiliary equipment in the powers, chemical,machinery and other fields plays an essential part in thermal power plants for itsproviding the cold source for the thermodynamic cycle in the large steam turbine.Therefore, the performance of its work has direct influence on the economy andsecurity in the entire steam turbine operation process. However, once the condenser isput into operation, there will be some fouling inside the condenser, which will do harmto the operation of condensers. In recent years many scholars from home and broadhave proposed a lot of new technologies of predicting fouling in order to solve theproblems discussed above and offer the basis for washing. The paper chooses thecondenser scaling as the main research object, its modeling and predicting method insoft measurement has been studied.Firstly, this paper explained the cause of fouling accumulation, the reason why itinfluences the performance of condenser, the important significance of its prediction,working principle and the status of prediction for it.Secondly, the prediction model of fouling in condenser is instructed thoroughlyaccording to its formation mechanism, classification and each phase when it is forming.Then, the overall fouling is separated into two parts soft fouling and hard fouling,which can be respectively got by phased prediction and mathematical statistics model,and the final fouling factor can be attained by adding up them together. Though, theconventional measurement methods of fouling are widely used, they have their ownshortcoming from the point of view of economics and accuracy. This paper designs anonline fouling prediction strategy based on K-means algorithm and Chebyshev neuralnetwork to conquer the problem discussed above, and introduces the multiple inputsingle output (MISO) models instead of the traditional dynamic single input singleoutput (SISO) Chebyshev neural network to adapt to the condenser fouling which isaffected by various of condition parameters. The fouling factor is largely influenced byworking conditions and big disturbance in the process of prediction, so sampling datapretreatment, which controls the data under the limit of the threshold to maintain thecontinuity of fouling data, is given to keep the prediction value in the default precisionand equal the actual value. At last the gray neural networks based on gray models andneural networks is proposed for fouling prediction because of Grey models and neuralnetworks has their own characteristics. In this model, gray models play a main role rather than neural network. By contrast, the simulation results show that gray modelscan achieve more accurate fouling prediction than traditional fouling models underdifferent work conditions in the short-term prediction process.Finally, we simulate the proposed algorithms to verify the validity and feasibilityof the algorithms. The results show that the algorithms are effective and feasible.更多还原