基于神经网络的电厂锅炉故障诊断研究

发布时间：2018-11-03 14:44

【摘要】：随着电力工业的迅速发展,大型电站锅炉的应用日趋广泛且锅炉系统的结构更复杂,且运行参数更多,对故障诊断技术应用的迫切性也与日俱增。在众多故障诊断技术中,基于神经网络的故障诊断技术由于学习能力强、容错性好、快捷方便以及能够处理复杂的非线性关系等优点而被广泛应用于电厂系统故障诊断研究。在基于神经网络故障诊断方法中,故障特征的识别与分类是影响故障诊断系统安全、可靠、高效的发挥功能关键步骤之一。因此对故障特征识别与分类精度的研究至关重要。论文针对电厂屏式过热器在发生泄漏时,故障特征规律难以总结,特征知识提取困难,特征参数变化快等问题,为了克服传统单一锅炉故障诊断方法和人工监控故障诊断所存在的缺点,设计了小波神经网络故障诊断模型。并针对网络模型训练参数取值问题,提出粒子群算法对参数进行优化。通过MATLAB仿真对比研究表明,基于粒子群优化小波神经网络故障诊断模型,在精度和训练时间上优于其它算法。此外,针对多种故障模式并存情况下,故障特征维数高等特点,设计了概率神经网络故障诊断模型,并采用粒子群算法对网络进行改进,形成自适应概率神经网络故障诊断模型。MATLAB仿真验证了改进算法的有效性。论文最后采用组态王软件完成了火电厂故障诊断系统的组态监控系统设计,并利用OPC技术建立了MATLAB与组态王之间的数据通讯,实现了在MATLAB环境下的故障监控。
[Abstract]:With the rapid development of power industry, the application of large power plant boilers is becoming more and more extensive, the structure of boiler system is more complex, and the operation parameters are more and more, and the application of fault diagnosis technology is becoming more and more urgent. Among many fault diagnosis technologies, neural network-based fault diagnosis technology is widely used in power plant system fault diagnosis research because of its strong learning ability, good fault tolerance, fast and convenient, and the ability to deal with complex nonlinear relationships. In the fault diagnosis method based on neural network, the identification and classification of fault features is one of the key steps to affect the safety, reliability and efficiency of fault diagnosis system. Therefore, it is very important to study the accuracy of fault feature identification and classification. In this paper, the fault feature rule is difficult to summarize, the feature knowledge is difficult to be extracted, and the characteristic parameters change quickly when the power plant screen superheater is leaking. In order to overcome the shortcomings of traditional single boiler fault diagnosis method and manual monitoring fault diagnosis, a wavelet neural network fault diagnosis model is designed. The particle swarm optimization (PSO) algorithm is proposed to optimize the network model training parameters. The simulation results of MATLAB show that the fault diagnosis model based on particle swarm optimization wavelet neural network is superior to other algorithms in accuracy and training time. In addition, a probabilistic neural network fault diagnosis model is designed and improved by particle swarm optimization (PSO). An adaptive probabilistic neural network fault diagnosis model is formed and the effectiveness of the improved algorithm is verified by MATLAB simulation. Finally, the configuration monitoring system of power plant fault diagnosis system is designed by using Kingview software, and the data communication between MATLAB and Kingview is established by using OPC technology, and the fault monitoring under MATLAB environment is realized.
【学位授予单位】：河北科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM621.2

【参考文献】