智能故障诊断技术在核动力装置中的应用研究

Application research of intelligent fault diagnosis technology in the nuclear power plant

核动力装置对系统的安全性能要求较高.为了让操作员在发生故障时,避免产生错误的判断及操作,一种较好的方法是将智能故障诊断技术应用到核动力装置故障诊断系统中.利用粗糙集的数据约简技术提取精简的规则,基于这些规则建立的模糊神经网络具有更好的拓扑结构,学习速度大大提高,容错能力强;RBF神经网络有着良好的局部性能,诊断单个故障的能力要优于模糊神经网络,且网络不用训练,诊断实时性好.将粗糙集理论所构建的模糊神经网络与RBF神经网络相结合,能充分发挥各自的优点.为了验证该方法的有效性,以核动力装置蒸汽发生器U形管破裂等故障为例,进行了仿真实验研究.研究结果表明该邦联网络具有良好的诊断准确性、实时性和可扩充性,得到了预期的效果.

The security performance requirement is higher for the system in nuclear power plants. In order to let the operator avoid wrong judgment and operation when fault happens, one better method is to apply the intelligent fault diagnosis technology in the fault diagnosis system of nuclear power plant. Using the reduction technology of rough set （RS） method to draw a simple rule from a large number of initial data, the fuzzy neural network （FNN） set up on the basis of these rules has better topological structure, the speed of study is improved, and the fault-tolerant ability is strong. The radial basis function （RBF） network has very good partial performance, the ability of diagnosis for single fault has surpassed the FNN. The network doesn＇t need the training, and the diagnosis has good real-time character. The combination of FNN based on RS theory and RBF neural networks can take full advantage of one＇s own. In order to test the validity of the method, the inverted U-tubes break accident of steam generator, etc. were used as examples and many simulation experiments were performed. The result of study indicates that confederate network has good diagnosis accuracy, real-time character and expandability, and has obtained the anticipated effect.