【作者】 刘岗；

【导师】 丁萃菁；

【作者基本信息】 重庆大学 ， 动力机械及工程， 2002， 硕士

【摘要】 本文针对重庆发电厂670t/h锅炉故障诊断问题，在原有的研究基础上采用分层混合专家网络，对其故障诊断系统做了进一步的开发研究工作，增添了分层模糊神经网络模块与新样本，拓展了锅炉故障诊断数据库及故障诊断处理和解释子系统。并采用Visual C++编制了系统程序，首先通过分层混合专家网络对故障样本进行训练，在训练中对分层混合专家网络各参数（规则数、学习率、门网络中止误差、专家网络初试迭代误差、循环次数、加权指数等）与学习误差平方和的关系进行了探讨，其结果为样本网络参数的合理选取提供了依据。然后，在计算机上对锅炉满水、缺水等故障进行了仿真实验。其实验结果表明：只要合理选取网络参数，本文所编辑的样本均能收敛于一个较小的稳定值，训练误差也较小，并且通过分层与不分层混合专家网络的样本训练结果比较可知，采用这个具有树结构型式的分层混合专家网络，由于其中两层门网络对系统的控制作用加强，因此样本训练的精度与稳定性有所提高。同时，实验结果还表明将模糊网络技术与基于知识的专家系统技术结合的分层混合专家网络应用于锅炉故障诊断，能实现符号推理与神经网络推理的有机结合，层次清晰、推理效率较高。该诊断系统能较好地模拟人类专家的逻辑思维和形象思维能力，能对锅炉故障进行较为准确及时的诊断。更多还原

【Abstract】 In connection with the diagnoses problem of 670t/h boiler for Chongqing power plant, the hierarchical mixture expert (HME) network is developed in this paper. The HME module is employed and the new samples are added and the database for fault diagnosis, fault treatment and interpretation subsystem are broadened to further develop the diagnosis system. The diagnosis program is made in Visual C++ .The samples are trained with this network, during which the relationships between the network parameters (such as rule number, study rate, expert network initial error and gate network error, circulation times and power index) and learning error were probed into, and the outcome can provide the basis for network parameter selecting. Then the simulation tests on drum excess water and lack of water were practiced; the results indicate that: if the network parameters are selected correctly，the training samples can be convergent to a relative small and stable value with a small learning error. Some comparisons can be made through sample training under the circumstance of ME and the HME, after adopting HME with the tree-style structure, two layers of the Gate network can enhance the control of the algorithm system. As a result the precision and stability is improved. Also, the experiments show that when the HME, which is a mixture of the fuzzy network and the expert system based on knowledge, is used to diagnose the fault of the boiler, some points can be attained: both symbol reasoning and nerve network reasoning can be combined very well. This system has the quality of a clear hierarchy and a high efficiency for reasoning. In a word it has such a capability of simulating the speculation of human expert to a certain degree that it can diagnose the faults for boiler accurately and timely.更多还原