Fault intelligent diagnosis for high-pressure feedwater heater system of a 300 MW coal-fired power unit based on improved BP neural network

In this paper, the multi-layer forward-direction ANN is used for fault intelligent diagnosis for thermal systems in power stations. In order to overcome the demerit of overlong training time and slow convergence rate of the BP algorithm, an improved BP algorithm, "learning rate self-adaptive adjusting method based on constant error correction rate", is put forward, with which the network training efficiency can be greatly improved. Besides the ANN structure and its training algorithm, another important factor to realize fault diagnosis with neural network is the fault symptom calculation. The calculating methods for different fault symptoms are discussed in detail. Finally, the high-pressure feed-water heater system of a 300 MW thermal power generating unit is taken as an example of fault diagnosis. The fault knowledge library of the system is summarized with the fault symptom calculation method, and the fault diagnosis is further realized based on the above improved BP neural network.