Study On Intelligent Diagnosis Method Of Vibration Fault For Hydropower Generating Unit

With the hydropower generating unit capacity increasing rapidly and occupying an increasingly proportion in power system, how to increase the operation reliability and stability of hydropower generating units is focus problem in power industry. As one of the most common fault in operation of hydropower generating unit, violent vibration will endanger the safe operation of hydropower generating unit or even the power system. So, it is very important for the hydropower generating unit to research vibration fault diagnosis and ensure its stable operation. This dissertation prime study on the method of hydropower generating unit vibration fault diagnosis and the studies mainly consist of the following aspects.The vibration signal and noises of hydropower generating unit separation method is investigated, and a denoise method of vibration signal is proposed by the second generation wavelet. The wavelet basis function with some special characteristic is acquired by means of designing prediction and lift coefficient in the second generation wavelet transform, the corresponding wavelet basis function has been constructed to different type fault feature. The denoise simulation result is reasonable, and its computing speed is also rapid. The presented method solves how to select the wavelet basis function for wavelet denoise, and the signal denoising theory is improved.The method of vibration fault feature extraction is investigated to the hydropower generating unit, and a wavelet packet method of vibration fault feature extraction is proposed based on Parseval energy integral equation. The mapping relationship between the fault types and variation energy is constructed and a fault diagnosis method is realized based on energy-fault. It is improved that traditional spectrum analysis can't carry out local signal analysis.Based on research on fault feature extraction, the neural network vibration fault diagnosis method for hydropower generating unit is studied by integrating GA and BP neural network, and a hybrid fault diagnosis method is introduced by optimizing the weights of neural network using genetic algorithm. It can develop the generation and mapping capability, and achieve a rapidly convergence rate and a great learning ability. The problems of convergence and local minimums are solved, and the BP network can achieve the global optimal solutions at a great probability.At last, the support vector machine vibration method for the hydropower generating unit fault diagnosis is researched, and a new fault diagnosis method is proposed by combining the rough sets and support vector machine. This method firstly applies rough sets to discretize and deduct the decision table, then classify the faulty types. The results show that the method has well robustness, its diagnosis speed is fast, and the requirement of online fault diagnosis can be satisfied.