Study On Fault Diagnosis Of Converter In Double-fed Wind Generator System Based On Wavelet Neural Network

Under the circumstance of energy crisis and protection of ecological environment, the status of renewable energy has become increasingly important. Wind power is one of the renewable energy members, which plays an important role. Converter is one of the main components of double-fed wind power generation system, but it has relatively poor stability. Therefore, condition monitoring and fault diagnosis of the converter is essential for improving the system stability.At first, the paper discusses mathematical model of the Double-fed wind power generation system, then studies the effects of open-switch faults on wind power system and new open-switch fault diagnosis methods on power converter with problems. Modeling process focuses on the control strategy of the rotor side. By using Matlab/Simulink, establishes a double-fed wind power generation simulation platform.Subsequently, the paper describes the wavelet neural network theory and construction methods as well as introduces several commonly used algorithms in wavelet neural network. In the process of discrete wavelet neural network training, figures out that the traditional recursive least squares algorithm (RLS) with problems like easy to fall into local minimum or slow convergence, which decreases its training efficiency. Therefore, based on the recursive least squares algorithm, this paper make some improvements in proposing variable learning rate and variable weighted recursive least squares algorithm and introduces the algorithm into discrete wavelet neural network training. Comparative experiments show that accuracy and rate of convergence variable learning rate and variable weighted RLS algorithm is superior to the traditional RLS algorithm.Finally, variable learning rate and variable weighted RLS algorithm is applied to training wavelet neural network (WNN) to detecting the open-circuit fault of converter, and then this paper describe the specific steps of converter fault diagnosis. Final experimental results prove that the discrete wavelet neural network(DWNN), which use variable learning rate variable weighted recursive least squares algorithm can accurately identify the type of fault.