| At present,the increasing environmental pollution is a serious threat to the generation and development of human beings.As a sustainable green energy source,the share of wind energy in energy continues to increase.The wind power converter ensures that the wind turbine can still output a stable frequency and amplitude current to the grid when the wind speed changes.As the installed capacity of wind power increases year by year,correct and timely diagnosis of converter faults not only relates to the stability of wind turbines,but also relates to the safety of the entire power grid system.There are many types of converter faults,including semiconductor faults,capacitor faults and sensor faults.The fault feature space is complex.Traditional methods have limited ability to express features and it is difficult to accurately diagnose faults.This paper takes the open circuit fault of wind power converter as the research object,and applies the deep learning model to open circuit fault diagnosis of wind power converter.The main research work of the dissertation includes:1)In view of the problem that the fault features of the wind power converter are low in time domain and frequency domain characteristic area,starting from the three-phase current signal,the wavelet analysis method is used to effectively extract the time frequency characteristics of the fault features and improve the effective extraction of fault information.2)The fault features of the wind power converter are effectively expressed and the fault diagnosis rate is improved by using the nonlinear fitting ability of the deep belief network and the learning ability of the characteristic manifolds in the time frequency domain characteristics of the fault features.Based on the qualitative analysis of the structure and parameters of the deep belief network,the depth belief network model is optimized through experimental design.Through experimental comparison and analysis,the method based on wavelet extraction and deep belief network achieves better classification results in wind power converter fault diagnosis.3)Using the convolutional neural network to classify the original fault signal.Compared with the deep belief network,the convolutional neural network can directly classify three-phase current fault signals while maintaining a small amount of computation.More importantly,by arranging the three-phase current fault signals in three rows,the convolutional neural network can identify the correlation characteristics between the fault signals,thereby improving fault diagnosis accuracy.In the end,some excellent convolutional neural network design ideas are merged.The convolutional neural network is designed for open circuit fault signals of wind power converters,which improves the classification accuracy of faults. |
