Detection And Classification Research Of Transient Power Quality Disturbance

With the development of modern science and technology, more and more power electronics devices and nonlinear loads are taken into the power system, such loads make the transient power quality problem more and more serious, power supply departments and electrical consumers pay more attention to transient power quality issues.The primary prerequisite of improving the transient power quality is to detect, identificate and analyse it. Only by detecting, identifying and analysing the power quality timely and accurately, can we propose a way to solve these problems effectively.The thesis presents some methods to detect the characteristics of voltage sag, such as root-mean-square(RMS), dq conversation of instantaneous voltage and αβ-dq transform method, these methods are verified correctly. According to the delay problems of voltage sags detection, the thesis obtains the delay error curve of the starting time of voltage sags through calculation, these curves can be used to correct the error of the starting time to obtain more accurate voltage sags. This method was verified in the DSP, The result shows that the method improves accuracy of the starting time of voltage sags detection. On this basis, thesis presents a new way to detect voltage sag by combinating wavelet transform and ap transform. The method uses the real-time locating fault of wavelet and accurately detecting amplitude and phase-jump angle of voltage sag of the αβ-dq transform method, this way can detect characteristics of voltage sag.At the same time, the thesis proposes a way to classify the transient power quality by combinating wavelet transform and artificial neural network. Firstly, a lot of waveform data can be got from establishing mathematical model of disturbances, the waveforms are applied to wavelet transform and the multi-scale decomposition to different characteristic features of transient power quality of the disturbance. Then the sample data can be classified after these characteristic features training in the BP neural network. Although this method can classify transient power quality disturbance types, the result is easy to local minimum which may make low accuracy and slow convergence speed. Therefore, this thesis presents a method using GA to optimize BP algorithm. The simulation results show that this method can improve classification accuracy of transient power quality.