Islanding Detection Of Distribution Grid-connected Generation System Based On Wavelet Analysis And Neural Network

Unplanned island is that the distributed generation disconnects from the grid,when a fault occurs.It is possible to form a state that distributed power generation system operating with a separate load.This state is unpredictable and endanger the safety of personnel and equipment of grid,due to the island's power mismatches,voltage and frequency varies.Currently,the local island detection method is mainly divided into the two categories that include passive and active islanding detection method.This paper mainly aims to the shortcomings of passive detection method which is difficult to set a threshold value,the low detection reliability caused by its NDZ.Wavelet analysis and neural network research are used to improve the passive islanding detection.Through the wavelet transform decomposes the sampled voltage signal at the point PCC.Comparing with a variety of wavelet basis functions,simulation determines the db4 mother wavelet as wavelet function which can effectively extract information that judges the system state.The obtained wavelet coefficients process by the algorithms,then treat the decomposition coefficients as the input feature of neural network.To use neural network self-learning and adaptive ability identifies the system state.Aiming to BP neural network convergence is slow and easily to fall into local optimal solution,PSO algorithm is used to improve BP neural network and the topology of neural network is determined which is used to islanding detection.The simulation confirms that the PSO algorithm can improve the convergence speed of BP neural network and avoid the local optimum condition.Neural network is used to train and learn the characteristic values,and identify the system.The islanding detection method uses wavelet analysis to extract characteristic values and the neural network to learn the classification characteristic amount,therefore,there is no need to set the detection threshold.Finally,the paper sets up the islanding detection circuit with the MATLAB simulation and analyzes the simulation results of this method which is under five different working states.Then wavelet coefficients are sampled and processed which obtain 20 groups of different load conditions characteristic values to train the neural,and get a load simulation waveforms which verifies the effectiveness of the detection method.Through 10 groups of test samples,the detecting success rate of the system is obtained.The passive islanding detection has a bigger calculating NDZ on the National Standards,and any islanding state can not be detected in the NDZ.But the islanding detection method with the enriching learning samples,theoretically,can eliminate the NDZ and increase the reliability of detection.